The predictive value of radiomics and deep learning for synchronous distant metastasis in clear cell renal cell carcinoma

The predictive value of radiomics and deep learning for synchronous distant metastasis in clear cell renal cell carcinoma | 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 The predictive value of radiomics and deep learning for synchronous distant metastasis in clear cell renal cell carcinoma Wanbin He, Winbo Zhang, Yunfeng Zhang, Zhijun Yang, Han He, Jia Wang, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5368462/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 9 You are reading this latest preprint version Abstract Objective The objective of this research was to devise and authenticate a predictive model that employs CT radiomics and deep learning methodologies for the accurate prediction of synchronous distant metastasis (SDM) in clear cell renal cell carcinoma (ccRCC). Methods A total of 143 ccRCC patients were included in the training cohort, and 62 ccRCC patients were included in the validation cohort. The CT images from all patients were normalized, and the tumor regions were manually segmented via ITK-SNAP software. Radiomic features were extracted via the FAE toolkit. The least absolute shrinkage and selection operator (LASSO) algorithm was employed to select features and build various machine learning models. Additionally, the largest cross-section of the tumor was cropped to train the deep learning model. Multiple deep learning models were trained to predict SDM in ccRCC patients. The results of the best machine learning model were then fused with those of the deep learning model to create a combined model. Results Of the 944 radiomic features identified, 15 were closely associated with SDM. With these 15 features, the support vector machine (SVM) model emerged as the most effective, demonstrating areas under the curve (AUC) of 0.860 and 0.813 in the training and validation cohort, respectively. Among the deep learning models, ResNet101 performed optimally, achieving AUC of 0.815 and 0.743 in the training and validation cohort, respectively. The combined model yielded an AUC of 0.863. Decision curve analysis suggested that the combined model offers superior clinical applicability. Conclusion This model, which combines radiomics and deep learning, has substantial potential the prediction of SDM in patients with ccRCC. This study is anticipated to bolster clinical decision-making processes and enhance the prognostic outcomes for individuals diagnosed with ccRCC. Radiomics Deep Learning Renal Clear Cell Carcinoma Simultaneous Distant Metastasis Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 1. Introduction Clear cell renal cell carcinoma (ccRCC) represents the predominant form of malignant kidney tumor, constituting approximately 70–90% of all renal cancers. This variant is distinguished by its pronounced aggressiveness and elevated metastatic rate, thereby presenting substantial challenges for therapeutic intervention [1] . At the time of diagnosis, approximately 30% of patients exhibit synchronous distant metastasis (SDM), wherein both primary and metastatic lesions coexist. Furthermore, approximately 20% of patients who initially undergo nephrectomy without evidence of distant metastatic lesions subsequently develop metachronous metastasis [2][3] . Consequently, early detection of metastasis is imperative for increasing both patient survival rates and overall quality of life. Despite significant advances in surgery, radiotherapy, and targeted therapies in recent years, tumor metastasis remains a critical factor affecting the prognosis of patients with ccRCC [4] . Previous studies have relied primarily on tumor size, some clinical indicators, and the experience of clinicians to assess the risk of metastasis. However, these methods have limitations and fail to fully exploit and utilize the rich information contained in tumor imaging. Furthermore, even with similar tumor sizes, there can be variations in the risk of SDM [5][6] . Radiomics provides a new perspective for studying the biological characteristics of tumors by extracting quantitative features from medical images via advanced computational methods and machine learning techniques [7] . Among these, CT radiomics has demonstrated significant potential for application in renal cancer research [8–10] . CT technology not only clearly displays the morphological features of tumors but also, through its high-resolution advantages, accurately captures the microscopic structural details within the tumors, providing strong support for in-depth research and precision treatment of renal cancer. Deep learning, as the most advanced machine learning algorithm currently available, has become an important tool in the field of image analysis. By constructing convolutional neural networks, it is possible to learn features directly from raw CT images, demonstrating strong learning and generalization capabilities [11] . Compared with traditional machine learning methods, the advantage of deep learning lies in its ability to automatically learn complex nonlinear features from images, thereby reducing the need for manual intervention [12] . Integrating radiomics with deep learning offers new possibilities for more accurately predicting the systemic disease management (SDM) of ccRCC. This study aimed to construct and validate a prediction model for ccRCC SDM based on CT radiomics and deep learning. 2. Materials and methods 2.1 Patients This study complies with the ethical principles outlined in the Declaration of Helsinki and has been approved by the Ethics Committee of Gansu Provincial People's Hospital, bearing approval number 2023 − 524. Owing to the nature of the research, the need for informed consent was waived. The study involved two cohorts: a training cohort and a validation cohort. The training cohort included a retrospective cohort of patients who were diagnosed with ccRCC through histological confirmation at Gansu Provincial People's Hospital from January 2019 to February 2024. The validation cohort was assembled by enrolling patients from Gansu Second People's Hospital between June 2022 and February 2024. To reduce bias caused by sample imbalance, SDM patients were continuously c enrolled, and ccRCC patients without SDM were randomly selected on the basis of the tumor size of the SDM cohort. The inclusion criteria were as follows: a: diagnosed with ccRCC via pathology; b: all CT examinations completed within 2 weeks prior to the pathological diagnosis; c: availability of complete contrast-enhanced CT imaging data. The exclusion criteria included the following: a: the presence of other malignant tumors; b: a history of treatment for renal parenchymal diseases; c: poor CT image quality or delayed enhancement that rendered the images unsuitable for feature extraction and analysis. The training cohort included a total of 143 patients, comprising 58 SDM patients and 85 non-SDM patients; the validation cohort included 62 patients, consisting of 19 SDM patients and 43 non-SDM patients(Fig. 1 ). 2.2 CT image acquisition All patients underwent CT scans before surgery. The CT parameters are detailed in Supplemental material 1. All eligible patients’ CT imaging data were screened and collected according to strict inclusion criteria. To ensure consistency of image quality, all the images to be processed were normalized before annotation. [13] . 2.3 Lesion segmentation and feature extraction Guided by the deputy chief physician of radiology, two physicians utilized ITK-SNAP software (version 4.0) to manually delineate tumor margins on arterial phase images from enhanced CT scans, thereby defining regions of interest (ROI). In instances of disagreement, consensus was achieved through discussion(Fig. 2 ). To evaluate the consistency of these annotations, both physicians repeated the segmentation process on CT images from 20 randomly chosen patients after a one-month interval. The intraclass correlation coefficient (ICC) was calculated to assess agreement, with robust radiomic features having an ICC value of ≥ 0.8 selected for further analysis. Using the delineated ROI, the FAE toolbox was employed to extract both original and wavelet transformation features. A cropping tool was also used to obtain images of the largest cross-section of the tumor for training the deep learning model. 2.4 Feature selection and model construction All radiomic features were subjected to Z-score normalization. Only those features with an ICC score of 0.8 or higher were retained for subsequent analysis. An optimal feature combination was identified via a stepwise search method grounded in the least absolute shrinkage and selection operator (LASSO) algorithm. Predictive models were formulated for both the training and validation cohorts, employing various machine learning classifiers. The model demonstrating superior performance was chosen. The relevant models underwent pretraining via DenseNet and ResNet series deep learning architectures, which were executed over 50 iterations. This pretraining leverages the ImageNet dataset as a regularization technique. To consolidate the predictions from the prime machine learning model and the deep learning model, a weighted averaging approach was adopted, culminating in the creation of a composite model. 2.5 Model evaluation The efficacy of the model was assessed through the construction of a receiver operating characteristic (ROC) curve and the computation of the area under the curve (AUC). Furthermore, clinical decision curve analysis (DCA) and calibration curves were employed to juxtapose the net benefits and goodness-of-fit across various models. 2.6 Statistical analysis The baseline patient data were evaluated via R software (version 4.3.1) and SPSS 23.0. Continuous variables that were normally distributed are presented as ( x̅ ± s ) and were analyzed with t-tests. For variables that were not normally distributed, the median (interquartile range) was reported, and the Mann‒Whitney U test was employed for analysis. Categorical variables were examined via the chi-square test. A p-value of < 0.05 was considered statistically significant. 