Evaluating Vision and Pathology Foundation Models for Computational Pathology: A Comprehensive Benchmark Study

Evaluating Vision and Pathology Foundation Models for Computational Pathology: A Comprehensive Benchmark Study | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Evaluating Vision and Pathology Foundation Models for Computational Pathology: A Comprehensive Benchmark Study Olivier Gevaert, Rohan Bareja, Francisco Carrillo-Perez, Yuanning Zheng, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6823810/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted You are reading this latest preprint version Abstract To advance precision medicine in pathology, robust AI-driven foundation models are increasingly needed to uncover complex patterns in large-scale pathology datasets, enabling more accurate disease detection, classification, and prognostic insights. However, despite substantial progress in deep learning and computer vision, the comparative performance and generalizability of these pathology foundation models across diverse histopathological datasets and tasks remain largely unexamined. In this study, we conduct a comprehensive benchmarking of 31 AI foundation models for computational pathology, including general vision models (VM), general vision-language models (VLM), pathology-specific vision models (Path-VM), and pathology-specific vision-language models (Path-VLM), evaluated over 41 tasks sourced from TCGA, CPTAC, external benchmarking datasets, and out-of-domain datasets. Our study demonstrates that Virchow2, a pathology foundation model, delivered the highest performance across TCGA, CPTAC, and external tasks, highlighting its effectiveness in diverse histopathological evaluations. We also show that Path-VM outperformed both Path-VLM and VM, securing top rankings across tasks despite lacking a statistically significant edge over vision models. Our findings reveal that model size and data size did not consistently correlate with improved performance in pathology foundation models, challenging assumptions about scaling in histopathological applications. Lastly, our study demonstrates that a fusion model, integrating top-performing foundation models, achieved superior generalization across external tasks and diverse tissues in histopathological analysis. These findings emphasize the need for further research to understand the underlying factors influencing model performance and to develop strategies that enhance the generalizability and robustness of pathology-specific vision foundation models across different tissue types and datasets. PathBench : https://pathbench.stanford.edu/ Health sciences/Medical research Health sciences/Health care/Medical imaging Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction Pathology plays a crucial role in cancer research, providing insights that drive the understanding, diagnosis, and treatment of cancer 1 . It involves thorough examination of tissue specimens for tumor diagnosis and identification of important prognostic features such as tumor stage and grade, as well as biomarkers suggestive of treatment response 2 . Pathologists study digital pathology slides to analyze tumor histologic features, including architectural patterns and cellular morphology, in order to detect and classify tumors 3 . Consequently, oncologic pathology plays an integral role in personalized medicine and treatment selection. Computational pathology (CPath) has enhanced the field of pathology by integrating automated workflows and processes and advanced computational techniques 1 . It has the potential to reduce the workload of pathologists considerably and can aid in speeding up the overall diagnostic process. With the advent of machine learning and artificial intelligence, CPath has proven potential to facilitate detection, classification and predictive modeling of cancer 4 . Further, CPath proved effective in assisting pathologists in tumor grading, staging, and subtyping 5 , while supporting diagnostic decision-making. Also, it has been employed as a computational technique to study patterns, segment tumors 6 , and extract intricate features, which enhances cancer detection 7 . The use of Artificial Intelligence (AI) has been at the forefront of medical research for the past decade 8 . AI algorithms based on convolutional neural networks and vision transformers have significantly advanced computational pathology image analysis, especially in automated tumor detection and classification 9 . AI-driven predictive models can analyze pathology images along with clinical and genomic data to predict patient outcomes, including disease progression, recurrence risk, and treatment response 10 – 12 . Integrating gene expression, spatial omics, single-cell omics, and digital pathology data with AI enables a comprehensive multi-modal approach to cancer diagnosis and biomarker discovery, leveraging deep learning to uncover intricate spatial and molecular relationships within the tumor microenvironment 13 , 14 . AI models can automatically segment and delineate tumor regions, extract complex features and perform risk based stratification 15 . Additionally, AI has accelerated Computational Pathology (CPath) by automating image analysis, streamlining workflows, integrating multimodal data 16 , 17 , and enhancing diagnostic accuracy 18 . Despite these advancements, several challenges persist, particularly regarding data limitations and model generalizability. Pathology datasets suffer from shortage of labeled data as manual pathologist annotations are time-consuming and also cost ineffective 19 . In contrast, large-scale datasets like ImageNet 20 benefit from extensive labeled data, enabling supervised learning approaches to achieve high accuracy across diverse natural image tasks. To address this, Self-Supervised Learning (SSL) has emerged as a machine learning paradigm where models are trained on large-scale datasets to learn meaningful representations without the need for labeled examples 21 . This approach reduces the burden associated with data labeling and speeds up development of AI models. SSL based foundation models can be pre-trained on diverse and large unlabeled datasets, 22 , 23 to learn generalized patterns, and then fine-tuned on specific pathology datasets of interest 24 . SSL enables these foundation models to adapt to different types of pathologies, domains, and tasks, and improve their generalizability. 25 , 26 Therefore there has been a large effort in developing general purpose pathology foundation models that are trained on millions of whole slide images 26 – 28 , and can be adapted to tumor classification, segmentation, grading tasks with smaller amounts of labeled data. One of the first prominent models to be released was CTransPath 29 , which was pretrained on the TCGA 30 and PAIP datasets 31 , 32 and trained on approximately 15 million images. Lunit’s DINO-based ViT-Small model 24 also utilized TCGA, along with internal datasets comprising over 32.6 million patches and 36,000 slides. Phikon 33 , another SSL model based on the iBOT framework, was pretrained on 40 million images from TCGA, covering 16 cancer types. More recently, several models such as UNI 25 , Virchow 26 , and Hibou 27 have adopted the DINOv2 SSL method with ViT-Large architectures for their pretraining, using proprietary datasets. Next, pathology language models such as PLIP 34 , Quilt-Net 35 , and CONCH 36 , which are trained on both pathology slides and paired text, have demonstrated significant performance in vision-based tasks as well. Vision encoders like CLIP 37 , iBOT 23 that are built into these models have shown competitive results in natural image classification and image retrieval tasks. Additionally, imagenet pre-trained models have been widely used as feature extractors for digital pathology. 38 – 40 Despite these advancements in SSL, current foundation models may still be biased toward datasets they are trained on, limiting their transferability across diverse institutions and staining protocols 41 . Additionally, biases can arise from imbalanced representation of tissue types, and the predominance of certain cancer subtypes in training data, all of which can impact the model’s generalization and clinical applicability 42 , 43 . Many pathology foundation models rely on self-supervised pre-training using datasets like TCGA 24 , 33 , 44 or large proprietary datasets from a single institution 25 , 26 , which can inadvertently result in domain-specific overfitting rather than true generalization across diverse clinical settings. 43 This is further compounded by the limited availability of large-scale, diverse pathology datasets that span multiple institutions, staining variations, and scanner types 45 . Unlike general computer vision SSL models that benefit from extensive public datasets such as ImageNet 20 or LAION-5B 46 , pathology-specific SSL models often lack sufficient pre-training diversity, making them susceptible to dataset-specific artifacts. A systematic comparison is needed to evaluate pathology foundation models against both domain-specific SSL models and general computer vision SSL models, assessing their robustness across independent external datasets. Additionally, benchmarking should consider model generalization across diverse tissue types and cancer subtypes, ensuring that performance gains are not restricted to specific datasets. In this study, we compare general vision models (VM), general vision-language models (VLM), pathology-specific vision models (Path-VM), and pathology-specific vision-language models (Path-VLM), to analyze their generalization capability and task-specific strengths. Altogether, over 20 pathology or pathology-language models have been released. In addition, several studies 47 have used off-the-shelf imagenet models such as ResNet, DINO and DINOv2 38,39,48,49 for medical imaging including digital pathology. However, a comprehensive comparison of these models across diverse datasets and tasks remains lacking. Here, we present such a benchmark of four different types of models: pathology image, vision (i.e. natural images), pathology-language and vision-language models to identify the best-performing foundation models on 41 downstream tasks, 53 datasets and over 17.5k samples. We conduct a comprehensive evaluation by testing them across identical downstream tasks within a standardized experimental framework, ensuring a fair comparison. Our analysis evaluates the consistency of top-performing models, identifies those excelling across diverse benchmarking tasks and tissue types, compares the efficacy of pathology image models against language and vision models, and examines the impact of model size and dataset size on performance. Finally, we explored the potential of fusion models to enhance generalization across external tasks and diverse tissues. This work aims to provide insights into the factors driving effective pathology foundation models, offering a foundation for optimizing their application in precision medicine. Results Benchmarking models on TCGA: Task-specific performance highlights variability in top model rankings We conducted an extensive benchmarking study of 31 models across diverse domains, including general vision, general vision-language, pathology-specific vision, and pathology specific vision-language, evaluating them over 41 distinct tasks. To ensure consistent and concise terminology throughout the study, we assigned model abbreviations based on their training domains and modalities (Table 1): general vision models (VM), general vision-language models (VLM), pathology-specific vision models (Path-VM), and pathology-specific vision-language models (Path-VLM). Among these, 19 tasks were curated from the TCGA dataset, which includes multiple tumor types such as brain, breast, bladder, and others. We evaluated model performance using the average of balanced accuracy, precision, recall, and F1 score, collectively defined as the average performance metric for a comprehensive assessment. Initially, we focused on 15 pathology-specific vision models (Path-VM, Table 2), tested across 19 tasks related to tumor subtyping, molecular subtyping, and grading, and calculated the mean of average performance across 19 tasks. Among these, the Virchow2 28 model consistently ranked first, achieving a mean average performance of 0.706 ± 0.10. The next best-performing models were H-optimus-0 (0.702 ± 0.11), Prov-GigaPath 50 (0.702 ± 0.10) and UNI (0.70 ± 0.10), comprising the top four rankings (Figure 1b). Among the Path-VM, Virchow2 and UNI2 ranked first in 5 out of 19 TCGA-based tasks (Figure 1c), while H-optimus-0 and Prov-GigaPath ranked first in 3/19 tasks. We further investigated whether specific models excel in tumor-type-specific classification and subtyping tasks, aiming to identify tailored strengths across diverse cancer contexts. H-optimus-0 demonstrated superior performance on brain and breast tumor classification tasks (Figure 1d, Supplementary Figure 1a,b). In contrast, Virchow2 outperformed others on bladder and prostate cancer subtyping tasks, showcasing its precision in delineating subtle histological subtypes critical for these cancers (Supplementary Figure 1b). Meanwhile, Prov-GigaPath achieved the highest rankings on lung cancer and pan-cancer assessments, excelling on both organ-specific and cross-cancer evaluations (Supplementary Figure 1a, b). Notably, UNI (0.70 ± 0.10) outperformed UNI2 (0.69 ± 0.11) when evaluated by the mean average performance metric across all 19 tasks(Figure 1b), reflecting its consistent strength across diverse challenges. However, UNI2 secured the top rank on 5 individual tasks (Figure 1c), surpassing UNI, which achieved the top rank on only 1 task, highlighting a striking contrast in task-specific excellence versus overall consistency across the 19-task evaluation. A detailed ranking of the top 10 Path-VM across all TCGA tasks is provided in Supplementary Figure 2a, offering a comprehensive perspective on model performance. Next, when broadening the evaluation to include all models—spanning general vision (VM), general vision-language (VLM), pathology-specific vision models (Path-VM), and pathology-specific vision language (Path-VLM) architectures—the top 10 rankings remained dominated by Path-VM. The only exception was iBOT-L16 23 , a vision model, which emerged as the sole non-pathology model in this group (Figure 1b, bottom panel). Overall, our results show significant variability in top model rankings across tumor-type classification and subtyping challenges. We also developed a live benchmarking dashboard PathBench, accessible at https://pathbench.stanford.edu/ , to enable real-time visualization and comparison of model performance. Virchow2 is the most consistent pathology model in CPTAC and external benchmarking tasks Next, we evaluated these foundation model categories (VM, Path-VM, VLM, Path-VLM) on CPTAC cancer cases spanning various tumor types, focusing on specific tasks such as tumor grading, tumor classification, and pathway activity prediction (e.g., p53, MTOR, SWI). Among the 31 models tested, Virchow2 demonstrated the highest overall performance (Figure 2a, Supplementary Figure 3c), followed closely by Prov-GigaPath and UNI, when averaged across seven CPTAC-based tasks. Virchow2 achieved the best performance in 2 out of 6 individual tasks (Supplementary Figure 3a), with an average score of 0.766 ± 0.103, while Prov-GigaPath and UNI followed with 0.763 ± 0.103 and 0.760 ± 0.103, respectively. Notably, Prov-GigaPath also excelled in 2 out of the 6 tasks, with an average performance of 0.763 ± 0.11. Both Virchow2 and Prov-GigaPath consistently ranked among the top three models across CPTAC tasks (Supplementary Figure 3a). H-optimus-0, the second-best model in TCGA-based tasks, ranked fourth in CPTAC tasks. In the tumor type classification task, Prov-GigaPath achieved the highest performance, followed by Virchow and Virchow2. Next, we evaluated these model categories (VM, Path-VM, VLM, Path-VLM) across eight external benchmarking tasks spanning multiple cancer types, including breast, prostate, and colorectal cancer. Virchow2 outperformed all other models, achieving an average performance of 0.82 ± 0.17 (Figure 2b, Supplementary Figure 3d), followed by UNI2 (0.79 ± 0.18). Prov-GigaPath ranked third, with a mean performance of 0.787 ± 0.17 across these tasks. Notably, Virchow2 consistently ranked among the top three in 6 out of 8 external benchmarking tasks. In contrast, H-optimus-0, which ranked second in TCGA-based tasks based on mean performance, performed poorly in external benchmarking tasks, placing seventh (0.766 ± 0.18). Interestingly, UNI2 outperformed its counterpart UNI in external tasks (Supplementary Figure 3d), where UNI ranked fourth (0.784 ± 0.18). However, in CPTAC tasks, UNI (0.760 ± 0.10, ranked third) outperformed UNI2 (0.751 ± 0.10, ranked sixth). We assessed whether the top five models from TCGA-based tasks demonstrated statistically significant performance differences compared to the remaining models across the three dataset categories. Virchow2 was the only model to show statistically significant performance gains in both CPTAC and external benchmarking tasks (Supplementary Figure 4a). UNI also exhibited statistically significant differences in CPTAC tasks; however, it did not show significant improvement in external benchmarking tasks (Supplementary Figure 4b). Overall, Virchow2 demonstrated consistently strong performance across both CPTAC and external benchmarking tasks. Pathology-specific vision models (Path-VM) achieve superior performance over pathology-specific vision language models (Path-VLM) and general vision models (VM) Next, we systematically evaluated the performance of pathology-specific vision models (Path-VM), pathology-specific vision-language models (Path-VLM), and general vision models (VM) across classification tasks from TCGA, CPTAC, and eight external benchmarking datasets (Figure 3a). In comparative analysis of Path-VM and VM, no statistically significant differences were observed between Path-VM and VM across these datasets (Figure 3a). To further assess their practical effectiveness, we analyzed model rankings, where Path-VM’s superior visual feature processing translated into consistent dominance across tasks (Figure 3b,c). Path-VM consistently dominated the top 10 positions in combined comparisons of VM and Path-VM, underscoring their practical superiority (Figure 3b,c, Supplementary figure 8a, Supplementary figure 9a,b). In 19 TCGA tasks, Path-VM held nearly all top 10 positions, with only a single VM appearing in most cases (Supplementary Figure 8a, Supplementary figure 10b). Similarly, in CPTAC and external tasks (Figure 3b,c), Path-VM were prevalent among the top 10, with only a few VM achieving high rankings (Supplementary Figure 8a-b). For example, iBOT VM achieved the second-highest performance in the TCGA brain histology subtype classification task and TCGA gastrointestinal subtype classification (Supplementary Figure 8a). A few VM’s(iBOT and DINO based general vision models) also ranked among the top three in the colorectal cancer challenge (MHIST), and the UnitoPatho challenge, and the DINO-B16 VM ranked fourth in the CPTAC-chromatin modification pathway classification task associated with gastrointestinal and colorectal tumors (Supplementary Figures 9a-b). These isolated successes suggest that VM’s can learn meaningful features for specific tasks, such as colorectal and gastrointestinal tumors, but Path-VM’s consistent top rankings highlight their broader superiority. In a comparative analysis of Path-VM and Path-VLM models across pathology tasks, Path-VM consistently outperformed Path-VLM in TCGA, CPTAC, and external benchmarking datasets, with statistical significance (Figure 3a). In TCGA tasks, only a few Path-VLM were ranked in top 10 in the combined ranking of Path-VM and Path-VLM (Supplementary figure 5a, Supplementary figure 10a) such as BiomedCLIP and QuiltNet-B-16 in bladder histology subtyping and MGMT methylation prediction, TITAN in brain histology, brain subtype(transcriptomic) and gi subtyping tasks. Similarly, in CPTAC tasks, Path-VLM failed to consistently rank within the top 10. The only exception was TITAN, a Path-VLM, which achieved a 6th-place ranking in the tumor label classification task (Supplementary Figure 6a). In external benchmarking tasks, Path-VLM again showed limited success. The only exception,TITAN, did achieve top-10 rankings in five out of eight tasks and ranked among the top three in the UnitoPatho challenge (Supplementary Figure 7a). A comparative analysis of vision-only models (VM and Path-VM), and vision-language models (VLM and Path-VLM), revealed that vision-only models, both VM and Path-VM significantly outperformed VLM and Path-VLM across TCGA, CPTAC, and external benchmarking tasks (Supplementary Figures 11a-c). While Path-VLM exhibited a statistically significant performance advantage over VLM due to pathology-specific textual data, both vision-language models lagged behind their vision-only counterparts. Even though VLMs are trained on large datasets, they might have underperformed due to diluted visual feature learning from multimodal objectives and insufficient pathology-specific pretraining. This highlights the pivotal role of high-resolution visual features in pathology image analysis, where models relying solely on image data consistently excel over those incorporating textual information. Overall, these results highlight the practical advantages of pathology-specific vision models, which leverage robust visual representations of tissue morphology to achieve superior classification performance, as demonstrated by Path-VM’s statistically significant outperformance of Path-VLM. Path-VM’s dominance in top 10 rankings, even in the absence of statistical significance over VM, underscores the value of pathology-specific training on larger, more diverse datasets that better represent the target tasks. While VM demonstrate potential in specific tasks, their limited presence in top rankings compared to Path-VM highlights the advantage of specialized models. Path-VLM, trained on histopathology slides and textual data (For example, CONCH 36 trained on 1.17 million image-text pairs, and PLIP 34 on ~200k image-text pairs), generally underperformed, likely due to smaller, less diverse training datasets that may not fully leverage the complementary nature of image and text modalities for classification tasks. The underperformance of Path-VLM may result from limitations in their current architectures or training data, which may hinder their effectiveness in classification tasks, suggesting a need for improved multimodal integration to achieve performance comparable to Path-VM. Model size and data size do not guarantee better performance in pathology foundation models We first investigated model size and specifically whether larger vision transformer (ViT) architecture (i.e. ViT-large, ViT-huge, and ViT-giant collectively referred to as ViT-L and model parameters > 300M) offer superior performance compared to small(model parameters = 21M) and base ViT (model parameters = 85M) among pathology models. In TCGA-based tasks, larger ViT architectures outperformed both ViT-S and ViT-B models (Figure 4a). However, for CPTAC and external benchmarking datasets, no statistically significant performance differences were observed among the different ViT sizes (Figure 4a). Similarly, when evaluating vision models, we found no statistically significant differences in performance among ViT-S, ViT-B, and ViT-L across TCGA, CPTAC, and external benchmarking tasks (Supplementary Figures 12a-c). To assess the impact of pre-training dataset size on model performance, we categorized pathology models based on the number of whole-slide images (WSIs) used for training: small (1 million WSIs). Our analysis revealed statistically significant differences in average performance between models trained on small versus medium datasets, as well as small versus large datasets, across TCGA, CPTAC, and external benchmarking tasks (Figure 4b). However, no statistically significant difference was observed between models trained on medium and large datasets (Figure 4b). For example, models such as UNI (trained on ~100k WSIs) and UNI2 (~300k WSIs) achieved top performance across multiple tasks, were close behind the performance of Virchow2, which was trained on 3 million WSIs, (Figure 1b,c, Figure 2a,b). Additionally, UNI, trained on ~100k WSIs, and H-optimus-0, trained on ~500k WSIs, outperformed both Hibou and Virchow, which were trained on ~1 million WSIs and 1.5 million WSI’s respectively, across all three dataset categories (Figure 1b,c and Figure 2a,b). These findings suggest that beyond a certain threshold, increasing training data size does not necessarily lead to improved classification performance and that diversity in the training set may play a bigger role in models’ performance. Fusion model generalizes better across external tasks and tissues Next, we evaluated model performance using a combination of proprietary and out-of-domain datasets (i.e. Stanford, NLST, and DHMC) as part of an out-of-domain assessment, alongside external tasks to assess generalizability. Since these datasets were not included in model training, they served as an external evaluation of model robustness. We ranked the models based on their cumulative average performance across TCGA, CPTAC, and a combined set of out-of-domain and external benchmarking tasks. While H-Optimus-0 (0.702 ± 0.11) and Prov-GigaPath (0.702 ± 0.10) ranked among the top-performing models in 19 TCGA tasks (Figure 5a), their average performance showed a moderate decline in out-of-domain and external benchmarking tasks, with scores of 0.662 ± 0.19 and 0.662 ± 0.21, respectively (Figure 5b), leading to lower rankings. Similarly, UNI2 (0.696 ± 0.11) and Virchow (0.683 ± 0.11) exhibited a slight decrease in average performance from TCGA tasks (Figure 5a) to 0.664 ± 0.21 and 0.662 ± 0.20 in external evaluations (Figure 5b). These shifts in rankings across datasets highlight variability in model generalizability. To explore whether combining top-performing foundation models could improve predictive performance, we applied a late fusion strategy that aggregates predictions using majority voting. We selected the top five models based on mean average performance across TCGA tasks (Figure 1b): Virchow2, H-Optimus-0, Prov-GigaPath, UNI, and UNI2. We then compared the fusion model’s performance against individual models across multiple evaluation categories, including TCGA, CPTAC, and external benchmarking datasets (which included out-of-domain tasks). In TCGA tasks, the best-performing individual model, Virchow2, surpassed the fusion model in cumulative average performance across 19 tasks (Figure 5a). However, in CPTAC and external evaluations, the fusion model outperformed all individual models (Figures 5b,c), suggesting stronger generalizability. To further analyze model specialization, we calculated z-scores for each model’s performance and identified the top 10 models per task. Grouping all 41 tasks by tissue type (Figure 5d), we found that the fusion model excelled in breast, colon, prostate, and pan-cancer tasks, while Virchow2 performed best in bladder and lung cancer tasks. A direct comparison of Virchow2 and the fusion model (Figure 5e) showed that in TCGA tasks, Virchow2 outperformed the fusion model in 11 of 19 tasks. However, in CPTAC and external evaluations, the fusion model ranked higher in 5 of 7 CPTAC tasks and 12 of 15 out-of-domain + external tasks, suggesting its ability to generalize across diverse datasets. The fusion model exhibited superior generalizability, outperforming individual models in CPTAC and external tasks, particularly in breast, colon, prostate, and pan-cancer classifications. It achieved the highest overall performance across datasets, emphasizing the strength of ensemble-based predictions. Discussion This study presents an in-depth benchmarking analysis of 31 models spanning multiple domains—general vision models (VM), general vision-language models (VLM), pathology-specific vision models (Path-VM), and pathology-specific vision-language models (Path-VLM)—evaluated across 41 distinct tasks, which include a comprehensive range of cancer classification and subtyping challenges derived from TCGA, CPTAC, and external benchmarking datasets. The goal of this study was to investigate the task-specific strengths of various models and assess their overall performance on common cancer-specific tasks. Our findings reveal notable variations in model performance, reflecting the inherent complexity of pathology tasks and emphasizing the necessity for tailored approaches depending on the type of cancer and the specific challenge. In our study, it is crucial to also acknowledge that TCGA (The Cancer Genome Atlas) 30 likely serves as a major source of training data for many of the models evaluated. Despite the use of self-supervised learning (SSL) techniques aimed at minimizing reliance on labeled data, some degree of overfitting on TCGA-specific characteristics may still occur. This is particularly relevant when evaluating models on TCGA-based tasks, as the overlap between training and evaluation datasets can lead to inflated performance results, making it essential to consider this factor when interpreting the findings. Virchow2 and UNI emerged as the top-performing pathology-specific vision models (Path-VM) across TCGA, CPTAC, and external tasks, with Virchow2 demonstrating the highest consistency and excelling in distinguishing subtle histological subtypes critical for cancers such as bladder and prostate. Meanwhile, H-optimus-0 and Prov-GigaPath exhibited strong performance in specific tumor types—H-optimus-0 excelled in brain and breast cancer classification, while Prov-GigaPath was highly effective in lung cancer and pan-cancer evaluations. A key observation from this study is the variability in model performance across out-of-domain and external tasks; while models such as H-optimus-0, Prov-GigaPath, and UNI2 performed well on TCGA tasks, their performance declined on external datasets, underscoring the challenge of ensuring generalizability beyond the training dataset. This task-specific specialization suggests that models optimized for particular cancers may exhibit strong performance in niche domains, even if their overall rankings do not fully capture these targeted strengths. Fusion models proved particularly valuable in addressing these generalizability challenges—by aggregating predictions from top-performing individual models using majority voting, they demonstrated superior performance across external datasets, including CPTAC, even though they did not surpass the best individual model (Virchow2) on TCGA tasks. Their strong results in breast, colon, prostate, and pan-cancer classifications further highlight their ability to generalize across diverse tissue types and datasets, making them a promising avenue for enhancing model robustness and real-world applicability. Given that ensemble methods are often more resilient to unseen data 51–53 , fusion models offer a compelling strategy for improving diagnostic AI in pathology. The comparatively lower performance of pathology-specific vision language models (Path-VLM) likely stems from their reliance on smaller datasets, as fewer sources contain both pathology images and paired textual annotations. Even though VLMs (general vision-language models) are trained on large datasets, such as those with hundreds of millions of image-text pairs, they underperform due to the mismatch between general-domain pretraining data and pathology-specific high-resolution images, noisy or irrelevant textual annotations, and limitations in handling the high-resolution visual details critical for pathology tasks. While these models integrate visual and linguistic information, their current architectures may not fully harness the complementary nature of these modalities, limiting their effectiveness in classification tasks. Advancements in multimodal learning, particularly with larger, well-curated datasets and improved fusion techniques, could enhance their capabilities and make them more competitive with image-specific models. Meanwhile, general vision-only models (VM) trained on large-scale datasets like ImageNet 20 exhibited competitive performance in certain tasks but were generally outperformed by pathology-specific vision models (Path-VM) across TCGA, CPTAC, and external benchmarks. This highlights the unique domain-specific features of pathology vision data that general-purpose vision models, trained on non-medical images, fail to fully capture. While VM excelled in specific cancer subtypes, such as gastrointestinal and colorectal tumors, Path-VM demonstrated a more consistent advantage across a broader range of tasks, underscoring the importance of in-domain pre-training for developing more specialized and effective computational pathology models. This study challenges the assumption that larger models and more training data always lead to better performance. While larger vision transformer architectures showed advantages in TCGA tasks, their gains did not extend to CPTAC and external datasets. Similarly, increasing the number of whole-slide images beyond a certain threshold did not consistently improve classification accuracy. Notably, smaller models performed competitively with or even outperformed some models trained on much larger datasets. These findings highlight that beyond a certain scale, model architecture, training strategies, and data quality are more critical than sheer size. Overall, these findings provide valuable insights into the strengths of various model architectures and strategies for pathology-related tasks. The results highlight that, while generalization is a key challenge, task-specific model optimization, along with ensemble-based approaches, can significantly enhance model performance. Future work could focus on exploring the considerable potential of multimodal models that integrate both image and textual data, alongside strategies such as ensemble approaches, addressing data diversity 16,54 and optimization techniques for pathology vision models (Path-VM), to enhance model transferability across diverse datasets. Methods Datasets for downstream tasks TCGA dataset : The Cancer Genome Atlas (TCGA) project aimed at cataloging and generating comprehensive molecular landscapes responsible for cancer. The Cancer Genome Atlas project 30 generated multiple omics modalities, including exome-seq, RNA-seq, and methylation data, along with digital pathology slides, covering over 32 tumor types and 10,000 patients. With respect to digital pathology, TCGA provided a vast collection of high-resolution whole-slide images of tumor sections stained with Hematoxylin and Eosin (H&E). This has facilitated the development of machine learning and deep learning algorithms for tumor subtype classification, prognostication, and treatment response prediction using pathology imaging data. We used this rich resource of data for evaluation of foundation models. We evaluated the models based on tumor-specific molecular subtyping, such as distinguishing between IDH mutant and wild-type brain tumors, as well as pan-cancer molecular subtyping, like differentiating between TP53 mutant and wild-type tumors. This approach provided an estimate of the models' performance in specific tumors and across a pan-cancer setting. Additionally, we evaluated these models on a diverse range of tasks, including pathway activity, immune group subtyping, mRNA cluster classification, and histological subtyping, across various cancer types. Overall, we curated around 19 downstream tasks from TCGA data, encompassing bladder, brain, prostate tumors, and more. CPTAC + External benchmarking datasets : The Clinical Proteomic Tumor Analysis Consortium 55 (CPTAC) has provided pan-cancer datasets that include omics and digital pathology data, similar to the TCGA project. We also used this resource to evaluate our models. Since most pathology models have been trained on the TCGA dataset, testing their performance on CPTAC pathology data will offer a true measure of their efficacy on an independent test dataset. From CPTAC, we curated 7 pan-cancer tasks 56 : grade, SWI pathway alteration, p53, chromatin modification pathway alteration, mTOR pathway alteration, MYC pathway alteration, tumor label. In addition to CPTAC, we curated 8 publicly available external benchmark datasets, including BRACS 57 , BACH 58 , UnitoPatho 59 , SICAPv2 60 , BreakHis 61 , LC2500 62 , MHIST 63 , and NCT-CRC-HE 64 . These datasets contain representative images from diverse tumor types and are completely independent of the training data, ensuring a robust and unbiased evaluation of the foundation models. Out of domain datasets: We used proprietary pathology datasets from Stanford, including lung and brain tumor samples. Given that some foundation models may have used CPTAC data during their pre-training, it was crucial to evaluate these models on our in-house datasets to test their generalizability and estimate their performance on real-world data. With access to detailed clinical information and molecular subtypes, we assessed these models on specific subtyping tasks, such as distinguishing between lung squamous cell carcinoma and adenocarcinoma, and differentiating between MGMT methylated and wild-type brain tumors. We also added NLST pathology 65 and DHMC lung cancer datasets 66 to this category as a part of our out-of-domain evaluation. This approach helped us determine the models' true effectiveness on practical clinical tasks. Metadata collection for downstream tasks For the TCGA tasks, we collected molecular subtype, gene mutation status, histologic subtype, and transcriptomic information from TCGA publications specific to the following cancer types: brain 67 , prostate 68 , lung 69,70 , breast 71 , and bladder 72 . For the pan-cancer tasks, we collected pathway information on the samples from the TCGA pan-cancer pathway study 73 and immune groups from the TCGA immune landscape study 74 . For the CPTAC tasks, we gathered molecular subtypes and activated pathways on cases from the CPTAC pan-cancer study 56 . Whole Slide Image preprocessing Whole-slide images (WSIs), initially in the SVS format, needed to be preprocessed to a machine learning-ready format. Due to the massive size of WSIs, which can often exceed dimensions of 10,000 x 10,000 pixels, it was necessary to divide them into smaller, non-overlapping tiles of 256 x 256 pixels each, at an objective magnification of 20x (0.5 microns per pixel). To accomplish this, a tissue mask was first obtained using the Otsu thresholding method 75 , which effectively separates the foreground (tissue) from the white background. A multiple instance learning (MIL) strategy was employed, where each WSI was represented by a selection of up to 4,000 tiles. This approach was adopted to address the challenge of variable tissue density and distribution within WSIs. Tiles with excessive background or low contrast were excluded to ensure the selection of informative and representative tiles for each WSI.The selected tiles were then saved in an HDF5 database, one database per slide, to facilitate fast input/output (I/O) operations during the training and inference stages. This database organization allows for efficient access to the tiles, reducing the computational overhead associated with loading and processing large WSIs. During the training phase, tiles were grouped into bags of G size. Then, in each bag, a feature vector was extracted for each tile using a given foundation model. A maximum of G tiles was considered in each bag, and the corresponding feature vectors were averaged. This averaged feature vector was then forwarded through a linear layer to obtain the final classification result for each WSI.This approach effectively leverages the MIL framework to model the variability within WSIs and enables the classification of slides based on the collective information extracted from their constituent tiles. The use of the HDF5 database and the efficient tile-based representation contribute to the scalability and computational efficiency of the overall pipeline. Vision transformers Vision Transformers (ViTs) have significantly advanced SSL in the realm of computer vision. Originally designed for natural language processing, ViTs adapt the transformer architecture to visual tasks by dividing images into smaller patches treated as tokens. The self-attention mechanisms in ViTs enable the model to capture intricate details and long-range dependencies within images, providing a superior contextual understanding compared to traditional convolutional neural network-based methods, especially in complex domains like pathology 76 . Foundation models In the past couple of years, several ViT based pathology models like Lunit-Dino 24 , Phikon(Owkin) 33 , UNI 25 , Virchow 26 have been trained on huge datasets on multiple tissue types and evaluated on a variety of downstream tasks. We evaluated in-domain foundation models (i.e., models trained on digital pathology data such as Lunit-Dino), as well as out-of-domain vision models, trained on natural images (ImageNet data) such as DINO from Meta AI. We further categorized these models as general vision only (natural image based), general vision-language (natural images with text), pathology vision models (digital pathology image-based) and pathology vision-language models (pathology images and medical text). We conducted a comparative analysis across the different model categories to assess their relative performance. The pathology domain-specific models have been primarily trained on publicly available datasets like TCGA, and/or on proprietary datasets (for example, in the case of UNI). These foundation models are built upon ViT-small, ViT-base, and ViT-large/ViT-huge architectures as their core backbones. In our work, we benchmark these models using identical tasks, datasets, and hyperparameters. Computational resources For our evaluation experiments, we used the Polaris supercomputer at the Argonne National Laboratory and the Stanford University Sherlock cluster (Stanford Research Computing Center). Model Evaluation We assessed the performance of these models by applying a linear classifier to frozen extracted features, using consistent evaluation settings across all models. We froze the pretrained model and extracted features for the downstream task of interest, before doing linear probing. For our evaluation protocol, we used the same hyperparameters across all models and tasks and did not perform any data augmentation. For linear evaluation, we used a learning rate of 0.001, batch size of 16 per gpu, and trained the linear layer for 30 epochs. All tasks were run on a single node equipped with four NVIDIA A100 GPUs, each with 40 GB of memory. The foundation models varied based on different architecture sizes like ViT-S, ViT-B, ViT-L as well as patch sizes. We divided datasets for TCGA, CPTAC and out-of-domain evaluation downstream tasks into training, validation and test sets with a 70/15/15 split. We also implemented 5-fold cross-validation, complemented by evaluation on a separate independent test set, to robustly assess model performance for these tasks. We report the performance of models on independent test sets in our study. For external public benchmarking tasks, the datasets are usually available as training and test set or validation set. We report the performance on validation or test sets as per their availability. We also evaluated these models on slide level tasks from TCGA, CPTAC and out of domain evaluation datasets as well as patch based classification tasks from external benchmarking tasks separately. Performance is evaluated using metrics including balanced accuracy, precision, recall, and F1 score. We computed the arithmetic mean of the four metrics to derive the ‘average performance’ providing a balanced summary of overall model effectiveness. Declarations Code availability All code for evaluation and benchmarking dashboard is available at https://github.com/gevaertlab/benchmarking-path-models Acknowledgements Research reported here was further supported by the National Cancer Institute (NCI) under awards: R01 CA260271. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. This research was further supported by the Argonne Leadership Computing Facility, a U.S. Department of Energy (DOE) Office of Science user facility at Argonne National Laboratory. References Verghese, G. et al. Computational pathology in cancer diagnosis, prognosis, and prediction - present day and prospects. J. Pathol. 260, 551–563 (2023). Connolly, J. L. et al. Role of the Surgical Pathologist in the Diagnosis and Management of the Cancer Patient. (BC Decker, 2003). Hanna, M. G. & Ardon, O. Digital pathology systems enabling quality patient care. Genes Chromosomes Cancer 62, 685–697 (2023). Louis, D. N. et al. Computational pathology: A path ahead. Arch. Pathol. Lab. Med. 140, 41–50 (2016). Pizurica, M. et al. Whole slide imaging-based prediction of TP53 mutations identifies an aggressive disease phenotype in prostate cancer. Cancer Res. 83, 2970–2984 (2023). Kurc, T. et al. Segmentation and classification in digital pathology for glioma research: Challenges and deep learning approaches. Front. Neurosci. 14, 27 (2020). Cheng, J., Huang, K. & Xu, J. Editorial: Computational pathology for precision diagnosis, treatment, and prognosis of cancer. Front. Med. 10, 1209666 (2023). McGenity, C. et al. Artificial intelligence in digital pathology: a systematic review and meta-analysis of diagnostic test accuracy. NPJ Digit. Med. 7, 114 (2024). Springenberg, M. et al. From modern CNNs to vision transformers: Assessing the performance, robustness, and classification strategies of deep learning models in histopathology. Med. Image Anal. 87, 102809 (2023). Phillips, S. P., Spithoff, S. & Simpson, A. Artificial intelligence and predictive algorithms in medicine: Promise and problems. Can. Fam. Physician 68, 570–572 (2022). Steyaert, S. et al. Multimodal data fusion for cancer biomarker discovery with deep learning. Nat. Mach. Intell. 5, 351–362 (2023). Boehm, K. M. et al. Multimodal data integration using machine learning improves risk stratification of high-grade serous ovarian cancer. Nat. Cancer 3, 723–733 (2022). Zheng, Y., Carrillo-Perez, F., Pizurica, M., Heiland, D. H. & Gevaert, O. Spatial cellular architecture predicts prognosis in glioblastoma. Nat. Commun. 14, 4122 (2023). Levy-Jurgenson, A., Tekpli, X., Kristensen, V. N. & Yakhini, Z. Spatial transcriptomics inferred from pathology whole-slide images links tumor heterogeneity to survival in breast and lung cancer. Sci. Rep. 10, 18802 (2020). Krauze, A. V., Zhuge, Y., Zhao, R., Tasci, E. & Camphausen, K. AI-driven image analysis in central nervous system tumors-traditional Machine Learning, Deep Learning and hybrid models. J. Biotechnol. Biomed. 