AI-based anatomical recognition for bladder neck dissection in robot-assisted radical prostatectomy

Area of Science:

• Robotics and Artificial Intelligence in Surgery
• Surgical Anatomy and Imaging

Background:

• Robot-assisted radical prostatectomy (RARP) involves complex bladder neck dissection.
• Accurate identification of anatomical structures is crucial for successful RARP outcomes.
• Current methods for anatomical recognition can be time-consuming and require significant expertise.

Purpose of the Study:

• To develop an artificial intelligence (AI) system for automated anatomical recognition during RARP.
• To identify key anatomical structures, including the prostate and bladder, during bladder neck dissection.
• To enhance surgical precision and potentially improve training for novice surgeons.

Main Methods:

• An AI model, specifically DeepLabV3+, was developed for anatomical recognition.
• The model was trained and evaluated using 210 RARP videos from two institutions.
• An active learning approach with iterative retraining was employed to enhance model performance.
• Segmentation accuracy was assessed using Intersection over Union (IoU) and Dice similarity coefficient.

Main Results:

• The initial AI model achieved IoU values of 0.41 for the prostate and 0.44 for the bladder.
• After retraining, the AI model's IoU improved to 0.75 for the prostate and 0.68 for the bladder.
• The retrained model demonstrated consistent segmentation performance on an independent test set from seven surgeons.

Conclusions:

• An accurate AI system for anatomical recognition during RARP bladder neck dissection was successfully developed.
• Iterative retraining significantly improved the AI model's segmentation accuracy.
• The developed AI system is expected to assist surgeons and facilitate the education of new surgeons in RARP.