3. Results 3.1 Clinical characteristics As presented in Table 1 , the training cohort comprised a total of 143 patients (92 males and 51 females), of whom 58 had ccRCC with SDM and 85 had ccRCC without SDM. The validation cohort consisted of 62 patients (37 males and 25 females), including 19 with ccRCC and SDM and 43 without SDM. In the training cohort, no significant differences were found in the clinical indicators apart from the maximum tumor diameter ( P > 0.05). Similarly, in the validation cohort, no significant differences were observed in any other indicators except for body mass index ( P > 0.05). 3.2 Radiomic feature selection and model construction A total of 944 radiomic features were extracted from the arterial phase of the enhanced CT images of each patient. Fifteen radiomic features closely associated with SDM were selected via the LASSO regression model (Fig. 3 ). Models based on these 15 features were constructed via different classifiers, and the support vector machine (SVM) was found to be the best-performing model. The area under the AUC of the SVM model was 0.860 (95% CI: 0.789–0.931) in the training cohort and 0.813 (95% CI: 0.694–0.931) in the validation cohort (Fig. 4 ). DCA revealed that this model provides significant net clinical benefits, and the calibration curve indicated a high level of accuracy for the model (Fig. 5 ). 3.3 Deep learning model construction Features were extracted from the original images via various deep learning models to develop prediction models. These models were subsequently evaluated on the basis of several performance metrics, such as the AUC, sensitivity, specificity, and F1 score. The findings revealed that the ResNet101 model was the most effective overall (Table 2 ). It demonstrated superior discriminative ability, predictive accuracy, and balance, with AUC values of 0.815 (95% CI: 0.742–0.887) and 0.743 (95% CI: 0.614–0.872) for the training and validation cohorts, respectively. Furthermore, the calibration curve revealed that the model had a minimal prediction error, suggesting that both the accuracy and stability of the model were satisfactory (Fig. 6 ). 3.4 Construction of the combined model A combined model was developed by integrating the highest-performing radiomic features with deep learning features. The AUC of the combined model reached 0.863 (95% CI: 0.804–0.921), indicating excellent discriminatory ability. Additionally, the calibration curve of the combined model shows high consistency between the predicted probabilities and actual outcomes. DCA indicates that the combined model outperforms other individual models in providing clinical utility. Therefore, this combined model demonstrates optimal performance and applicability in clinical settings( Fig. 7 ). 4. Discussion In this study, we investigated the novel integration of deep learning and radiomics to predict the probability of SDM occurrence in patients with ccRCC. Our findings suggest that this method holds substantial promise in enhancing the precision of metastatic risk prediction for ccRCC. The synergy of deep learning and radiomics is anticipated to emerge as a potent instrument, offering clinicians more accurate risk evaluations and facilitating informed treatment decisions. This, in turn, may significantly improve treatment outcomes and prognosis for cancer patients. The early identification of patients with high-risk SDM has substantial clinical implications. This not only facilitates the development of personalized treatment plans but also significantly enhances patient survival rates and quality of life [14] . Additionally, the use of noninvasive CT imaging data can effectively mitigate patient discomfort [15] . Traditionally, the prediction of metastatic risk in ccRCC patients primarily depends on tumor size, patient symptoms, and the judgment of experienced clinicians. Research conducted by Monda et al. [16] indicates that when tumors exceed 3 cm in size, the risk of metastasis progressively increases in proportion to the tumor size. Gallan et al. [17] also highlighted that even patients with small tumors are still at risk for SDM. Therefore, while these methods offer some guidance, an overreliance on tumor size neglects potential biological characteristics present in imaging data and their potential impact on metastasis. The integration of radiomics and deep learning facilitates the extraction of a multitude of quantitative features from CT images, offering a more holistic representation of tumor characteristics beyond just volume or diameter [18–19] . Previous research has demonstrated significant advancements in predicting ccRCC SDM via radiomics. For instance, Xu et al. [20] transformed the optimal MRI radiomics features into a radiomic score and subsequently integrated it with pertinent clinical radiological features, achieving an AUC of 0.914 for predicting ccRCC SDM. Concurrently, Yu et al. [21] enhanced predictive efficacy by incorporating blood biochemical indicators into the clinical model alongside the radiomic model. However, in our study, clinical indicators were not associated with metastasis, potentially due to sample size and regional variations. The rapid advancement of deep learning technology has ushered in novel opportunities for medical image analysis, enabling the direct identification of morphological and texture patterns within images [22] . Xi et al. [23] utilized a model based on the ResNet50 architecture, achieving more accurate differentiation between benign and malignant renal lesions than did radiomics and radiologists. However, our study revealed that the AUC of the deep learning model was lower than that of the radiomics model, which may be due to the limited sample size and overfitting issues associated with small datasets [24] . Xv et al. [25] found that deep learning models based on ultrasound images performed excellently in diagnosing renal artery stenosis, with ResNet models outperforming XCiT models. This further highlights the application potential of ResNet models in the diagnosis of renal diseases. By comparing different deep convolutional neural network models, we found that the ResNet101 model performed best in predicting SDM, with AUCs of 0.815 and 0.743 for the training and validation cohorts, respectively. To further enhance predictive performance, we combined radiomics with deep learning techniques, fully integrating the quantitative features from radiomics and the high-level features extracted by deep learning to construct a more optimized combined model. This model achieved an AUC of 0.863, demonstrating high accuracy in predicting SDM risk in ccRCC patients. The combined model shows potential and practicality for clinical application, providing better clinical decision support for developing more personalized and precise treatment plans. This study has several limitations. First, the sample size was relatively small. Although the model performed well in the validation cohort, its robustness still needs further validation in larger, multicenter datasets. Second, owing to differences in CT scanning parameters, the standardization of image data processing still needs optimization to ensure the model's applicability across different devices and settings. Future research should explore further optimization and improvement of deep learning models based on enhancing data diversity. Additionally, combining data from other modalities (such as genetic information and molecular markers) should be considered when constructing a more comprehensive predictive model. 5. Conclusion In summary, by integrating deep learning and radiomics, this study provides a powerful tool for predicting SDM risk in ccRCC patients. The results demonstrate that this method has significant advantages in improving the efficiency and accuracy of metastatic prediction. Abbreviations SDM: synchronous distant metastasis ccRCC: clear cell renal cell carcinoma LASSO: least absolute shrinkage and selection operator SVM: support vector machine AUC: area under the curve ROI: regions of interest ICC: intraclass correlation coefficient DCA: decision curve analysis Declarations Acknowledgments We thank Gansu Provincial Hospital, Lanzhou University and Gansu University of Chinese Medicine for their guidance and advice during the implementation of this project; we thank the onekey AI platform for providing technical support for this study. Funding This work was funded by grants from the following sources: The Natural Science Foundation of Gansu Province (No.22JR5RA650) Gansu Provincial Hospital Scientific Research Foundation (No.23GSSYD-12) Authors’ contributions Guarantor of integrity of the entire study: ZFH; Design of the research: ZFH, HWB, ZWB; literature retrieve: HWB, ZWB; information collection of patients: ZWB, ZYF, YZJ; supervised learning and statistical analysis: ZYF, YZJ, HH, WJ; manuscript preparation: HWB; manuscript review: ZFH. All the authors read and approved the final manuscript. Availability of data and materials All data supporting the findings of this study are available within the paper and its Supplementary material. Ethical standards and informed consent This retrospective study was approved by the Ethics Committee of Gansu Provincial Hospital and the requirement for obtaining informed consent was waived by the institutional review board. Patients’ records were anonymized and deidentified prior to analysis. The study was conducted according to the guidelines of the Declaration of Helsinki. Competing interests: All authors declare that they have no competing interests. References Bahadoram S, Davoodi M, Hassanzadeh S, Bahadoram M, Barahman M, Mafakher L. Renal cell carcinoma: an overview of the epidemiology, diagnosis, and treatment. G Ital Nefrol. 2022;39(3):2022-vol3. Published 2022 Jun 20. Baston C, Parosanu AI, Stanciu IM, Nitipir C. Metastatic Kidney Cancer: Does the Location of the Metastases Matter? Moving towards Personalized Therapy for Metastatic Renal Cell Carcinoma. Biomedicines. 