AUA 2019: Deep Learning Computer Vision Algorithm For Detecting Kidney Stone Composition: Towards An Automated Future
Pure kidney stones of greater than 90% composition of Calcium Oxalate Monohydrate, Uric Acid, Magnesium Ammonium Phosphate Hexahydrate, Calcium Hydrogen Phosphate Dihydrate and Cystine stones were obtained from human patients and each photographed twice. In each image, the background was removed and presented to the deep learning computer to detect and categorize a type of kidney stone. Once the machine was able to learn each feature, a total of 63 stones were then photographed and presented to the computer to test accuracy in detection of stone composition.
The performance of this deep learning machine produced an accuracy of 90% for Calcium Oxalate Monohydrate, 94% for Uric Acid, 86% for Magnesium Ammonium Phosphate Hexahydrate, 71% for Calcium Hydrogen Phosphate Dihydrate, and 75% for Cystine stones. Given these results, there is great potential for machine learning to accurately identify various pure kidney stones.
There are various limitations for this study. For example, only pure kidney stones were examined and this study does not account for the various hybrids of stone compositions produced by our general population. Further, these stones are only examined by their exterior surface, as opposed to the ideal cross-section analysis of stones. Though these limitations exist, there is huge potential to the applicability of this algorithm such as attaching the machine to a ureteroscope that would allow evaluation of these stones during surgery.
Authors: Ali H Aldoukhi1, Hei Law2, Kristian M Black1, William W Roberts1, Jiang Deng2, Khurshid R Ghani1
Authors Affiliation: 1University of Michigan, 2Princeton University
Presented by: Kristian Black, medical student, University of Michigan Medical School
Written by: Sherry Lu, Department of Urology, University of California-Irvine, at the American Urological Association's 2019 Annual Meeting (AUA 2019), May 3 – 6, 2019 in Chicago, Illinois