PD02-11 UNLOCKING THE POWER OF MRI: DEEP LEARNING FOR ASSESSMENT OF MUSCLE-INVASIVE BLADDER CANCER IN VI-RADS 3

You have accessJournal of UrologyBladder Cancer: Invasive I (PD02)1 May 2024PD02-11 UNLOCKING THE POWER OF MRI: DEEP LEARNING FOR ASSESSMENT OF MUSCLE-INVASIVE BLADDER CANCER IN VI-RADS 3 Qiang Lyu, Xiao Yang, Qiang Cao, and Lingkai Cai Qiang LyuQiang Lyu , Xiao YangXiao Yang , Qiang CaoQiang Cao , and Lingkai CaiLingkai Cai View All Author Informationhttps://doi.org/10.1097/01.JU.0001008836.73392.92.11AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: There is a lack of clarity in the relationship between tumors and the muscle layer in VI-RADS 3. This study aims to develop deep learning (DL) models based on MRI to predict muscle-invasive bladder cancer (MIBC) in VI-RADS 3. METHODS: Preoperative MRI images of patients with bladder cancer from 2012 to 2023 were collected, and a total of 233 patients of VI-RADS 3 were included in this study according to the include and exclude criteria. The models were based on a Rennet-50 framework and constructed using T2, DWI, and T2&DWI imaging modalities. A cut off for p-value 0.5 was used for assessing MIBC. The best-performing model in five-fold cross-validation were reported in test set. The performance of the model was verified by AUC, accuracy, sensitivity and specificity, and compared with radiologist (reader 1) and senior urologist (reader 2). RESULTS: The muscle-invasive states of 233 patients were 81(34.7%) in MIBC and 152(65.2%) in NMIBC. In the validation set, the mean AUC, accuracy, sensitivity, and specificity of five-fold cross-validation were 0.825, 75.4%, 50.7%, 88.4% in T2 model, 0.717, 66.9%, 50.8%, 75.4%, and 0.782, 73.8%, 66.8%, 52.3%, 85.2%. In the test set, the mean AUC, accuracy, sensitivity, and specificity of five-fold cross-validation are 0.749, 74.7%, 53.7%, 86.2% in T2 model, 0.818, 74.2%, 60.0%, 82.1% in DWI model, and 0.750, 74.7%, 53.7%, 86.2% in T2&DWI model respectively. The results assessed by clinicians showed that the accuracy, sensitivity, and specificity were 64.3%, 57.1%, 67.9% in reader 1 and 59.5%, 64.3%, 57.1% in reader 2. CONCLUSIONS: The deep learning model, based on MRI, has been developed to detect the muscle infiltration status of patients with VI-RADS 3 and the performance of the model surpasses that of both radiologist and senior urologist. Download PPT Source of Funding: This study has received funding by the National Natural Science Foundation of China. (No: 82072832, 82273152) © 2024 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 211Issue 5SMay 2024Page: e74 Advertisement Copyright & Permissions© 2024 by American Urological Association Education and Research, Inc.Metrics Author Information Qiang Lyu More articles by this author Xiao Yang More articles by this author Qiang Cao More articles by this author Lingkai Cai More articles by this author Expand All Advertisement PDF downloadLoading ...