2024;12(5):1111. Published 2024 May 16. Donskov F, Xie W, Overby A, et al. Synchronous Versus Metachronous Metastatic Disease: Impact of Time to Metastasis on Patient Outcome-Results from the International Metastatic Renal Cell Carcinoma Database Consortium. Eur Urol Oncol. 2020;3(4):530-539. Nizam A, Schindelheim JA, Ornstein MC. The role of active surveillance and cytoreductive nephrectomy in metastatic renal cell carcinoma. Cancer Treat Res Commun. Published online February 22, 2020. Monda SM, Carney BW, May AM, et al. Differences in mutations across tumour sizes in clear-cell renal cell carcinoma. BJU Int. Published online September 12, 2024. Jiang W, Shou J, Shi H, et al. Impact of Primary Tumor Size on Prognosis in Patients With Metastatic Renal Cell Carcinoma Receiving Cytoreductive Nephrectomy: A Population Study of a Chinese Center and the US SEER Database. Technol Cancer Res Treat. 2021;20:15330338211019507. Mu W, Schabath MB, Gillies RJ. Images are data: challenges and opportunities in the clinical translation of radiomics. Cancer Res. 2022;82(11):2066-2068. Nie P, Liu S, Zhou R, et al. A preoperative CT-based deep learning radiomics model in predicting the stage, size, grade and necrosis score and outcome in localized clear cell renal cell carcinoma: A multicenter study. Eur J Radiol. 2023;166: 111018. Yu S, Yang Y, Wang Z, et al. CT-based conventional radiomics and quantification of intratumoral heterogeneity for predicting benign and malignant renal lesions. Cancer Imaging. 2024;24(1):130. Published 2024 Oct 2. Zhu P, Dai C, Xiong Y, et al. Tumor contour irregularity on preoperative CT predicts prognosis in renal cell carcinoma: a multi-institutional study. EClinicalMedicine. 2024;75:102775. Published 2024 Aug 16. Dai C, Xiong Y, Zhu P, et al. Deep Learning Assessment of Small Renal Masses at Contrast-enhanced Multiphase CT. Radiology. 2024;311(2):e232178. Salvi M, Acharya UR, Molinari F, Meiburger KM. The impact of pre- and post-image processing techniques on deep learning frameworks: A comprehensive review for digital pathology image analysis. Comput Biol Med. 2021;128:104129. Tong N, Xu Y, Zhang J, Gou S, Li M. Robust and efficient abdominal CT segmentation using shape constrained multi-scale attention network. Phys Med. 2023;110:102595. Yang Y, Wang J, Ren Q, et al. Multimodal data integration using machine learning to predict the risk of clear cell renal cancer metastasis: a retrospective multicentre study. Abdom Radiol (NY). 2024;49(7):2311-2324. Mühlbauer J, Egen L, Kowalewski KF, et al. Radiomics in Renal Cell Carcinoma-A Systematic Review and Meta-Analysis. Cancers (Basel) . 2021; 13(6):1348. Published 2021 Mar 17. Monda SM, Lui HT, Pratsinis MA, Chandrasekar T, Evans CP, Dall'Era MA. The Metastatic Risk of Renal Cell Carcinoma by Primary Tumor Size and Subtype. Eur Urol Open Sci. 2023;52:137-144. Published 2023 May 10. Gallan AJ, Antic T. Clinicopathologic Features of Small Renal Masses Associated With Distant Metastatic Disease . Am J Clin Pathol. 2020;153(5):613-617. Arndt C, Güttler F, Heinrich A, Bürckenmeyer F, Diamantis I, Teichgräber U. Deep Learning CT Image Reconstruction in Clinical Practice. CT-Bildrekonstruktion mit Deep Learning in der klinischen Praxis. Rofo. 2021;193(3):252-261. Koetzier LR, Mastrodicasa D, Szczykutowicz TP, et al. Deep Learning Image Reconstruction for CT: Technical Principles and Clinical Prospects. Radiology. 2023;306(3):e221257. Bai X, Huang Q, Zuo P, et al. MRI radiomics-based nomogram for individualised prediction of synchronous distant metastasis in patients with clear cell renal cell carcinoma. Eur Radiol. 2021;31(2):1029-1042. Yu X, Gao L, Zhang S, et al. Development and validation of A CT-based radiomics nomogram for prediction of synchronous distant metastasis in clear cell renal cell carcinoma. Front Oncol. 2023;12:1016583. Published 2023 Jan 4. Xu Y, Quan R, Xu W, Huang Y, Chen X, Liu F. Advances in Medical Image Segmentation: A Comprehensive Review of Traditional, Deep Learning and Hybrid Approaches. Bioengineering (Basel). 2024;11(10):1034. Published 2024 Oct 16. Xi IL, Zhao Y, Wang R, et al. Deep Learning to Distinguish Benign from Malignant Renal Lesions Based on Routine MR Imaging. Clin Cancer Res. 2020;26(8):1944-1952. Demircioğlu A. Are deep models in radiomics performing better than generic models? A systematic review. Eur Radiol Exp. 2023;7(1):11. Published 2023 Mar 15. Xv Y, Wei Z, Jiang Q, et al. Three-Dimensional (3D) deep learning model complements existing models for preoperative disease-free survival prediction (DFS) in localized clear cell renal cell carcinoma (ccRCC): A multicenter retrospective cohort study. Int J Surg. Published online June 19, 2024. Tables Table 1 Comparison of clinical data between training and validation cohort patients Variables Training group (n = 143) Validation group (n = 62) SDM(+)(n = 58) SDM(-)(n = 85) P SDM(+)(n = 19) SDM(-)(n = 43) P Sex,n(%) 0.190 0.710 Male 41 (70.69) 51 (60.00) 12 (63.16) 25 (58.14) Female 17 (29.31) 34 (40.00) 7 (36.84) 18 (41.86) Age 61.50 (51.00, 72.00) 58.00 (51.00, 67.00) 0.140 59.00 (50.50, 63.50) 58.00 (52.00, 65.00) 0.994 Size 6.15 (5.00, 7.07) 5.00 (3.00, 7.00) 0.003 5.80 (4.75, 6.80) 5.05 (3.42, 7.00) 0.739 BUN 5.96 (4.88, 6.68) 5.90 (4.42, 7.27) 0.702 5.32 (4.02, 6.38) 4.94 (4.10, 6.29) 0.658 Cr 71.15 (61.73, 83.78) 69.22 (57.70, 85.00) 0.353 71.70 (58.15, 88.70) 74.50 (61.60, 90.50) 0.551 BMI 24.77 (21.36, 26.85) 24.21 (21.12, 26.20) 0.355 25.39 (23.99, 27.16) 22.59 (21.05, 25.16) 0.023 Note: BUN: Blood Urea Nitrogen; Cr: Creatinine; BMI: Body Mass Index. Table 2 Predictive performance of different models Model AUC(95%CI) Sensitivity Specificity PPV NPV F1 Group densenet121 0.735(0.652–0.818) 0.656 0.774 0.831 0.569 0.733 Training densenet121 0.664(0.522–0.807) 0.579 0.792 0.815 0.543 0.677 Validation densenet169 0.665(0.575–0.755) 0.433 0.849 0.830 0.469 0.569 Training densenet169 0.681(0.538–0.825) 0.684 0.625 0.743 0.556 0.712 Validation resnet101 0.815(0.742–0.887) 0.800 0.698 0.818 0.673 0.809 Training resnet101 0.743(0.614–0.872) 0.684 0.750 0.812 0.600 0.743 Validation resnet152 0.726(0.643–0.809) 0.444 0.868 0.851 0.479 0.584 Training resnet152 0.727(0.593–0.861) 0.658 0.750 0.806 0.581 0.725 Validation resnet18 0.694(0.604–0.785) 0.811 0.528 0.745 0.622 0.777 Training resnet18 0.678(0.538–0.818) 0.684 0.625 0.743 0.556 0.712 Validation resnet34 0.786(0.709–0.863) 0.789 0.679 0.807 0.655 0.798 Training resnet34 0.617(0.462–0.772) 0.921 0.333 0.686 0.727 0.787 Validation resnet50 0.690(0.603–0.777) 0.478 0.811 0.811 0.478 0.601 Training resnet50 0.697(0.551–0.843) 0.737 0.625 0.757 0.600 0.747 Validation Additional Declarations No competing interests reported. Supplementary Files SupplementaryMaterial1.docx Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 03 Dec, 2024 Reviews received at journal 30 Nov, 2024 Reviews received at journal 23 Nov, 2024 Reviewers agreed at journal 23 Nov, 2024 Reviewers agreed at journal 21 Nov, 2024 Reviewers invited by journal 21 Nov, 2024 Editor assigned by journal 14 Nov, 2024 Submission checks completed at journal 12 Nov, 2024 First submitted to journal 31 Oct, 2024 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-5368462","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":385654256,"identity":"63f7cf58-6a17-42fd-a3e2-bfa1220160ad","order_by":0,"name":"Wanbin He","email":"","orcid":"","institution":"The First Clinical Medical College of Lanzhou University","correspondingAuthor":false,"prefix":"","firstName":"Wanbin","middleName":"","lastName":"He","suffix":""},{"id":385654257,"identity":"43b9ff44-9221-4996-b6bf-f87d477727b1","order_by":1,"name":"Winbo Zhang","email":"","orcid":"","institution":"The First Clinical Medical College of Gansu University of Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Winbo","middleName":"","lastName":"Zhang","suffix":""},{"id":385654258,"identity":"73457186-d7fd-4906-9d46-d7004399153d","order_by":2,"name":"Yunfeng Zhang","email":"","orcid":"","institution":"The First Clinical Medical College of Lanzhou University","correspondingAuthor":false,"prefix":"","firstName":"Yunfeng","middleName":"","lastName":"Zhang","suffix":""},{"id":385654259,"identity":"da83c96a-e496-473e-a58d-5d66b9558f5f","order_by":3,"name":"Zhijun Yang","email":"","orcid":"","institution":"The First Clinical Medical College of Lanzhou University","correspondingAuthor":false,"prefix":"","firstName":"Zhijun","middleName":"","lastName":"Yang","suffix":""},{"id":385654260,"identity":"da3c04ad-775f-4abe-aadb-a76888734d97","order_by":4,"name":"Han He","email":"","orcid":"","institution":"The First Clinical Medical College of Lanzhou University","correspondingAuthor":false,"prefix":"","firstName":"Han","middleName":"","lastName":"He","suffix":""},{"id":385654261,"identity":"89a6f2d0-cdd3-4a01-bf2e-378f356fc8e8","order_by":5,"name":"Jia Wang","email":"","orcid":"","institution":"The First Clinical Medical College of Gansu University of Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Jia","middleName":"","lastName":"Wang","suffix":""},{"id":385654262,"identity":"01982b66-f4da-42e5-9b31-2c071b25907c","order_by":6,"name":"Fenghai Zhou","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA6UlEQVRIie3NMWrDMBTG8c8Y7Ayv9qoQ8BlecTCFDr3KK4Fm8RDo0hOkS3KXZsvoItDkA3go1FDwlMFrQEPtFEonJd0K1X+QeEI/HuDz/cESyHhVQFgHbXt6EzeJvklUhiy/JZG6jMRloWDfMjZX5kmsRhqXjOPeQegwEOpyNslDI6Qx3Rw42NYOosYtSt+/vG+KRpQGNyWHwfoc4YEYKlbCGneXEfkiEBm2qHOEuscbVDqfGsqVVEtSdbd63TpIGi92DazOEkPXfW9vs/R5sWuPDgJM5jP7Y6TxqFwAiD969wefz+f7930CkDhNgxe+ZowAAAAASUVORK5CYII=","orcid":"","institution":"The First Clinical Medical College of Lanzhou University","correspondingAuthor":true,"prefix":"","firstName":"Fenghai","middleName":"","lastName":"Zhou","suffix":""}],"badges":[],"createdAt":"2024-10-31 17:23:09","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5368462/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5368462/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":71614386,"identity":"d3cbcbf9-5838-481e-bae7-22a012b3c7f8","added_by":"auto","created_at":"2024-12-17 07:12:30","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":73328,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFlowchart of patient recruitment\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"image1.png","url":"https://assets-eu.researchsquare.com/files/rs-5368462/v1/4f8ed1f12c361e375cae01f2.png"},{"id":71612860,"identity":"96a00f0c-d674-45b0-8873-600bbf0ae3d4","added_by":"auto","created_at":"2024-12-17 07:04:30","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":125971,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSketching flowchart. a: Original image of the lesion; b: Sketch diagram; c: 3D volume of interest.