Artificial Intelligence for treatment outcomes in Pancreatic Cancer: a Scoping Review

Artificial Intelligence for treatment outcomes in Pancreatic Cancer: a Scoping Review | 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 Research Article Artificial Intelligence for treatment outcomes in Pancreatic Cancer: a Scoping Review Tigran Gevorkyan, Yana Belenkaya, Valeriya Pavlova, Rukiyat Abdulaeva, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8940110/v1 This work is licensed under a CC BY 4.0 License Status: Under Revision Version 1 posted 15 You are reading this latest preprint version Abstract Background Pancreatic cancer is among the most aggressive malignancies, characterized by poor survival outcomes. The potential of artificial intelligence (AI) to enhance prognostic predictions is increasingly recognized, as traditional tools have demonstrated limited accuracy. Objective This review aimed to evaluate the existing literature on the application of AI for predicting treatment outcomes in pancreatic cancer. The primary objective was to assess various AI models, data types, and their advantages. Methods A systematic literature search was conducted, encompassing studies published during 2017–2025. The review focused on research utilizing AI methodologies for predicting pancreatic cancer progression. The analysis followed a three-stage process: initial search, title and abstract screening, and full-text review. Data synthesis included the evaluation of model performance, data types, and validation strategies. Results From an initial pool of 35,577 articles, 33 met the inclusion criteria. The random forest was the most frequently applied machine learning (ML) algorithm (14/33, 42.4%). Three types of data were used: clinical data from electronic health records in 9 (27.3%) studies, radiomics in 11 (33.3%) studies, and genomics in 13 (39.4%) studies. The number of patients varied between 45 and 4846 for clinical data based models, between 70 and 1711 for genomics and between 64 and 1516 for radiomics. The resulting AUC varied between 0.933 and 0.732 for clinical data based models, between 0.671 and 0.938 for radiomics, between 0.571 and 0.92 for radiomics. All studies were heterogenous in terms of design, data feature selection and endpoints. Only 14 studies (42.4%) reported external validation of prognostic models. Ten out of thirteen genomics studies used data from open-source databases. Only 3/11 radiomics studies used unsupervised ML methods. High risk of bias was detected in 7 (21.2%) of studies. Conclusions AI demonstrates substantial potential for improving the accuracy of recurrence prediction in pancreatic cancer. However, standardization and improved accessibility are critical for facilitating clinical implementation. Further research is required to refine AI models for routine clinical use. pancreatic cancer early recurrence artificial intelligence machine learning deep learning prognostic factors Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction The incidence and mortality rates of pancreatic cancer (PC) are steadily increasing worldwide, with a five-year survival rate of less than 10% [ 1 ]. One potential approach to improving survival rates is the early prediction of disease progression. Emerging evidence suggests that, despite the overall poor prognosis, early treatment of disease progression may positively impact overall survival (OS) [ 2 ]. Several nomograms have been developed, demonstrating C-indexes ranging from 0.656 to 0.734. While these models provide some predictive utility, their modest performance limits their application in routine clinical practice [ 3 – 5 ]. Achieving higher precision in identifying prognostic factors and understanding their interactions requires more detailed and nuanced investigation. Conventional clinical studies face limitations due to patient heterogeneity, the need for large sample sizes, and unaccounted factors influencing prognosis [ 6 ]. Emerging data sources, such as radiomics and genomic analyses, offer new opportunities for identifying prognostic factors. However, the complexity of these datasets necessitates advanced analytical methods. Given the rapid advancements in computational sciences and expanding technical capabilities, artificial intelligence (AI), represents a promising solution for addressing these challenges[ 7 , 8 ]. Recent studies applying AI to predict early recurrence of PC have demonstrated high accuracy, though it remains unclear whether these approaches significantly outperform existing nomograms [ 9 – 11 ]. Despite the growing body of research on AI applications in PC prognosis, several questions remain unresolved. It is not yet clear which data types or AI methodologies yield the most reliable outcomes. Furthermore, the magnitude of AI's advantage over traditional predictive tools remains uncertain, as does whether this advantage justifies the substantial intellectual and financial investments required for its implementation. In this review, we summarize the findings of various studies on AI applications for predicting recurrence in PC. Our aim was to investigate the optimal ML algorithms, sample sizes, and data sources used for prognostic modelling. Additionally, we highlight the most promising research directions and discuss the potential impact of these advancements on clinical practice. Materials and Methods We conducted a literature search covering articles published during 2017–2025, in six online databases: PubMed, ScienceDirect, NATURE, MedRXiv, BioRXiv, and Google Scholar. The primary search query was ("Artificial intelligence" OR "Machine learning" OR "Deep learning" OR "supervised learning" OR "unsupervised learning" OR "reinforcement learning") AND ("Pancreatic Cancer" OR "Pancreatic adenocarcinoma") AND (diagnos* OR detect* OR predict* OR screen*). We used this query for the PubMed and Google Scholar databases, but it was not applicable to other databases due to exceeding number of letters. The query «(Artificial intelligence OR Machine learning OR Deep learning) AND (Pancreatic Cancer) AND (predict)» was used for ScienceDirect, NATURE, MedRXiv and BioRXiv searches. This review included only studies focusing on AI methods used for predicting treatment outcomes of PC. We included only journal articles and conference proceedings, excluding case reports, reviews, proposals, conference abstracts, editorial articles, studies that employed non-AI methods, research providing a theoretical basis for AI models applied to PC. Only studies published in English were considered for this review. We didn’t impose any restrictions regarding study conditions and design, outcomes, month, or country of publication. The selection process for studies followed three stages. In the first step, we performed the literature search in the aforementioned databases and used Rayyan to remove duplicates. In the second step, two reviewers independently screened titles and abstracts, excluding studies that were not relevant to the review topic. Finally, the reviewers independently examined the full text of the articles that passed the previous stage. Any discrepancies between the two reviewers were resolved through discussion. To measure the agreement between reviewers, we calculated the Cohen kappa, which was 0.93 for the screening of the full text. Considering the heterogeneity of the source data and AI models, we’ve distinguished clinical data, radiomics, and genomics-based models. Data Synthesis Following the extraction of data from the included studies, a narrative synthesis approach was employed. The synthesis summarized and described the AI techniques utilized in the studies, focusing on their objectives, characteristics, data sources, and AI model types (e.g., SVM, NN, RF). Furthermore, the programming languages used for implementing these AI techniques, the nature of the data (clinical, genomic, or radiomic), and the statistical metrics reported (accuracy, specificity, sensitivity, precision) were analyzed. Data management throughout the synthesis process was facilitated using Microsoft Excel. For calculations comparing the performance of different prognostic models, the best results in the test cohort were used. We analyzed the studies quality using a risk of bias evaluating tool PROBAST [ 12 ]. Results Search Results We initially identified 35 577 articles using 6 databases: PubMed (n = 883), Science Direct (n = 11162), NATURE (n = 644), Google Scholar (n = 18200), BioRXiv (n = 3905), and MedRxiv (n = 783). All articles from PubMed, NATURE and MedRXiv were analyzed based on the specified query. Due to the large volume of articles from Science Direct, BioRxiv, and Google Scholar, only the first 1000 results (sorted by relevance) were reviewed. In total, 5310 articles were included for detailed analysis (Fig. 1 ). A total of 35577 articles were identified. Of these, 5310 articles were analyzed, while 5275 were excluded for the following reasons: 1960 articles weren’t related to AI, 789 articles didn’t focus on PC, 847 were literature reviews, and 1681 weren’t relevant to recurrence risk (3 studies were excluded because they used unconventional data features (serial free-text CT reports, marital status and body composition) and would be incomparable to other studies [ 13 – 15 ]). Ultimately, 33 were included in this scoping review. Included articles characteristics All of the included studies were published in peer-reviewed journals. The included studies were published between 2017 and 2025, with 61.8% (21/34) of the studies published in 2023–2025. The number of participants ranged from 45 to 4846, with an average of 541.89 (SD 875.91) participants. Characteristics of the used AI Techniques Types of the used AI techniques are presented in the Table 1 . Table 1 Types of the used artificial intelligence techniques (n = 33 studies). Types Clinical data N = 9 (100%) Genomics N = 13 (100%) Radiomics N = 11 (100%) AI* type ● DL* 1 (11.1%) 0 (0.0%) 5 (45.5%) ● ML* 8 (88.9%) 13 (100.0%) 6 (54.5%) AI* algorithms/models/methods ● RF* 2 (22.2%) 7 (53.8%) 5 (45.5%) ● LR* 2 (22.2%) 1 (7.7%) 1 (9.1%) ● XG-Boost 2 (22.2%) 0 (0.0%) 0 (0.0%) ● Cox-Boost 0 (0.0%) 1 (7.7%) 0 (0.0%) ● plsRcox 0 (0.0%) 0 (0.0%) 0 (0.0%) ● SuperPC 0 (0.0%) 0 (0.0%) 0 (0.0%) ● Decision Tree 0 (0.0%) 0 (0.0%) 0 (0.0%) ● NN* 2 (22.2%) 0 (0.0%) 0 (0.0%) ● AX-Unet 0 (0.0%) 0 (0.0%) 1 (9.1%) ● CPH* 1 (11.1%) 0 (0.0%) 0 (0.0%) ● SVM* 1 (11.1%) 2 (15.4%) 0 (0.0%) ● Gradient boosting 0 (0.0%) 0 (0.0%) 0 (0.0%) ● LASSO* 1 (11.1%) 2 (15.4%) 0 (0.0%) ● Scalable deep segmentation and prognostic models (via self-learning) 0 (0.0%) 0 (0.0%) 1 (9.1%) ● L2 regularized LR* 0 (0.0%) 0 (0.0%) 0 (0.0%) ● Ridge 0 (0.0%) 0 (0.0%) 0 (0.0%) ● 3D-CNN* 0 (0.0%) 0 (0.0%) 2 (18.2%) ● Enet 0 (0.0%) 1 (7.7%) 0 (0.0%) Validation method • Internal 6 (66.7%) 7 (53.8%) 6 (54.5%) ● External 3 (33.3%) 6 (46.2%) 5 (45.5%) Number of data features ● < 50 9 (100.0%) 12 (92.3%) 7 (63.6%) ● 50-1000 0 (0.0%) 1 (7.7%) 4 (36.4%) *AI – artificial intelligence, DL – deep learning, ML – machine learning, RF – random forest, LR – logistic regression, NN – neural network, CPH - Cox proportional hazards, SVM - support vector machine, LASSO – least absolute shrinkage and selection operator, 3D-CNN – 3-Dimensional Convolutional Neural Network, LOOCV – leave-one-out cross-validation Most of the studies (27/33, 81.8%) used ML algorithms: 8/9 (88.9%) - in clinical data studies, 13/13 (100.0%) – in genomics, 6/11 (54.5%) – in radiomics. The other studies used DL algorithms. The most commonly used AI algorithms were RF (2/9, 22.2%), NN (2/9, 22.2%), XG-Boost (2/9, 22.2%), and LR (2/9, 22.2%) – in clinical data studies, RF (7/13, 53.8%) and LASSO (2/13, 15.4%) – in genomics, and RF (5/11, 45.5%) – in radiomics. The number of data features was less than 50 in the most studies (9/9, 100.0%; 12/13, 92.3%; 7/11, 63.6%, in clinical data, genomics and radiomics studies, respectively). The characteristics of the AI techniques used in each study are presented in Tables 2 , 3 , and 4 . The studies reviewed utilized various types of data to predict PC recurrence. Radiological images, including CT, MRI, and PET scans, were used in 33.3% (11/33), and genomic or molecular data were incorporated in 39.4% (13/33). Clinical data were least frequently employed, appearing in 27.3% (9/33) of the analyses. Table 2 Detailed characteristics of the included studies with AI models based on clinical data features Author, year Patient population AI* type Algorithm used Outcome measure Results External validation Number of data features Number of patients Lee K. et al., 2021 [ 10 ] Patients undergoing surgery with curative intent ML* CPH* DFS* C-Index 0.7738 No 6 4846 Elarre S.P. et al., 2019 [ 16 ] Patients undergoing induction CT and CRT before surgery with curative intent ML* LR* relapse at 2 years after surgery sensitivity 0.70, specificity 0.73, accuracy 0.71 (95%CI* 0.56–0.84, p = 0.005), AUC* 0.75 Yes 4 45 Nopour R. et al., 2024 [ 17 ] Any stage ML* XG-Boost OS* sensitivity 94.96%, specificity 93.62%, accuracy 94.55%, F-Score 96.0%, AUC* 0.933 (95%CI* 0.906–0.958) Yes 18 654 Kun Huang et al., 2025 [ 18 ] Stage IV ML* Cox regression + LASSO* + RF* CSS*, OS* CSS*: C-index 0.716, AUC 0.785; OS*: C-index 0.719, AUC 0.785 No 9 1662 Jin-Can Huang et al., 2025 [ 19 ] borderline resectable with upfront surgery ML* LR* OS*: Long-term vs Short-term Survival (≤ 2 years vs > 2 years) AUC 0.875 No 4 104 Walczak S. et al., 2017 [ 20 ] Any stage DL* NN* 7-months OS* Sensitivity 91%, specificity 38% No 13 219 Baig Z. et al., 2021 [ 21 ] Patients undergoing surgery with curative intent ML* Nonlinear SVM* 2-year OS* Accuracy 75%, sensitivity 41,9%, specificity 98% No 11 93 Wei Xiao et al., 2025 [ 22 ] Unresectable PC treated with CRT ML* XGBoost OS* C-index 0.949, AUC 0.732 No 21 214 Tong Z. et al., 2020 [ 23 ] Unresectable PC ML* 3 NN* models 8-months OS* 3 NN* models vs LR* AUC* (0.811 vs. 0.680, 0.844 vs. 0.722, 0.921 vs. 0.849, all p < 0.05), sensitivity 0.8241, specificity 0.8961. Yes 3/7/32 221 *AI – artificial intelligence, ML – machine learning, PC- pancreatic cancer, CPH – Cox proportional hazards, DFS – disease-free survival, LR – logistic regression, CSS - cancer-specific survival, CI – confidence interval, AUC – area under curve, SVM – support vector machine, DL – deep learning, OS – overall survival, NN – neural network, CRT- chemoradiotherapy, LASSO – least absolute shrinkage and selection operator Table 3 Detailed characteristics of the included studies with AI models based on genomics Author, year Patient population AI* type Algorithm used Outcome measure Results External validation Number of data features Data source Number of patients Yokoyama S. et al., 2020 [ 24 ] Any stage ML* SVM* 5-yearOS* p < 0.001, HR* 0.322 (CI* 0.17–0.61) No Genes – 3 Clinical- 6 Patients 191 Wang L. et al., 2022 [ 25 ] Any stage ML* CoxBoost and Survival-SVM* OS* AUCs* of 1-, 2-, 3-year OS* were 0.715, 0.748, 0.671; С-index 0.675 Yes Genes – 32 Clinical − 10 TCGA, open-source datasetss 1280 Zou Q.et al., 2025 [ 26 ] NS* ML* RF*, Cox regression OS* C-index 0.854 Yes Genes – 3 TCGA, GEO 219 Sun Y. et al., 2025 [ 27 ] Any stage ML* LASSO*, RF* 5-year OS* C-index 0.98, AUC 0.938, sensitivity 0.861, specificity 0.987 Yes Genes – 9 TCGA, GEO 279 Baek B. et al., 2020 [ 28 ] NS* ML* LR* OS*, DFS* DFS* AUC* 0.795 (LR), OS* AUC* 0.769 (LR), No Genes – 5, Clinical − 7 TCGA 134 Ojha A. et al., 2024 [ 29 ] NS* ML* RF* 3-year OS* accuracy 92.62% (males), 91.96% (females), 87.84% (overall) No Genes – 999 Clinical – 1 (sex) TCGA 178 Alsharoh H. et al., 2023 [ 30 ] NS* ML* RF* metastatic progression 76% accuracy No Genes − 38 TCGA 151 Chen D. et al., 2023 [ 31 ] NS* ML* Enet + SVM OS* C-index 0.678 Yes Genes − 38 TCGA, ICGC, ArrayExpress, CPTAC, GEO 1711 Zhang X-J. et al., 2025 [ 32 ] NS* ML* RF* OS* С-index 0.779 Yes Genes − 13 TCGA 226 Huang L. et al., 2023 [ 33 ] Stage I-II PC undergoing surgery with curative intent ML* SVM* OS* AUC* (1-year OS*) 0.9 No Genes – 15 Clinical − 15 Patients 70 Chang-L. et al., 2025 [ 34 ] Any stage ML* RF* OS* C-index: 0.777; AUC 0.974 No Genes − 9 TCGA 286 Ahmed Y. et al., 2023 [ 35 ] NS* ML* RF* OS* Accuracy 0.86 No Genes – 4 cBioportal 346 Xiaohong L. et al., 2025 [ 36 ] NS* ML* RSF + LASSO* OS* C-index 0.777; 1-year: 89.8%, 2-year: 93.2%, 3-year: 92.2% accuracy Yes Genes- 4 TCGA, ICGC, GTEx 690 *AI – artificial intelligence, ML – machine learning, SVM – support vector machine, NN – neural network, OS – overall survival, HR – hazard ratio, AUC – area under curve, RF – random forest, LR – logistic regression, DFS – disease-free survival, LASSO – least absolute shrinkage and selection operator, NS – not specified Table 4 Detailed characteristics of the included studies with AI models based on radiomics Author, year Patient population AI* type Algorithm used Outcome measure Results External validation Number of data features Imaging modality Clustering method Number of patients Lee W. et al., 2022 [ 37 ] Patients undergoing surgery with curative intent ML* (clinical) DL*(radiomics) Ensemble model: ANN*, LR*, RF*, GB*, SVM* (clinical) 3D ResNet-18, R (2 + 1) D-18, 3D ResNeXt-50, 3D DenseNet-121 (radiomics) 2-year OS* AUC* 0.76 (95% CI*, 0.62–0.89), sensitivity 69.0%, specificity 83.3%, F1-score 0.755 Yes Clinical and radiomics, NS* CT* supervised 282 Yao J. et al., 2023 [ 38 ] Patients undergoing surgery with curative intent DL* scalable deep segmentation and prognostic models (via self-learning) OS* HR* 2.03, 95%CI*: 1.50–2.75 (internal validation); HR* 2.47, 95%CI* 1.35–4.53 (external validation) Yes NS* CT* unsupervised 1516 Kaissis G. et al., 2019 [ 39 ] Any stage ML* RF* median OS* 87% sensitivity (95%CI* 67.3–92.7); 80% specificity (95%CI* 74.0–86.7); AUC* 90% No Radiomics − 504 MRI* supervised 102 Du X. et al., 2025 [ 40 ] Resectable PC* DL* CNN* recurrence within 9 months post-surgery AUC 0.895 No Radiomics − 112 CT* supervised 250 Keyl J. et al., 2022 [ 41 ] Unresectable PC* ML* RF* OS* c-index 0.71 (clinical data) 0.73 (clinical data+radiomics) 0.76 (clinical data+radiomics+KRAS status) Yes Clinical – 8 Radiomics − 3 CT* unsupervised 203 Toyama Y. et al., 2020 [ 42 ] Any stage ML* RF* 1-year OS* HR* 2.0; 95%CI*1.2–3.4; p = 0.0094 No Radiomics– 10 FDG-PET* semi-supervised 138 Palumbo D. et al., 2021 [ 9 ] Patients undergoing surgery with curative intent ML* LR* OS* AUC* 0.736 Sensitivity 58.6 Specifity 86.44 No Radiomics – 8 Clinical − 2 CT* supervised 147 Q. Gu et al., 2025 [ 43 ] Resectable PC* ML* RF* Recurrence-free survival AUCs 0.71, 0.83, and 0.79 at 0.5, 1, and 2 years Yes Radiomics – 416 CT* supervised 455 Schuurmans M. et al., 2025 [ 44 ] Any stage DL* CNN* OS* Results: AUC internal: 0.637; External: 0.571–0.675 Yes Clinical – 12, radiomics − 859 CT* supervised 762 Ni H. et al., 2023 [ 45 ] Patients undergoing surgery with curative intent DL* AX-Unet recurrence AUC* 0.92 (95%CI* 0.78–0.99); c-index 0.62 (95%CI* 0.48–0.78); accuracy 85.9% No Clinical − 10 Radiomics - NS* CT* supervised 64 Zhang Y. et al., 2021 [ 46 ] Resectable PC* DL* RF* OS* AUC* 0.84 No Radiomics– 1463 CT* Unsupervised 98 *AI – artificial intelligence, ML – machine learning, DL – deep learning, ANN – artificial neural network, LR – logistic regression, RF – random forest, CNN – convolutional neural network, GB – gradient boosting, SVM – support vector machine, OS – overall survival, AUC – area under curve, CI – confidence interval, HR – hazard ratio, MRI-DWI – diffusion-weighted magnetic resonance imaging, FDG-PET – fluorodeoxyglucose-positron emission tomography, CT – computed tomography, LASSO – least absolute shrinkage and selection operator, NS – not specified, PC – pancreatic cancer The number of data features extracted for the radiomics models varied significantly across studies (3–1463) and did not seem to correlate with final performance outcomes. Among the genomics studies, 10 (76.9%) used data from TCGA. Among the radiomics studies, 3 (27.3%) used unsupervised models, 7 (63.6%) – supervised models, and 1 (9.1%) study used semi-supervised ML method. CT was the most commonly used modality for radiomics (9/11, 81.8%). Performance measures in the included articles are showed in the Table 5 . Table 5 Performance measures (n = 33 studies) Performance measures Studies N = 33 (100%) AUC* 19 (57.6%) C-index 15 (45.5%) Accuracy 13 (39.4%) Sensitivity 7 (21.2%) Specificity 6 (18.2%) HR* 4 (12.1%) F1-score 3 (9.1%) NPV* 2 (6.1%) PPV* 2 (6.1%) * AUC – area under the curve, HR – hazard ratio, PPV – positive predictive value, NPV – negative predictive value The most commonly used performance measure was AUC (25/33, 75.8%). Table 6 and Figs. 2 – 5 present the results of the PROBAST quality assessment. Table 6 Suggested Tabular Presentation for PROBAST Results Author, year Risk of Bias Applicability Overall Participants Predictors Outcome Analysis Participants Predictors Outcome Risk of Bias Applicability Palumbo D. et al., 2021 [ 9 ] + + + ? + + + + + Lee K.S. et al., 2021 [ 10 ] + - + ? + + - + + Sala Elarre P. et al., 2019 [ 16 ] ? + + - ? + + - + Nopour R. et al., 2024 [ 17 ] + + - + + + - + + Huang K. et al., 2025 [ 18 ] + + ? ? + + + + + Huang J.-C. et al., 2025 [ 19 ] + + + ? + + + + + Walczak S. et al., 2017 [ 20 ] - - + + - - + - - Baig Z. et al., 2021 [ 21 ] ? + + ? + + + + + Xiao W. et al., 2025 [ 22 ] + + ? + + + + + + Tong Z. et al., 2020 [ 23 ] + + + ? + + + + + Yokoyama S. et al., 2020 [ 24 ] - + + ? + + + + + Wang L. et al., 2022 [ 25 ] - + ? - - + + - - Zou Q. et al., 2025 [ 26 ] - + ? + + + + - + Sun Y. et al., 2025 [ 27 ] - + + ? + + + + + Baek B. et al., 2020 [ 28 ] - + - ? - + + - + Ojha A. et al., 2024 [ 29 ] - - + + + ? + - - Alsharoh H. et al., 2023 [ 30 ] + + ? + + + + + + Chen D.et al., 2023 [31] ? + + + + + + + + Zhang X.-J. et al., 2025 [ 32 ] - ? - ? ? + + - ? Huang L. et al., 2023 [ 33 ] + + + ? + + - + + Chang-L. et al., 2025 [ 34 ] ? + + ? + + + ? + Ahmed Y. et al., 2023 [ 35 ] - + + + ? + + + - Xiaohong L. et al., 2025 [ 36 ] - ? ? ? ? + + ? ? Lee W. et al., 2022 [ 37 ] + + + ? + + + + + Yao J. et al., 2023 [ 38 ] + + - + + ? + + + Kaissis G. et al., 2019 [ 39 ] + + ? + + + + + + Du X. et al., 2025 [ 40 ] + + + ? + + + + + Keyl J. et al., 2022 [ 41 ] + - ? + + - + ? - Toyama Y. et al., 2020 [ 42 ] + + + ? + + + + + Gu Q. et al., 2025 [ 43 ] + + + + + ? + + + Schuurmans M. et al., 2025 [ 44 ] + + ? + + + + + + Ni H. et al., 2023 [ 45 ] + + ? ? + + + + + Zhang Y. et al., 2021 [ 46 ] - + ? ? ? + + ? + Overall the highest risk of bias was detected in genomics studies – 30.8% (4/13). High risk of bias was detected in 0/11 (0.0%) of radiomics studies and in 2/9 (22.2%) of studies using data from electronic health records. Discussion ML is an emerging tool for analyzing large volumes of clinical data to predict cancer progression. Our review revealed substantial heterogeneity among the included studies in terms of study endpoints, outcome measures, feature selection criteria, and choice of ML algorithms. Study sample size did not appear to influence the performance of prognostic models. Studies including only 45–70 patients achieved AUC values of 0.75–0.92 for OS prediction across all data types [ 45 ] [ 33 ] [ 16 ]. Larger sample sizes did not appear to correlate with better prognostic performance. The same was true for the number of data features. Four out of nine studies using clinical data from electronic health records relied on fewer than 10 data features after pre-selection from the original clinical datasets. Only one study reported that increasing the number of data features improved the performance of the prognostic model [ 23 ]. The optimal choice of clinical data features is an important question. Most of the studies used uni- or multivariate analysis to preselect features for AI-based prognostic models [ 10 , 20 , 21 ]. Tong Z. et al investigated an alternative approach, comparing different methods of feature selection. Interestingly, non-significant variables played important roles in prediction, the ANN-based prediction model with no restrictions on feature selection demonstrated good outcomes with 0.921 AUC [ 23 ]. Our review revealed that RF was the most commonly used ML algorithm and demonstrated high performance when using both clinical and genomics data. XG-Boost model showed some of the best results using clinical data from electronic health records. In the study by Nopour et al., 2024, XG-Boost achieved 94.96% accuracy, 93.62% specificity, and an AUC of 0.933 in OS prediction [ 17 ]. ANN was another successful model used for working with electronic health records. In the study by Tong et al., 2020, ANN achieved an AUC of 0.921, with 82.41% accuracy and 89.61% specificity in predicting OS in patients with inoperable cancer [ 23 ]. KNN algorithm demonstrated the poorest results in clinical data analysis, as its performance greatly depends on data volume and feature structure. For example, in a study by Sala Elarre et al., KNN showed accuracy only of 71% [ 16 ]. For the genomics studies, RF and SVM algorithms demonstrated superior performance. RF achieved an AUC of 0.974 in the study by Li C.-L. et al. and 0.938 in the study of Yuan Sun et al. [ 27 , 34 ]. Yokoyama et al., 2020, reported that SVM achieved an AUC of 0.795, with 76% accuracy and 98% specificity in predicting OS [ 24 ]. Hasan Alsharoh et al., 2023, confirmed SVM's reliability, reporting identical metrics—76% accuracy and 98% specificity—for recurrence prediction [ 30 ]. In the study by Huang et al., SVM also demonstrated high predictive power with an AUC of 0.900 in OS prediction [ 33 ]. However, linear versions of SVM and logistic regression are limited in adapting to complex patterns and are often applied to simple prediction models but are less effective for data with many nonlinear relationships. The integration of molecular data into clinical practice remains limited by unequal access to advanced technologies across healthcare institutions. The majority of genomics studies rely on open-source data, like TCGA, which may not be reproducible in real-world practice. Unlike studies based on electronic health records and radiomics, the majority of genomics studies (12/13 in this review) included unselected patient populations with either unspecified or any stage of pancreatic cancer. Information on treatment modalities in these studies was also limited, which makes them difficult to compare with the other studies in this review. In radiomics studies, increasing complexity of the models did not always lead to better prognostic performance. Schuurmans et al., reported that CNN-based models integrating both clinical (N = 12 features) and CT radiomic (N = 859 features) data achieved only moderate performance (internal AUC 0.637, external 0.571–0.675), despite the computational complexity and large feature space [ 44 ]. These contrasting outcomes underscore a critical finding: sophisticated deep learning architectures do not necessarily outperform traditional ML methods, particularly when applied to heterogeneous radiomics data. The success of simpler models with carefully selected data features suggests that feature quality and biological relevance may be more important than model complexity or data volume in achieving superior prognostic accuracy. This observation highlights the need for further research into optimal feature selection strategies and standardized imaging protocols to maximize the clinical utility of radiomics-based AI models. AI allowed the use of new data types for prognosis prediction, including radiomics. Despite growing interest in them, currently available models do not outperform the ones based on electronic health records. Approaches to handling radiomics data types are too heterogenous to draw any direct conclusions. Surprisingly, unsupervised AI radiomics models achieved some of the best results, which allows to save human resources and presents one of the most reproducible data types, free of potential data input and human error-associated biases [ 45 ]. Our study provides a comprehensive review and comparative analysis of a wide range of ML and DL algorithms and data combinations. By integrating results from various approaches, we present a holistic perspective on the application of AI in predicting PC outcomes. Reproducibility remains a major challenge in AI-based prognostic models. In our review, 14/33 (42.4%) of the studies had a high risk of bias, and 14 (42.4%) studies reported external validation. Among the genomics studies, 11 used data exclusively from open-source libraries. All included studies relied on retrospective data. Given the rapidly evolving treatment landscape for PC, these outcomes may not be reproducible in the future. The quality of data in electronic health records may vary substantially across centers, both in terms of reporting quality and the accuracy of assessing prognostic parameters (e.g. frequency of lymphovascular or perineural invasion detection). All these issues may limit the applicability of AI-based prognostic models in the near future. This analysis has certain limitations. The included studies are highly heterogeneous in terms of outcome measures, patient populations, and analytical methods. Comparison of outcomes across these studies, even indirectly, is therefore subject to considerable bias. Our aim was to outline the most promising approaches to AI-based prognostic modelling and to estimate the minimum prerequisites (such as sample size and number of data features) required for these models. Including additional databases in our search could potentially have increased the number of eligible studies. Nevertheless, we believe that a systematic search of PubMed, ScienceDirect, Nature, MedRxiv, BioRxiv, and Google Scholar provides a sufficient basis for this review and reflects the current status of AI-based prediction of long-term outcomes in pancreatic cancer. Overall, the application of AI for identifying predictors of survival and disease progression in PC represents a significant and promising area of research. Based on the findings, the use of clinical data appears to be both the easiest and the most reproducible method for predictive modeling, while using more complex data does not yield superior results. DL models leveraging boosting techniques demonstrate superior performance in such scenarios. To achieve high-quality outcomes, large volumes of structured clinical data are essential. Declarations Funding This research has been financially supported by The Analytical Center for the Government of the Russian Federation (Agreement No. 70-2025-000121 dd 29.8.2025. IGK 000000D730324P540002) Author Contribution T.G. Gevorkyan and S.S. Gordeyev created the study conception and design. Material preparation, data collection and analysis were performed by Y.V. Belenkaya, V.I. Pavlova, M.Sh. Manukyan, M.O. Mandrina, R.Sh. Abdulaeva and G.G. Makiev. The first draft of the manuscript was written by Y.V. Belenkaya, V.I. Pavlova, M.Sh. Manukyan, M.O. Mandrina, R.Sh. Abdulaeva and G.G. Makiev. The manuscript editing was performed by Y.V. Belenkaya and S.S.Gordeyev. All authors read and approved the final manuscript. Data Availability All data supporting the findings of this study are available within the paper and its Supplementary Information and References. References Ilic I, Ilic M. International patterns in incidence and mortality trends of pancreatic cancer in the last three decades: A joinpoint regression analysis. World J Gastroenterol. 