AUA 2018: Automated and Dynamic Classification of Bladder Cancer using Deep Learning on Real-Time Confocal Laser Endomicroscopy Images

San Francisco, CA USA ( Dr. Timothy Chang, urology resident at Stanford University, presented on the use of artificial intelligence to classify bladder cancer by analyzing real-time confocal laser endomicroscopy (CLE) images. As an introduction, the author explained the function of CLE and its ability to provide real time microscopic tissue characterization. The goal of this study was to apply deep learning, which is a form of machine learning that involves object detection, to identify bladder cancer in tissues using these CLE images.

CLE images were obtained from 81 patients who underwent cystoscopy/TURBT. These CLE images were then using to train the algorithm to provide proper diagnostic of bladder tissue. The accuracy of the deep learning algorithm was test by radiologists to ensure it was outputting accurate results. The presenter also displayed a video of the algorithm being used in the operating room, providing instant feedback for classifying bladder cancer during cystoscopy. The author noted that the software was running on an older Mac model to show that the software has room for improvement if a stronger computer with better processing power is utilized.

Dr. Chang noted the speed in which the algorithm output analysis of the bladder tissue is 1 frame per second. The speed and accuracy of diagnosis was stressed for this could allow very accurate determination of tumor border and margins. The author concluded that the use of CLE with deep learning was shown to accurately diagnose malignancy versus benign in 87% of cases.

Possible future routes using this technology was discussed, specifically in terms of real-time diagnosis of the upper urinary tract as well as more challenging diagnosis such as high and low-grade bladder carcinoma in situ (CIS). The moderator mentioned in the question section that truly ensure that the artificial intelligence can accurately diagnose bladder cancer, there would need to be an extremely large sample size. Another future route was to use this technology to check residual tumor after resection of bladder tumors.

Presented by Egor Parkhomenko

Written by: Luke Limfueco, MD, Department of Urology, University of California-Irvine at the 2018 AUA Annual Meeting - May 18 - 21, 2018 – San Francisco, CA USA