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"image2.png","url":"https://assets-eu.researchsquare.com/files/rs-5368462/v1/1097859e80c2ee4bb0f2fd32.png"},{"id":71612854,"identity":"77ddf51d-2c4a-41c9-8842-5d895817bd81","added_by":"auto","created_at":"2024-12-17 07:04:30","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":78442,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFeature selection via the LASSO regression model. a: Factors selected for regression modeling via LASSO; b: Selection of the tuning parameter (λ) in the LASSO model; c: The 15 selected features that contribute to the model and the role of each feature.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"image3.png","url":"https://assets-eu.researchsquare.com/files/rs-5368462/v1/88cd9fda2c3854be5cfad29a.png"},{"id":71612858,"identity":"a0d5d0ef-75a2-4004-8332-ce214f0d4edc","added_by":"auto","created_at":"2024-12-17 07:04:30","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":54260,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eROC curves of different prediction models. a: ROC curves of different machine learning models in the external validation cohort; b: ROC curve of the SVM model.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"image4.png","url":"https://assets-eu.researchsquare.com/files/rs-5368462/v1/5dcada840c38da6acf0f84f1.png"},{"id":71612856,"identity":"145dff1a-f620-400f-9d2c-5b24a660dbeb","added_by":"auto","created_at":"2024-12-17 07:04:30","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":5713,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eEvaluation of the SVW model. a: DCA of the SVM model; b: Calibration curve of the SVM model.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-5368462/v1/6027d63d45cd65e97f22c060.png"},{"id":71614389,"identity":"5f33f864-61fb-4c10-812d-4b9ef1846bda","added_by":"auto","created_at":"2024-12-17 07:12:30","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":53197,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eEvaluation of the ResNet101 Model. a: ROC curve of the ResNet101 model; b: Calibration curve of the ResNet101 model.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"image6.png","url":"https://assets-eu.researchsquare.com/files/rs-5368462/v1/855b7ed4a42be46bc021b0f1.png"},{"id":71612861,"identity":"64ef0342-755f-4d96-a820-dc37beb2c687","added_by":"auto","created_at":"2024-12-17 07:04:30","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":65995,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eEvaluation of the joint model. a: ROC curve of the combined model; b: Comparison of DCA between the combined model and other models; c: Calibration curve of the combined model.\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"image7.png","url":"https://assets-eu.researchsquare.com/files/rs-5368462/v1/36f102ca5cad726addda0527.png"},{"id":71614672,"identity":"18a96652-d577-49d7-b588-001b365d181b","added_by":"auto","created_at":"2024-12-17 07:20:30","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1207074,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5368462/v1/2cd11132-3ba0-46df-a291-f11b572f7163.pdf"},{"id":71614388,"identity":"ab2cfed3-5468-4e05-919d-874626327bb7","added_by":"auto","created_at":"2024-12-17 07:12:30","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":14824,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterial1.docx","url":"https://assets-eu.researchsquare.com/files/rs-5368462/v1/1c216ae6b8c36799f1d794db.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"The predictive value of radiomics and deep learning for synchronous distant metastasis in clear cell renal cell carcinoma","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eClear cell renal cell carcinoma (ccRCC) represents the predominant form of malignant kidney tumor, constituting approximately 70\u0026ndash;90% of all renal cancers. This variant is distinguished by its pronounced aggressiveness and elevated metastatic rate, thereby presenting substantial challenges for therapeutic intervention\u003csup\u003e[1]\u003c/sup\u003e. At the time of diagnosis, approximately 30% of patients exhibit synchronous distant metastasis (SDM), wherein both primary and metastatic lesions coexist. Furthermore, approximately 20% of patients who initially undergo nephrectomy without evidence of distant metastatic lesions subsequently develop metachronous metastasis\u003csup\u003e[2][3]\u003c/sup\u003e. Consequently, early detection of metastasis is imperative for increasing both patient survival rates and overall quality of life.\u003c/p\u003e \u003cp\u003eDespite significant advances in surgery, radiotherapy, and targeted therapies in recent years, tumor metastasis remains a critical factor affecting the prognosis of patients with ccRCC\u003csup\u003e[4]\u003c/sup\u003e. Previous studies have relied primarily on tumor size, some clinical indicators, and the experience of clinicians to assess the risk of metastasis. However, these methods have limitations and fail to fully exploit and utilize the rich information contained in tumor imaging. Furthermore, even with similar tumor sizes, there can be variations in the risk of SDM\u003csup\u003e[5][6]\u003c/sup\u003e. Radiomics provides a new perspective for studying the biological characteristics of tumors by extracting quantitative features from medical images via advanced computational methods and machine learning techniques\u003csup\u003e[7]\u003c/sup\u003e. Among these, CT radiomics has demonstrated significant potential for application in renal cancer research\u003csup\u003e[8\u0026ndash;10]\u003c/sup\u003e. CT technology not only clearly displays the morphological features of tumors but also, through its high-resolution advantages, accurately captures the microscopic structural details within the tumors, providing strong support for in-depth research and precision treatment of renal cancer.\u003c/p\u003e \u003cp\u003eDeep learning, as the most advanced machine learning algorithm currently available, has become an important tool in the field of image analysis. By constructing convolutional neural networks, it is possible to learn features directly from raw CT images, demonstrating strong learning and generalization capabilities\u003csup\u003e[11]\u003c/sup\u003e. Compared with traditional machine learning methods, the advantage of deep learning lies in its ability to automatically learn complex nonlinear features from images, thereby reducing the need for manual intervention\u003csup\u003e[12]\u003c/sup\u003e. Integrating radiomics with deep learning offers new possibilities for more accurately predicting the systemic disease management (SDM) of ccRCC. This study aimed to construct and validate a prediction model for ccRCC SDM based on CT radiomics and deep learning.\u003c/p\u003e"},{"header":"2. Materials and methods","content":"\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.1 Patients\u003c/h2\u003e \u003cp\u003e This study complies with the ethical principles outlined in the Declaration of Helsinki and has been approved by the Ethics Committee of Gansu Provincial People's Hospital, bearing approval number 2023\u0026thinsp;\u0026minus;\u0026thinsp;524. Owing to the nature of the research, the need for informed consent was waived. The study involved two cohorts: a training cohort and a validation cohort. The training cohort included a retrospective cohort of patients who were diagnosed with ccRCC through histological confirmation at Gansu Provincial People's Hospital from January 2019 to February 2024. The validation cohort was assembled by enrolling patients from Gansu Second People's Hospital between June 2022 and February 2024.\u003c/p\u003e \u003cp\u003eTo reduce bias caused by sample imbalance, SDM patients were continuously c enrolled, and ccRCC patients without SDM were randomly selected on the basis of the tumor size of the SDM cohort. The inclusion criteria were as follows: a: diagnosed with ccRCC via pathology; b: all CT examinations completed within 2 weeks prior to the pathological diagnosis; c: availability of complete contrast-enhanced CT imaging data. The exclusion criteria included the following: a: the presence of other malignant tumors; b: a history of treatment for renal parenchymal diseases; c: poor CT image quality or delayed enhancement that rendered the images unsuitable for feature extraction and analysis. The training cohort included a total of 143 patients, comprising 58 SDM patients and 85 non-SDM patients; the validation cohort included 62 patients, consisting of 19 SDM patients and 43 non-SDM patients(Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003e2.2 CT image acquisition\u003c/h3\u003e\n\u003cp\u003eAll patients underwent CT scans before surgery. The CT parameters are detailed in Supplemental material 1. All eligible patients\u0026rsquo; CT imaging data were screened and collected according to strict inclusion criteria. To ensure consistency of image quality, all the images to be processed were normalized before annotation.\u003csup\u003e[13]\u003c/sup\u003e.