(UroToday.com) The 2022 American Urological Association (AUA) Annual Meeting included a session on non-invasive bladder cancer and a presentation by Dr. Atsushi Ikeda discussing real-time bladder tumor detection at clinics in flexible cystoscopy with white light and narrow band imaging using deep learning. Non-muscle invasive bladder cancer (NMIBC) is the most common clinical condition observed during the diagnosis of bladder cancer and is often a chronic condition that requires frequent cystoscopy. Dr. Ikeda and colleagues developed a cystoscopy diagnostic support system using artificial intelligence to reduce tumor lesion oversight during flexible cystoscopy in clinics. The objective of this study was to validate the quality of probability maps showing lesion distribution detected by deep learning-based image identification methods for tumors in white light imaging and narrow-band imaging.
The authors used a dataset of 6,523 white-light imaging and 1,636 narrow-band imaging cystoscopy images (5,480 normal and 1,045 tumor images; and 1,354 normal and 282 tumor images, respectively), and divided them into training and testing datasets in an 8:2 ratio to create our artificial intelligence model based on the deep learning model ResNet50 pre-trained using ImageNet. A probability map showing the distribution of the abnormal lesions in the analyzed cystoscopy images was compared with the annotation data pixel-by-pixel to measure the diagnostic accuracy of the proposed artificial intelligence model. The results of artificial intelligence were considered accurate if the ratio of overlap between the estimated area and the area annotated by the doctor exceeded 50%, otherwise the results were considered inaccurate. The evaluation was performed using a five-fold cross-validation test. A summary of the evaluation method is as follows:
For the test dataset, containing 1,650 cystoscopy images (white-light imaging 217 and narrow-band imaging 64 images of bladders with tumors and white-light imaging 1,104 and narrow-band imaging 265 images of normal bladders), and at the threshold where the F1 score in the training dataset was the maximum, the proposed artificial intelligence model exhibited average sensitivity, specificity, and F1 scores of 84.4%, 81.2%, and 0.678, respectively.
Dr. Ikeda concluded this presentation by discussing real-time bladder tumor detection at clinics in flexible cystoscopy with white light and narrow band imaging using deep learning with the following take-home messages:
- By applying the analysis of still images to video, real-time detection of bladder tumor lesions for both white-light imaging and narrow-band imaging is possible, with artificial intelligence showing the probability map of the lesion sites using the proposed method
- This may improve diagnostic accuracy by creating awareness among urologists of all skill levels and by encouraging more robust observations
Presented by: Atsushi Ikeda, MD, PhD, Department of Urology, University of Tsukuba, Tsukuba, Japan
Co-Authors: Shogo Takaoka, Hirokazu Nosato, Hiromitsu Negoro, Hidenori Sakanashi, Masahiro Murakawa, Hiroyuki Nishiyama, Tsukuba, Japan
Written by: Zachary Klaassen, MD, MSc – Urologic Oncologist, Assistant Professor of Urology, Georgia Cancer Center, Augusta University/Medical College of Georgia, @zklaassen_md on Twitter during the 2022 American Urological Association (AUA) Annual Meeting, New Orleans, LA, Fri, May 13 – Mon, May 16, 2022.