2022;28(32):4698–715. Halle-Smith JM, et al. Clinical benefit of surveillance after resection of pancreatic ductal adenocarcinoma: A systematic review and meta-analysis. Eur J Surg Oncol. 2021;47(9):2248–55. Song W, Miao DL, Chen L. Nomogram for predicting survival in patients with pancreatic cancer. Onco Targets Ther. 2018;11:539–45. Goldstein D, et al. Nomogram for Estimating Overall Survival in Patients With Metastatic Pancreatic Cancer. Pancreas. 2020;49(6):744–50. Choi SH, Park SW, Seong J. A nomogram for predicting survival of patients with locally advanced pancreatic cancer treated with chemoradiotherapy. Radiother Oncol. 2018;129(2):340–6. Leonhardt CS, et al. Prognostic Factors for Early Recurrence After Resection of Pancreatic Cancer: A Systematic Review and Meta-Analysis. Gastroenterology. 2024;167(5):977–92. Tandon R, et al. A systematic review on deep learning-based automated cancer diagnosis models. J Cell Mol Med. 2024;28(6):e18144. Tran KA, et al. Deep learning in cancer diagnosis, prognosis and treatment selection. Genome Med. 2021;13(1):152. Palumbo D et al. Prediction of Early Distant Recurrence in Upfront Resectable Pancreatic Adenocarcinoma: A Multidisciplinary, Machine Learning-Based Approach. Cancers (Basel), 2021. 13(19). Lee KS, et al. Usefulness of artificial intelligence for predicting recurrence following surgery for pancreatic cancer: Retrospective cohort study. Int J Surg. 2021;93:106050. Lee JH, et al. Preoperative prediction of early recurrence in resectable pancreatic cancer integrating clinical, radiologic, and CT radiomics features. Cancer Imaging. 2024;24(1):6. Wolff RF, et al. PROBAST: a tool to assess the risk of bias and applicability of prediction model studies. Ann Intern Med. 2019;170(1):51–8. Chen Q, et al. Studying the impact of marital status on diagnosis and survival prediction in pancreatic ductal carcinoma using machine learning methods. Sci Rep. 2024;14(1):5273. Hsu TH, et al. Artificial intelligence to assess body composition on routine abdominal CT scans and predict mortality in pancreatic cancer- A recipe for your local application. Eur J Radiol. 2021;142:109834. Kim S, et al. Deep-Transfer-Learning-Based Natural Language Processing of Serial Free-Text Computed Tomography Reports for Predicting Survival of Patients With Pancreatic Cancer. JCO Clin Cancer Inf. 2024;8:e2400021. Sala Elarre P et al. Use of Machine-Learning Algorithms in Intensified Preoperative Therapy of Pancreatic Cancer to Predict Individual Risk of Relapse. Cancers (Basel), 2019. 11(5). Nopour R. Establishment of prediction model for mortality risk of pancreatic cancer: a retrospective study. BMC Med Inf Decis Mak. 2024;24(1):181. Huang K, et al. Development and validation of machine learning nomograms for predicting survival in stage IV pancreatic cancer: A retrospective study. World J Gastrointest Oncol. 2025;17(5):102459. Huang J-C, et al. A logistic regression model to predict long-term survival for borderline resectable pancreatic cancer patients with upfront surgery. Cancer Imaging. 2025;25(1):10. Walczak S, Velanovich V. An Evaluation of Artificial Neural Networks in Predicting Pancreatic Cancer Survival. J Gastrointest Surg. 2017;21(10):1606–12. Baig Z, et al. Prognosticating Outcome in Pancreatic Head Cancer With the use of a Machine Learning Algorithm. Technol Cancer Res Treat. 2021;20:15330338211050767. Xiao W, Yang B, Ke S. Application of machine learning for prognostic modeling in unresectable pancreatic cancer treated with chemoradiotherapy. Front Oncol. 2025;15:1644141. Tong Z, et al. Development, Validation and Comparison of Artificial Neural Network Models and Logistic Regression Models Predicting Survival of Unresectable Pancreatic Cancer. Front Bioeng Biotechnol. 2020;8:196. Yokoyama S, et al. Predicted Prognosis of Patients with Pancreatic Cancer by Machine Learning. Clin Cancer Res. 2020;26(10):2411–21. Wang L et al. Comprehensive machine-learning survival framework develops a consensus model in large-scale multicenter cohorts for pancreatic cancer. Elife, 2022. 11. Zou Q, et al. Machine learning-driven prognostic model based on sphingolipid-related gene signature in pancreatic cancer: development and validation. Translational Cancer Res. 2025;14(5):2779. Sun Y, et al. Prognostic model identification of ribosome biogenesis-related genes in pancreatic cancer based on multiple machine learning analyses. Discover Oncol. 2025;16(1):905. Baek B, Lee H. Prediction of survival and recurrence in patients with pancreatic cancer by integrating multi-omics data. Sci Rep. 2020;10(1):18951. Ojha A et al. Gap-App: A sex-distinct AI-based predictor for pancreatic ductal adenocarcinoma survival as a web application open to patients and physicians. bioRxiv, 2024. Alsharoh H. Machine learning predicts metastatic progression using novel differentially expressed lncRNAs as potential markers in pancreatic cancer. medRxiv, 2023: p. 2023.11.01.23297724. Chen D, et al. Comprehensive machine learning-generated classifier identifies pro-metastatic characteristics and predicts individual treatment in pancreatic cancer: A multicenter cohort study based on super-enhancer profiling. Theranostics. 2023;13(10):3290–309. Zhang X-J, et al. Improving molecular subtypes and prognosis of pancreatic cancer through multi group analysis and machine learning. Discover Oncol. 2025;16(1):96. Huang L et al. Gene signature developed for predicting early relapse and survival in early-stage pancreatic cancer. BJS Open, 2023. 7(3). Li C-L, et al. Machine learning model reveals the risk, prognosis, and drug response of histamine-related signatures in pancreatic cancer. Discover Oncol. 2025;16(1):155. Ahmed Y et al. A prognostic machine learning model for the prediction of pancreatic adenocarcinoma prognosis based on genomic expression of four cell-cycle associated hub genes. Annals Pancreat Cancer, 2023. 6. Liu X, et al. Machine learning based identification of an amino acid metabolism related signature for predicting prognosis and immune microenvironment in pancreatic cancer. BMC Cancer. 2025;25(1):6. Lee W, et al. Preoperative data-based deep learning model for predicting postoperative survival in pancreatic cancer patients. Int J Surg. 2022;105:106851. Yao J, et al. Deep Learning for Fully Automated Prediction of Overall Survival in Patients Undergoing Resection for Pancreatic Cancer: A Retrospective Multicenter Study. Ann Surg. 2023;278(1):e68–79. Kaissis G, et al. A machine learning model for the prediction of survival and tumor subtype in pancreatic ductal adenocarcinoma from preoperative diffusion-weighted imaging. Eur Radiol Exp. 2019;3(1):41. Du X et al. Novel CT radiomics models for the postoperative prediction of early recurrence of resectable pancreatic adenocarcinoma: A single-center retrospective study in China. J Appl Clin Med Phys, 2025: p. e70092. Keyl J, et al. Multimodal survival prediction in advanced pancreatic cancer using machine learning. ESMO Open. 2022;7(5):100555. Toyama Y, et al. Prognostic value of FDG-PET radiomics with machine learning in pancreatic cancer. Sci Rep. 2020;10(1):17024. Gu Q, et al. Interpretable Prognostic Modeling for Postoperative Pancreatic Cancer Using Multi-machine Learning and Habitat Radiomics: A Multi-center Study. Acad Radiol. 2025;32(9):5231–41. Schuurmans M et al. End-to-end prognostication in pancreatic cancer by multimodal deep learning: a retrospective, multicenter study. Eur Radiol, 2025: pp. 1–12. Ni H et al. Predicting Recurrence in Pancreatic Ductal Adenocarcinoma after Radical Surgery Using an AX-Unet Pancreas Segmentation Model and Dynamic Nomogram. Bioeng (Basel), 2023. 10(7). Zhang Y, et al. Improving prognostic performance in resectable pancreatic ductal adenocarcinoma using radiomics and deep learning features fusion in CT images. Sci Rep. 2021;11(1):1378. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Revision Version 1 posted Editorial decision: Revision requested 28 Apr, 2026 Reviews received at journal 17 Apr, 2026 Reviews received at journal 06 Apr, 2026 Reviews received at journal 31 Mar, 2026 Reviewers agreed at journal 23 Mar, 2026 Reviewers agreed at journal 23 Mar, 2026 Reviewers agreed at journal 22 Mar, 2026 Reviewers agreed at journal 20 Mar, 2026 Reviews received at journal 20 Mar, 2026 Reviewers agreed at journal 18 Mar, 2026 Reviewers agreed at journal 18 Mar, 2026 Reviewers invited by journal 18 Mar, 2026 Editor assigned by journal 14 Mar, 2026 Submission checks completed at journal 14 Mar, 2026 First submitted to journal 22 Feb, 2026 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-8940110","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":610153378,"identity":"a3dbb4eb-34fd-4630-93c3-7b7ab492742f","order_by":0,"name":"Tigran Gevorkyan","email":"","orcid":"","institution":"N.N.Blokhin Russian Cancer Research Center","correspondingAuthor":false,"prefix":"","firstName":"Tigran","middleName":"","lastName":"Gevorkyan","suffix":""},{"id":610153380,"identity":"9df827af-90c1-475a-a297-e72dd6039b29","order_by":1,"name":"Yana Belenkaya","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA60lEQVRIiWNgGAWjYLACxgYGGQYGNhDTJkGCWC08UC1ppGs5TFiLwfGzBz8w7rDh4ZduS/zMU3M+T7KB9xlebQZn8pIlGM+k8UjOOXZYmufY7WJpBnYz/FoO5BhI/207zGNwI71BmoftduI8BjY2/FrOvzH+wdj2n8f+Rnrzb55/54jQciPHTIKx7QCPgUTaMWnetgOJswlpkbzxxsyC8Uwyj8SNtDTLuX3JxZLNbMwW+LTwnc8xvsG4w06Of0aa8Y033+zyJI63Md7Ap0XhABKHiQdEMuNTDwTyDUgcxh8EVI+CUTAKRsHIBAAEw0WA7/PiWAAAAABJRU5ErkJggg==","orcid":"","institution":"N.N.Blokhin Russian Cancer Research Center","correspondingAuthor":true,"prefix":"","firstName":"Yana","middleName":"","lastName":"Belenkaya","suffix":""},{"id":610153383,"identity":"db9a0e7a-f30a-4305-a155-bc73d1a1967e","order_by":2,"name":"Valeriya Pavlova","email":"","orcid":"","institution":"Tyumen State Medical University","correspondingAuthor":false,"prefix":"","firstName":"Valeriya","middleName":"","lastName":"Pavlova","suffix":""},{"id":610153384,"identity":"765a5e90-5b74-46ef-9a26-0aa33990e3ea","order_by":3,"name":"Rukiyat Abdulaeva","email":"","orcid":"","institution":"N.N.Blokhin Russian Cancer Research Center","correspondingAuthor":false,"prefix":"","firstName":"Rukiyat","middleName":"","lastName":"Abdulaeva","suffix":""},{"id":610153385,"identity":"be787995-f7c5-42f1-8669-e5b355d7b1dc","order_by":4,"name":"Mariam Manukyan","email":"","orcid":"","institution":"N.N.Blokhin Russian Cancer Research Center","correspondingAuthor":false,"prefix":"","firstName":"Mariam","middleName":"","lastName":"Manukyan","suffix":""},{"id":610153386,"identity":"153becce-c47b-43ec-8437-49133eddb16a","order_by":5,"name":"Maryana Mandrina","email":"","orcid":"","institution":"N.N.Blokhin Russian Cancer Research Center","correspondingAuthor":false,"prefix":"","firstName":"Maryana","middleName":"","lastName":"Mandrina","suffix":""},{"id":610153388,"identity":"2f06d9e3-9909-4852-a9c8-13f703eef48c","order_by":6,"name":"Georgii Makiev","email":"","orcid":"","institution":"N.N.Blokhin Russian Cancer Research Center","correspondingAuthor":false,"prefix":"","firstName":"Georgii","middleName":"","lastName":"Makiev","suffix":""},{"id":610153394,"identity":"6807a455-df64-4b83-bfac-4237270f35da","order_by":7,"name":"Sergey Gordeyev","email":"","orcid":"","institution":"N.N.Blokhin Russian Cancer Research Center","correspondingAuthor":false,"prefix":"","firstName":"Sergey","middleName":"","lastName":"Gordeyev","suffix":""}],"badges":[],"createdAt":"2026-02-22 15:38:14","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8940110/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8940110/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":105282455,"identity":"7ed5e607-7c16-496b-9f29-3876d43f1a03","added_by":"auto","created_at":"2026-03-24 10:28:03","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":27065,"visible":true,"origin":"","legend":"\u003cp\u003ePRISMA-ScR (Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews) flowchart of the study selection.