\u003c/p\u003e\n\u003ch3\u003e2.3 Lesion segmentation and feature extraction\u003c/h3\u003e\n\u003cp\u003eGuided by the deputy chief physician of radiology, two physicians utilized ITK-SNAP software (version 4.0) to manually delineate tumor margins on arterial phase images from enhanced CT scans, thereby defining regions of interest (ROI). In instances of disagreement, consensus was achieved through discussion(Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). To evaluate the consistency of these annotations, both physicians repeated the segmentation process on CT images from 20 randomly chosen patients after a one-month interval. The intraclass correlation coefficient (ICC) was calculated to assess agreement, with robust radiomic features having an ICC value of \u0026ge;\u0026thinsp;0.8 selected for further analysis. Using the delineated ROI, the FAE toolbox was employed to extract both original and wavelet transformation features. A cropping tool was also used to obtain images of the largest cross-section of the tumor for training the deep learning model.\u003c/p\u003e\n\u003ch3\u003e2.4 Feature selection and model construction\u003c/h3\u003e\n\u003cp\u003eAll radiomic features were subjected to Z-score normalization. Only those features with an ICC score of 0.8 or higher were retained for subsequent analysis. An optimal feature combination was identified via a stepwise search method grounded in the least absolute shrinkage and selection operator (LASSO) algorithm. Predictive models were formulated for both the training and validation cohorts, employing various machine learning classifiers. The model demonstrating superior performance was chosen.\u003c/p\u003e \u003cp\u003eThe relevant models underwent pretraining via DenseNet and ResNet series deep learning architectures, which were executed over 50 iterations. This pretraining leverages the ImageNet dataset as a regularization technique.\u003c/p\u003e \u003cp\u003eTo consolidate the predictions from the prime machine learning model and the deep learning model, a weighted averaging approach was adopted, culminating in the creation of a composite model.\u003c/p\u003e\n\u003ch3\u003e2.5 Model evaluation\u003c/h3\u003e\n\u003cp\u003eThe efficacy of the model was assessed through the construction of a receiver operating characteristic (ROC) curve and the computation of the area under the curve (AUC). Furthermore, clinical decision curve analysis (DCA) and calibration curves were employed to juxtapose the net benefits and goodness-of-fit across various models.\u003c/p\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e2.6 Statistical analysis\u003c/h2\u003e \u003cp\u003eThe baseline patient data were evaluated via R software (version 4.3.1) and SPSS 23.0. Continuous variables that were normally distributed are presented as (\u003cem\u003ex̅\u003c/em\u003e \u0026plusmn; \u003cem\u003es\u003c/em\u003e) and were analyzed with t-tests. For variables that were not normally distributed, the median (interquartile range) was reported, and the Mann‒Whitney \u003cem\u003eU\u003c/em\u003e test was employed for analysis. Categorical variables were examined via the chi-square test. A p-value of \u0026lt;\u0026thinsp;0.05 was considered statistically significant.\u003c/p\u003e \u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Clinical characteristics\u003c/h2\u003e \u003cp\u003eAs presented in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, the training cohort comprised a total of 143 patients (92 males and 51 females), of whom 58 had ccRCC with SDM and 85 had ccRCC without SDM. The validation cohort consisted of 62 patients (37 males and 25 females), including 19 with ccRCC and SDM and 43 without SDM. In the training cohort, no significant differences were found in the clinical indicators apart from the maximum tumor diameter (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.05). Similarly, in the validation cohort, no significant differences were observed in any other indicators except for body mass index (\u003cem\u003eP\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.05).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Radiomic feature selection and model construction\u003c/h2\u003e \u003cp\u003eA total of 944 radiomic features were extracted from the arterial phase of the enhanced CT images of each patient. Fifteen radiomic features closely associated with SDM were selected via the LASSO regression model (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e). Models based on these 15 features were constructed via different classifiers, and the support vector machine (SVM) was found to be the best-performing model. The area under the AUC of the SVM model was 0.860 (95% CI: 0.789\u0026ndash;0.931) in the training cohort and 0.813 (95% CI: 0.694\u0026ndash;0.931) in the validation cohort (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e). DCA revealed that this model provides significant net clinical benefits, and the calibration curve indicated a high level of accuracy for the model (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Deep learning model construction\u003c/h2\u003e \u003cp\u003eFeatures were extracted from the original images via various deep learning models to develop prediction models. These models were subsequently evaluated on the basis of several performance metrics, such as the AUC, sensitivity, specificity, and F1 score. The findings revealed that the ResNet101 model was the most effective overall (Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). It demonstrated superior discriminative ability, predictive accuracy, and balance, with AUC values of 0.815 (95% CI: 0.742\u0026ndash;0.887) and 0.743 (95% CI: 0.614\u0026ndash;0.872) for the training and validation cohorts, respectively. Furthermore, the calibration curve revealed that the model had a minimal prediction error, suggesting that both the accuracy and stability of the model were satisfactory (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003e3.4 Construction of the combined model\u003c/h2\u003e \u003cp\u003eA combined model was developed by integrating the highest-performing radiomic features with deep learning features. The AUC of the combined model reached 0.863 (95% CI: 0.804\u0026ndash;0.921), indicating excellent discriminatory ability. Additionally, the calibration curve of the combined model shows high consistency between the predicted probabilities and actual outcomes. DCA indicates that the combined model outperforms other individual models in providing clinical utility. Therefore, this combined model demonstrates optimal performance and applicability in clinical settings(\u003cb\u003eFig.\u0026nbsp;7\u003c/b\u003e).\u003c/p\u003e \u003c/div\u003e "},{"header":"4. Discussion","content":"\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003cp\u003eIn this study, we investigated the novel integration of deep learning and radiomics to predict the probability of SDM occurrence in patients with ccRCC. Our findings suggest that this method holds substantial promise in enhancing the precision of metastatic risk prediction for ccRCC. The synergy of deep learning and radiomics is anticipated to emerge as a potent instrument, offering clinicians more accurate risk evaluations and facilitating informed treatment decisions. This, in turn, may significantly improve treatment outcomes and prognosis for cancer patients.\u003c/p\u003e \u003cp\u003eThe early identification of patients with high-risk SDM has substantial clinical implications. This not only facilitates the development of personalized treatment plans but also significantly enhances patient survival rates and quality of life\u003csup\u003e[14]\u003c/sup\u003e. Additionally, the use of noninvasive CT imaging data can effectively mitigate patient discomfort\u003csup\u003e[15]\u003c/sup\u003e. Traditionally, the prediction of metastatic risk in ccRCC patients primarily depends on tumor size, patient symptoms, and the judgment of experienced clinicians. Research conducted by Monda et al.\u003csup\u003e[16]\u003c/sup\u003e indicates that when tumors exceed 3 cm in size, the risk of metastasis progressively increases in proportion to the tumor size. Gallan et al.\u003csup\u003e[17]\u003c/sup\u003e also highlighted that even patients with small tumors are still at risk for SDM. Therefore, while these methods offer some guidance, an overreliance on tumor size neglects potential biological characteristics present in imaging data and their potential impact on metastasis.\u003c/p\u003e \u003cp\u003eThe integration of radiomics and deep learning facilitates the extraction of a multitude of quantitative features from CT images, offering a more holistic representation of tumor characteristics beyond just volume or diameter\u003csup\u003e[18\u0026ndash;19]\u003c/sup\u003e. Previous research has demonstrated significant advancements in predicting ccRCC SDM via radiomics. For instance, Xu et al.\u003csup\u003e[20]\u003c/sup\u003e transformed the optimal MRI radiomics features into a radiomic score and subsequently integrated it with pertinent clinical radiological features, achieving an AUC of 0.914 for predicting ccRCC SDM. Concurrently, Yu et al.\u003csup\u003e[21]\u003c/sup\u003e enhanced predictive efficacy by incorporating blood biochemical indicators into the clinical model alongside the radiomic model. However, in our study, clinical indicators were not associated with metastasis, potentially due to sample size and regional variations.\u003c/p\u003e \u003cp\u003eThe rapid advancement of deep learning technology has ushered in novel opportunities for medical image analysis, enabling the direct identification of morphological and texture patterns within images\u003csup\u003e[22]\u003c/sup\u003e. Xi et al.\u003csup\u003e[23]\u003c/sup\u003e utilized a model based on the ResNet50 architecture, achieving more accurate differentiation between benign and malignant renal lesions than did radiomics and radiologists. However, our study revealed that the AUC of the deep learning model was lower than that of the radiomics model, which may be due to the limited sample size and overfitting issues associated with small datasets\u003csup\u003e[24]\u003c/sup\u003e. Xv et al.\u003csup\u003e[25]\u003c/sup\u003e found that deep learning models based on ultrasound images performed excellently in diagnosing renal artery stenosis, with ResNet models outperforming XCiT models. This further highlights the application potential of ResNet models in the diagnosis of renal diseases. By comparing different deep convolutional neural network models, we found that the ResNet101 model performed best in predicting SDM, with AUCs of 0.815 and 0.743 for the training and validation cohorts, respectively.