5, 1–19 (2022). Carrillo-Perez, F. et al. Generation of synthetic whole-slide image tiles of tumours from RNA-sequencing data via cascaded diffusion models. Nat. Biomed. Eng. 9, 320–332 (2025). Pizurica, M. et al. Digital profiling of gene expression from histology images with linearized attention. Nat. Commun. 15, 9886 (2024). Thakur, N., Yoon, H. & Chong, Y. Current trends of artificial intelligence for colorectal cancer pathology image analysis: A systematic review. Cancers (Basel) 12, 1884 (2020). Marini, N. et al. Unleashing the potential of digital pathology data by training computer-aided diagnosis models without human annotations. NPJ Digit. Med. 5, 102 (2022). Deng, J. et al. ImageNet: A large-scale hierarchical image database. in 2009 IEEE Conference on Computer Vision and Pattern Recognition (IEEE, 2009). doi: 10.1109/cvpr.2009.5206848 . Rani, V., Nabi, S. T., Kumar, M., Mittal, A. & Kumar, K. Self-supervised learning: A succinct review. Arch. Comput. Methods Eng. 30, 2761–2775 (2023). Caron, M. et al. Emerging properties in self-supervised vision transformers. arXiv [cs.CV] (2021) doi: 10.48550/ARXIV.2104.14294 . Zhou, J. et al. iBOT: Image BERT Pre-Training with Online Tokenizer. arXiv [cs.CV] (2021) doi: 10.48550/ARXIV.2111.07832 . Kang, M., Song, H., Park, S., Yoo, D. & Pereira, S. Benchmarking self-supervised learning on diverse pathology datasets. arXiv [cs.CV] (2022). Chen, R. J. et al. Towards a general-purpose foundation model for computational pathology. Nat. Med. 30, 850–862 (2024). Vorontsov, E. et al. Virchow: A million-slide digital pathology foundation model. arXiv [eess.IV] (2023). Nechaev, D., Pchelnikov, A. & Ivanova, E. Hibou: A family of foundational vision transformers for pathology. arXiv [eess.IV] (2024) doi: 10.48550/ARXIV.2406.05074 . Zimmermann, E. et al. Virchow2: Scaling self-supervised mixed magnification models in pathology. arXiv [cs.CV] (2024) doi: 10.48550/ARXIV.2408.00738 . Wang, X. et al. Transformer-based unsupervised contrastive learning for histopathological image classification. Med. Image Anal. 81, 102559 (2022). Cancer Genome Atlas Research Network et al. The Cancer Genome Atlas Pan-Cancer analysis project. Nat. Genet. 45, 1113–1120 (2013). Kim, K. et al. PAIP 2020: Microsatellite instability prediction in colorectal cancer. Med. Image Anal. 89, 102886 (2023). Kim, Y. J. et al. PAIP 2019: Liver cancer segmentation challenge. Med. Image Anal. 67, 101854 (2021). Filiot, A. et al. Scaling self-Supervised Learning for histopathology with Masked Image Modeling. medRxiv (2023) doi: 10.1101/2023.07.21.23292757 . Huang, Z., Bianchi, F., Yuksekgonul, M., Montine, T. J. & Zou, J. A visual-language foundation model for pathology image analysis using medical Twitter. Nat. Med. 29, 2307–2316 (2023). Ikezogwo, W. O. et al. Quilt-1M: One million image-text pairs for histopathology. arXiv [cs.CV] (2023). Lu, M. Y. et al. A visual-language foundation model for computational pathology. Nat. Med. 30, 863–874 (2024). Radford, A. et al. Learning transferable visual models from natural language supervision. arXiv [cs.CV] (2021) doi: 10.48550/ARXIV.2103.00020 . Sikaroudi, M., Hosseini, M., Gonzalez, R., Rahnamayan, S. & Tizhoosh, H. R. Generalization of vision pre-trained models for histopathology. Sci. Rep. 13, 6065 (2023). Papadopoulos, K. M. & Stathaki, T. Comparing ImageNet pre-training with digital pathology foundation models for whole slide image-based survival analysis. arXiv [eess.IV] (2024) doi: 10.48550/ARXIV.2405.17446 . Campanella, G. et al. Computational pathology at health system scale -- self-supervised foundation models from three billion images. arXiv [cs.CV] (2023) doi: 10.48550/ARXIV.2310.07033 . Lin, W., Liu, S., Zhu, R. & Wang, L. Unveiling institution-specific bias in pathology foundation models: Detriments, causes, and potential solutions. arXiv [cs.CV] (2025) doi: 10.48550/ARXIV.2502.16889 . Yang, Y. et al. A survey of recent methods for addressing AI fairness and bias in biomedicine. J. Biomed. Inform. 154, 104646 (2024). Vaidya, A. et al. Demographic bias in misdiagnosis by computational pathology models. Nat. Med. 30, 1174–1190 (2024). Yang, Z. et al. A foundation model for generalizable cancer diagnosis and survival prediction from histopathological images. Nat. Commun. 16, 2366 (2025). Li, D. et al. A survey on computational pathology foundation models: Datasets, adaptation strategies, and evaluation tasks. arXiv [cs.CV] (2025) doi: 10.48550/ARXIV.2501.15724 . Schuhmann, C. et al. LAION-5B: An open large-scale dataset for training next generation image-text models. arXiv [cs.CV] (2022) doi: 10.48550/ARXIV.2210.08402 . Ai, K. et al. Towards large-scale training of pathology foundation models. arXiv [cs.CV] (2024) doi: 10.48550/ARXIV.2404.15217 . Huang, Y. et al. Comparative analysis of ImageNet pre-trained deep learning models and DINOv2 in medical imaging classification. arXiv [eess.IV] (2024) doi: 10.48550/ARXIV.2402.07595 . Huix, J. P. et al. Are natural domain foundation models useful for medical image classification? arXiv [cs.CV] (2023) doi: 10.48550/ARXIV.2310.19522 . Xu, H. et al. A whole-slide foundation model for digital pathology from real-world data. Nature 630, 181–188 (2024). Gong, H., Wang, M., Zhang, H., Elahe, M. F. & Jin, M. An explainable AI approach for the rapid diagnosis of COVID-19 using ensemble learning algorithms. Front. Public Health 10, 874455 (2022). Yang, J., Wang, G., Xiao, X., Bao, M. & Tian, G. Explainable ensemble learning method for OCT detection with transfer learning. PLoS One 19, e0296175 (2024). Liu, Y. et al. Deep learning based digital pathology for predicting treatment response to first-line PD-1 blockade in advanced gastric cancer. J. Transl. Med. 22, 438 (2024). Carrillo-Perez, F. et al. Synthetic whole-slide image tile generation with gene expression profile-infused deep generative models. Cell Rep. Methods 3, 100534 (2023). Li, Y. et al. Proteogenomic data and resources for pan-cancer analysis. Cancer Cell 41, 1397–1406 (2023). Zhang, Y., Chen, F., Chandrashekar, D. S., Varambally, S. & Creighton, C. J. Proteogenomic characterization of 2002 human cancers reveals pan-cancer molecular subtypes and associated pathways. Nat. Commun. 13, 2669 (2022). Brancati, N. et al. BRACS: A dataset for BReAst Carcinoma Subtyping in H&E histology images. Database 2022, (2022). Aresta, G. et al. BACH: Grand challenge on breast cancer histology images. Med. Image Anal. 56, 122–139 (2019). Bertero, L. et al. UNITOPATHO. IEEE DataPort https://doi.org/10.21227/9FSV-TM25 (2021). Silva-Rodríguez, J., Colomer, A., Sales, M. A., Molina, R. & Naranjo, V. Going deeper through the Gleason scoring scale: An automatic end-to-end system for histology prostate grading and cribriform pattern detection. arXiv [eess.IV] (2021). Spanhol, F. A., Oliveira, L. S., Petitjean, C. & Heutte, L. A dataset for breast cancer histopathological image classification. IEEE Trans. Biomed. Eng. 63, 1455–1462 (2016). Borkowski, A. A. et al. Lung and Colon Cancer Histopathological Image Dataset (LC25000). arXiv [eess.IV] (2019) doi: 10.48550/ARXIV.1912.12142 . Wei, J. et al. A Petri Dish for Histopathology Image Analysis. arXiv [eess.IV] (2021) doi: 10.48550/ARXIV.2101.12355 . Kather, J. N., Halama, N. & Marx, A. 100,000 histological images of human colorectal cancer and healthy tissue. Zenodo https://doi.org/10.5281/zenodo.1214456 (2018). National Lung Screening Trial Research Team. Data from the National Lung Screening Trial (NLST). The Cancer Imaging Archive https://doi.org/10.7937/TCIA.HMQ8-J677 (2013). Wei, J. W. et al. Pathologist-level classification of histologic patterns on resected lung adenocarcinoma slides with deep neural networks. Sci. Rep. 9, 3358 (2019). Brennan, C. W. et al. The somatic genomic landscape of glioblastoma. Cell 155, 462–477 (2013). Cancer Genome Atlas Research Network. The molecular taxonomy of primary prostate cancer. Cell 163, 1011–1025 (2015). Cancer Genome Atlas Research Network. Comprehensive genomic characterization of squamous cell lung cancers. Nature 489, 519–525 (2012). Cancer Genome Atlas Research Network. Comprehensive molecular profiling of lung adenocarcinoma. Nature 511, 543–550 (2014). Berger, A. C. et al. A comprehensive pan-cancer molecular study of gynecologic and breast cancers. Cancer Cell 33, 690–705.e9 (2018). Cancer Genome Atlas Research Network. Comprehensive molecular characterization of urothelial bladder carcinoma. Nature 507, 315–322 (2014). Sanchez-Vega, F. et al. Oncogenic signaling pathways in The Cancer Genome Atlas. Cell 173, 321–337.e10 (2018). Thorsson, V. et al. The immune landscape of cancer. Immunity 48, 812–830.e14 (2018). Otsu, N. A threshold selection method from gray-level histograms. IEEE Trans. Syst. Man Cybern. 9, 62–66 (1979). He, K. et al. Transformers in Medical Image Analysis: A Review. arXiv [cs.CV] (2022). Yu, J. et al. CoCa: Contrastive Captioners are Image-Text Foundation Models. arXiv [cs.CV] (2022) doi: 10.48550/ARXIV.2205.01917 . Tables Table 1. Abbreviations of Model Categories Model Category Model Category Abbreviation Training data General vision models VM Natural images General vision language models VLM Natural images and text Pathology-specific vision models Path-VM Pathology-specific images Pathology-specific vision language models Path-VLM Pathology-specific images and text Table 2. Pathology-specific vision model (Path-VM) summary Model Name SSL method Model Architecture Parameters in million No. of WSI’s in million No. of patches in million No. of cancer/tissue types Data Source cTranspath Contrastive learning CNN + SwinT ~27 .032 15 32 cancer types TCGA, PAIP Kaiko DINO ViT-base ~85 .029 NA 32 cancer types TCGA HIPT DINO ViT-small ~21 .01 104 33 cancer types TCGA BEPH BEiT ViT-base ~85 .012 11 32 cancer types TCGA Virchow DINOv2 Vit-huge ~632 1.5 2000 NA Proprietary Lunit DINO ViT-small ~21 .036 32.5 NA TCGA, TULIP UNI DINOv2 ViT-large ~307 0.1 100 20 tissue types Proprietary, GTEx Hibou DINOv2 ViT-base ~85 1 1200 NA Proprietary Phikon iBOT ViT-base ~85 0.006 40 16 cancer types TCGA GPFM DINOv2 ViT-large ~307 0.086 190 34 tissue types TCGA, PAIP and 49 public datasets H-optimus-O DINOv2 Vit-giant ~1100 0.5 NA NA Proprietary UNI2 DINOv2 Vit-huge ~632 0.35 200 NA Proprietary Phikon-v2 DINOv2 ViT-large ~307 0.058 460 30 cancer sites TCGA, GTEx, CPTAC, TCIA datasets Virchow2 DINOv2 Vit-huge ~632 3.1 NA 17 tissue types Proprietary Prov-GigaPath DINOv2 Vit-giant ~1100 0.171 1300 31 tissue types Proprietary Table 3. Pathology-specific vision-language model (Path-VLM) summary Model Name Pretraining method Model Architecture Parameters in million Data source PLIP CLIP ViT-base ~85 208k pathology image-text pairs QuiltNet-B16 CLIP ViT-base ~85 1M pathology image-text pairs BiomedCLIP CLIP ViT-base ~85 14M pathology image-text pairs Mizero pubmedbert Contrastive learning ViT-small ~21 33k pathology image-text pairs Mizero clinmedbert Contrastive learning ViT-small ~21 33k pathology image-text pairs CONCH CoCa 77 ViT-base ~85 1.17 pathology image-text pairs TITAN iBOT + CoCa 77 ViT-base ~85 336k whole slide images & 423k image captions Table 4. General Vision Model (VM) summary Model Name SSL method Model Architecture Parameters in million Data source Dino-S16 DINO ViT-small ~21 ImageNet (14M) Dino-B16 DINO ViT-base ~85 ImageNet (14M) Dinov2 DINOv2 ViT-large ~300 LVD-142M iBOT-B16 iBOT ViT-base ~85 ImageNet (14M) iBOT-L16 iBOT ViT-large ~307 ImageNet (14M) Table 5. General Vision-language model (VLM) summary Model Name Pretraining method Model Architecture Parameters in million Data source CLIP-B16 CLIP (Contrastive learning) ViT-base ~115 400 million (image, text) pairs BLIP-B16-14M BLIP ViT-base ~85 ImageNet (14M) Align-base EfficientNet + BERT (Contrastive learning) ViT-base ~746 700 million (image, text) pairs beit3-L16 iBOT ViT-large ~307 ImageNet (14M) Table 6. Downstream tasks Task Cancer Binary or Multi Dataset Slide or patch classification tissue 1 TCGA-Grade Pan-cancer multi-class TCGA Slide pan-cancer 2 PAM50 status BRCA multi-class TCGA Slide breast 3 prad mrna cluster PRAD multi-class TCGA Slide prostate 4 IDH mutation GBM + LGG binary TCGA Slide brain 5 brain hist subtype GBM + LGG multi-class TCGA Slide brain 6 PI3K Pan-cancer binary TCGA Slide pan-cancer 7 ER status BRCA binary TCGA Slide breast 8 MYC Pan-cancer binary TCGA Slide pan-cancer 9 Immunegroup Pan-cancer multi-class TCGA Slide pan-cancer 10 MGMT GBM + LGG binary TCGA Slide brain 11 blca hist subtype BLCA multi-class TCGA Slide brain 12 gi subtype GI multi-class TCGA Slide GI 13 brain transcriptome subtype GBM + LGG multi-class TCGA Slide brain 14 lung transcriptome subtype LUSC + LUAD multi-class TCGA Slide lung 15 blca mrna cluster BLCA multi-class TCGA Slide bladder 16 lusc vs luad LUSC + LUAD binary TCGA Slide lung 17 ERG status PRAD binary TCGA Slide prostate 18 kidney subtype KIRC + KIRP + KICH multi-class TCGA Slide kidney 19 PR status BRCA binary TCGA Slide breast 20 CPTAC-grade Pan-cancer multi-class CPTAC Slide pan-cancer 21 SWI Pan-cancer binary CPTAC Slide pan-cancer 22 P53 Pan-cancer binary CPTAC Slide pan-cancer 23 Chrom Pan-cancer binary CPTAC Slide pan-cancer 24 MTOR Pan-cancer binary CPTAC Slide pan-cancer 25 MYC Pan-cancer binary CPTAC Slide pan-cancer 26 tumor label Pan-cancer multi-class CPTAC Slide pan-cancer 27 SICAPv2 PRAD multi-class External benchmarking Patch prostate 28 MHIST COAD binary External benchmarking Patch colon 29 UniToPatho COAD multi-class External benchmarking Patch colon 30 BACH BRCA multi-class External benchmarking Patch breast 31 BRACS BRCA multi-class External benchmarking Patch breast 32 LC25000 Pan-cancer multi-class External benchmarking Patch pan-cancer 33 BreakHis BRCA multi-class External benchmarking Patch breast 34 NCT-CRC-HE COAD multi-class External benchmarking Patch colon 35 lung stage Lung multi-class Out of Domain (NLST) Slide lung 36 lung grade Lung multi-class Out of Domain (NLST) Slide lung 37 lung hist OOD1 Lung binary Out of Domain (Stanford) Slide lung 38 lung hist OOD2 Lung multi-class Out of Domain (DHMC) Slide lung 39 Lung stage OOD Lung multi-class Out of Domain (Stanford) Slide lung 40 Brain necrosis Brain binary Out of Domain (Stanford) Slide brain 41 MGMT OOD Brain binary Out of Domain (Stanford) Slide brain Additional Declarations There is NO Competing Interest. Supplementary Files SupplementaryFigures.pdf Supplementary Figure Legends Supplementary Figure 1. Classification performance of pathology-specific vision models (Path-VM) on TCGA tasks. (a) Top 5 performing Path-VM in each TCGA task. (b) Model-wise classification performance across individual TCGA tumor types, sorted by average performance per model. This figure highlights the relative strengths of each model across different tumor types and underscores variability in model generalization. Supplementary Figure 2. Classification performance of top 10 pathology- specific vision models (Path-VM) on each TCGA task. (a) Top 10 performing Path-VM in each TCGA task, ranked by average performance. The top three models are highlighted in purple for emphasis. Supplementary Figure 3. Classification performance in CPTAC and external benchmarking tasks. (a) Top 10 performing pathology-specific vision models (Path-VM) in each CPTAC task, ranked by average performance. The top three models are highlighted in purple for emphasis. (b) Top 10 performing Path-VM in each external task, ranked by average performance, with the top three models highlighted in purple for emphasis. (c) Performance of all 31 models across, sorted by mean of average performance across CPTAC tasks. (d) Performance of all 31 models, sorted by mean of average performance, across external tasks. (e) Performance of Path-VM models on external tasks, grouped by tumor types BRCA and COAD. Supplementary Figure 4. Classification Performance of top pathology-specific vision(Path-VM) models vs other models. (a) Performance comparison between Virchow2 vs rest of other models. (b) Performance comparison between UNI vs rest of other models. (c) Performance comparison between Prov-GigaPath vs rest of other models. (d) Performance comparison between H-optimus-0 vs rest of other models. (e) Performance comparison between UNI2 vs rest of other models. Significance levels: * p < 0.05, ** p < 0.01, *** p < 0.001 (t-tests with Bonferroni correction) Supplementary Figure 5. Classification performance of pathology-specific vision models ( Path-VM )and pathology-specific vision-language models (Path-VLM) in TCGA tasks. (a) Performance of all Path-VM and Path-VLM in each TCGA task, ranked by average performance, with the top three models highlighted in pink for emphasis. Supplementary Figure 6. Classification performance of pathology-specific vision models ( Path-VM )and pathology-specific vision-language models (Path-VLM) in CPTAC tasks. (a) Performance of all Path-VM and Path-VLM in each CPTAC task, ranked by average performance, with the top three models highlighted in pink for emphasis. Supplementary Figure 7. Classification performance of pathology-specific vision models ( Path-VM )and pathology-specific vision-language models (Path-VLM) in external tasks. (a) Performance of all Path-VM and Path-VLM in each external task, ranked by average performance, with the top three models highlighted in pink for emphasis. Supplementary Figure 8. Classification performance of pathology-specific vision models ( Path-VM ) and general vision models (VM) in TCGA tasks. (a) Top 10 performing pathology and vision models in each TCGA task, ranked by average performance, with the top three models visually emphasized in pink. Supplementary Figure 9. Classification performance of pathology-specific vision and general vision models in CPTAC and external tasks. (a) Top 10 performing Path-VM and VM in each CPTAC task, ranked by average performance, with the top three models visually emphasized in pink. (b) Top 10 performing Path-VM and VM in each external task, ranked by average performance, with the top three models visually emphasized in pink. Supplementary Figure 10. Classification performance in TCGA. (a) Performance of all Path-VM and Path-VLM in each TCGA task, based on average performance (b) Performance of all Path-VM and VM in each TCGA task, based on average performance. Supplementary Figure 11. Comparison between model categories. Performance comparison of general vision, pathology-specific vision models, general vision language models and pathology-specific vision-language models in (a) TCGA (b) CPTAC (c) External benchmarking tasks. Significance levels: * p < 0.05, ** p < 0.01, *** p < 0.001 (pairwise t-tests with Bonferroni correction) (d) Heatmap of performance of all models in all tasks including out of domain tasks. Supplementary Figure 12. Comparison between Model Architectures and performance of pathology models in external tasks. Performance comparison of vision model architectures ViT-S, Vit-B , ViT-L in (a) TCGA (b) CPTAC (c) External benchmarking tasks (d) Cumulative average performance of Path-VM across out of domain tasks, sorted in descending order. (e) Cumulative average performance of Path-VM across external benchmarking tasks and out of domain tasks, sorted in descending order. Cite Share Download PDF Status: Under Review Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6823810","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":471069404,"identity":"bbeaf1f8-3034-4485-aaed-7f93f7ab1d3b","order_by":0,"name":"Olivier Gevaert","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA6ElEQVRIie3NsQrCMBCA4UggU2rWhPoQkUJB7MMcFLoquDiICIKT4FrBzRdJKegidC0IohSc3ApSEMWqm0Osm2B+SA7CfQQhk+mXY8/bK49CiFcRXIxwOYJviFRVCWPLFS/6u6GTJFHeBa9TVzjaUg0R4TEQ002Pu6mP7RCC1lwRv60jMt040ppASTCyrSKWUlHX1pMkb15vwJ1ZjC8UHoSd9SSZ1jJrBFwin9gvQomWiJA0s8YKRJj6bptCIEVMnNZCQxiL99FpAIzNomxLwZP19fiQnjQEcXh/wbr15zfq04bJZDL9fXfE+0oxYGJDtQAAAABJRU5ErkJggg==","orcid":"https://orcid.org/0000-0002-9965-5466","institution":"Stanford University School of Medicine","correspondingAuthor":true,"prefix":"","firstName":"Olivier","middleName":"","lastName":"Gevaert","suffix":""},{"id":471069405,"identity":"1ea1e4ef-4a8b-4865-a902-f912830f56b4","order_by":1,"name":"Rohan Bareja","email":"","orcid":"","institution":"Stanford University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Rohan","middleName":"","lastName":"Bareja","suffix":""},{"id":471069406,"identity":"57bee3de-d81c-441e-9d35-04ad1dbd4a94","order_by":2,"name":"Francisco Carrillo-Perez","email":"","orcid":"","institution":"Stanford University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Francisco","middleName":"","lastName":"Carrillo-Perez","suffix":""},{"id":471069407,"identity":"91b44088-fca7-4e09-9e54-94a8f0006640","order_by":3,"name":"Yuanning Zheng","email":"","orcid":"https://orcid.org/0000-0002-0018-3252","institution":"Stanford University","correspondingAuthor":false,"prefix":"","firstName":"Yuanning","middleName":"","lastName":"Zheng","suffix":""},{"id":471069408,"identity":"16ed3f3d-22aa-4b6e-b52f-e60707715b04","order_by":4,"name":"Marija Pizurica","email":"","orcid":"","institution":"Ghent University","correspondingAuthor":false,"prefix":"","firstName":"Marija","middleName":"","lastName":"Pizurica","suffix":""},{"id":471069409,"identity":"dfc14499-481d-4068-b2b4-566bfe745e72","order_by":5,"name":"Tarak Nandi","email":"","orcid":"","institution":"Data Science and Learning Division, Argonne National Laboratory","correspondingAuthor":false,"prefix":"","firstName":"Tarak","middleName":"","lastName":"Nandi","suffix":""},{"id":471069410,"identity":"d4510732-6fe2-4bdd-ac2c-3c1f0c4018b8","order_by":6,"name":"Jeanne Shen","email":"","orcid":"https://orcid.org/0000-0002-1519-0308","institution":"Stanford University","correspondingAuthor":false,"prefix":"","firstName":"Jeanne","middleName":"","lastName":"Shen","suffix":""},{"id":471069411,"identity":"3cff1088-f6f5-4ed4-ba81-ae0eb597783f","order_by":7,"name":"Ravi Madduri","email":"","orcid":"https://orcid.org/0000-0003-2130-2887","institution":"Argonne National Laboratory","correspondingAuthor":false,"prefix":"","firstName":"Ravi","middleName":"","lastName":"Madduri","suffix":""}],"badges":[],"createdAt":"2025-06-04 23:10:27","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6823810/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6823810/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":85997871,"identity":"ce79fc63-8eea-49f2-a183-fb7eaf0e0a6d","added_by":"auto","created_at":"2025-07-04 06:42:12","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":451373,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ea)\u003c/strong\u003e. \u003cstrong\u003eBenchmarking Workflow \u003c/strong\u003eA schematic representation of the benchmarking workflow used in this study. The framework includes data preprocessing, model feature extraction, evaluation on multiple histopathology datasets, and performance comparison across tumor classification, molecular subtyping, and grading tasks.\u003cstrong\u003e b).