Digital Staining of Unpaired White and Blue Light Cystoscopy Videos for Bladder Cancer Detection in the Clinic

Shuang Chang, Haoli Yin, Kristen R Scarpato, Amy N Luckenbaugh, Sam Chang, Christian Bolenz, Maximilian C Kriegmair, Nikolaos C Deliolanis, Soheil Kolouri, Audrey Bowden

Keywords: Bladder cancer, cystoscopy, style transfer, deep learning

TL;DR: We performed digital staining of white light cystoscopy images and demonstrate that the resulting digitally generated blue light cystoscopy (BLC) images show strong resemblance to ground truth BLC images.

Abstract: Blue light cystoscopy (BLC) has been shown to detect bladder tumors with better sensitivity than white light cystoscopy (WLC); however, its increased cost and dye administration time have challenged widespread adoption of the technology. Here, we demonstrate a low-cost strategy to generate BLC images directly from WLC images. We performed digital staining of WLC images obtained from tumor resection procedures and demonstrate that the resulting digitally generated BLC images show strong resemblance to ground truth BLC images, with negligible degradation of the image quality.