PD47-07 THE STUDY OF AUTOMATIC RECOGNITION OF STONE COMPONENTS USING DIGITAL IMAGES FROM INTRAOPERATIVE FLEXIBLE URETEROSCOPY

You have accessJournal of UrologyStone Disease: Surgical Therapy (Including ESWL) IV (PD47)1 May 2024PD47-07 THE STUDY OF AUTOMATIC RECOGNITION OF STONE COMPONENTS USING DIGITAL IMAGES FROM INTRAOPERATIVE FLEXIBLE URETEROSCOPY Daxun Luo, Bixiao Wang, Yubao Liu, Haifeng Song, Weiguo Hu, and Jianxing Li Daxun LuoDaxun Luo , Bixiao WangBixiao Wang , Yubao LiuYubao Liu , Haifeng SongHaifeng Song , Weiguo HuWeiguo Hu , and Jianxing LiJianxing Li View All Author Informationhttps://doi.org/10.1097/01.JU.0001008652.62443.0a.07AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: This study aimed to investigate whether urinary stones can be predicted by convolutional neural network (CNN) using flexible ureteroscopic (fURS) images. METHODS: This study retrospectively used surgical videos of fURS performed by a single surgeon using single-use electronic ureteroscopes (Zebra) between 2022 and 2023 . Digital images of the surfaces of the stones before laser fragmentation and relatively intact cross-sections of stones were captured from the surgical video. The cases were categorized into five groups based on postoperative infrared spectroscopy analysis: calcium oxalate group, calcium oxalate mixed with uric acid group, calcium oxalate mixed with carbonate apatite group, struvite mixed with calcium oxalate mixed with carbonate apatite group, and a control group without stones. A total of 372 images were finally included and divided into training, validation and test sets. In the CNN model, ResNet-152-V2 model was used, and to enhance the network's generalization capability, data augmentation was applied to expand the training dataset. Only endoscopic digital images and stone classification data were input to achieve minimal supervised learning (Fig 1). RESULTS: There were 113 cases in the calcium oxalate group, 19 cases in the calcium oxalate mixed with uric acid group, 134 cases in the calcium oxalate mixed with carbonate apatite group, 19 cases in the struvite mixed with calcium oxalate mixed with carbonate apatite group, and 67 cases in the control group without stones. After whole training, the total accuracy was 98.0% on validation set. After training and validation, the model was tested using the test set consisting of 26 cases with a total accuracy of 80.8%. The recall, specificity and precision of the test result were 75%, 88.9%, and75% in calcium oxalate group, 50%, 100%, and 100% in the calcium oxalate mixed with uric acid group, 83.3%, 100%, and 100% in the calcium oxalate mixed with carbonate apatite group, and 100% for all three metrics in the struvite mixed with calcium oxalate mixed with carbonate apatite group. The control group without stones had values of 100%, 85%, and 66.7% for the three metrics, respectively. CONCLUSIONS: This preliminary study suggests that deep CNN is a promising method for identifying the composition of renal stones from endoscopic images obtained during surgery. Both pure and mixed stone compositions can be distinguished. Surface and cross-sectional images collected in a clinical setting analyzed by deep CNN can provide valuable information about the morphology of stones for computer-aided diagnosis. Download PPT Source of Funding: No source of funds © 2024 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 211Issue 5SMay 2024Page: e982 Advertisement Copyright & Permissions© 2024 by American Urological Association Education and Research, Inc.Metrics Author Information Daxun Luo More articles by this author Bixiao Wang More articles by this author Yubao Liu More articles by this author Haifeng Song More articles by this author Weiguo Hu More articles by this author Jianxing Li More articles by this author Expand All Advertisement PDF downloadLoading ...