\u003c/p\u003e","description":"","filename":"Picture1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8940110/v1/d70c58f826cd09f1285e4765.jpg"},{"id":105282457,"identity":"3124a0bc-64b8-46b2-86ed-ccdba6640dde","added_by":"auto","created_at":"2026-03-24 10:28:04","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":27024,"visible":true,"origin":"","legend":"\u003cp\u003eOverall Risk of Bias Assessment\u003c/p\u003e","description":"","filename":"Picture2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8940110/v1/fbecd5a5b39feeaa44b76214.jpg"},{"id":105282486,"identity":"a262d5bd-082f-47be-98a4-71ba726d16ef","added_by":"auto","created_at":"2026-03-24 10:28:10","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":29254,"visible":true,"origin":"","legend":"\u003cp\u003eClinical data studies Risk of Bias Assessment\u003c/p\u003e","description":"","filename":"Picture3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8940110/v1/1d8eea17f9697e6f6aa474bd.jpg"},{"id":105282459,"identity":"e2e66b88-5515-4c1b-b45c-9c1bf96d4425","added_by":"auto","created_at":"2026-03-24 10:28:04","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":28817,"visible":true,"origin":"","legend":"\u003cp\u003eGenomics studies Risk of Bias Assessment\u003c/p\u003e","description":"","filename":"Picture4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8940110/v1/103db6ea258b0bb004d3f9ce.jpg"},{"id":105282466,"identity":"61ec818d-e17d-477d-afd5-9dafd4dd5ced","added_by":"auto","created_at":"2026-03-24 10:28:07","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":29448,"visible":true,"origin":"","legend":"\u003cp\u003eRadiomics studies Risk of Bias Assessment\u003c/p\u003e","description":"","filename":"Picture5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8940110/v1/bff1b8bcd45170b617aed1fa.jpg"},{"id":105564473,"identity":"36e65767-30e3-4ae3-9b04-3dd60843e40d","added_by":"auto","created_at":"2026-03-27 12:49:42","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1415152,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8940110/v1/f6ae908f-e434-4cb5-8851-f20fcb1dbe22.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Artificial Intelligence for treatment outcomes in Pancreatic Cancer: a Scoping Review","fulltext":[{"header":"Introduction","content":"\u003cp\u003eThe incidence and mortality rates of pancreatic cancer (PC) are steadily increasing worldwide, with a five-year survival rate of less than 10% [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eOne potential approach to improving survival rates is the early prediction of disease progression. Emerging evidence suggests that, despite the overall poor prognosis, early treatment of disease progression may positively impact overall survival (OS) [\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eSeveral nomograms have been developed, demonstrating C-indexes ranging from 0.656 to 0.734. While these models provide some predictive utility, their modest performance limits their application in routine clinical practice [\u003cspan additionalcitationids=\"CR4\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAchieving higher precision in identifying prognostic factors and understanding their interactions requires more detailed and nuanced investigation. Conventional clinical studies face limitations due to patient heterogeneity, the need for large sample sizes, and unaccounted factors influencing prognosis [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]. Emerging data sources, such as radiomics and genomic analyses, offer new opportunities for identifying prognostic factors. However, the complexity of these datasets necessitates advanced analytical methods. Given the rapid advancements in computational sciences and expanding technical capabilities, artificial intelligence (AI), represents a promising solution for addressing these challenges[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]. Recent studies applying AI to predict early recurrence of PC have demonstrated high accuracy, though it remains unclear whether these approaches significantly outperform existing nomograms [\u003cspan additionalcitationids=\"CR10\" citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eDespite the growing body of research on AI applications in PC prognosis, several questions remain unresolved. It is not yet clear which data types or AI methodologies yield the most reliable outcomes. Furthermore, the magnitude of AI's advantage over traditional predictive tools remains uncertain, as does whether this advantage justifies the substantial intellectual and financial investments required for its implementation.\u003c/p\u003e \u003cp\u003eIn this review, we summarize the findings of various studies on AI applications for predicting recurrence in PC. Our aim was to investigate the optimal ML algorithms, sample sizes, and data sources used for prognostic modelling. Additionally, we highlight the most promising research directions and discuss the potential impact of these advancements on clinical practice.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cp\u003eWe conducted a literature search covering articles published during 2017\u0026ndash;2025, in six online databases: PubMed, ScienceDirect, NATURE, MedRXiv, BioRXiv, and Google Scholar. The primary search query was (\"Artificial intelligence\" OR \"Machine learning\" OR \"Deep learning\" OR \"supervised learning\" OR \"unsupervised learning\" OR \"reinforcement learning\") AND (\"Pancreatic Cancer\" OR \"Pancreatic adenocarcinoma\") AND (diagnos* OR detect* OR predict* OR screen*). We used this query for the PubMed and Google Scholar databases, but it was not applicable to other databases due to exceeding number of letters. The query \u0026laquo;(Artificial intelligence OR Machine learning OR Deep learning) AND (Pancreatic Cancer) AND (predict)\u0026raquo; was used for ScienceDirect, NATURE, MedRXiv and BioRXiv searches. This review included only studies focusing on AI methods used for predicting treatment outcomes of PC.\u003c/p\u003e \u003cp\u003eWe included only journal articles and conference proceedings, excluding case reports, reviews, proposals, conference abstracts, editorial articles, studies that employed non-AI methods, research providing a theoretical basis for AI models applied to PC. Only studies published in English were considered for this review. We didn\u0026rsquo;t impose any restrictions regarding study conditions and design, outcomes, month, or country of publication.\u003c/p\u003e \u003cp\u003eThe selection process for studies followed three stages. In the first step, we performed the literature search in the aforementioned databases and used Rayyan to remove duplicates. In the second step, two reviewers independently screened titles and abstracts, excluding studies that were not relevant to the review topic. Finally, the reviewers independently examined the full text of the articles that passed the previous stage. Any discrepancies between the two reviewers were resolved through discussion. To measure the agreement between reviewers, we calculated the Cohen kappa, which was 0.93 for the screening of the full text.\u003c/p\u003e \u003cp\u003eConsidering the heterogeneity of the source data and AI models, we\u0026rsquo;ve distinguished clinical data, radiomics, and genomics-based models.\u003c/p\u003e \u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eData Synthesis\u003c/h2\u003e \u003cp\u003eFollowing the extraction of data from the included studies, a narrative synthesis approach was employed. The synthesis summarized and described the AI techniques utilized in the studies, focusing on their objectives, characteristics, data sources, and AI model types (e.g., SVM, NN, RF). Furthermore, the programming languages used for implementing these AI techniques, the nature of the data (clinical, genomic, or radiomic), and the statistical metrics reported (accuracy, specificity, sensitivity, precision) were analyzed. Data management throughout the synthesis process was facilitated using Microsoft Excel. For calculations comparing the performance of different prognostic models, the best results in the test cohort were used.\u003c/p\u003e \u003cp\u003eWe analyzed the studies quality using a risk of bias evaluating tool PROBAST [\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e].\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003eSearch Results\u003c/h2\u003e \u003cp\u003eWe initially identified 35 577 articles using 6 databases: PubMed (n\u0026thinsp;=\u0026thinsp;883), Science Direct (n\u0026thinsp;=\u0026thinsp;11162), NATURE (n\u0026thinsp;=\u0026thinsp;644), Google Scholar (n\u0026thinsp;=\u0026thinsp;18200), BioRXiv (n\u0026thinsp;=\u0026thinsp;3905), and MedRxiv (n\u0026thinsp;=\u0026thinsp;783). All articles from PubMed, NATURE and MedRXiv were analyzed based on the specified query. Due to the large volume of articles from Science Direct, BioRxiv, and Google Scholar, only the first 1000 results (sorted by relevance) were reviewed. In total, 5310 articles were included for detailed analysis (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eA total of 35577 articles were identified. Of these, 5310 articles were analyzed, while 5275 were excluded for the following reasons: 1960 articles weren\u0026rsquo;t related to AI, 789 articles didn\u0026rsquo;t focus on PC, 847 were literature reviews, and 1681 weren\u0026rsquo;t relevant to recurrence risk (3 studies were excluded because they used unconventional data features (serial free-text CT reports, marital status and body composition) and would be incomparable to other studies [\u003cspan additionalcitationids=\"CR14\" citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]). Ultimately, 33 were included in this scoping review.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eIncluded articles characteristics\u003c/h3\u003e\n\u003cp\u003eAll of the included studies were published in peer-reviewed journals. The included studies were published between 2017 and 2025, with 61.8% (21/34) of the studies published in 2023\u0026ndash;2025. The number of participants ranged from 45 to 4846, with an average of 541.89 (SD 875.91) participants.\u003c/p\u003e\n\u003ch3\u003eCharacteristics of the used AI Techniques\u003c/h3\u003e\n\u003cp\u003eTypes of the used AI techniques are presented in the Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eTypes of the used artificial intelligence techniques (n\u0026thinsp;=\u0026thinsp;33 studies).\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTypes\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eClinical data\u003c/p\u003e \u003cp\u003eN\u0026thinsp;=\u0026thinsp;9 (100%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eGenomics\u003c/p\u003e \u003cp\u003eN\u0026thinsp;=\u0026thinsp;13 (100%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRadiomics\u003c/p\u003e \u003cp\u003eN\u0026thinsp;=\u0026thinsp;11 (100%)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"4\" nameend=\"c4\" namest=\"c1\"\u003e \u003cp\u003eAI* type\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e● DL*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1 (11.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e5 (45.5%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e● ML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e8 (88.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e13 (100.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e6 (54.5%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"4\" nameend=\"c4\" namest=\"c1\"\u003e \u003cp\u003eAI* algorithms/models/methods\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e● RF*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2 (22.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7 (53.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e5 (45.5%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e● LR*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2 (22.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (7.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1 (9.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e● XG-Boost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2 (22.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e● Cox-Boost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (7.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e● plsRcox\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e● SuperPC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e● Decision Tree\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e● NN*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e2 (22.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e● AX-Unet\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1 (9.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e● CPH*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1 (11.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e● SVM*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1 (11.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2 (15.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e● Gradient boosting\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e● LASSO*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1 (11.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2 (15.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e● Scalable deep segmentation and prognostic models (via self-learning)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e1 (9.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e● L2 regularized LR*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e● Ridge\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e● 3D-CNN*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e2 (18.2%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e● Enet\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (7.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"4\" nameend=\"c4\" namest=\"c1\"\u003e \u003cp\u003eValidation method\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e\u0026bull; Internal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e6 (66.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e7 (53.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e6 (54.5%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e● External\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e3 (33.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e6 (46.2%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e5 (45.5%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"4\" nameend=\"c4\" namest=\"c1\"\u003e \u003cp\u003eNumber of data features\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e● \u0026lt;\u0026thinsp;50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e9 (100.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e12 (92.