Computer vision for evaluating retraction of the neurovascular bundle during nerve-sparing prostatectomy

Area of Science:

• Urology
• Computer Vision
• Surgical Robotics

Background:

• Nerve-sparing prostatectomy aims to preserve sexual function.
• Excessive retraction on the neurovascular bundle (NVB) negatively impacts post-operative sexual recovery.
• Objective quantification of NVB retraction is needed to understand its effects.

Purpose of the Study:

• To quantitatively assess NVB countertraction using computer vision during prostatectomy.
• To evaluate the association between NVB retraction metrics and post-operative sexual function.
• To improve predictive models for sexual recovery using computer vision-derived features.

Main Methods:

• Sixty-four nerve-sparing prostatectomy videos were analyzed.
• Computer vision and semantic segmentation (CVAT) measured tissue stretch and grasp.
• Machine learning models (Random Forest, MLP, XGBoost) predicted erectile function at 12 months.

Main Results:

• Greater NVB angular deviation and tissue stretch were observed in patients with insufficient erectile function.
• Integrating computer vision features improved the predictive accuracy (AUC) of all machine learning models.
• The Random Forest model's AUC increased from 0.78 to 0.83 with added segmentation features.

Conclusions:

• Objective assessment of NVB retraction via computer vision provides valuable insights into sexual recovery.
• Minimizing NVB stretch and grasp during surgery is crucial for preserving erectile function.
• Computer vision enhances predictive capabilities beyond traditional clinical and surgical factors.