\u003c/p\u003e \u003cp\u003eTo further enhance predictive performance, we combined radiomics with deep learning techniques, fully integrating the quantitative features from radiomics and the high-level features extracted by deep learning to construct a more optimized combined model. This model achieved an AUC of 0.863, demonstrating high accuracy in predicting SDM risk in ccRCC patients. The combined model shows potential and practicality for clinical application, providing better clinical decision support for developing more personalized and precise treatment plans.\u003c/p\u003e \u003cp\u003eThis study has several limitations. First, the sample size was relatively small. Although the model performed well in the validation cohort, its robustness still needs further validation in larger, multicenter datasets. Second, owing to differences in CT scanning parameters, the standardization of image data processing still needs optimization to ensure the model's applicability across different devices and settings. Future research should explore further optimization and improvement of deep learning models based on enhancing data diversity. Additionally, combining data from other modalities (such as genetic information and molecular markers) should be considered when constructing a more comprehensive predictive model.\u003c/p\u003e \u003c/div\u003e"},{"header":"5. Conclusion","content":"\u003cp\u003eIn summary, by integrating deep learning and radiomics, this study provides a powerful tool for predicting SDM risk in ccRCC patients. The results demonstrate that this method has significant advantages in improving the efficiency and accuracy of metastatic prediction.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eSDM: synchronous distant metastasis\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eccRCC:\u0026nbsp;\u003c/strong\u003eclear cell renal cell carcinoma\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eLASSO:\u0026nbsp;\u003c/strong\u003eleast absolute shrinkage and selection operator\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSVM:\u0026nbsp;\u003c/strong\u003esupport vector machine\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAUC:\u003c/strong\u003e area under the curve\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eROI:\u0026nbsp;\u003c/strong\u003eregions of interest\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eICC:\u0026nbsp;\u003c/strong\u003eintraclass correlation coefficient \u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eDCA:\u003c/strong\u003e decision curve analysis\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe thank Gansu Provincial Hospital, Lanzhou University and Gansu University of Chinese Medicine for their guidance and advice during the implementation of this project; we thank the onekey AI platform for providing technical support for this study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis work was funded by grants from the following sources:\u003c/p\u003e\n\u003cp\u003eThe Natural Science Foundation of Gansu Province (No.22JR5RA650)\u003c/p\u003e\n\u003cp\u003eGansu Provincial Hospital Scientific Research Foundation (No.23GSSYD-12)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors\u0026rsquo; contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eGuarantor of integrity of the entire study: ZFH; Design of the research: ZFH, HWB, ZWB; literature retrieve: HWB, ZWB; information collection of patients: ZWB, ZYF, YZJ; supervised learning and statistical analysis: ZYF, YZJ, HH, WJ; manuscript preparation: HWB; manuscript review: ZFH. All the authors read and approved the final manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll data supporting the findings of this study are available within the paper and its Supplementary material.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eEthical standards and informed consent\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis retrospective study was approved by the Ethics Committee of Gansu Provincial Hospital and the requirement for obtaining informed consent was waived by the institutional review board. Patients\u0026rsquo; records were anonymized and deidentified prior to analysis. The study was conducted according to the guidelines of the Declaration of Helsinki.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests:\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAll authors declare that they have no competing interests.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eBahadoram S, Davoodi M, Hassanzadeh S, Bahadoram M, Barahman M, Mafakher L. Renal cell carcinoma: an overview of the epidemiology, diagnosis, and treatment. G Ital Nefrol. 2022;39(3):2022-vol3. Published 2022 Jun 20.\u003c/li\u003e\n \u003cli\u003eBaston C, Parosanu AI, Stanciu IM, Nitipir C. Metastatic Kidney Cancer: Does the Location of the Metastases Matter? Moving towards Personalized Therapy for Metastatic Renal Cell Carcinoma. Biomedicines. 2024;12(5):1111. Published 2024 May 16.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eDonskov F, Xie W, Overby A, et al. Synchronous Versus Metachronous Metastatic Disease: Impact of Time to Metastasis on Patient Outcome-Results from the International Metastatic Renal Cell Carcinoma Database Consortium. Eur Urol Oncol. 2020;3(4):530-539.\u003c/li\u003e\n \u003cli\u003eNizam A, Schindelheim JA, Ornstein MC. The role of active surveillance and cytoreductive nephrectomy in metastatic renal cell carcinoma. Cancer Treat Res Commun. Published online February 22, 2020.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eMonda SM, Carney BW, May AM, et al. Differences in mutations across tumour sizes in clear-cell renal cell carcinoma. BJU Int. Published online September 12, 2024.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eJiang W, Shou J, Shi H, et al. Impact of Primary Tumor Size on Prognosis in Patients With Metastatic Renal Cell Carcinoma Receiving Cytoreductive Nephrectomy: A Population Study of a Chinese Center and the US SEER Database. Technol Cancer Res Treat. 2021;20:15330338211019507.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eMu W, Schabath MB, Gillies RJ. Images are data: challenges and opportunities in the clinical translation of radiomics. Cancer Res. 2022;82(11):2066-2068.\u003c/li\u003e\n \u003cli\u003eNie P, Liu S, Zhou R, et al. A preoperative CT-based deep learning radiomics model in predicting the stage, size, grade and necrosis score and outcome in localized clear cell renal cell carcinoma: A multicenter study. Eur J Radiol. 2023;166: 111018.\u003c/li\u003e\n \u003cli\u003eYu S, Yang Y, Wang Z, et al. CT-based conventional radiomics and quantification of intratumoral heterogeneity for predicting benign and malignant renal lesions. Cancer Imaging. 2024;24(1):130. Published 2024 Oct 2.\u003c/li\u003e\n \u003cli\u003eZhu P, Dai C, Xiong Y, et al. Tumor contour irregularity on preoperative CT predicts prognosis in renal cell carcinoma: a multi-institutional study. EClinicalMedicine. 2024;75:102775. Published 2024 Aug 16.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eDai C, Xiong Y, Zhu P, et al. Deep Learning Assessment of Small Renal Masses at Contrast-enhanced Multiphase CT. Radiology. 2024;311(2):e232178.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eSalvi M, Acharya UR, Molinari F, Meiburger KM. The impact of pre- and post-image processing techniques on deep learning frameworks: A comprehensive review for digital pathology image analysis. Comput Biol Med. 2021;128:104129.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eTong N, Xu Y, Zhang J, Gou S, Li M. Robust and efficient abdominal CT segmentation using shape constrained multi-scale attention network. Phys Med. 2023;110:102595.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eYang Y, Wang J, Ren Q, et al. Multimodal data integration using machine learning to predict the risk of clear cell renal cancer metastasis: a retrospective multicentre study. Abdom Radiol (NY). 2024;49(7):2311-2324.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eM\u0026uuml;hlbauer J, Egen L, Kowalewski KF, et al. Radiomics in Renal Cell Carcinoma-A Systematic Review and Meta-Analysis. \u003cem\u003eCancers (Basel)\u003c/em\u003e. 2021; 13(6):1348. Published 2021 Mar 17. \u0026nbsp;\u003c/li\u003e\n \u003cli\u003eMonda SM, Lui HT, Pratsinis MA, Chandrasekar T, Evans CP, Dall\u0026apos;Era MA. The Metastatic Risk of Renal Cell Carcinoma by Primary Tumor Size and Subtype. Eur Urol Open Sci. 2023;52:137-144. Published 2023 May 10.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eGallan AJ, Antic T. Clinicopathologic Features of Small Renal Masses Associated With Distant Metastatic Disease . Am J Clin Pathol. 2020;153(5):613-617.\u003c/li\u003e\n \u003cli\u003eArndt C, G\u0026uuml;ttler F, Heinrich A, B\u0026uuml;rckenmeyer F, Diamantis I, Teichgr\u0026auml;ber U. Deep Learning CT Image Reconstruction in Clinical Practice. CT-Bildrekonstruktion mit Deep Learning in der klinischen Praxis. Rofo. 2021;193(3):252-261.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eKoetzier LR, Mastrodicasa D, Szczykutowicz TP, et al. Deep Learning Image Reconstruction for CT: Technical Principles and Clinical Prospects. Radiology. 2023;306(3):e221257.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eBai X, Huang Q, Zuo P, et al. MRI radiomics-based nomogram for individualised prediction of synchronous distant metastasis in patients with clear cell renal cell carcinoma. Eur Radiol. 2021;31(2):1029-1042.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eYu X, Gao L, Zhang S, et al. Development and validation of A CT-based radiomics nomogram for prediction of synchronous distant metastasis in clear cell renal cell carcinoma. Front Oncol. 2023;12:1016583. Published 2023 Jan 4.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eXu Y, Quan R, Xu W, Huang Y, Chen X, Liu F. Advances in Medical Image Segmentation: A Comprehensive Review of Traditional, Deep Learning and Hybrid Approaches. Bioengineering (Basel). 