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e7 (63.6%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e● 50-1000\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (7.7%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e4 (36.4%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"4\"\u003e\u003cem\u003e*AI \u0026ndash; artificial intelligence, DL \u0026ndash; deep learning, ML \u0026ndash; machine learning, RF \u0026ndash; random forest, LR \u0026ndash; logistic regression, NN \u0026ndash; neural network, CPH - Cox proportional hazards, SVM - support vector machine, LASSO \u0026ndash; least absolute shrinkage and selection operator, 3D-CNN \u0026ndash; 3-Dimensional Convolutional Neural Network, LOOCV \u0026ndash; leave-one-out cross-validation\u003c/em\u003e\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eMost of the studies (27/33, 81.8%) used ML algorithms: 8/9 (88.9%) - in clinical data studies, 13/13 (100.0%) \u0026ndash; in genomics, 6/11 (54.5%) \u0026ndash; in radiomics. The other studies used DL algorithms. The most commonly used AI algorithms were RF (2/9, 22.2%), NN (2/9, 22.2%), XG-Boost (2/9, 22.2%), and LR (2/9, 22.2%) \u0026ndash; in clinical data studies, RF (7/13, 53.8%) and LASSO (2/13, 15.4%) \u0026ndash; in genomics, and RF (5/11, 45.5%) \u0026ndash; in radiomics. The number of data features was less than 50 in the most studies (9/9, 100.0%; 12/13, 92.3%; 7/11, 63.6%, in clinical data, genomics and radiomics studies, respectively). The characteristics of the AI techniques used in each study are presented in Tables\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e, \u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e, and \u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eThe studies reviewed utilized various types of data to predict PC recurrence. Radiological images, including CT, MRI, and PET scans, were used in 33.3% (11/33), and genomic or molecular data were incorporated in 39.4% (13/33). Clinical data were least frequently employed, appearing in 27.3% (9/33) of the analyses.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDetailed characteristics of the included studies with AI models based on clinical data features\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"9\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAuthor, year\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePatient population\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAI* type\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAlgorithm used\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOutcome measure\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eResults\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eExternal validation\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eNumber of data features\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eNumber of patients\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLee K. et al., 2021 [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePatients undergoing surgery with curative intent\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eCPH*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eDFS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eC-Index 0.7738\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e4846\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eElarre S.P. et al., 2019 [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePatients undergoing induction CT and CRT before surgery with curative intent\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLR*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003erelapse at 2 years after surgery\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003esensitivity 0.70, specificity 0.73, accuracy 0.71 (95%CI* 0.56\u0026ndash;0.84, p\u0026thinsp;=\u0026thinsp;0.005), AUC* 0.75\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e45\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNopour R. et al., 2024 [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAny stage\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eXG-Boost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003esensitivity 94.96%, specificity 93.62%, accuracy 94.55%, F-Score 96.0%, AUC* 0.933 (95%CI* 0.906\u0026ndash;0.958)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e18\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e654\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eKun Huang et al., 2025 [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage IV\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eCox regression\u0026thinsp;+\u0026thinsp;LASSO* + RF*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eCSS*, OS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eCSS*: C-index 0.716, AUC 0.785;\u003c/p\u003e \u003cp\u003eOS*: C-index 0.719, AUC 0.785\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e1662\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eJin-Can Huang et al.,\u0026nbsp;2025 [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eborderline resectable with upfront surgery\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLR*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOS*: Long-term vs Short-term Survival (\u0026le;\u0026thinsp;2 years vs\u0026thinsp;\u0026gt;\u0026thinsp;2 years)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAUC 0.875\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e104\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWalczak S. et al., 2017 [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAny stage\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDL*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNN*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e7-months OS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSensitivity 91%, specificity 38%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e219\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBaig Z. et al., 2021 [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePatients undergoing surgery with curative intent\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNonlinear SVM*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2-year OS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAccuracy 75%, sensitivity 41,9%, specificity 98%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e11\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e93\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWei Xiao et al., 2025 [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUnresectable PC treated with CRT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eXGBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eC-index 0.949,\u003c/p\u003e \u003cp\u003eAUC 0.732\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e21\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e214\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTong Z. et al., 2020 [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUnresectable PC\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e3 NN* models\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e8-months OS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e3 NN* models vs LR* AUC* (0.811 vs. 0.680, 0.844 vs. 0.722, 0.921 vs. 0.849, all p\u0026thinsp;\u0026lt;\u0026thinsp;0.05), sensitivity 0.8241, specificity 0.8961.\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e3/7/32\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c9\"\u003e \u003cp\u003e221\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"9\"\u003e\u003cem\u003e*AI \u0026ndash; artificial intelligence, ML \u0026ndash; machine learning, PC- pancreatic cancer, CPH \u0026ndash; Cox proportional hazards, DFS \u0026ndash; disease-free survival, LR \u0026ndash; logistic regression, CSS - cancer-specific survival, CI \u0026ndash; confidence interval, AUC \u0026ndash; area under curve, SVM \u0026ndash; support vector machine, DL \u0026ndash; deep learning, OS \u0026ndash; overall survival, NN \u0026ndash; neural network, CRT- chemoradiotherapy, LASSO \u0026ndash; least absolute shrinkage and selection operator\u003c/em\u003e\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDetailed characteristics of the included studies with AI models based on genomics\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"10\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAuthor, year\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePatient population\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAI* type\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAlgorithm used\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOutcome measure\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eResults\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eExternal validation\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eNumber of data features\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eData source\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c10\"\u003e \u003cp\u003eNumber of patients\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYokoyama S. et al., 2020 [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAny stage\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSVM*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e5-yearOS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003ep\u0026thinsp;\u0026lt;\u0026thinsp;0.001, HR* 0.322 (CI* 0.17\u0026ndash;0.61)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eGenes \u0026ndash; 3\u003c/p\u003e \u003cp\u003eClinical- 6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003ePatients\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e191\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWang L. et al., 2022 [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAny stage\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eCoxBoost and Survival-SVM*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAUCs* of 1-, 2-, 3-year OS* were 0.715, 0.748, 0.671; С-index 0.675\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eGenes \u0026ndash; 32\u003c/p\u003e \u003cp\u003eClinical\u0026thinsp;\u0026minus;\u0026thinsp;10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eTCGA, open-source datasetss\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e1280\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZou Q.et al., 2025 [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRF*, Cox regression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eC-index 0.854\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eGenes \u0026ndash; 3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eTCGA, GEO\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e219\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSun Y. et al., 2025 [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAny stage\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLASSO*, RF*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e5-year OS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eC-index 0.98, AUC 0.938, sensitivity 0.861, specificity 0.987\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eGenes \u0026ndash; 9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eTCGA, GEO\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e279\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBaek B. et al., 2020 [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLR*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOS*, DFS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eDFS* AUC* 0.795 (LR),\u003c/p\u003e \u003cp\u003eOS* AUC* 0.769 (LR),\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eGenes \u0026ndash; 5,\u003c/p\u003e \u003cp\u003eClinical\u0026thinsp;\u0026minus;\u0026thinsp;7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eTCGA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e134\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOjha A. et al., 2024 [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRF*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e3-year OS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eaccuracy 92.62% (males), 91.96% (females), 87.84% (overall)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eGenes \u0026ndash; 999\u003c/p\u003e \u003cp\u003eClinical \u0026ndash; 1 (sex)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eTCGA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e178\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlsharoh H. et al., 2023 [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRF*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003emetastatic progression\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e76% accuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eGenes\u0026thinsp;\u0026minus;\u0026thinsp;38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eTCGA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e151\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChen D. et al., 2023 [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eEnet\u0026thinsp;+\u0026thinsp;SVM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eC-index 0.678\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eGenes\u0026thinsp;\u0026minus;\u0026thinsp;38\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eTCGA, ICGC, ArrayExpress, CPTAC, GEO\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e1711\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZhang X-J. et al., 2025 [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRF*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eС-index 0.779\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eGenes\u0026thinsp;\u0026minus;\u0026thinsp;13\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eTCGA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e226\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHuang L. et al., 2023 [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStage I-II PC undergoing surgery with curative intent\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSVM*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAUC* (1-year OS*) 0.