Automated Surgical Urethral Length Estimation for Robot-Assisted Radical Prostatectomy

Abstract

Urinary Incontinence (UI) is a major side effect of Robot-Assisted Radical Prostatectomy (RARP). The surgical Urethra Length (SUL) emerges as a crucial predictive factor for postoperative-RARP UI. In response, this study introduces a novel approach for the automated estimation of the SUL from surgical video frames. A dedicated RARP dataset was meticulously curated, placing emphasis on the segmentation of structures crucial for accurate SUL estimation. The dataset contains 282 frames extracted from 114 patients' videos and extra care was taken that all frames of the test set were suited for SUL estimation. Notably, each frame in the test set was annotated by both an expert urologist and a medical research fellow. Eventually, the predictions of the segmentation model are integrated into a heuristic method to determine the SUL. Despite training on a relatively small dataset, we have found a small mean difference between prediction and expert annotation SUL (1.86 ± 3.56 mm). This shows the future potential for automated SUL estimation from video, particularly when sufficient samples per patient are available.