2024;11(10):1034. Published 2024 Oct 16.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eXi IL, Zhao Y, Wang R, et al. Deep Learning to Distinguish Benign from Malignant Renal Lesions Based on Routine MR Imaging. Clin Cancer Res. 2020;26(8):1944-1952.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eDemircioğlu A. Are deep models in radiomics performing better than generic models? A systematic review. Eur Radiol Exp. 2023;7(1):11. Published 2023 Mar 15.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eXv Y, Wei Z, Jiang Q, et al. Three-Dimensional (3D) deep learning model complements existing models for preoperative disease-free survival prediction (DFS) in localized clear cell renal cell carcinoma (ccRCC): A multicenter retrospective cohort study. Int J Surg. Published online June 19, 2024.\u0026nbsp;\u003c/li\u003e\n\u003c/ol\u003e"},{"header":"Tables","content":" \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 \u003cdiv class=\"SimplePara\"\u003eComparison of clinical data between training and validation cohort patients\u003c/div\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"7\"\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 \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\" morerows=\"1\" rowspan=\"2\"\u003e \u003cdiv class=\"SimplePara\"\u003eVariables\u003c/div\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eTraining group (n\u0026thinsp;=\u0026thinsp;143)\u003c/div\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c7\" namest=\"c5\"\u003e \u003cdiv class=\"SimplePara\"\u003eValidation group (n\u0026thinsp;=\u0026thinsp;62)\u003c/div\u003e \u003c/th\u003e \u003c/tr\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eSDM(+)(n\u0026thinsp;=\u0026thinsp;58)\u003c/div\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cdiv class=\"SimplePara\"\u003eSDM(-)(n\u0026thinsp;=\u0026thinsp;85)\u003c/div\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cdiv class=\"SimplePara\"\u003e\u003cspan type=\"Italic\" class=\"Italic\" name=\"Emphasis\"\u003eP\u003c/span\u003e\u003c/div\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cdiv class=\"SimplePara\"\u003eSDM(+)(n\u0026thinsp;=\u0026thinsp;19)\u003c/div\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cdiv class=\"SimplePara\"\u003eSDM(-)(n\u0026thinsp;=\u0026thinsp;43)\u003c/div\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cdiv class=\"SimplePara\"\u003e\u003cspan type=\"Italic\" class=\"Italic\" name=\"Emphasis\"\u003eP\u003c/span\u003e\u003c/div\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003eSex,n(%)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.190\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.710\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003eMale\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003e41 (70.69)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cdiv class=\"SimplePara\"\u003e51 (60.00)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cdiv class=\"SimplePara\"\u003e12 (63.16)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cdiv class=\"SimplePara\"\u003e25 (58.14)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003eFemale\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003e17 (29.31)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cdiv class=\"SimplePara\"\u003e34 (40.00)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cdiv class=\"SimplePara\"\u003e7 (36.84)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cdiv class=\"SimplePara\"\u003e18 (41.86)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e\u0026nbsp;\u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003eAge\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003e61.50\u003c/div\u003e \u003cdiv class=\"SimplePara\"\u003e(51.00, 72.00)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cdiv class=\"SimplePara\"\u003e58.00\u003c/div\u003e \u003cdiv class=\"SimplePara\"\u003e(51.00, 67.00)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.140\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cdiv class=\"SimplePara\"\u003e59.00\u003c/div\u003e \u003cdiv class=\"SimplePara\"\u003e(50.50, 63.50)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cdiv class=\"SimplePara\"\u003e58.00\u003c/div\u003e \u003cdiv class=\"SimplePara\"\u003e(52.00, 65.00)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.994\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003eSize\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003e6.15 (5.00, 7.07)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cdiv class=\"SimplePara\"\u003e5.00 (3.00, 7.00)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.003\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cdiv class=\"SimplePara\"\u003e5.80 (4.75, 6.80)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cdiv class=\"SimplePara\"\u003e5.05 (3.42, 7.00)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.739\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003eBUN\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003e5.96 (4.88, 6.68)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cdiv class=\"SimplePara\"\u003e5.90 (4.42, 7.27)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.702\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cdiv class=\"SimplePara\"\u003e5.32 (4.02, 6.38)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cdiv class=\"SimplePara\"\u003e4.94 (4.10, 6.29)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.658\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003eCr\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003e71.15\u003c/div\u003e \u003cdiv class=\"SimplePara\"\u003e(61.73, 83.78)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cdiv class=\"SimplePara\"\u003e69.22\u003c/div\u003e \u003cdiv class=\"SimplePara\"\u003e(57.70, 85.00)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.353\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cdiv class=\"SimplePara\"\u003e71.70\u003c/div\u003e \u003cdiv class=\"SimplePara\"\u003e(58.15, 88.70)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cdiv class=\"SimplePara\"\u003e74.50\u003c/div\u003e \u003cdiv class=\"SimplePara\"\u003e(61.60, 90.50)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.551\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003eBMI\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003e24.77\u003c/div\u003e \u003cdiv class=\"SimplePara\"\u003e(21.36, 26.85)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cdiv class=\"SimplePara\"\u003e24.21\u003c/div\u003e \u003cdiv class=\"SimplePara\"\u003e(21.12, 26.20)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.355\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cdiv class=\"SimplePara\"\u003e25.39\u003c/div\u003e \u003cdiv class=\"SimplePara\"\u003e(23.99, 27.16)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cdiv class=\"SimplePara\"\u003e22.59\u003c/div\u003e \u003cdiv class=\"SimplePara\"\u003e(21.05, 25.16)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.023\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"7\"\u003eNote: BUN: Blood Urea Nitrogen; Cr: Creatinine; BMI: Body Mass Index.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003cbr/\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 \u003cdiv class=\"SimplePara\"\u003ePredictive performance of different models\u003c/div\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\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 \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003eModel\u003c/div\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003eAUC(95%CI)\u003c/div\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cdiv class=\"SimplePara\"\u003eSensitivity\u003c/div\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cdiv class=\"SimplePara\"\u003eSpecificity\u003c/div\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cdiv class=\"SimplePara\"\u003ePPV\u003c/div\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cdiv class=\"SimplePara\"\u003eNPV\u003c/div\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cdiv class=\"SimplePara\"\u003eF1\u003c/div\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cdiv class=\"SimplePara\"\u003eGroup\u003c/div\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003edensenet121\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.735(0.652\u0026ndash;0.818)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.656\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.774\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.831\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.569\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.733\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cdiv class=\"SimplePara\"\u003eTraining\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003edensenet121\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.664(0.522\u0026ndash;0.807)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.579\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.792\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.815\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.543\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.677\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cdiv class=\"SimplePara\"\u003eValidation\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003edensenet169\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.665(0.575\u0026ndash;0.755)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.433\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.849\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.830\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.469\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.569\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cdiv class=\"SimplePara\"\u003eTraining\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003edensenet169\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.681(0.538\u0026ndash;0.825)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.684\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.625\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.743\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.556\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.712\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cdiv