PD14-01 AUTOMATED MACHINE LEARNING SEGMENTATION AND MEASUREMENT OF URINARY STONES ON CT SCAN

Abstract

You have accessJournal of UrologyStone Disease: Epidemiology & Evaluation I (PD14)1 Sep 2021PD14-01 AUTOMATED MACHINE LEARNING SEGMENTATION AND MEASUREMENT OF URINARY STONES ON CT SCAN Rilwan Babajide, Katerina Lembrikova, Justin Ziemba, Yong Fan, and Gregory Tasian Rilwan BabajideRilwan Babajide More articles by this author , Katerina LembrikovaKaterina Lembrikova More articles by this author , Justin ZiembaJustin Ziemba More articles by this author , Yong FanYong Fan More articles by this author , and Gregory TasianGregory Tasian More articles by this author View All Author Informationhttps://doi.org/10.1097/JU.0000000000001990.01AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: Treatment decisions for patients presenting with urinary stones depend on multiple factors including stone size, location, and renal anatomy. Current methods depend on manual measurement of these parameters by humans, which introduces inter- and intra-observer variation, is laborious, and time-consuming. The objective of this study is to evaluate the performance of a machine learning algorithm to quickly and accurately automate measurement of stone features and renal anatomy. METHODS: A sample of 95 CT scans from patients who underwent assessment and imaging for suspected kidney stones were included in the study. Two raters manually measured kidney stones in 3 orthogonal dimensions, renal pelvis width, and ureter diameter from 46 scans. A two-way random intraclass correlation (ICC) score was calculated to quantify intrarater agreement. The remaining 49 scans were used to train a deep learning model to segment kidney stones from the surrounding kidney. Times for manual and machine calculations were recorded. RESULTS: The sample included 19 scans with kidney stones, 17 with ureteral stones, and 10 with both. Median time to measure stones in 3 dimensions was longer manually than with the machine algorithm (16.1 vs. 2.1 seconds). Intrarater reliability of manual measurements was poor for pelvis width (0.44, 95% CI 0.21 – 0.62) and ureter diameter (0.40, 95% CI 0.16 – 0.59), and good for stone size (0.79, 95% CI 0.75 – 0.83). The algorithm identified all stones present (100% sensitivity) with no false positive stones (100% specificity). Analyzing at the individual voxel level, the sensitivity of the algorithm for stone detection fell to 58%, while the specificity remained at 100%, using manual measurements as ground truth. Although the algorithm reliably captured the centers of stones, the total area of kidney stones identified by the machine was smaller than that identified by human raters (1,019.4 vs. 1,256.3 mm3; Figure 1). CONCLUSIONS: Manual measurements of kidney stones and anatomy on CT are limited by the time required and poor reproducibility. The more rapid and accurate measurements provided by the machine learning algorithm has a high probability to transform clinical care as it enhances and standardizes assessment across patients, institutions, and providers. Source of Funding: NA © 2021 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 206Issue Supplement 3September 2021Page: e215-e215 Advertisement Copyright & Permissions© 2021 by American Urological Association Education and Research, Inc.MetricsAuthor Information Rilwan Babajide More articles by this author Katerina Lembrikova More articles by this author Justin Ziemba More articles by this author Yong Fan More articles by this author Gregory Tasian More articles by this author Expand All Advertisement Loading ...