\u003c/strong\u003e \u003cstrong\u003eModel Performance on TCGA Dataset\u003c/strong\u003e Performance comparison of foundation models on TCGA-derived tasks. The top panel displays the average performance of Path-VM models, while the bottom panel includes all evaluated models including Path-VM, VM, VLM, Path-VLM. Average performance is defined as the mean of balanced accuracy, F1-score, recall, and precision. Each black dot indicates the mean performance across TCGA-derived tasks for a given model, while the red bars represent the standard deviation across those tasks. \u003cstrong\u003ec). Top-Ranked Path-VM Models\u003c/strong\u003e The most top ranked Path-VM model based on average performance metrics (marked with asterisk), demonstrating superior predictive performance across tumor classification, molecular subtyping, and grading tasks. The figure illustrates the frequency with which each model achieved the top ranking across 19 TCGA tasks. Each bar represents a specific TCGA task. \u003cstrong\u003ed)\u003c/strong\u003e. \u003cstrong\u003ePath-VM Performance Across Tumor Types\u003c/strong\u003e Evaluation of pathology-specific vision models across different tumor types, illustrating variations in predictive power depending on tissue origin and diagnostic complexity. Performance trends highlight strengths and limitations in pathology model generalization across tumor subtypes.\u003c/p\u003e","description":"","filename":"BenchmarkingFigures1.png","url":"https://assets-eu.researchsquare.com/files/rs-6823810/v1/a39a1da74feb9923ad15f880.png"},{"id":85997272,"identity":"df3b9b68-533a-43cd-9393-75f964667812","added_by":"auto","created_at":"2025-07-04 06:34:12","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":228076,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ea). Path-VM performance in CPTAC\u003c/strong\u003e – Performance comparison of pathology-specific vision models on CPTAC datasets. Models are evaluated based on average performance across multiple classification tasks, highlighting their robustness and generalizability within the CPTAC domain. \u003cstrong\u003eb). Path-VM performance in External benchmarking\u003c/strong\u003e – Evaluation of pathology models on external benchmarking datasets, assessing their ability to generalize beyond training distributions and perform effectively on independent datasets. \u003cstrong\u003ec).\u003c/strong\u003e \u003cstrong\u003eTop 5 Path-VM performance in CPTAC\u003c/strong\u003e – Performance analysis of the five highest-ranking Path-VM on CPTAC datasets, demonstrating their relative strengths across classification tasks. The top performing model for a task is highlighted with an asterisk. \u003cstrong\u003ed). Top 5 Path-VM Performance in External Benchmarking\u003c/strong\u003e – Comparative performance of the top five Path-VM on external benchmarking datasets, illustrating their generalizability across diverse histopathology data sources. The top performing model for a task is highlighted with an asterisk. \u003cstrong\u003ee). Most top-ranked Path-VM in CPTAC\u003c/strong\u003e among all Path-VM– The number of times each Path-VM achieved the highest ranking across CPTAC classification tasks. Each bar represents a specific CPTAC task. The asterisk represents the top performing model in that task. \u003cstrong\u003ef). Most top-ranked Path-VM in External benchmarking\u003c/strong\u003e – The frequency of top rankings achieved by Path-VM across external benchmarking tasks. Each bar represents a distinct classification task, highlighting the best-performing models in out-of-domain evaluations.\u003c/p\u003e","description":"","filename":"BenchmarkingFigures2.png","url":"https://assets-eu.researchsquare.com/files/rs-6823810/v1/74f14af334400b1d60feb6a3.png"},{"id":85997188,"identity":"7c70014a-13b8-4c80-818e-c444a412f242","added_by":"auto","created_at":"2025-07-04 06:26:12","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":456984,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ea). Path-VM, Path-VLM and VM in TCGA. \u003c/strong\u003eComparison of average performance between pathology-specific vision models (Path-VM), pathology-specific vision-language models (Path-VLM) and general vision models (VM) across TCGA , CPTAC and External benchmarking tasks. Path-VM perform better than Path-VLM across TCGA, CPTAC and External benchmarking tasks, with statistically significant differences. No statistically significant differences observed between Path-VM and VM. \u003cstrong\u003eb). Path-VM, Path-VLM and VM in each CPTAC task. \u003c/strong\u003ePer-task performance of Path-VM, Path-VLM and VM across CPTAC datasets. Each dot represents a model’s performance in a given task, illustrating the consistent advantage of pathology-specific vision models across most tasks. \u003cstrong\u003ec)\u003c/strong\u003e. \u003cstrong\u003ePath-VM, Path-VLM and VM in each External benchmarking task. \u003c/strong\u003ePer-task performance comparison between Path-VM, Path-VLM and VM on external datasets. Path-VM consistently outperform and dominate top rankings, demonstrating superior performance.\u003c/p\u003e","description":"","filename":"BenchmarkingFigures3.png","url":"https://assets-eu.researchsquare.com/files/rs-6823810/v1/0453ca1ea8c997752aff356d.png"},{"id":85997192,"identity":"dea97b65-3868-41d2-ab70-863b9bd82ca2","added_by":"auto","created_at":"2025-07-04 06:26:12","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":453227,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ea)\u003c/strong\u003e \u003cstrong\u003ePath-VM :\u003c/strong\u003e \u003cstrong\u003eComparison of ViT Architectures in TCGA, CPTAC and External benchmarking\u003c/strong\u003e – Model performance analysis of ViT architectures on TCGA tasks reveals statistically significant differences between ViT-small, and ViT-large models, as well as between ViT-base and ViT-large models, while no significant difference is observed between ViT-small, and ViT-base models. Model performance evaluation on CPTAC tasks indicates no statistically significant differences among ViT-small, ViT-base, and ViT-large architectures. Performance comparison of ViT architectures across external benchmarking datasets shows no statistically significant differences among ViT-small, ViT-base, and ViT-large models. \u003cstrong\u003eb)\u003c/strong\u003e \u003cstrong\u003ePath-VM\u003c/strong\u003e: \u003cstrong\u003eImpact of pretraining data Size on performance in TCGA, CPTAC and External benchmarking\u003c/strong\u003e – Model performance in TCGA classification tasks as a function of the number of whole-slide images used during pretraining, highlighting the influence of data scale on downstream accuracy. Data sizes are categorized as small (\u0026lt;0.1 million slides), medium (0.1 - 1 million slides), and large (≥1 million slides). Statistical analysis reveals significant differences between small and large as well as small and medium data sizes, while no significant difference is observed between medium and large data sizes. Statistical analysis of model performance in CPTAC tasks reveals significant differences between small and large data sizes, as well as between small and medium, but no significant difference between medium and large. Statistical analysis of model performance in External benchmarking indicates significant differences between small and large data sizes, as well as between small and medium, with no significant difference observed between medium and large data sizes in the external benchmarking tasks. Significance levels: * p \u0026lt; 0.05, ** p \u0026lt; 0.01, *** p \u0026lt; 0.001 (pairwise t-tests with Bonferroni correction)\u003c/p\u003e","description":"","filename":"BenchmarkingFigures4.png","url":"https://assets-eu.researchsquare.com/files/rs-6823810/v1/f8e1aa8f025e13dd190f1fee.png"},{"id":85997191,"identity":"2b5e7bb0-719f-4064-94f2-85074d8c3259","added_by":"auto","created_at":"2025-07-04 06:26:12","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":422801,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ea) Model Rankings: TCGA\u003c/strong\u003e – Cumulative average performance of pathology models across TCGA classification tasks, with rankings reflecting overall model performance. The fusion model ranks second in this category, demonstrating strong performance across TCGA tasks. \u003cstrong\u003eb)\u003c/strong\u003e \u003cstrong\u003eModel Rankings: External Benchmarking + Out-of-Domain\u003c/strong\u003e – Cumulative average performance of pathology models across external benchmarking and out-of-domain classification tasks. This ranking emphasizes the ability of models to generalize across datasets that were not used during pre-training, with the fusion model achieving the top ranking in this category. \u003cstrong\u003ec)\u003c/strong\u003e \u003cstrong\u003eModel Rankings: CPTAC\u003c/strong\u003e – Cumulative average performance of pathology models across CPTAC classification tasks, showcasing the overall model rankings within the CPTAC dataset. The fusion model performs the best in this category, leading the rankings across CPTAC tasks. \u003cstrong\u003ed)\u003c/strong\u003e \u003cstrong\u003eHeatmap of Model Performance Across All Tasks\u003c/strong\u003e – Heatmap showing model performance across all classification tasks, with z-scores indicating performance relative to other models within each specific task. This provides insight into how models perform across a broad range of pathological challenges. \u003cstrong\u003ee)\u003c/strong\u003e \u003cstrong\u003eHeatmap of Model Performance Across Tissues\u003c/strong\u003e – Heatmap displaying model performance across various tissue types, with z-scores calculated for each model’s performance within individual tissue categories. This illustrates how each model performs relative to other models for each tissue type.\u003c/p\u003e","description":"","filename":"BenchmarkingFigures5.png","url":"https://assets-eu.researchsquare.com/files/rs-6823810/v1/54061cae096ec12de975670f.png"},{"id":87516644,"identity":"3a89a9ce-1a93-44a3-b81e-c32d54da06bd","added_by":"auto","created_at":"2025-07-24 16:39:14","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3832395,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6823810/v1/466fd46b-e9e8-4fee-b7d0-dcb812549f1f.pdf"},{"id":85997193,"identity":"9b027065-fdc8-4f1b-89fb-1076c328e218","added_by":"auto","created_at":"2025-07-04 06:26:12","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":5375418,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSupplementary Figure Legends\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSupplementary Figure 1.\u003c/strong\u003e \u003cem\u003eClassification performance of pathology-specific vision models (Path-VM) on TCGA tasks. \u003c/em\u003e\u003cstrong\u003e(a)\u003c/strong\u003e Top 5 performing Path-VM in each TCGA task. \u003cstrong\u003e(b)\u003c/strong\u003e Model-wise classification performance across individual TCGA tumor types, sorted by average performance per model. This figure highlights the relative strengths of each model across different tumor types and underscores variability in model generalization.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSupplementary Figure 2. \u003c/strong\u003e\u003cem\u003eClassification performance of top 10 \u003c/em\u003epathology-\u003cem\u003especific vision models\u003c/em\u003e \u003cem\u003e(Path-VM) on each TCGA task. \u003c/em\u003e(a) Top 10 performing Path-VM in each TCGA task, ranked by average performance. The top three models are highlighted in purple for emphasis.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSupplementary Figure 3. \u003c/strong\u003e\u003cem\u003eClassification performance in CPTAC and external benchmarking tasks.\u003c/em\u003e (a) Top 10 performing pathology-specific vision models (Path-VM)\u003cem\u003e \u003c/em\u003ein each CPTAC task, ranked by average performance. The top three models are highlighted in purple for emphasis. (b) Top 10 performing Path-VM in each external task, ranked by average performance, with the top three models highlighted in purple for emphasis. (c) Performance of all 31 models across, sorted by mean of average performance across CPTAC tasks. (d) Performance of all 31 models, sorted by mean of average performance, across external tasks. (e) Performance of Path-VM models on external tasks, grouped by tumor types BRCA and COAD.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSupplementary Figure 4. \u003c/strong\u003e\u003cem\u003eClassification Performance of top pathology-specific vision(Path-VM) models vs other models. \u003c/em\u003e\u0026nbsp;(a) Performance comparison between Virchow2 vs rest of other models. (b) Performance comparison between UNI vs rest of other models. (c) Performance comparison between Prov-GigaPath vs rest of other models. (d) Performance comparison between H-optimus-0 vs rest of other models. (e) Performance comparison between UNI2 vs rest of other models. Significance levels: * p \u0026lt; 0.05, ** p \u0026lt; 0.01, *** p \u0026lt; 0.001 (t-tests with Bonferroni correction)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSupplementary Figure 5. \u003c/strong\u003e\u003cem\u003eClassification performance of pathology-specific vision models (\u003c/em\u003ePath-VM\u003cem\u003e)and pathology-specific vision-language models (Path-VLM) in TCGA tasks. \u003c/em\u003e(a) Performance of all Path-VM and Path-VLM in each TCGA task, ranked by average performance, with the top three models highlighted in pink for emphasis.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSupplementary Figure 6. \u003c/strong\u003e\u003cem\u003eClassification performance of pathology-specific vision models (\u003c/em\u003ePath-VM\u003cem\u003e)and pathology-specific vision-language models (Path-VLM) in CPTAC tasks. \u003c/em\u003e(a) Performance of all Path-VM and Path-VLM in each CPTAC task, ranked by average performance, with the top three models highlighted in pink for emphasis.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSupplementary Figure 7. \u003c/strong\u003e\u003cem\u003eClassification performance of pathology-specific vision models (\u003c/em\u003ePath-VM\u003cem\u003e)and pathology-specific vision-language models (Path-VLM) in external tasks. \u003c/em\u003e(a) Performance of all Path-VM and Path-VLM in each external task, ranked by average performance, with the top three models highlighted in pink for emphasis.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSupplementary Figure 8. \u003c/strong\u003e\u003cem\u003eClassification performance of pathology-specific vision models (\u003c/em\u003ePath-VM\u003cem\u003e) and general vision models (VM) in TCGA tasks. \u003c/em\u003e(a) Top 10 performing pathology and vision models in each TCGA task, ranked by average performance, with the top three models visually emphasized in pink.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSupplementary Figure 9. \u003c/strong\u003e\u003cem\u003eClassification performance of pathology-specific vision and general vision models in CPTAC and external tasks. \u003c/em\u003e(a) Top 10 performing Path-VM and VM in each CPTAC task, ranked by average performance, with the top three models visually emphasized in pink. (b) Top 10 performing Path-VM and VM in each external task, ranked by average performance, with the top three models visually emphasized in pink.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSupplementary Figure 10. \u003c/strong\u003e\u003cem\u003eClassification performance in TCGA.\u003c/em\u003e\u003cstrong\u003e \u003c/strong\u003e(a) Performance of all Path-VM and Path-VLM in each TCGA task, based on average performance (b) Performance of all Path-VM and VM in each TCGA task, based on average performance.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSupplementary Figure 11. \u003c/strong\u003e\u003cem\u003eComparison between model categories.\u003c/em\u003e Performance comparison of general vision, pathology-specific vision models, general vision language models and pathology-specific vision-language models in (a) TCGA (b) CPTAC (c) External benchmarking tasks. Significance levels: * p \u0026lt; 0.05, ** p \u0026lt; 0.01, *** p \u0026lt; 0.001 (pairwise t-tests with Bonferroni correction) (d) Heatmap of performance of all models in all tasks including out of domain tasks.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSupplementary Figure 12. \u003c/strong\u003e\u003cem\u003eComparison between Model Architectures and performance of pathology models in external tasks. \u003c/em\u003ePerformance comparison of vision model architectures ViT-S, Vit-B , ViT-L in (a) TCGA (b) CPTAC (c) External benchmarking tasks (d) Cumulative average performance of Path-VM across out of domain tasks, sorted in descending order. (e) Cumulative average performance of Path-VM across external benchmarking tasks and out of domain tasks, sorted in descending order.\u003c/p\u003e","description":"","filename":"SupplementaryFigures.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6823810/v1/47a0f29d447a444ce8d25051.pdf"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Evaluating Vision and Pathology Foundation Models for Computational Pathology: A Comprehensive Benchmark Study","fulltext":[{"header":"Introduction","content":"\u003cp\u003ePathology plays a crucial role in cancer research, providing insights that drive the understanding, diagnosis, and treatment of cancer\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. It involves thorough examination of tissue specimens for tumor diagnosis and identification of important prognostic features such as tumor stage and grade, as well as biomarkers suggestive of treatment response\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e. Pathologists study digital pathology slides to analyze tumor histologic features, including architectural patterns and cellular morphology, in order to detect and classify tumors\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e. Consequently, oncologic pathology plays an integral role in personalized medicine and treatment selection. Computational pathology (CPath) has enhanced the field of pathology by integrating automated workflows and processes and advanced computational techniques\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. It has the potential to reduce the workload of pathologists considerably and can aid in speeding up the overall diagnostic process. With the advent of machine learning and artificial intelligence, CPath has proven potential to facilitate detection, classification and predictive modeling of cancer\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. Further, CPath proved effective in assisting pathologists in tumor grading, staging, and subtyping\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e, while supporting diagnostic decision-making. Also, it has been employed as a computational technique to study patterns, segment tumors\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e, and extract intricate features, which enhances cancer detection\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThe use of Artificial Intelligence (AI) has been at the forefront of medical research for the past decade\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. AI algorithms based on convolutional neural networks and vision transformers have significantly advanced computational pathology image analysis, especially in automated tumor detection and classification \u003csup\u003e\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e. AI-driven predictive models can analyze pathology images along with clinical and genomic data to predict patient outcomes, including disease progression, recurrence risk, and treatment response\u003csup\u003e\u003cspan additionalcitationids=\"CR11\" citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. Integrating gene expression, spatial omics, single-cell omics, and digital pathology data with AI enables a comprehensive multi-modal approach to cancer diagnosis and biomarker discovery, leveraging deep learning to uncover intricate spatial and molecular relationships within the tumor microenvironment\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e,\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e. AI models can automatically segment and delineate tumor regions, extract complex features and perform risk based stratification\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e. Additionally, AI has accelerated Computational Pathology (CPath) by automating image analysis, streamlining workflows, integrating multimodal data\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e,\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e, and enhancing diagnostic accuracy\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eDespite these advancements, several challenges persist, particularly regarding data limitations and model generalizability. Pathology datasets suffer from shortage of labeled data as manual pathologist annotations are time-consuming and also cost ineffective\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e. In contrast, large-scale datasets like ImageNet\u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e benefit from extensive labeled data, enabling supervised learning approaches to achieve high accuracy across diverse natural image tasks. To address this, Self-Supervised Learning (SSL) has emerged as a machine learning paradigm where models are trained on large-scale datasets to learn meaningful representations without the need for labeled examples \u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. This approach reduces the burden associated with data labeling and speeds up development of AI models. SSL based foundation models can be pre-trained on diverse and large unlabeled datasets,\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e,\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e to learn generalized patterns, and then fine-tuned on specific pathology datasets of interest \u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e. SSL enables these foundation models to adapt to different types of pathologies, domains, and tasks, and improve their generalizability.\u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e,\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e Therefore there has been a large effort in developing general purpose pathology foundation models that are trained on millions of whole slide images\u003csup\u003e\u003cspan additionalcitationids=\"CR27\" citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e, and can be adapted to tumor classification, segmentation, grading tasks with smaller amounts of labeled data.\u003c/p\u003e \u003cp\u003eOne of the first prominent models to be released was CTransPath\u003csup\u003e\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e, which was pretrained on the TCGA\u003csup\u003e\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e\u003c/sup\u003e and PAIP datasets\u003csup\u003e\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e,\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e\u003c/sup\u003e and trained on approximately 15\u0026nbsp;million images. Lunit\u0026rsquo;s DINO-based ViT-Small model \u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e\u003c/sup\u003e also utilized TCGA, along with internal datasets comprising over 32.6\u0026nbsp;million patches and 36,000 slides. Phikon\u003csup\u003e\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e\u003c/sup\u003e, another SSL model based on the iBOT framework, was pretrained on 40\u0026nbsp;million images from TCGA, covering 16 cancer types. More recently, several models such as UNI\u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e, Virchow\u003csup\u003e\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e, and Hibou\u003csup\u003e\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e\u003c/sup\u003e have adopted the DINOv2 SSL method with ViT-Large architectures for their pretraining, using proprietary datasets. Next, pathology language models such as PLIP \u003csup\u003e\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u003c/sup\u003e, Quilt-Net \u003csup\u003e\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e\u003c/sup\u003e, and CONCH \u003csup\u003e\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e\u003c/sup\u003e, which are trained on both pathology slides and paired text, have demonstrated significant performance in vision-based tasks as well. Vision encoders like CLIP\u003csup\u003e\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e\u003c/sup\u003e, iBOT\u003csup\u003e\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e that are built into these models have shown competitive results in natural image classification and image retrieval tasks. Additionally, imagenet pre-trained models have been widely used as feature extractors for digital pathology.