class=\"SimplePara\"\u003eValidation\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003eresnet101\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.815(0.742\u0026ndash;0.887)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.800\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.698\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.818\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.673\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.809\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cdiv class=\"SimplePara\"\u003eTraining\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003eresnet101\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.743(0.614\u0026ndash;0.872)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.684\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.750\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.812\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.600\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.743\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cdiv class=\"SimplePara\"\u003eValidation\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003eresnet152\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.726(0.643\u0026ndash;0.809)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.444\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.868\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.851\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.479\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.584\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cdiv class=\"SimplePara\"\u003eTraining\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003eresnet152\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.727(0.593\u0026ndash;0.861)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.658\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.750\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.806\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.581\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.725\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cdiv class=\"SimplePara\"\u003eValidation\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003eresnet18\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.694(0.604\u0026ndash;0.785)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.811\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.528\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.745\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.622\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.777\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cdiv class=\"SimplePara\"\u003eTraining\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003eresnet18\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.678(0.538\u0026ndash;0.818)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.684\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.625\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.743\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.556\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.712\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cdiv class=\"SimplePara\"\u003eValidation\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003eresnet34\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.786(0.709\u0026ndash;0.863)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.789\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.679\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.807\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.655\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.798\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cdiv class=\"SimplePara\"\u003eTraining\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003eresnet34\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.617(0.462\u0026ndash;0.772)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.921\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.333\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.686\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.727\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.787\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cdiv class=\"SimplePara\"\u003eValidation\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003eresnet50\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.690(0.603\u0026ndash;0.777)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.478\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.811\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.811\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.478\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.601\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cdiv class=\"SimplePara\"\u003eTraining\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cdiv class=\"SimplePara\"\u003eresnet50\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.697(0.551\u0026ndash;0.843)\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.737\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.625\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.757\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.600\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cdiv class=\"SimplePara\"\u003e0.747\u003c/div\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cdiv class=\"SimplePara\"\u003eValidation\u003c/div\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003cbr/\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"discover-oncology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"dion","sideBox":"Learn more about [Discover Oncology](https://www.springer.com/12672)","snPcode":"","submissionUrl":"","title":"Discover Oncology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Radiomics, Deep Learning, Renal Clear Cell Carcinoma, Simultaneous Distant Metastasis","lastPublishedDoi":"10.21203/rs.3.rs-5368462/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5368462/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eObjective\u003c/h2\u003e \u003cp\u003eThe objective of this research was to devise and authenticate a predictive model that employs CT radiomics and deep learning methodologies for the accurate prediction of synchronous distant metastasis (SDM) in clear cell renal cell carcinoma (ccRCC).\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eA total of 143 ccRCC patients were included in the training cohort, and 62 ccRCC patients were included in the validation cohort. The CT images from all patients were normalized, and the tumor regions were manually segmented via ITK-SNAP software. Radiomic features were extracted via the FAE toolkit. The least absolute shrinkage and selection operator (LASSO) algorithm was employed to select features and build various machine learning models. Additionally, the largest cross-section of the tumor was cropped to train the deep learning model. Multiple deep learning models were trained to predict SDM in ccRCC patients. The results of the best machine learning model were then fused with those of the deep learning model to create a combined model.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eOf the 944 radiomic features identified, 15 were closely associated with SDM. With these 15 features, the support vector machine (SVM) model emerged as the most effective, demonstrating areas under the curve (AUC) of 0.860 and 0.813 in the training and validation cohort, respectively. Among the deep learning models, ResNet101 performed optimally, achieving AUC of 0.815 and 0.743 in the training and validation cohort, respectively. The combined model yielded an AUC of 0.863. Decision curve analysis suggested that the combined model offers superior clinical applicability.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eThis model, which combines radiomics and deep learning, has substantial potential the prediction of SDM in patients with ccRCC. This study is anticipated to bolster clinical decision-making processes and enhance the prognostic outcomes for individuals diagnosed with ccRCC.\u003c/p\u003e","manuscriptTitle":"The predictive value of radiomics and deep learning for synchronous distant metastasis in clear cell renal cell carcinoma","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-12-17 07:04:25","doi":"10.21203/rs.3.rs-5368462/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2024-12-03T11:25:00+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-11-30T13:52:34+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-11-23T22:39:33+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"60450139113600049972957436333127300810","date":"2024-11-23T12:01:45+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"300735158308813737528850322766738046795","date":"2024-11-21T20:10:27+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-11-21T09:29:14+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-11-14T05:56:16+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-11-13T04:42:26+00:00","index":"","fulltext":""},{"type":"submitted","content":"Discover Oncology","date":"2024-10-31T17:08:13+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"discover-oncology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"dion","sideBox":"Learn more about [Discover Oncology](https://www.springer.com/12672)","snPcode":"","submissionUrl":"","title":"Discover Oncology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Discover Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"455a4318-e265-4a31-a923-4bc6bba966c5","owner":[],"postedDate":"December 17th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2025-01-13T09:09:19+00:00","versionOfRecord":[],"versionCreatedAt":"2024-12-17 07:04:25","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5368462","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5368462","identity":"rs-5368462","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}