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eGenes \u0026ndash; 15\u003c/p\u003e \u003cp\u003eClinical\u0026thinsp;\u0026minus;\u0026thinsp;15\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003ePatients\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e70\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChang-L. et al.,\u0026nbsp;2025 [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAny stage\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRF*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eC-index: 0.777; AUC 0.974\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eGenes\u0026thinsp;\u0026minus;\u0026thinsp;9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eTCGA\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e286\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAhmed Y. et al., 2023 [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRF*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAccuracy 0.86\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eGenes \u0026ndash; 4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003ecBioportal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e346\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eXiaohong L. et al.,\u0026nbsp;2025 [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eNS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRSF\u0026thinsp;+\u0026thinsp;LASSO*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eC-index 0.777; 1-year: 89.8%, 2-year: 93.2%, 3-year: 92.2% accuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eGenes- 4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eTCGA, ICGC, GTEx\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c10\"\u003e \u003cp\u003e690\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"10\"\u003e\u003cem\u003e*AI \u0026ndash; artificial intelligence, ML \u0026ndash; machine learning, SVM \u0026ndash; support vector machine, NN \u0026ndash; neural network, OS \u0026ndash; overall survival, HR \u0026ndash; hazard ratio, AUC \u0026ndash; area under curve, RF \u0026ndash; random forest, LR \u0026ndash; logistic regression, DFS \u0026ndash; disease-free survival, LASSO \u0026ndash; least absolute shrinkage and selection operator, NS \u0026ndash; not specified\u003c/em\u003e\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDetailed characteristics of the included studies with AI models based on radiomics\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"11\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c11\" colnum=\"11\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAuthor, year\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePatient population\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAI* type\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAlgorithm used\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOutcome measure\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eResults\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eExternal validation\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eNumber of data features\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eImaging modality\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c10\"\u003e \u003cp\u003eClustering method\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c11\"\u003e \u003cp\u003eNumber of patients\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLee W. et al., 2022 [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePatients undergoing surgery with curative intent\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003cp\u003e (clinical)\u003c/p\u003e \u003cp\u003eDL*(radiomics)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eEnsemble model: ANN*, LR*, RF*, GB*, SVM* (clinical)\u003c/p\u003e \u003cp\u003e3D ResNet-18, R (2\u0026thinsp;+\u0026thinsp;1) D-18, 3D ResNeXt-50, 3D DenseNet-121 (radiomics)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e2-year OS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAUC* 0.76 (95% CI*, 0.62\u0026ndash;0.89), sensitivity 69.0%, specificity 83.3%, F1-score 0.755\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eClinical and radiomics, NS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eCT*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003esupervised\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e282\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYao J. et al., 2023 [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePatients undergoing surgery with curative intent\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDL*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003escalable deep segmentation and prognostic models (via self-learning)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eHR* 2.03, 95%CI*: 1.50\u0026ndash;2.75 (internal validation); HR* 2.47, 95%CI* 1.35\u0026ndash;4.53 (external validation)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eNS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eCT*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003eunsupervised\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e1516\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eKaissis G. et al., 2019 [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAny stage\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRF*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003emedian OS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e87% sensitivity (95%CI* 67.3\u0026ndash;92.7); 80% specificity (95%CI* 74.0\u0026ndash;86.7); AUC* 90%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eRadiomics\u0026thinsp;\u0026minus;\u0026thinsp;504\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eMRI*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003esupervised\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e102\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDu X. et al., 2025 [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eResectable PC*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDL*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eCNN*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003erecurrence within 9 months post-surgery\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAUC 0.895\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eRadiomics\u0026thinsp;\u0026minus;\u0026thinsp;112\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eCT*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003esupervised\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e250\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eKeyl J. et al., 2022 [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eUnresectable PC*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRF*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003ec-index 0.71 (clinical data)\u003c/p\u003e \u003cp\u003e0.73 (clinical data+radiomics)\u003c/p\u003e \u003cp\u003e0.76 (clinical data+radiomics+KRAS status)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eClinical \u0026ndash; 8\u003c/p\u003e \u003cp\u003eRadiomics\u0026thinsp;\u0026minus;\u0026thinsp;3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eCT*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003eunsupervised\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e203\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eToyama Y. et al., 2020 [\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAny stage\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRF*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e1-year OS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eHR* 2.0; 95%CI*1.2\u0026ndash;3.4; p\u0026thinsp;=\u0026thinsp;0.0094\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eRadiomics\u0026ndash; 10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eFDG-PET*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003esemi-supervised\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e138\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePalumbo D. et al., 2021 [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePatients undergoing surgery with curative intent\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eLR*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAUC* 0.736\u003c/p\u003e \u003cp\u003eSensitivity 58.6\u003c/p\u003e \u003cp\u003eSpecifity 86.44\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eRadiomics \u0026ndash; 8\u003c/p\u003e \u003cp\u003eClinical\u0026thinsp;\u0026minus;\u0026thinsp;2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eCT*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003esupervised\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e147\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eQ. Gu et al., 2025 [\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eResectable PC*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eML*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRF*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eRecurrence-free survival\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAUCs 0.71, 0.83, and 0.79 at 0.5, 1, and 2 years\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eRadiomics \u0026ndash; 416\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eCT*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003esupervised\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e455\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSchuurmans M. et al., 2025 [\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAny stage\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDL*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eCNN*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eResults:\u0026nbsp;AUC internal: 0.637; External: 0.571\u0026ndash;0.675\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eYes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eClinical \u0026ndash; 12, radiomics \u0026minus;\u0026thinsp;859\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eCT*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003esupervised\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e762\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNi H. et al.,\u003c/p\u003e \u003cp\u003e2023 [\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePatients undergoing surgery with curative intent\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDL*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eAX-Unet\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003erecurrence\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAUC* 0.92 (95%CI* 0.78\u0026ndash;0.99);\u003c/p\u003e \u003cp\u003ec-index 0.62 (95%CI* 0.48\u0026ndash;0.78);\u003c/p\u003e \u003cp\u003eaccuracy 85.9%\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eClinical\u0026thinsp;\u0026minus;\u0026thinsp;10 Radiomics - NS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eCT*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003esupervised\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e64\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZhang Y. et al., 2021 [\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eResectable PC*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003eDL*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRF*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eOS*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAUC* 0.84\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003eNo\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003eRadiomics\u0026ndash; 1463\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003eCT*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003eUnsupervised\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c11\"\u003e \u003cp\u003e98\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"11\"\u003e\u003cem\u003e*AI \u0026ndash; artificial intelligence, ML \u0026ndash; machine learning, DL \u0026ndash; deep learning, ANN \u0026ndash; artificial neural network, LR \u0026ndash; logistic regression, RF \u0026ndash; random forest, CNN \u0026ndash; convolutional neural network, GB \u0026ndash; gradient boosting, SVM \u0026ndash; support vector machine, OS \u0026ndash; overall survival, AUC \u0026ndash; area under curve, CI \u0026ndash; confidence interval, HR \u0026ndash; hazard ratio, MRI-DWI \u0026ndash; diffusion-weighted magnetic resonance imaging, FDG-PET \u0026ndash; fluorodeoxyglucose-positron emission tomography, CT \u0026ndash; computed tomography, LASSO \u0026ndash; least absolute shrinkage and selection operator, NS \u0026ndash; not specified, PC \u0026ndash; pancreatic cancer\u003c/em\u003e\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThe number of data features extracted for the radiomics models varied significantly across studies (3\u0026ndash;1463) and did not seem to correlate with final performance outcomes.\u003c/p\u003e \u003cp\u003eAmong the genomics studies, 10 (76.9%) used data from TCGA. Among the radiomics studies, 3 (27.3%) used unsupervised models, 7 (63.6%) \u0026ndash; supervised models, and 1 (9.1%) study used semi-supervised ML method. CT was the most commonly used modality for radiomics (9/11, 81.8%).\u003c/p\u003e \u003cp\u003ePerformance measures in the included articles are showed in the Table\u0026nbsp;\u003cspan refid=\"Tab5\" class=\"InternalRef\"\u003e5\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab5\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 5\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePerformance measures (n\u0026thinsp;=\u0026thinsp;33 studies)\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePerformance measures\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eStudies\u003c/p\u003e \u003cp\u003eN\u0026thinsp;=\u0026thinsp;33 (100%)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAUC*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e19 (57.6%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eC-index\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e15 (45.5%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e13 (39.4%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSensitivity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e7 (21.2%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSpecificity\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e6 (18.2%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHR*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e4 (12.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eF1-score\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3 (9.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNPV*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2 (6.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePPV*\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2 (6.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"2\"\u003e\u003cem\u003e* AUC \u0026ndash; area under the curve, HR \u0026ndash; hazard ratio, PPV \u0026ndash; positive predictive value, NPV \u0026ndash; negative predictive value\u003c/em\u003e\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThe most commonly used performance measure was AUC (25/33, 75.8%).\u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab6\" class=\"InternalRef\"\u003e6\u003c/span\u003e and Figs.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e\u0026ndash;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e present the results of the PROBAST quality assessment.