\u003csup\u003e\u003cspan additionalcitationids=\"CR39\" citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e\u003c/sup\u003e\u003c/p\u003e \u003cp\u003eDespite these advancements in SSL, current foundation models may still be biased toward datasets they are trained on, limiting their transferability across diverse institutions and staining protocols\u003csup\u003e\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e\u003c/sup\u003e. Additionally, biases can arise from imbalanced representation of tissue types, and the predominance of certain cancer subtypes in training data, all of which can impact the model\u0026rsquo;s generalization and clinical applicability\u003csup\u003e\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e,\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e\u003c/sup\u003e. Many pathology foundation models rely on self-supervised pre-training using datasets like TCGA\u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e,\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e,\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e\u003c/sup\u003e or large proprietary datasets from a single institution\u003csup\u003e\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e,\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u003c/sup\u003e, which can inadvertently result in domain-specific overfitting rather than true generalization across diverse clinical settings.\u003csup\u003e\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e\u003c/sup\u003e This is further compounded by the limited availability of large-scale, diverse pathology datasets that span multiple institutions, staining variations, and scanner types\u003csup\u003e\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e\u003c/sup\u003e. Unlike general computer vision SSL models that benefit from extensive public datasets such as ImageNet \u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e or LAION-5B \u003csup\u003e\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e\u003c/sup\u003e, pathology-specific SSL models often lack sufficient pre-training diversity, making them susceptible to dataset-specific artifacts.\u003c/p\u003e \u003cp\u003eA systematic comparison is needed to evaluate pathology foundation models against both domain-specific SSL models and general computer vision SSL models, assessing their robustness across independent external datasets. Additionally, benchmarking should consider model generalization across diverse tissue types and cancer subtypes, ensuring that performance gains are not restricted to specific datasets. In this study, we compare general vision models (VM), general vision-language models (VLM), pathology-specific vision models (Path-VM), and pathology-specific vision-language models (Path-VLM), to analyze their generalization capability and task-specific strengths.\u003c/p\u003e \u003cp\u003eAltogether, over 20 pathology or pathology-language models have been released. In addition, several studies\u003csup\u003e\u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e\u003c/sup\u003e have used off-the-shelf imagenet models such as ResNet, DINO and DINOv2\u003csup\u003e38,39,48,49\u003c/sup\u003e for medical imaging including digital pathology. However, a comprehensive comparison of these models across diverse datasets and tasks remains lacking. Here, we present such a benchmark of four different types of models: pathology image, vision (i.e. natural images), pathology-language and vision-language models to identify the best-performing foundation models on 41 downstream tasks, 53 datasets and over 17.5k samples. We conduct a comprehensive evaluation by testing them across identical downstream tasks within a standardized experimental framework, ensuring a fair comparison. Our analysis evaluates the consistency of top-performing models, identifies those excelling across diverse benchmarking tasks and tissue types, compares the efficacy of pathology image models against language and vision models, and examines the impact of model size and dataset size on performance. Finally, we explored the potential of fusion models to enhance generalization across external tasks and diverse tissues. This work aims to provide insights into the factors driving effective pathology foundation models, offering a foundation for optimizing their application in precision medicine.\u003c/p\u003e"},{"header":"Results","content":"\u003ch3\u003e\u003cstrong\u003eBenchmarking models on TCGA: Task-specific performance highlights variability in top model rankings\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eWe conducted an extensive benchmarking study of 31 models across diverse domains, including general vision, general vision-language, pathology-specific vision, and pathology specific vision-language, evaluating them over 41 distinct tasks. To ensure consistent and concise terminology throughout the study, we assigned model abbreviations based on their training domains and modalities (Table 1): general vision models (VM), general vision-language models (VLM), pathology-specific vision models (Path-VM), and pathology-specific vision-language models (Path-VLM). Among these, 19 tasks were curated from the TCGA dataset, which includes multiple tumor types such as brain, breast, bladder, and others. We evaluated model performance using the average of balanced accuracy, precision, recall, and F1 score, collectively defined as the average performance metric for a comprehensive assessment. Initially, we focused on 15 pathology-specific vision models (Path-VM, Table 2), tested across 19 tasks related to tumor subtyping, molecular subtyping, and grading, and calculated the mean of average performance across 19 tasks. \u003c/p\u003e\n\u003cp\u003eAmong these, the Virchow2\u003csup\u003e28\u003c/sup\u003e model consistently ranked first, achieving a mean average performance of 0.706 ± 0.10. The next best-performing models were H-optimus-0 (0.702 ± 0.11), Prov-GigaPath\u003csup\u003e50\u003c/sup\u003e (0.702 ± 0.10) and UNI (0.70 ± 0.10), comprising the top four rankings (Figure 1b). Among the Path-VM, Virchow2 and UNI2 ranked first in 5 out of 19 TCGA-based tasks (Figure 1c), while H-optimus-0 and Prov-GigaPath ranked first in 3/19 tasks. \u003c/p\u003e\n\u003cp\u003eWe further investigated whether specific models excel in tumor-type-specific classification and subtyping tasks, aiming to identify tailored strengths across diverse cancer contexts. H-optimus-0 demonstrated superior performance on brain and breast tumor classification tasks (Figure 1d, Supplementary Figure 1a,b). In contrast, Virchow2 outperformed others on bladder and prostate cancer subtyping tasks, showcasing its precision in delineating subtle histological subtypes critical for these cancers (Supplementary Figure 1b). Meanwhile, Prov-GigaPath achieved the highest rankings on lung cancer and pan-cancer assessments, excelling on both organ-specific and cross-cancer evaluations (Supplementary Figure 1a, b). \u003c/p\u003e\n\u003cp\u003eNotably, UNI (0.70 ± 0.10) outperformed UNI2 (0.69 ± 0.11) when evaluated by the mean average performance metric across all 19 tasks(Figure 1b), reflecting its consistent strength across diverse challenges. However, UNI2 secured the top rank on 5 individual tasks (Figure 1c), surpassing UNI, which achieved the top rank on only 1 task, highlighting a striking contrast in task-specific excellence versus overall consistency across the 19-task evaluation. A detailed ranking of the top 10 Path-VM across all TCGA tasks is provided in Supplementary Figure 2a, offering a comprehensive perspective on model performance. Next, when broadening the evaluation to include all models—spanning general vision (VM), general vision-language (VLM), pathology-specific vision models (Path-VM), and pathology-specific vision language (Path-VLM) architectures—the top 10 rankings remained dominated by Path-VM. The only exception was iBOT-L16\u003csup\u003e23\u003c/sup\u003e, a vision model, which emerged as the sole non-pathology model in this group (Figure 1b, bottom panel). Overall, our results show significant variability in top model rankings across tumor-type classification and subtyping challenges. We also developed a live benchmarking dashboard PathBench, accessible at \u003cem\u003ehttps://pathbench.stanford.edu/\u003c/em\u003e, to enable real-time visualization and comparison of model performance.\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003eVirchow2 is the most consistent pathology model in CPTAC and external benchmarking tasks\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eNext, we evaluated these foundation model categories (VM, Path-VM, VLM, Path-VLM) on CPTAC cancer cases spanning various tumor types, focusing on specific tasks such as tumor grading, tumor classification, and pathway activity prediction (e.g., p53, MTOR, SWI). Among the 31 models tested, Virchow2 demonstrated the highest overall performance (Figure 2a, Supplementary Figure 3c), followed closely by Prov-GigaPath and UNI, when averaged across seven CPTAC-based tasks. Virchow2 achieved the best performance in 2 out of 6 individual tasks (Supplementary Figure 3a), with an average score of 0.766 ± 0.103, while Prov-GigaPath and UNI followed with 0.763 ± 0.103 and 0.760 ± 0.103, respectively. Notably, Prov-GigaPath also excelled in 2 out of the 6 tasks, with an average performance of 0.763 ± 0.11. Both Virchow2 and Prov-GigaPath consistently ranked among the top three models across CPTAC tasks (Supplementary Figure 3a). H-optimus-0, the second-best model in TCGA-based tasks, ranked fourth in CPTAC tasks. In the tumor type classification task, Prov-GigaPath achieved the highest performance, followed by Virchow and Virchow2.\u003c/p\u003e\n\u003cp\u003eNext, we evaluated these model categories (VM, Path-VM, VLM, Path-VLM) across eight external benchmarking tasks spanning multiple cancer types, including breast, prostate, and colorectal cancer. Virchow2 outperformed all other models, achieving an average performance of 0.82 ± 0.17 (Figure 2b, Supplementary Figure 3d), followed by UNI2 (0.79 ± 0.18). Prov-GigaPath ranked third, with a mean performance of 0.787 ± 0.17 across these tasks. Notably, Virchow2 consistently ranked among the top three in 6 out of 8 external benchmarking tasks. In contrast, H-optimus-0, which ranked second in TCGA-based tasks based on mean performance, performed poorly in external benchmarking tasks, placing seventh (0.766 ± 0.18). Interestingly, UNI2 outperformed its counterpart UNI in external tasks (Supplementary Figure 3d), where UNI ranked fourth (0.784 ± 0.18). However, in CPTAC tasks, UNI (0.760 ± 0.10, ranked third) outperformed UNI2 (0.751 ± 0.10, ranked sixth). We assessed whether the top five models from TCGA-based tasks demonstrated statistically significant performance differences compared to the remaining models across the three dataset categories. Virchow2 was the only model to show statistically significant performance gains in both CPTAC and external benchmarking tasks (Supplementary Figure 4a). UNI also exhibited statistically significant differences in CPTAC tasks; however, it did not show significant improvement in external benchmarking tasks (Supplementary Figure 4b). Overall, Virchow2 demonstrated consistently strong performance across both CPTAC and external benchmarking tasks.\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003ePathology-specific vision models (Path-VM) achieve superior performance over pathology-specific vision language models (Path-VLM) and general vision models (VM)\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eNext, we systematically evaluated the performance of pathology-specific vision models (Path-VM), pathology-specific vision-language models (Path-VLM), and general vision models (VM) across classification tasks from TCGA, CPTAC, and eight external benchmarking datasets (Figure 3a). In comparative analysis of Path-VM and VM, no statistically significant differences were observed between Path-VM and VM across these datasets (Figure 3a). To further assess their practical effectiveness, we analyzed model rankings, where Path-VM’s superior visual feature processing translated into consistent dominance across tasks (Figure 3b,c). Path-VM consistently dominated the top 10 positions in combined comparisons of VM and Path-VM, underscoring their practical superiority (Figure 3b,c, Supplementary figure 8a, Supplementary figure 9a,b). In 19 TCGA tasks, Path-VM held nearly all top 10 positions, with only a single VM appearing in most cases (Supplementary Figure 8a, Supplementary figure 10b). Similarly, in CPTAC and external tasks (Figure 3b,c), Path-VM were prevalent among the top 10, with only a few VM achieving high rankings (Supplementary Figure 8a-b). For example, iBOT VM achieved the second-highest performance in the TCGA brain histology subtype classification task and TCGA gastrointestinal subtype classification (Supplementary Figure 8a). A few VM’s(iBOT and DINO based general vision models) also ranked among the top three in the colorectal cancer challenge (MHIST), and the UnitoPatho challenge, and the DINO-B16 VM ranked fourth in the CPTAC-chromatin modification pathway classification task associated with gastrointestinal and colorectal tumors (Supplementary Figures 9a-b). These isolated successes suggest that VM’s can learn meaningful features for specific tasks, such as colorectal and gastrointestinal tumors, but Path-VM’s consistent top rankings highlight their broader superiority. \u003c/p\u003e\n\u003cp\u003eIn a comparative analysis of Path-VM and Path-VLM models across pathology tasks, Path-VM consistently outperformed Path-VLM in TCGA, CPTAC, and external benchmarking datasets, with statistical significance (Figure 3a). In TCGA tasks, only a few Path-VLM were ranked in top 10 in the combined ranking of Path-VM and Path-VLM (Supplementary figure 5a, Supplementary figure 10a) such as BiomedCLIP and QuiltNet-B-16 in bladder histology subtyping and MGMT methylation prediction, TITAN in brain histology, brain subtype(transcriptomic) and gi subtyping tasks. Similarly, in CPTAC tasks, Path-VLM failed to consistently rank within the top 10. The only exception was TITAN, a Path-VLM, which achieved a 6th-place ranking in the tumor label classification task (Supplementary Figure 6a). In external benchmarking tasks, Path-VLM again showed limited success. The only exception,TITAN, did achieve top-10 rankings in five out of eight tasks and ranked among the top three in the UnitoPatho challenge (Supplementary Figure 7a).\u003c/p\u003e\n\u003cp\u003eA comparative analysis of vision-only models (VM and Path-VM), and vision-language models (VLM and Path-VLM), revealed that vision-only models, both VM and Path-VM significantly outperformed VLM and Path-VLM across TCGA, CPTAC, and external benchmarking tasks (Supplementary Figures 11a-c). While Path-VLM exhibited a statistically significant performance advantage over VLM due to pathology-specific textual data, both vision-language models lagged behind their vision-only counterparts. Even though VLMs are trained on large datasets, they might have underperformed due to diluted visual feature learning from multimodal objectives and insufficient pathology-specific pretraining. This highlights the pivotal role of high-resolution visual features in pathology image analysis, where models relying solely on image data consistently excel over those incorporating textual information.\u003c/p\u003e\n\u003cp\u003eOverall, these results highlight the practical advantages of pathology-specific vision models, which leverage robust visual representations of tissue morphology to achieve superior classification performance, as demonstrated by Path-VM’s statistically significant outperformance of Path-VLM. Path-VM’s dominance in top 10 rankings, even in the absence of statistical significance over VM, underscores the value of pathology-specific training on larger, more diverse datasets that better represent the target tasks. While VM demonstrate potential in specific tasks, their limited presence in top rankings compared to Path-VM highlights the advantage of specialized models. Path-VLM, trained on histopathology slides and textual data (For example, CONCH\u003csup\u003e36\u003c/sup\u003e trained on 1.17 million image-text pairs, and PLIP\u003csup\u003e34\u003c/sup\u003e on ~200k image-text pairs), generally underperformed, likely due to smaller, less diverse training datasets that may not fully leverage the complementary nature of image and text modalities for classification tasks. The underperformance of Path-VLM may result from limitations in their current architectures or training data, which may hinder their effectiveness in classification tasks, suggesting a need for improved multimodal integration to achieve performance comparable to Path-VM.\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003eModel size and data size do not guarantee better performance in pathology foundation models \u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eWe first investigated model size and specifically whether larger vision transformer (ViT) architecture (i.e. ViT-large, ViT-huge, and ViT-giant collectively referred to as ViT-L and model parameters \u0026gt; 300M) offer superior performance compared to small(model parameters = 21M) and base ViT (model parameters = 85M) among pathology models. In TCGA-based tasks, larger ViT architectures outperformed both ViT-S and ViT-B models (Figure 4a). However, for CPTAC and external benchmarking datasets, no statistically significant performance differences were observed among the different ViT sizes (Figure 4a). Similarly, when evaluating vision models, we found no statistically significant differences in performance among ViT-S, ViT-B, and ViT-L across TCGA, CPTAC, and external benchmarking tasks (Supplementary Figures 12a-c).\u003c/p\u003e\n\u003cp\u003eTo assess the impact of pre-training dataset size on model performance, we categorized pathology models based on the number of whole-slide images (WSIs) used for training: small (\u0026lt;0.1 million WSIs), medium (0.1–1 million WSIs), and large (\u0026gt;1 million WSIs). Our analysis revealed statistically significant differences in average performance between models trained on small versus medium datasets, as well as small versus large datasets, across TCGA, CPTAC, and external benchmarking tasks (Figure 4b). However, no statistically significant difference was observed between models trained on medium and large datasets (Figure 4b). For example, models such as UNI (trained on ~100k WSIs) and UNI2 (~300k WSIs) achieved top performance across multiple tasks, were close behind the performance of Virchow2, which was trained on 3 million WSIs, (Figure 1b,c, Figure 2a,b). Additionally, UNI, trained on ~100k WSIs, and H-optimus-0, trained on ~500k WSIs, outperformed both Hibou and Virchow, which were trained on ~1 million WSIs and 1.5 million WSI’s respectively, across all three dataset categories (Figure 1b,c and Figure 2a,b). These findings suggest that beyond a certain threshold, increasing training data size does not necessarily lead to improved classification performance and that diversity in the training set may play a bigger role in models’ performance.\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003eFusion model generalizes better across external tasks and tissues\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eNext, we evaluated model performance using a combination of proprietary and out-of-domain datasets (i.e. Stanford, NLST, and DHMC) as part of an out-of-domain assessment, alongside external tasks to assess generalizability. Since these datasets were not included in model training, they served as an external evaluation of model robustness. We ranked the models based on their cumulative average performance across TCGA, CPTAC, and a combined set of out-of-domain and external benchmarking tasks. While H-Optimus-0 (0.702 ± 0.11) and Prov-GigaPath (0.702 ± 0.10) ranked among the top-performing models in 19 TCGA tasks (Figure 5a), their average performance showed a moderate decline in out-of-domain and external benchmarking tasks, with scores of 0.662 ± 0.19 and 0.662 ± 0.21, respectively (Figure 5b), leading to lower rankings. Similarly, UNI2 (0.696 ± 0.11) and Virchow (0.683 ± 0.11) exhibited a slight decrease in average performance from TCGA tasks (Figure 5a) to 0.664 ± 0.21 and 0.662 ± 0.20 in external evaluations (Figure 5b). These shifts in rankings across datasets highlight variability in model generalizability. To explore whether combining top-performing foundation models could improve predictive performance, we applied a late fusion strategy that aggregates predictions using majority voting. We selected the top five models based on mean average performance across TCGA tasks (Figure 1b): Virchow2, H-Optimus-0, Prov-GigaPath, UNI, and UNI2. We then compared the fusion model’s performance against individual models across multiple evaluation categories, including TCGA, CPTAC, and external benchmarking datasets (which included out-of-domain tasks). In TCGA tasks, the best-performing individual model, Virchow2, surpassed the fusion model in cumulative average performance across 19 tasks (Figure 5a). However, in CPTAC and external evaluations, the fusion model outperformed all individual models (Figures 5b,c), suggesting stronger generalizability. \u003c/p\u003e\n\u003cp\u003eTo further analyze model specialization, we calculated z-scores for each model’s performance and identified the top 10 models per task. Grouping all 41 tasks by tissue type (Figure 5d), we found that the fusion model excelled in breast, colon, prostate, and pan-cancer tasks, while Virchow2 performed best in bladder and lung cancer tasks. A direct comparison of Virchow2 and the fusion model (Figure 5e) showed that in TCGA tasks, Virchow2 outperformed the fusion model in 11 of 19 tasks. However, in CPTAC and external evaluations, the fusion model ranked higher in 5 of 7 CPTAC tasks and 12 of 15 out-of-domain + external tasks, suggesting its ability to generalize across diverse datasets. The fusion model exhibited superior generalizability, outperforming individual models in CPTAC and external tasks, particularly in breast, colon, prostate, and pan-cancer classifications. It achieved the highest overall performance across datasets, emphasizing the strength of ensemble-based predictions.