(UroToday.com) Cystoscopy is a vital step that is involved in many urological procedures. With the advancement of artificial intelligence (AI) and its incorporation as a medical tool, Dr. Maya Srinath and her team believe that AI can be integrated with cystoscopy. As such, she brilliantly presents her work on the testing and feasibility of an AI-based image analysis model to detect anatomic landmarks in the male urethra and bladder during cystoscopy.
For her study, Dr. Srinath included 276 individual images from 6 cystourethroscopy videos obtained from 6 male patients. These videos were segmented on the Roboflow web platform and used to train the Roboflow 3.0 Object Detection model. Of these images, 70% of the images were used to train the model, and 20% of the images were used to validate the model to ensure that the model was truly learning rather than memorizing. The remaining 10% of the images were used to test the model’s precision in identifying these anatomical landmarks. 2 reviewers reviewed the videos and annotated the following anatomical landmarks: bladder neck, bladder wall, left ureteral orifice, right ureteral orifice, prostatic urethra, ureteral sphincter, verumontanum, trigone, and penile urethra (Figure 1).
Figure 1. Example of the annotated anatomical landmarks of the bladder and urethra on cystoscopy.
Dr. Srinath stated that the results were very promising as the computer vision model achieved high precision for most anatomical landmarks with a mean average precision of 79.9% and a recall of 74.4% (Figure 2). Of note, trigone is 0% due to poor visualization and quality of the trigone images, that made trigone identification hard to distinguish from the bladder wall.
Figure 2. Mean average precision of each anatomical landmark.
Dr. Srinath concludes with a strong statement that an AI image analysis model can be successful in identifying anatomical landmarks, with its potential being limitless.
Afterwards, a member of the audience asked what the angles of the images were, to which Dr. Srinath replied that all the images were observed from different angles. Additionally, she thought that looking more specifically at the image angles, as well as teaching the AI model with image angles, would be interesting moving forward. Moderator Dr. Robert Sweet then asked how this model would be used and what other dimensions could be added to the model. Dr. Srinath answers that the next step is in progress with the application of the model for bladder tumor identification, and lastly, the model can be combined with pathological specimens to possibly create a risk stratification model.
Presented by: Maya Srinath, PGY-5, The Smith Institute for Urology, Northwell Health, @SrinathMaya on X
Written by: Victor Pham, BS, University of California Irvine, @victorpham01 on X during the 2025 World Congress of Endourology and Uro-Technology (WCET) Annual Meeting: September 8 – September 12, 2025, Phoenix, Arizona