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab6\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 6\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eSuggested Tabular Presentation for PROBAST Results\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"10\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c9\" colnum=\"9\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c10\" colnum=\"10\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cp\u003eAuthor, year\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"4\" nameend=\"c5\" namest=\"c2\"\u003e \u003cp\u003eRisk of Bias\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c8\" namest=\"c6\"\u003e \u003cp\u003eApplicability\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c10\" namest=\"c9\"\u003e \u003cp\u003eOverall\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eParticipants\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePredictors\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eOutcome\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAnalysis\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eParticipants\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003ePredictors\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eOutcome\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c9\"\u003e \u003cp\u003eRisk of Bias\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c10\"\u003e \u003cp\u003eApplicability\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePalumbo D. et al., 2021 [\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLee K.S. et al., 2021 [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSala Elarre P. et al., 2019 [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNopour R. et al., 2024 [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHuang K. et al., 2025 [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHuang J.-C. et al., 2025 [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWalczak S. et al., 2017 [\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBaig Z. et al., 2021 [\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eXiao W. et al., 2025 [\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTong Z. et al., 2020 [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYokoyama S. et al., 2020 [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eWang L. et al., 2022 [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZou Q. et al., 2025 [\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSun Y. et al., 2025 [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBaek B. et al., 2020 [\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eOjha A. et al., 2024 [\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlsharoh H. et al., 2023 [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChen D.et al., 2023 [31]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZhang X.-J. et al., 2025 [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHuang L. et al., 2023 [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eChang-L. et al.,\u0026nbsp;2025 [\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAhmed Y. et al., 2023 [\u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eXiaohong L. et al.,\u0026nbsp;2025 [\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLee W. et al., 2022 [\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eYao J. et al., 2023 [\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eKaissis G. et al., 2019 [\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDu X. et al., 2025 [\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eKeyl J. et al., 2022 [\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eToyama Y. et al., 2020 [\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGu Q. et al., 2025 [\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSchuurmans M. et al., 2025 [\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNi H. et al., 2023 [\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eZhang Y. et al., 2021 [\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e]\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c9\"\u003e \u003cp\u003e?\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c10\"\u003e \u003cp\u003e+\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eOverall the highest risk of bias was detected in genomics studies \u0026ndash; 30.8% (4/13). High risk of bias was detected in 0/11 (0.0%) of radiomics studies and in 2/9 (22.2%) of studies using data from electronic health records.\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eML is an emerging tool for analyzing large volumes of clinical data to predict cancer progression. Our review revealed substantial heterogeneity among the included studies in terms of study endpoints, outcome measures, feature selection criteria, and choice of ML algorithms.\u003c/p\u003e \u003cp\u003eStudy sample size did not appear to influence the performance of prognostic models. Studies including only 45\u0026ndash;70 patients achieved AUC values of 0.75\u0026ndash;0.92 for OS prediction across all data types [\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e] [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e] [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]. Larger sample sizes did not appear to correlate with better prognostic performance. The same was true for the number of data features. Four out of nine studies using clinical data from electronic health records relied on fewer than 10 data features after pre-selection from the original clinical datasets. Only one study reported that increasing the number of data features improved the performance of the prognostic model [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe optimal choice of clinical data features is an important question. Most of the studies used uni- or multivariate analysis to preselect features for AI-based prognostic models [\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. Tong Z. et al investigated an alternative approach, comparing different methods of feature selection. Interestingly, non-significant variables played important roles in prediction, the ANN-based prediction model with no restrictions on feature selection demonstrated good outcomes with 0.921 AUC [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eOur review revealed that RF was the most commonly used ML algorithm and demonstrated high performance when using both clinical and genomics data. XG-Boost model showed some of the best results using clinical data from electronic health records. In the study by Nopour et al., 2024, XG-Boost achieved 94.96% accuracy, 93.62% specificity, and an AUC of 0.933 in OS prediction [\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e]. ANN was another successful model used for working with electronic health records. In the study by Tong et al., 2020, ANN achieved an AUC of 0.921, with 82.41% accuracy and 89.61% specificity in predicting OS in patients with inoperable cancer [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. KNN algorithm demonstrated the poorest results in clinical data analysis, as its performance greatly depends on data volume and feature structure. For example, in a study by Sala Elarre et al., KNN showed accuracy only of 71% [\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eFor the genomics studies, RF and SVM algorithms demonstrated superior performance. RF achieved an AUC of 0.974 in the study by Li C.-L. et al. and 0.938 in the study of Yuan Sun et al. [\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. Yokoyama et al., 2020, reported that SVM achieved an AUC of 0.795, with 76% accuracy and 98% specificity in predicting OS [\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. Hasan Alsharoh et al., 2023, confirmed SVM's reliability, reporting identical metrics\u0026mdash;76% accuracy and 98% specificity\u0026mdash;for recurrence prediction [\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]. In the study by Huang et al., SVM also demonstrated high predictive power with an AUC of 0.900 in OS prediction [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]. However, linear versions of SVM and logistic regression are limited in adapting to complex patterns and are often applied to simple prediction models but are less effective for data with many nonlinear relationships. The integration of molecular data into clinical practice remains limited by unequal access to advanced technologies across healthcare institutions. The majority of genomics studies rely on open-source data, like TCGA, which may not be reproducible in real-world practice. Unlike studies based on electronic health records and radiomics, the majority of genomics studies (12/13 in this review) included unselected patient populations with either unspecified or any stage of pancreatic cancer. Information on treatment modalities in these studies was also limited, which makes them difficult to compare with the other studies in this review.\u003c/p\u003e \u003cp\u003eIn radiomics studies, increasing complexity of the models did not always lead to better prognostic performance. Schuurmans et al., reported that CNN-based models integrating both clinical (N\u0026thinsp;=\u0026thinsp;12 features) and CT radiomic (N\u0026thinsp;=\u0026thinsp;859 features) data achieved only moderate performance (internal AUC 0.637, external 0.571\u0026ndash;0.675), despite the computational complexity and large feature space [\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]. These contrasting outcomes underscore a critical finding: sophisticated deep learning architectures do not necessarily outperform traditional ML methods, particularly when applied to heterogeneous radiomics data. The success of simpler models with carefully selected data features suggests that feature quality and biological relevance may be more important than model complexity or data volume in achieving superior prognostic accuracy. This observation highlights the need for further research into optimal feature selection strategies and standardized imaging protocols to maximize the clinical utility of radiomics-based AI models.\u003c/p\u003e \u003cp\u003eAI allowed the use of new data types for prognosis prediction, including radiomics. Despite growing interest in them, currently available models do not outperform the ones based on electronic health records. Approaches to handling radiomics data types are too heterogenous to draw any direct conclusions. Surprisingly, unsupervised AI radiomics models achieved some of the best results, which allows to save human resources and presents one of the most reproducible data types, free of potential data input and human error-associated biases [\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eOur study provides a comprehensive review and comparative analysis of a wide range of ML and DL algorithms and data combinations. By integrating results from various approaches, we present a holistic perspective on the application of AI in predicting PC outcomes. Reproducibility remains a major challenge in AI-based prognostic models. In our review, 14/33 (42.4%) of the studies had a high risk of bias, and 14 (42.4%) studies reported external validation. Among the genomics studies, 11 used data exclusively from open-source libraries. All included studies relied on retrospective data. Given the rapidly evolving treatment landscape for PC, these outcomes may not be reproducible in the future. The quality of data in electronic health records may vary substantially across centers, both in terms of reporting quality and the accuracy of assessing prognostic parameters (e.g. frequency of lymphovascular or perineural invasion detection). All these issues may limit the applicability of AI-based prognostic models in the near future.\u003c/p\u003e \u003cp\u003eThis analysis has certain limitations. The included studies are highly heterogeneous in terms of outcome measures, patient populations, and analytical methods. Comparison of outcomes across these studies, even indirectly, is therefore subject to considerable bias. Our aim was to outline the most promising approaches to AI-based prognostic modelling and to estimate the minimum prerequisites (such as sample size and number of data features) required for these models. Including additional databases in our search could potentially have increased the number of eligible studies. Nevertheless, we believe that a systematic search of PubMed, ScienceDirect, Nature, MedRxiv, BioRxiv, and Google Scholar provides a sufficient basis for this review and reflects the current status of AI-based prediction of long-term outcomes in pancreatic cancer. Overall, the application of AI for identifying predictors of survival and disease progression in PC represents a significant and promising area of research. Based on the findings, the use of clinical data appears to be both the easiest and the most reproducible method for predictive modeling, while using more complex data does not yield superior results. DL models leveraging boosting techniques demonstrate superior performance in such scenarios. To achieve high-quality outcomes, large volumes of structured clinical data are essential.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThis research has been financially supported by The Analytical Center for the Government of the Russian Federation (Agreement No. 70-2025-000121 dd 29.8.2025. IGK 000000D730324P540002)\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eT.G. Gevorkyan and S.S. Gordeyev created the study conception and design. Material preparation, data collection and analysis were performed by Y.V. Belenkaya, V.I. Pavlova, M.Sh. Manukyan, M.O. Mandrina, R.Sh. Abdulaeva and G.G. Makiev. The first draft of the manuscript was written by Y.V. Belenkaya, V.I. Pavlova, M.Sh. Manukyan, M.O. Mandrina, R.Sh. Abdulaeva and G.G. Makiev. The manuscript editing was performed by Y.V. Belenkaya and S.S.Gordeyev. All authors read and approved the final manuscript.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eAll data supporting the findings of this study are available within the paper and its Supplementary Information and References.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eIlic I, Ilic M. International patterns in incidence and mortality trends of pancreatic cancer in the last three decades: A joinpoint regression analysis. World J Gastroenterol. 2022;28(32):4698\u0026ndash;715.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHalle-Smith JM, et al. Clinical benefit of surveillance after resection of pancreatic ductal adenocarcinoma: A systematic review and meta-analysis. Eur J Surg Oncol. 2021;47(9):2248\u0026ndash;55.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSong W, Miao DL, Chen L. Nomogram for predicting survival in patients with pancreatic cancer. Onco Targets Ther. 2018;11:539\u0026ndash;45.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGoldstein D, et al. Nomogram for Estimating Overall Survival in Patients With Metastatic Pancreatic Cancer. Pancreas. 2020;49(6):744\u0026ndash;50.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChoi SH, Park SW, Seong J. A nomogram for predicting survival of patients with locally advanced pancreatic cancer treated with chemoradiotherapy. Radiother Oncol. 2018;129(2):340\u0026ndash;6.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLeonhardt CS, et al. Prognostic Factors for Early Recurrence After Resection of Pancreatic Cancer: A Systematic Review and Meta-Analysis. Gastroenterology. 2024;167(5):977\u0026ndash;92.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTandon R, et al. A systematic review on deep learning-based automated cancer diagnosis models. J Cell Mol Med. 2024;28(6):e18144.