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study presents an in-depth benchmarking analysis of 31 models spanning multiple domains\u0026mdash;general vision models (VM), general vision-language models (VLM), pathology-specific vision models (Path-VM), and pathology-specific vision-language models (Path-VLM)\u0026mdash;evaluated across 41 distinct tasks, which include a comprehensive range of cancer classification and subtyping challenges derived from TCGA, CPTAC, and external benchmarking datasets. The goal of this study was to investigate the task-specific strengths of various models and assess their overall performance on common cancer-specific tasks. Our findings reveal notable variations in model performance, reflecting the inherent complexity of pathology tasks and emphasizing the necessity for tailored approaches depending on the type of cancer and the specific challenge. In our study, it is crucial to also acknowledge that TCGA (The Cancer Genome Atlas)\u003csup\u003e30\u003c/sup\u003e likely serves as a major source of training data for many of the models evaluated. Despite the use of self-supervised learning (SSL) techniques aimed at minimizing reliance on labeled data, some degree of overfitting on TCGA-specific characteristics may still occur. This is particularly relevant when evaluating models on TCGA-based tasks, as the overlap between training and evaluation datasets can lead to inflated performance results, making it essential to consider this factor when interpreting the findings.\u003c/p\u003e\n\u003cp\u003eVirchow2 and UNI emerged as the top-performing pathology-specific vision models (Path-VM) across TCGA, CPTAC, and external tasks, with Virchow2 demonstrating the highest consistency and excelling in distinguishing subtle histological subtypes critical for cancers such as bladder and prostate. Meanwhile, H-optimus-0 and Prov-GigaPath exhibited strong performance in specific tumor types\u0026mdash;H-optimus-0 excelled in brain and breast cancer classification, while Prov-GigaPath was highly effective in lung cancer and pan-cancer evaluations. A key observation from this study is the variability in model performance across out-of-domain and external tasks; while models such as H-optimus-0, Prov-GigaPath, and UNI2 performed well on TCGA tasks, their performance declined on external datasets, underscoring the challenge of ensuring generalizability beyond the training dataset. This task-specific specialization suggests that models optimized for particular cancers may exhibit strong performance in niche domains, even if their overall rankings do not fully capture these targeted strengths. Fusion models proved particularly valuable in addressing these generalizability challenges\u0026mdash;by aggregating predictions from top-performing individual models using majority voting, they demonstrated superior performance across external datasets, including CPTAC, even though they did not surpass the best individual model (Virchow2) on TCGA tasks. Their strong results in breast, colon, prostate, and pan-cancer classifications further highlight their ability to generalize across diverse tissue types and datasets, making them a promising avenue for enhancing model robustness and real-world applicability. Given that ensemble methods are often more resilient to unseen data\u003csup\u003e51\u0026ndash;53\u003c/sup\u003e, fusion models offer a compelling strategy for improving diagnostic AI in pathology.\u003c/p\u003e\n\u003cp\u003eThe comparatively lower performance of pathology-specific vision language models (Path-VLM) likely stems from their reliance on smaller datasets, as fewer sources contain both pathology images and paired textual annotations. Even though VLMs (general vision-language models) are trained on large datasets, such as those with hundreds of millions of image-text pairs, they underperform due to the mismatch between general-domain pretraining data and pathology-specific high-resolution images, noisy or irrelevant textual annotations, and limitations in handling the high-resolution visual details critical for pathology tasks. While these models integrate visual and linguistic information, their current architectures may not fully harness the complementary nature of these modalities, limiting their effectiveness in classification tasks. Advancements in multimodal learning, particularly with larger, well-curated datasets and improved fusion techniques, could enhance their capabilities and make them more competitive with image-specific models. Meanwhile, general vision-only models (VM) trained on large-scale datasets like ImageNet\u003csup\u003e20\u003c/sup\u003e exhibited competitive performance in certain tasks but were generally outperformed by pathology-specific vision models (Path-VM) across TCGA, CPTAC, and external benchmarks. This highlights the unique domain-specific features of pathology vision data that general-purpose vision models, trained on non-medical images, fail to fully capture. While VM excelled in specific cancer subtypes, such as gastrointestinal and colorectal tumors, Path-VM demonstrated a more consistent advantage across a broader range of tasks, underscoring the importance of in-domain pre-training for developing more specialized and effective computational pathology models.\u003c/p\u003e\n\u003cp\u003eThis study challenges the assumption that larger models and more training data always lead to better performance. While larger vision transformer architectures showed advantages in TCGA tasks, their gains did not extend to CPTAC and external datasets. Similarly, increasing the number of whole-slide images beyond a certain threshold did not consistently improve classification accuracy. Notably, smaller models performed competitively with or even outperformed some models trained on much larger datasets. These findings highlight that beyond a certain scale, model architecture, training strategies, and data quality are more critical than sheer size.\u003c/p\u003e\n\u003cp\u003eOverall, these findings provide valuable insights into the strengths of various model architectures and strategies for pathology-related tasks. The results highlight that, while generalization is a key challenge, task-specific model optimization, along with ensemble-based approaches, can significantly enhance model performance. Future work could focus on exploring the considerable potential of multimodal models that integrate both image and textual data, alongside strategies such as ensemble approaches, addressing data diversity\u003csup\u003e16,54\u003c/sup\u003e and optimization techniques for pathology vision models (Path-VM), to enhance model transferability across diverse datasets.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e\u003cstrong\u003eDatasets for downstream tasks\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eTCGA dataset\u003c/em\u003e : The Cancer Genome Atlas (TCGA) project aimed at cataloging and generating comprehensive molecular landscapes responsible for cancer. The Cancer Genome Atlas project\u003csup\u003e30\u003c/sup\u003e generated multiple omics modalities, including exome-seq, RNA-seq, and methylation data, along with digital pathology slides, covering over 32 tumor types and 10,000 patients. With respect to digital pathology, TCGA provided a vast collection of high-resolution whole-slide images of tumor sections stained with Hematoxylin and Eosin (H\u0026amp;E). This has facilitated the development of machine learning and deep learning algorithms for tumor subtype classification, prognostication, and treatment response prediction using pathology imaging data. We used this rich resource of data for evaluation of foundation models. We evaluated the models based on tumor-specific molecular subtyping, such as distinguishing between IDH mutant and wild-type brain tumors, as well as pan-cancer molecular subtyping, like differentiating between TP53 mutant and wild-type tumors. This approach provided an estimate of the models' performance in specific tumors and across a pan-cancer setting. Additionally, we evaluated these models on a diverse range of tasks, including pathway activity, immune group subtyping, mRNA cluster classification, and histological subtyping, across various cancer types. Overall, we curated around 19 downstream tasks from TCGA data, encompassing bladder, brain, prostate tumors, and more.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eCPTAC + External benchmarking datasets\u003c/em\u003e: The Clinical Proteomic Tumor Analysis Consortium \u003csup\u003e55\u003c/sup\u003e(CPTAC) has provided pan-cancer datasets that include omics and digital pathology data, similar to the TCGA project. We also used this resource to evaluate our models. Since most pathology models have been trained on the TCGA dataset, testing their performance on CPTAC pathology data will offer a true measure of their efficacy on an independent test dataset. From CPTAC, we curated 7 pan-cancer tasks\u003csup\u003e56\u003c/sup\u003e : grade, SWI pathway alteration, p53, chromatin modification pathway alteration, mTOR pathway alteration, MYC pathway alteration, tumor label. In addition to CPTAC, we curated 8 publicly available external benchmark datasets, including BRACS\u003csup\u003e57\u003c/sup\u003e, BACH\u003csup\u003e58\u003c/sup\u003e, UnitoPatho \u003csup\u003e59\u003c/sup\u003e, SICAPv2\u003csup\u003e60\u003c/sup\u003e, BreakHis\u003csup\u003e61\u003c/sup\u003e, LC2500\u003csup\u003e62\u003c/sup\u003e, MHIST\u003csup\u003e63\u003c/sup\u003e, and NCT-CRC-HE\u003csup\u003e64\u003c/sup\u003e. These datasets contain representative images from diverse tumor types and are completely independent of the training data, ensuring a robust and unbiased evaluation of the foundation models.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eOut of domain datasets:\u003c/em\u003e We used proprietary pathology datasets from Stanford, including lung and brain tumor samples. Given that some foundation models may have used CPTAC data during their pre-training, it was crucial to evaluate these models on our in-house datasets to test their generalizability and estimate their performance on real-world data. With access to detailed clinical information and molecular subtypes, we assessed these models on specific subtyping tasks, such as distinguishing between lung squamous cell carcinoma and adenocarcinoma, and differentiating between MGMT methylated and wild-type brain tumors. We also added NLST pathology\u003csup\u003e65\u003c/sup\u003e and DHMC lung cancer datasets\u003csup\u003e66\u003c/sup\u003e to this category as a part of our out-of-domain evaluation. This approach helped us determine the models' true effectiveness on practical clinical tasks.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMetadata collection for downstream tasks\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFor the TCGA tasks, we collected molecular subtype, gene mutation status, histologic subtype, and transcriptomic information from TCGA publications specific to the following cancer types: brain\u003csup\u003e67\u003c/sup\u003e, prostate\u003csup\u003e68\u003c/sup\u003e, lung\u003csup\u003e69,70\u003c/sup\u003e, breast\u003csup\u003e71\u003c/sup\u003e, and bladder\u003csup\u003e72\u003c/sup\u003e. For the pan-cancer tasks, we collected pathway information on the samples from the TCGA pan-cancer pathway study\u003csup\u003e73\u003c/sup\u003e and immune groups from the TCGA immune landscape study\u003csup\u003e74\u003c/sup\u003e. For the CPTAC tasks, we gathered molecular subtypes and activated pathways on cases from the CPTAC pan-cancer study\u003csup\u003e56\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eWhole Slide Image preprocessing\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWhole-slide images (WSIs), initially in the SVS format, needed to be preprocessed to a machine learning-ready format. Due to the massive size of WSIs, which can often exceed dimensions of 10,000 x 10,000 pixels, it was necessary to divide them into smaller, non-overlapping tiles of 256 x 256 pixels each, at an objective magnification of 20x (0.5 microns per pixel). To accomplish this, a tissue mask was first obtained using the Otsu thresholding method\u003csup\u003e75\u003c/sup\u003e, which effectively separates the foreground (tissue) from the white background.\u003c/p\u003e\n\u003cp\u003eA multiple instance learning (MIL) strategy was employed, where each WSI was represented by a selection of up to 4,000 tiles. This approach was adopted to address the challenge of variable tissue density and distribution within WSIs. Tiles with excessive background or low contrast were excluded to ensure the selection of informative and representative tiles for each WSI.The selected tiles were then saved in an HDF5 database, one database per slide, to facilitate fast input/output (I/O) operations during the training and inference stages. This database organization allows for efficient access to the tiles, reducing the computational overhead associated with loading and processing large WSIs.\u003c/p\u003e\n\u003cp\u003eDuring the training phase, tiles were grouped into bags of \u003cem\u003eG \u003c/em\u003e size. Then, in each bag, a feature vector was extracted for each tile using a given foundation model. A maximum of \u003cem\u003eG \u003c/em\u003etiles was considered in each bag, and the corresponding feature vectors were averaged. This averaged feature vector was then forwarded through a linear layer to obtain the final classification result for each WSI.This approach effectively leverages the MIL framework to model the variability within WSIs and enables the classification of slides based on the collective information extracted from their constituent tiles. The use of the HDF5 database and the efficient tile-based representation contribute to the scalability and computational efficiency of the overall pipeline.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eVision transformers\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eVision Transformers (ViTs) have significantly advanced SSL in the realm of computer vision. Originally designed for natural language processing, ViTs adapt the transformer architecture to visual tasks by dividing images into smaller patches treated as tokens. The self-attention mechanisms in ViTs enable the model to capture intricate details and long-range dependencies within images, providing a superior contextual understanding compared to traditional convolutional neural network-based methods, especially in complex domains like pathology \u003csup\u003e76\u003c/sup\u003e. \u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFoundation models\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn the past couple of years, several ViT based pathology models like Lunit-Dino\u003csup\u003e24\u003c/sup\u003e, Phikon(Owkin)\u003csup\u003e33\u003c/sup\u003e, UNI\u003csup\u003e25\u003c/sup\u003e, Virchow\u003csup\u003e26\u003c/sup\u003e have been trained on huge datasets on multiple tissue types and evaluated on a variety of downstream tasks. We evaluated in-domain foundation models (i.e., models trained on digital pathology data such as Lunit-Dino), as well as out-of-domain vision models, trained on natural images (ImageNet data) such as DINO from Meta AI. We further categorized these models as general vision only (natural image based), general vision-language (natural images with text), pathology vision models (digital pathology image-based) and pathology vision-language models (pathology images and medical text). We conducted a comparative analysis across the different model categories to assess their relative performance. The pathology domain-specific models have been primarily trained on publicly available datasets like TCGA, and/or on proprietary datasets (for example, in the case of UNI). These foundation models are built upon ViT-small, ViT-base, and ViT-large/ViT-huge architectures as their core backbones. In our work, we benchmark these models using identical tasks, datasets, and hyperparameters. \u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eComputational resources\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFor our evaluation experiments, we used the Polaris supercomputer at the Argonne National Laboratory and the Stanford University Sherlock cluster (Stanford Research Computing Center).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eModel Evaluation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe assessed the performance of these models by applying a linear classifier to frozen extracted features, using consistent evaluation settings across all models. We froze the pretrained model and extracted features for the downstream task of interest, before doing linear probing. For our evaluation protocol, we used the same hyperparameters across all models and tasks and did not perform any data augmentation. For linear evaluation, we used a learning rate of 0.001, batch size of 16 per gpu, and trained the linear layer for 30 epochs. All tasks were run on a single node equipped with four NVIDIA A100 GPUs, each with 40 GB of memory. The foundation models varied based on different architecture sizes like ViT-S, ViT-B, ViT-L as well as patch sizes. We divided datasets for TCGA, CPTAC and out-of-domain evaluation downstream tasks into training, validation and test sets with a 70/15/15 split. We also implemented 5-fold cross-validation, complemented by evaluation on a separate independent test set, to robustly assess model performance for these tasks. We report the performance of models on independent test sets in our study. For external public benchmarking tasks, the datasets are usually available as training and test set or validation set. We report the performance on validation or test sets as per their availability. We also evaluated these models on slide level tasks from TCGA, CPTAC and out of domain evaluation datasets as well as patch based classification tasks from external benchmarking tasks separately. Performance is evaluated using metrics including balanced accuracy, precision, recall, and F1 score. We computed the arithmetic mean of the four metrics to derive the ‘average performance’ providing a balanced summary of overall model effectiveness.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eCode availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll code for evaluation and benchmarking dashboard is available at https://github.com/gevaertlab/benchmarking-path-models\u003c/p\u003e\n\u003ch3\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/h3\u003e\n\u003cp\u003eResearch reported here was further supported by the National Cancer Institute (NCI) under awards: R01 CA260271. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. This research was further supported by the Argonne Leadership Computing Facility, a U.S. Department of Energy (DOE) Office of Science user facility at Argonne National Laboratory.\u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eVerghese, G. et al. Computational pathology in cancer diagnosis, prognosis, and prediction - present day and prospects. J. Pathol. 260, 551\u0026ndash;563 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eConnolly, J. L. et al. Role of the Surgical Pathologist in the Diagnosis and Management of the Cancer Patient. (BC Decker, 2003).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHanna, M. G. \u0026amp; Ardon, O. Digital pathology systems enabling quality patient care. Genes Chromosomes Cancer 62, 685\u0026ndash;697 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLouis, D. N. et al. Computational pathology: A path ahead. Arch. Pathol. Lab. Med. 140, 41\u0026ndash;50 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePizurica, M. et al. Whole slide imaging-based prediction of TP53 mutations identifies an aggressive disease phenotype in prostate cancer. Cancer Res. 83, 2970\u0026ndash;2984 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKurc, T. et al. Segmentation and classification in digital pathology for glioma research: Challenges and deep learning approaches. Front. Neurosci. 14, 27 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCheng, J., Huang, K. \u0026amp; Xu, J. Editorial: Computational pathology for precision diagnosis, treatment, and prognosis of cancer. Front. Med. 10, 1209666 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMcGenity, C. et al. Artificial intelligence in digital pathology: a systematic review and meta-analysis of diagnostic test accuracy. NPJ Digit. Med. 7, 114 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSpringenberg, M. et al. From modern CNNs to vision transformers: Assessing the performance, robustness, and classification strategies of deep learning models in histopathology. Med. Image Anal. 87, 102809 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePhillips, S. P., Spithoff, S. \u0026amp; Simpson, A. Artificial intelligence and predictive algorithms in medicine: Promise and problems. Can. Fam. Physician 68, 570\u0026ndash;572 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSteyaert, S. et al. Multimodal data fusion for cancer biomarker discovery with deep learning. Nat. Mach. Intell. 5, 351\u0026ndash;362 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBoehm, K. M. et al. Multimodal data integration using machine learning improves risk stratification of high-grade serous ovarian cancer. Nat. Cancer 3, 723\u0026ndash;733 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZheng, Y., Carrillo-Perez, F., Pizurica, M., Heiland, D. H. \u0026amp; Gevaert, O. Spatial cellular architecture predicts prognosis in glioblastoma. Nat. Commun. 14, 4122 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLevy-Jurgenson, A., Tekpli, X., Kristensen, V. N. \u0026amp; Yakhini, Z. Spatial transcriptomics inferred from pathology whole-slide images links tumor heterogeneity to survival in breast and lung cancer. Sci. Rep. 10, 18802 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKrauze, A. V., Zhuge, Y., Zhao, R., Tasci, E. \u0026amp; Camphausen, K. AI-driven image analysis in central nervous system tumors-traditional Machine Learning, Deep Learning and hybrid models. J. Biotechnol. Biomed. 5, 1\u0026ndash;19 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCarrillo-Perez, F. et al. Generation of synthetic whole-slide image tiles of tumours from RNA-sequencing data via cascaded diffusion models. Nat. Biomed. Eng. 9, 320\u0026ndash;332 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePizurica, M. et al. Digital profiling of gene expression from histology images with linearized attention. Nat. Commun. 15, 9886 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eThakur, N., Yoon, H. \u0026amp; Chong, Y. Current trends of artificial intelligence for colorectal cancer pathology image analysis: A systematic review. Cancers (Basel) 12, 1884 (2020).