Machine Vision Augmentation to Detect Detrusor Overactivity in Overactive Bladder: A Frontier of Artificial Intelligence Application in Functional Urology-Proof of Concept Clinical Study.

Introduction: Overactive bladder (OAB) is a common urological condition with increasing prevalence, especially in an aging population. Diagnosing and treating OAB can be challenging. While urodynamic study (UDS) is useful to confirm involuntary detrusor overactivity (DO), it is invasive, time-consuming, and requires good patient coordination, which limits its clinical utility. In this proof-of-concept clinical trial, we propose a novel method in which cystoscopic images can be augmented by machine vision to identify DO and detect OAB based on differences in vascular network motion over time.

Materials and methods: We prospectively extracted 30-second clips from 112 videos that were relatively artifact-free. This cohort consisted of 34 UDS confirmed DO and 78 non-OAB videos. Over 85 000 frames were then processed in the following manner: (A) De-noised to remove artifacts, (B) Super-resolution enhancement, (C) Segmentation and identification of keypoints along the vascular network, (D) Mosaic stitching of frames to reconstitute a 3D bladder map after accounting for geometric distortions, (E) Tracking of keypoint motion differences over time as a surrogate for areas for DO.

Results: The structure-from-motion pipeline demonstrated satisfactory 3D reconstructions of processed cystoscopy videos. Videos from OAB patients showed a mean of 113.9 pixel-deviations per time frame (SD 32.8). This is 324.5% of those in the non-OAB group, which had an average of 35.1 pixel-deviations (SD 31.3) (p < 0.001). The heatmap generated provided a topographical representation of the cystoscopic views, thus helping to identify key areas of increased focal detrusor contractions.

Conclusion: We describe a novel model leveraging AI machine vision to demonstrate statistically increased keypoint deviations on the detrusor vascular network of OAB patients as compared to non-OAB patients. This technology may potentially streamline the diagnosis of OAB and identify localized areas of increased DO for targeted treatment.