Mp30-15 Machine Learning to Distinguish Benign and Malignant Renal Lesions Based on Routine Ct Imaging

You have accessJournal of UrologyImaging/Uroradiology II (MP30)1 May 2024MP30-15 MACHINE LEARNING TO DISTINGUISH BENIGN AND MALIGNANT RENAL LESIONS BASED ON ROUTINE CT IMAGING Shuanbao Yu, Gaurab Pokhrel, Jinshan Cui, Jin Tao, and Xuepei Zhang Shuanbao YuShuanbao Yu , Gaurab PokhrelGaurab Pokhrel , Jinshan CuiJinshan Cui , Jin TaoJin Tao , and Xuepei ZhangXuepei Zhang View All Author Informationhttps://doi.org/10.1097/01.JU.0001009416.90901.7b.15AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: To evaluate the performance of a radiomics model in distinguishing between benign and malignant renal lesions, and determine whether it improves radiologists' diagnostic performance. METHODS: Preoperative CT images of 1,395 renal lesions with definitive pathological diagnosis from three hospital sites were divided into development (941 cases, from December 2011 to June 2020) and test cohorts (454 cases, from July 2020 to December 2021). Radiomic features were extracted from three-phase 3D CT images to build a radiomics signature. The final radiomics model was developed by incorporating radiomics signature, clinical factors, and CT reported results. RESULTS: The radiomics signature based on combined features and stacking classifier, yielded the highest AUC of 0.847 among all combinations in the test cohort. After integrating clinical factors and CT reported results into the radiomics signature, the radiomics model achieved highest performance than the interpretation of radiologists (p=0.001) in both the whole test cohort (AUC=0.919) and small renal lesions (AUC=0.895). The model also demonstrated the highest concordance, and net benefit across threshold probabilities exceeding 60%. The AUC values for subgroup with predicted risk below or above cutoff values for 95% sensitivity or specificity were 0.964 and 0.963 in the whole test cohort and small renal lesions, respectively. CONCLUSIONS: The radiomics model effectively distinguished benign from malignant renal lesions with good discrimination, compared with radiologists' interpretation. The radiomics model could be utilized with more confidence in clinical practice, when the predicted risk is below or above cutoff values for 95% sensitivity or specificity. Download PPTDownload PPT Source of Funding: Henan Provincial Youth Science Foundation Project grant no. 232300420254 and Henan Provincial Medical Science and Technology Tackling Key Project of Province-Ministry Co-construction grant no. SBGJ202102140 © 2024 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 211Issue 5SMay 2024Page: e498 Advertisement Copyright & Permissions© 2024 by American Urological Association Education and Research, Inc.Metrics Author Information Shuanbao Yu More articles by this author Gaurab Pokhrel More articles by this author Jinshan Cui More articles by this author Jin Tao More articles by this author Xuepei Zhang More articles by this author Expand All Advertisement PDF downloadLoading ...