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMarini, N. et al. Unleashing the potential of digital pathology data by training computer-aided diagnosis models without human annotations. NPJ Digit. Med. 5, 102 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeng, J. et al. ImageNet: A large-scale hierarchical image database. in 2009 IEEE Conference on Computer Vision and Pattern Recognition (IEEE, 2009). doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1109/cvpr.2009.5206848\u003c/span\u003e\u003cspan address=\"10.1109/cvpr.2009.5206848\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRani, V., Nabi, S. T., Kumar, M., Mittal, A. \u0026amp; Kumar, K. Self-supervised learning: A succinct review. Arch. Comput. Methods Eng. 30, 2761\u0026ndash;2775 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCaron, M. et al. Emerging properties in self-supervised vision transformers. arXiv [cs.CV] (2021) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.48550/ARXIV.2104.14294\u003c/span\u003e\u003cspan address=\"10.48550/ARXIV.2104.14294\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhou, J. et al. iBOT: Image BERT Pre-Training with Online Tokenizer. arXiv [cs.CV] (2021) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.48550/ARXIV.2111.07832\u003c/span\u003e\u003cspan address=\"10.48550/ARXIV.2111.07832\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKang, M., Song, H., Park, S., Yoo, D. \u0026amp; Pereira, S. Benchmarking self-supervised learning on diverse pathology datasets. arXiv [cs.CV] (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen, R. J. et al. Towards a general-purpose foundation model for computational pathology. Nat. Med. 30, 850\u0026ndash;862 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVorontsov, E. et al. Virchow: A million-slide digital pathology foundation model. arXiv [eess.IV] (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNechaev, D., Pchelnikov, A. \u0026amp; Ivanova, E. Hibou: A family of foundational vision transformers for pathology. arXiv [eess.IV] (2024) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.48550/ARXIV.2406.05074\u003c/span\u003e\u003cspan address=\"10.48550/ARXIV.2406.05074\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZimmermann, E. et al. Virchow2: Scaling self-supervised mixed magnification models in pathology. arXiv [cs.CV] (2024) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.48550/ARXIV.2408.00738\u003c/span\u003e\u003cspan address=\"10.48550/ARXIV.2408.00738\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang, X. et al. Transformer-based unsupervised contrastive learning for histopathological image classification. Med. Image Anal. 81, 102559 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCancer Genome Atlas Research Network et al. The Cancer Genome Atlas Pan-Cancer analysis project. Nat. Genet. 45, 1113\u0026ndash;1120 (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, K. et al. PAIP 2020: Microsatellite instability prediction in colorectal cancer. Med. Image Anal. 89, 102886 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim, Y. J. et al. PAIP 2019: Liver cancer segmentation challenge. Med. Image Anal. 67, 101854 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFiliot, A. et al. Scaling self-Supervised Learning for histopathology with Masked Image Modeling. medRxiv (2023) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1101/2023.07.21.23292757\u003c/span\u003e\u003cspan address=\"10.1101/2023.07.21.23292757\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang, Z., Bianchi, F., Yuksekgonul, M., Montine, T. J. \u0026amp; Zou, J. A visual-language foundation model for pathology image analysis using medical Twitter. Nat. Med. 29, 2307\u0026ndash;2316 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eIkezogwo, W. O. et al. Quilt-1M: One million image-text pairs for histopathology. arXiv [cs.CV] (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLu, M. Y. et al. A visual-language foundation model for computational pathology. Nat. Med. 30, 863\u0026ndash;874 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRadford, A. et al. Learning transferable visual models from natural language supervision. arXiv [cs.CV] (2021) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.48550/ARXIV.2103.00020\u003c/span\u003e\u003cspan address=\"10.48550/ARXIV.2103.00020\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSikaroudi, M., Hosseini, M., Gonzalez, R., Rahnamayan, S. \u0026amp; Tizhoosh, H. R. Generalization of vision pre-trained models for histopathology. Sci. Rep. 13, 6065 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePapadopoulos, K. M. \u0026amp; Stathaki, T. Comparing ImageNet pre-training with digital pathology foundation models for whole slide image-based survival analysis. arXiv [eess.IV] (2024) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.48550/ARXIV.2405.17446\u003c/span\u003e\u003cspan address=\"10.48550/ARXIV.2405.17446\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCampanella, G. et al. Computational pathology at health system scale -- self-supervised foundation models from three billion images. arXiv [cs.CV] (2023) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.48550/ARXIV.2310.07033\u003c/span\u003e\u003cspan address=\"10.48550/ARXIV.2310.07033\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLin, W., Liu, S., Zhu, R. \u0026amp; Wang, L. Unveiling institution-specific bias in pathology foundation models: Detriments, causes, and potential solutions. arXiv [cs.CV] (2025) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.48550/ARXIV.2502.16889\u003c/span\u003e\u003cspan address=\"10.48550/ARXIV.2502.16889\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang, Y. et al. A survey of recent methods for addressing AI fairness and bias in biomedicine. J. Biomed. Inform. 154, 104646 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVaidya, A. et al. Demographic bias in misdiagnosis by computational pathology models. Nat. Med. 30, 1174\u0026ndash;1190 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang, Z. et al. A foundation model for generalizable cancer diagnosis and survival prediction from histopathological images. Nat. Commun. 16, 2366 (2025).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi, D. et al. A survey on computational pathology foundation models: Datasets, adaptation strategies, and evaluation tasks. arXiv [cs.CV] (2025) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.48550/ARXIV.2501.15724\u003c/span\u003e\u003cspan address=\"10.48550/ARXIV.2501.15724\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchuhmann, C. et al. LAION-5B: An open large-scale dataset for training next generation image-text models. arXiv [cs.CV] (2022) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.48550/ARXIV.2210.08402\u003c/span\u003e\u003cspan address=\"10.48550/ARXIV.2210.08402\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAi, K. et al. Towards large-scale training of pathology foundation models. arXiv [cs.CV] (2024) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.48550/ARXIV.2404.15217\u003c/span\u003e\u003cspan address=\"10.48550/ARXIV.2404.15217\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang, Y. et al. Comparative analysis of ImageNet pre-trained deep learning models and DINOv2 in medical imaging classification. arXiv [eess.IV] (2024) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.48550/ARXIV.2402.07595\u003c/span\u003e\u003cspan address=\"10.48550/ARXIV.2402.07595\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuix, J. P. et al. Are natural domain foundation models useful for medical image classification? arXiv [cs.CV] (2023) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.48550/ARXIV.2310.19522\u003c/span\u003e\u003cspan address=\"10.48550/ARXIV.2310.19522\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXu, H. et al. A whole-slide foundation model for digital pathology from real-world data. Nature 630, 181\u0026ndash;188 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGong, H., Wang, M., Zhang, H., Elahe, M. F. \u0026amp; Jin, M. An explainable AI approach for the rapid diagnosis of COVID-19 using ensemble learning algorithms. Front. Public Health 10, 874455 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang, J., Wang, G., Xiao, X., Bao, M. \u0026amp; Tian, G. Explainable ensemble learning method for OCT detection with transfer learning. PLoS One 19, e0296175 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu, Y. et al. Deep learning based digital pathology for predicting treatment response to first-line PD-1 blockade in advanced gastric cancer. J. Transl. Med. 22, 438 (2024).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCarrillo-Perez, F. et al. Synthetic whole-slide image tile generation with gene expression profile-infused deep generative models. Cell Rep. Methods 3, 100534 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi, Y. et al. Proteogenomic data and resources for pan-cancer analysis. Cancer Cell 41, 1397\u0026ndash;1406 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang, Y., Chen, F., Chandrashekar, D. S., Varambally, S. \u0026amp; Creighton, C. J. Proteogenomic characterization of 2002 human cancers reveals pan-cancer molecular subtypes and associated pathways. Nat. Commun. 13, 2669 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBrancati, N. et al. BRACS: A dataset for BReAst Carcinoma Subtyping in H\u0026amp;E histology images. Database 2022, (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAresta, G. et al. BACH: Grand challenge on breast cancer histology images. Med. Image Anal. 56, 122\u0026ndash;139 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBertero, L. et al. UNITOPATHO. IEEE DataPort \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.21227/9FSV-TM25\u003c/span\u003e\u003cspan address=\"10.21227/9FSV-TM25\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSilva-Rodr\u0026iacute;guez, J., Colomer, A., Sales, M. A., Molina, R. \u0026amp; Naranjo, V. Going deeper through the Gleason scoring scale: An automatic end-to-end system for histology prostate grading and cribriform pattern detection. arXiv [eess.IV] (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSpanhol, F. A., Oliveira, L. S., Petitjean, C. \u0026amp; Heutte, L. A dataset for breast cancer histopathological image classification. IEEE Trans. Biomed. Eng. 63, 1455\u0026ndash;1462 (2016).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBorkowski, A. A. et al. Lung and Colon Cancer Histopathological Image Dataset (LC25000). arXiv [eess.IV] (2019) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.48550/ARXIV.1912.12142\u003c/span\u003e\u003cspan address=\"10.48550/ARXIV.1912.12142\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWei, J. et al. A Petri Dish for Histopathology Image Analysis. arXiv [eess.IV] (2021) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.48550/ARXIV.2101.12355\u003c/span\u003e\u003cspan address=\"10.48550/ARXIV.2101.12355\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKather, J. N., Halama, N. \u0026amp; Marx, A. 100,000 histological images of human colorectal cancer and healthy tissue. Zenodo \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.5281/zenodo.1214456\u003c/span\u003e\u003cspan address=\"10.5281/zenodo.1214456\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNational Lung Screening Trial Research Team. Data from the National Lung Screening Trial (NLST). The Cancer Imaging Archive \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.7937/TCIA.HMQ8-J677\u003c/span\u003e\u003cspan address=\"10.7937/TCIA.HMQ8-J677\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWei, J. W. et al. Pathologist-level classification of histologic patterns on resected lung adenocarcinoma slides with deep neural networks. Sci. Rep. 9, 3358 (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBrennan, C. W. et al. The somatic genomic landscape of glioblastoma. Cell 155, 462\u0026ndash;477 (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCancer Genome Atlas Research Network. The molecular taxonomy of primary prostate cancer. Cell 163, 1011\u0026ndash;1025 (2015).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCancer Genome Atlas Research Network. Comprehensive genomic characterization of squamous cell lung cancers. Nature 489, 519\u0026ndash;525 (2012).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCancer Genome Atlas Research Network. Comprehensive molecular profiling of lung adenocarcinoma. Nature 511, 543\u0026ndash;550 (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBerger, A. C. et al. A comprehensive pan-cancer molecular study of gynecologic and breast cancers. Cancer Cell 33, 690\u0026ndash;705.e9 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCancer Genome Atlas Research Network. Comprehensive molecular characterization of urothelial bladder carcinoma. Nature 507, 315\u0026ndash;322 (2014).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSanchez-Vega, F. et al. Oncogenic signaling pathways in The Cancer Genome Atlas. Cell 173, 321\u0026ndash;337.e10 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eThorsson, V. et al. The immune landscape of cancer. Immunity 48, 812\u0026ndash;830.e14 (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOtsu, N. A threshold selection method from gray-level histograms. IEEE Trans. Syst. Man Cybern. 9, 62\u0026ndash;66 (1979).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHe, K. et al. Transformers in Medical Image Analysis: A Review. arXiv [cs.CV] (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYu, J. et al. CoCa: Contrastive Captioners are Image-Text Foundation Models. arXiv [cs.CV] (2022) doi:\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.48550/ARXIV.2205.01917\u003c/span\u003e\u003cspan address=\"10.48550/ARXIV.2205.01917\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"},{"header":"Tables","content":"\u003cp\u003e\u003cstrong\u003eTable 1. Abbreviations of Model Categories\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"720\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 240px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel Category\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 240px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel Category Abbreviation\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 240px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTraining data\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 240px;\"\u003e\n \u003cp\u003eGeneral vision models\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 240px;\"\u003e\n \u003cp\u003eVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 240px;\"\u003e\n \u003cp\u003eNatural images\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 240px;\"\u003e\n \u003cp\u003eGeneral vision language models\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 240px;\"\u003e\n \u003cp\u003eVLM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 240px;\"\u003e\n \u003cp\u003eNatural images and text\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 240px;\"\u003e\n \u003cp\u003ePathology-specific vision models\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 240px;\"\u003e\n \u003cp\u003ePath-VM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 240px;\"\u003e\n \u003cp\u003ePathology-specific images\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 240px;\"\u003e\n \u003cp\u003ePathology-specific vision language models\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 240px;\"\u003e\n \u003cp\u003ePath-VLM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 240px;\"\u003e\n \u003cp\u003ePathology-specific images and text\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eTable 2. Pathology-specific vision model (Path-VM) summary\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"712\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel Name\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSSL method\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel Architecture\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eParameters in million\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNo. of WSI\u0026rsquo;s in million\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNo. of patches in million\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNo. of cancer/tissue types\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eData Source\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003ecTranspath\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eContrastive learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eCNN + SwinT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e~27\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e.032\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e32 cancer types\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eTCGA, PAIP\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eKaiko\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eDINO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eViT-base\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e~85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e.029\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e32 cancer types\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eTCGA\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eHIPT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eDINO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eViT-small\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e~21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e104\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e33 cancer types\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eTCGA\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eBEPH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eBEiT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eViT-base\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e~85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e.012\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e32 cancer types\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eTCGA\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eVirchow\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eDINOv2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eVit-huge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e~632\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e1.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e2000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eProprietary\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eLunit\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eDINO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eViT-small\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e~21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e.036\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e32.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eTCGA, TULIP\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eUNI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eDINOv2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eViT-large\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e~307\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e0.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e20 tissue types\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eProprietary, GTEx\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eHibou\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eDINOv2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eViT-base\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e~85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e1200\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eProprietary\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003ePhikon\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eiBOT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eViT-base\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e~85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e0.006\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e40\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e16 cancer types\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eTCGA\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eGPFM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eDINOv2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eViT-large\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e~307\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e0.086\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e190\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e34 tissue types\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eTCGA, PAIP and 49 public datasets\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eH-optimus-O\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eDINOv2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eVit-giant\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e~1100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e0.5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eProprietary\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eUNI2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eDINOv2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eVit-huge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e~632\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e0.35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e200\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eProprietary\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003ePhikon-v2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eDINOv2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eViT-large\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e~307\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e0.058\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e460\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e30 cancer sites\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eTCGA, GTEx, CPTAC, TCIA datasets\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eVirchow2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eDINOv2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eVit-huge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e~632\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e3.