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTran KA, et al. Deep learning in cancer diagnosis, prognosis and treatment selection. Genome Med. 2021;13(1):152.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePalumbo D et al. \u003cem\u003ePrediction of Early Distant Recurrence in Upfront Resectable Pancreatic Adenocarcinoma: A Multidisciplinary, Machine Learning-Based Approach.\u003c/em\u003e Cancers (Basel), 2021. 13(19).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee KS, et al. Usefulness of artificial intelligence for predicting recurrence following surgery for pancreatic cancer: Retrospective cohort study. Int J Surg. 2021;93:106050.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee JH, et al. Preoperative prediction of early recurrence in resectable pancreatic cancer integrating clinical, radiologic, and CT radiomics features. Cancer Imaging. 2024;24(1):6.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWolff RF, et al. PROBAST: a tool to assess the risk of bias and applicability of prediction model studies. Ann Intern Med. 2019;170(1):51\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen Q, et al. Studying the impact of marital status on diagnosis and survival prediction in pancreatic ductal carcinoma using machine learning methods. Sci Rep. 2024;14(1):5273.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHsu TH, et al. Artificial intelligence to assess body composition on routine abdominal CT scans and predict mortality in pancreatic cancer- A recipe for your local application. Eur J Radiol. 2021;142:109834.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim S, et al. Deep-Transfer-Learning-Based Natural Language Processing of Serial Free-Text Computed Tomography Reports for Predicting Survival of Patients With Pancreatic Cancer. JCO Clin Cancer Inf. 2024;8:e2400021.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSala Elarre P et al. Use of Machine-Learning Algorithms in Intensified Preoperative Therapy of Pancreatic Cancer to Predict Individual Risk of Relapse. Cancers (Basel), 2019. 11(5).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNopour R. Establishment of prediction model for mortality risk of pancreatic cancer: a retrospective study. BMC Med Inf Decis Mak. 2024;24(1):181.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang K, et al. Development and validation of machine learning nomograms for predicting survival in stage IV pancreatic cancer: A retrospective study. World J Gastrointest Oncol. 2025;17(5):102459.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang J-C, et al. A logistic regression model to predict long-term survival for borderline resectable pancreatic cancer patients with upfront surgery. Cancer Imaging. 2025;25(1):10.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWalczak S, Velanovich V. An Evaluation of Artificial Neural Networks in Predicting Pancreatic Cancer Survival. J Gastrointest Surg. 2017;21(10):1606\u0026ndash;12.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBaig Z, et al. Prognosticating Outcome in Pancreatic Head Cancer With the use of a Machine Learning Algorithm. Technol Cancer Res Treat. 2021;20:15330338211050767.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eXiao W, Yang B, Ke S. Application of machine learning for prognostic modeling in unresectable pancreatic cancer treated with chemoradiotherapy. Front Oncol. 2025;15:1644141.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTong Z, et al. Development, Validation and Comparison of Artificial Neural Network Models and Logistic Regression Models Predicting Survival of Unresectable Pancreatic Cancer. Front Bioeng Biotechnol. 2020;8:196.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYokoyama S, et al. Predicted Prognosis of Patients with Pancreatic Cancer by Machine Learning. Clin Cancer Res. 2020;26(10):2411\u0026ndash;21.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang L et al. Comprehensive machine-learning survival framework develops a consensus model in large-scale multicenter cohorts for pancreatic cancer. Elife, 2022. 11.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZou Q, et al. Machine learning-driven prognostic model based on sphingolipid-related gene signature in pancreatic cancer: development and validation. Translational Cancer Res. 2025;14(5):2779.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSun Y, et al. Prognostic model identification of ribosome biogenesis-related genes in pancreatic cancer based on multiple machine learning analyses. Discover Oncol. 2025;16(1):905.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBaek B, Lee H. Prediction of survival and recurrence in patients with pancreatic cancer by integrating multi-omics data. Sci Rep. 2020;10(1):18951.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOjha A et al. Gap-App: A sex-distinct AI-based predictor for pancreatic ductal adenocarcinoma survival as a web application open to patients and physicians. bioRxiv, 2024.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlsharoh H. \u003cem\u003eMachine learning predicts metastatic progression using novel differentially expressed lncRNAs as potential markers in pancreatic cancer.\u003c/em\u003e medRxiv, 2023: p. 2023.11.01.23297724.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen D, et al. Comprehensive machine learning-generated classifier identifies pro-metastatic characteristics and predicts individual treatment in pancreatic cancer: A multicenter cohort study based on super-enhancer profiling. Theranostics. 2023;13(10):3290\u0026ndash;309.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang X-J, et al. Improving molecular subtypes and prognosis of pancreatic cancer through multi group analysis and machine learning. Discover Oncol. 2025;16(1):96.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang L et al. Gene signature developed for predicting early relapse and survival in early-stage pancreatic cancer. BJS Open, 2023. 7(3).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi C-L, et al. Machine learning model reveals the risk, prognosis, and drug response of histamine-related signatures in pancreatic cancer. Discover Oncol. 2025;16(1):155.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAhmed Y et al. A prognostic machine learning model for the prediction of pancreatic adenocarcinoma prognosis based on genomic expression of four cell-cycle associated hub genes. Annals Pancreat Cancer, 2023. 6.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu X, et al. Machine learning based identification of an amino acid metabolism related signature for predicting prognosis and immune microenvironment in pancreatic cancer. BMC Cancer. 2025;25(1):6.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee W, et al. Preoperative data-based deep learning model for predicting postoperative survival in pancreatic cancer patients. Int J Surg. 2022;105:106851.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYao J, et al. Deep Learning for Fully Automated Prediction of Overall Survival in Patients Undergoing Resection for Pancreatic Cancer: A Retrospective Multicenter Study. Ann Surg. 2023;278(1):e68\u0026ndash;79.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKaissis G, et al. A machine learning model for the prediction of survival and tumor subtype in pancreatic ductal adenocarcinoma from preoperative diffusion-weighted imaging. Eur Radiol Exp. 2019;3(1):41.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDu X et al. Novel CT radiomics models for the postoperative prediction of early recurrence of resectable pancreatic adenocarcinoma: A single-center retrospective study in China. J Appl Clin Med Phys, 2025: p. e70092.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKeyl J, et al. Multimodal survival prediction in advanced pancreatic cancer using machine learning. ESMO Open. 2022;7(5):100555.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eToyama Y, et al. Prognostic value of FDG-PET radiomics with machine learning in pancreatic cancer. Sci Rep. 2020;10(1):17024.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGu Q, et al. Interpretable Prognostic Modeling for Postoperative Pancreatic Cancer Using Multi-machine Learning and Habitat Radiomics: A Multi-center Study. Acad Radiol. 2025;32(9):5231\u0026ndash;41.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSchuurmans M et al. End-to-end prognostication in pancreatic cancer by multimodal deep learning: a retrospective, multicenter study. Eur Radiol, 2025: pp. 1\u0026ndash;12.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNi H et al. Predicting Recurrence in Pancreatic Ductal Adenocarcinoma after Radical Surgery Using an AX-Unet Pancreas Segmentation Model and Dynamic Nomogram. Bioeng (Basel), 2023. 10(7).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang Y, et al. Improving prognostic performance in resectable pancreatic ductal adenocarcinoma using radiomics and deep learning features fusion in CT images. Sci Rep. 2021;11(1):1378.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"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":"clinical-and-experimental-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"clem","sideBox":"Learn more about [Clinical and Experimental Medicine](https://www.springer.com/journal/10238)","snPcode":"10238","submissionUrl":"https://submission.nature.com/new-submission/10238/3","title":"Clinical and Experimental Medicine","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"pancreatic cancer, early recurrence, artificial intelligence, machine learning, deep learning, prognostic factors","lastPublishedDoi":"10.21203/rs.3.rs-8940110/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8940110/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003ePancreatic cancer is among the most aggressive malignancies, characterized by poor survival outcomes. The potential of artificial intelligence (AI) to enhance prognostic predictions is increasingly recognized, as traditional tools have demonstrated limited accuracy.\u003c/p\u003e\u003ch2\u003eObjective\u003c/h2\u003e \u003cp\u003eThis review aimed to evaluate the existing literature on the application of AI for predicting treatment outcomes in pancreatic cancer. The primary objective was to assess various AI models, data types, and their advantages.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eA systematic literature search was conducted, encompassing studies published during 2017\u0026ndash;2025. The review focused on research utilizing AI methodologies for predicting pancreatic cancer progression. The analysis followed a three-stage process: initial search, title and abstract screening, and full-text review. Data synthesis included the evaluation of model performance, data types, and validation strategies.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eFrom an initial pool of 35,577 articles, 33 met the inclusion criteria. The random forest was the most frequently applied machine learning (ML) algorithm (14/33, 42.4%). Three types of data were used: clinical data from electronic health records in 9 (27.3%) studies, radiomics in 11 (33.3%) studies, and genomics in 13 (39.4%) studies. The number of patients varied between 45 and 4846 for clinical data based models, between 70 and 1711 for genomics and between 64 and 1516 for radiomics. The resulting AUC varied between 0.933 and 0.732 for clinical data based models, between 0.671 and 0.938 for radiomics, between 0.571 and 0.92 for radiomics. All studies were heterogenous in terms of design, data feature selection and endpoints. Only 14 studies (42.4%) reported external validation of prognostic models. Ten out of thirteen genomics studies used data from open-source databases. Only 3/11 radiomics studies used unsupervised ML methods. High risk of bias was detected in 7 (21.2%) of studies.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eAI demonstrates substantial potential for improving the accuracy of recurrence prediction in pancreatic cancer. However, standardization and improved accessibility are critical for facilitating clinical implementation. Further research is required to refine AI models for routine clinical use.\u003c/p\u003e","manuscriptTitle":"Artificial Intelligence for treatment outcomes in Pancreatic Cancer: a Scoping Review","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-24 10:27:58","doi":"10.21203/rs.3.rs-8940110/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-04-28T16:07:49+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-18T02:49:53+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-06T12:00:36+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-31T14:33:11+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"61842547999823834340793372859184796340","date":"2026-03-23T18:29:28+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"254208751301432196286790966049100587062","date":"2026-03-23T05:02:23+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"20377615025011728299521866936509286990","date":"2026-03-22T14:09:58+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"65051421649825710010013041170626740989","date":"2026-03-20T22:28:22+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-20T12:25:17+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"278958693262422031522182702837853200919","date":"2026-03-19T03:26:57+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"150610685410155623663606167324108250436","date":"2026-03-18T17:35:42+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-03-18T16:08:05+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-14T09:44:16+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-03-14T09:43:17+00:00","index":"","fulltext":""},{"type":"submitted","content":"Clinical and Experimental Medicine","date":"2026-02-22T15:33:55+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"clinical-and-experimental-medicine","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"clem","sideBox":"Learn more about [Clinical and Experimental Medicine](https://www.springer.com/journal/10238)","snPcode":"10238","submissionUrl":"https://submission.nature.com/new-submission/10238/3","title":"Clinical and Experimental Medicine","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"ab1bc252-50ba-4b17-a959-d67c009336ff","owner":[],"postedDate":"March 24th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"in-revision","subjectAreas":[],"tags":[],"updatedAt":"2026-04-28T16:23:18+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-24 10:27:58","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8940110","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8940110","identity":"rs-8940110","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}