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eNA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e17 tissue types\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eProprietary\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 94px;\"\u003e\n \u003cp\u003eProv-GigaPath\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eDINOv2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eVit-giant\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e~1100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e0.171\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e1300\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e31 tissue types\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eProprietary\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eTable 3. Pathology-specific vision-language model (Path-VLM) \u0026nbsp;summary\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"720\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel Name\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePretraining method\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel Architecture\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eParameters in million\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eData source\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003ePLIP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eCLIP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eViT-base\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e~85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e208k pathology image-text pairs\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eQuiltNet-B16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eCLIP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eViT-base\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e~85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e1M pathology image-text pairs\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eBiomedCLIP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eCLIP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eViT-base\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e~85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e14M pathology image-text pairs\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eMizero pubmedbert\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eContrastive learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eViT-small\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e~21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e33k pathology image-text pairs\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eMizero clinmedbert\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eContrastive learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eViT-small\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e~21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e33k pathology image-text pairs\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eCONCH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eCoCa\u003csup\u003e77\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eViT-base\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e~85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e1.17 pathology image-text pairs\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eTITAN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eiBOT + CoCa\u003csup\u003e77\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eViT-base\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e~85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e336k whole slide images \u0026amp; 423k image captions\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eTable 4. General Vision Model (VM) summary\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"720\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel Name\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSSL method\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel Architecture\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eParameters in million\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eData source\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eDino-S16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eDINO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eViT-small\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e~21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eImageNet (14M)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eDino-B16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eDINO\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eViT-base\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e~85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eImageNet (14M)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eDinov2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eDINOv2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eViT-large\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e~300\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eLVD-142M\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eiBOT-B16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eiBOT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eViT-base\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e~85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eImageNet (14M)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eiBOT-L16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eiBOT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eViT-large\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e~307\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eImageNet (14M)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eTable 5. General Vision-language model (VLM) summary\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"720\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel Name\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePretraining method\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel Architecture\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eParameters in million\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eData source\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eCLIP-B16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eCLIP (Contrastive learning)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eViT-base\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e~115\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e400 million (image, text) pairs\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eBLIP-B16-14M\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eBLIP\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eViT-base\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e~85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eImageNet (14M)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eAlign-base\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eEfficientNet + BERT (Contrastive learning)\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eViT-base\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e~746\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e700 million (image, text) pairs\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003ebeit3-L16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eiBOT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eViT-large\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003e~307\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 144px;\"\u003e\n \u003cp\u003eImageNet (14M)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eTable 6. Downstream tasks\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"623\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTask\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCancer\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBinary or Multi\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDataset\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSlide or patch classification\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e\u003cstrong\u003etissue\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eTCGA-Grade\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ePan-cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eTCGA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003epan-cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003ePAM50 status\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eBRCA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eTCGA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003ebreast\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eprad mrna cluster\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ePRAD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eTCGA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eprostate\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eIDH mutation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eGBM + LGG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ebinary\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eTCGA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003ebrain\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003ebrain hist subtype\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eGBM + LGG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eTCGA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003ebrain\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003ePI3K\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ePan-cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ebinary\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eTCGA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003epan-cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e7\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eER status\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eBRCA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ebinary\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eTCGA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003ebreast\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eMYC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ePan-cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ebinary\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eTCGA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003epan-cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eImmunegroup\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ePan-cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eTCGA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003epan-cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eMGMT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eGBM + LGG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ebinary\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eTCGA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003ebrain\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eblca hist subtype\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eBLCA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eTCGA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003ebrain\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003egi subtype\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eGI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eTCGA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eGI\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e13\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003ebrain transcriptome subtype\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eGBM + LGG\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eTCGA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003ebrain\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003elung transcriptome subtype\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eLUSC + LUAD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eTCGA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003elung\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e15\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eblca mrna cluster\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eBLCA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eTCGA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003ebladder\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003elusc vs luad\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eLUSC + LUAD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ebinary\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eTCGA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003elung\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eERG status\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ePRAD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ebinary\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eTCGA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eprostate\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e18\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003ekidney subtype\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eKIRC + KIRP + KICH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eTCGA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003ekidney\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003ePR status\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eBRCA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ebinary\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eTCGA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003ebreast\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eCPTAC-grade\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ePan-cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eCPTAC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003epan-cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e21\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eSWI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ePan-cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ebinary\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eCPTAC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003epan-cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eP53\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ePan-cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ebinary\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eCPTAC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003epan-cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e23\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eChrom\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ePan-cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ebinary\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eCPTAC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003epan-cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e24\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eMTOR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ePan-cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ebinary\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eCPTAC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003epan-cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eMYC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ePan-cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ebinary\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eCPTAC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003epan-cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e26\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003etumor label\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ePan-cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eCPTAC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003epan-cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e27\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eSICAPv2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ePRAD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eExternal benchmarking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003ePatch\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003eprostate\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e28\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eMHIST\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eCOAD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ebinary\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eExternal benchmarking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003ePatch\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003ecolon\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eUniToPatho\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eCOAD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eExternal benchmarking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003ePatch\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003ecolon\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e30\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eBACH\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eBRCA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eExternal benchmarking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003ePatch\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003ebreast\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e31\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eBRACS\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eBRCA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eExternal benchmarking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003ePatch\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003ebreast\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eLC25000\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ePan-cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eExternal benchmarking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003ePatch\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003epan-cancer\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e33\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eBreakHis\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eBRCA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eExternal benchmarking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003ePatch\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003ebreast\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e34\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eNCT-CRC-HE\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eCOAD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eExternal benchmarking\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003ePatch\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003ecolon\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003elung stage\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eLung\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eOut of Domain (NLST)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003elung\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e36\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003elung grade\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eLung\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eOut of Domain (NLST)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003elung\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003elung hist OOD1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eLung\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ebinary\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eOut of Domain (Stanford)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003elung\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e38\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003elung hist OOD2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eLung\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eOut of Domain (DHMC)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003elung\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e39\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eLung stage OOD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eLung\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003emulti-class\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eOut of Domain (Stanford)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003elung\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e40\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eBrain necrosis\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eBrain\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ebinary\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eOut of Domain (Stanford)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003ebrain\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003e41\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eMGMT OOD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003eBrain\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 84px;\"\u003e\n \u003cp\u003ebinary\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 99px;\"\u003e\n \u003cp\u003eOut of Domain (Stanford)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eSlide\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 89px;\"\u003e\n \u003cp\u003ebrain\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"nature-portfolio","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"","title":"Nature Portfolio","twitterHandle":"","acdcEnabled":false,"dfaEnabled":false,"editorialSystem":"ejp","reportingPortfolio":"","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-6823810/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6823810/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eTo advance precision medicine in pathology, robust AI-driven foundation models are increasingly needed to uncover complex patterns in large-scale pathology datasets, enabling more accurate disease detection, classification, and prognostic insights. However, despite substantial progress in deep learning and computer vision, the comparative performance and generalizability of these pathology foundation models across diverse histopathological datasets and tasks remain largely unexamined. In this study, we conduct a comprehensive benchmarking of 31 AI foundation models for computational pathology, including general vision models (VM), general vision-language models (VLM), pathology-specific vision models (Path-VM), and pathology-specific vision-language models (Path-VLM), evaluated over 41 tasks sourced from TCGA, CPTAC, external benchmarking datasets, and out-of-domain datasets. Our study demonstrates that Virchow2, a pathology foundation model, delivered the highest performance across TCGA, CPTAC, and external tasks, highlighting its effectiveness in diverse histopathological evaluations. We also show that Path-VM outperformed both Path-VLM and VM, securing top rankings across tasks despite lacking a statistically significant edge over vision models. Our findings reveal that model size and data size did not consistently correlate with improved performance in pathology foundation models, challenging assumptions about scaling in histopathological applications. Lastly, our study demonstrates that a fusion model, integrating top-performing foundation models, achieved superior generalization across external tasks and diverse tissues in histopathological analysis. These findings emphasize the need for further research to understand the underlying factors influencing model performance and to develop strategies that enhance the generalizability and robustness of pathology-specific vision foundation models across different tissue types and datasets.\u003c/p\u003e \u003cp\u003ePathBench : \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://pathbench.stanford.edu/\u003c/span\u003e\u003cspan address=\"https://pathbench.stanford.edu/\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e","manuscriptTitle":"Evaluating Vision and Pathology Foundation Models for Computational Pathology: A Comprehensive Benchmark Study","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-07-04 06:26:07","doi":"10.21203/rs.3.rs-6823810/v1","editorialEvents":[],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"nature-communications","isNatureJournal":true,"hasQc":false,"allowDirectSubmit":false,"externalIdentity":"NCOMMS","sideBox":"Learn more about [Nature Communications](http://www.nature.com/ncomms/)","snPcode":"","submissionUrl":"https://mts-ncomms.nature.com/","title":"Nature Communications","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"ejp","reportingPortfolio":"Nature Communications","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"12d09597-c1a0-4add-afdb-ac8cb99ccd30","owner":[],"postedDate":"July 4th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[{"id":50031014,"name":"Health sciences/Medical research"},{"id":50031015,"name":"Health sciences/Health care/Medical imaging"}],"tags":[],"updatedAt":"2025-10-07T20:05:08+00:00","versionOfRecord":[],"versionCreatedAt":"2025-07-04 06:26:07","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6823810","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6823810","identity":"rs-6823810","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}