Surgical skill assessment using an AI-based surgical phase recognition model for laparoscopic cholecystectomy

Automated surgical skill assessment using artificial intelligence (AI) in laparoscopic cholecystectomy (Lap-C) can be a valuable method for improving the effectiveness of surgical education and enhancing the surgical outcomes of Lap-C. This study aimed to assess the applicability of automated surgical skill assessment using a developed AI-based surgical phase recognition model in Lap-C.

We collected Lap-C videos and categorized them into training and test datasets. The training dataset was used to develop a surgical phase recognition model that classified the Lap-C procedure into 10 phases. The test dataset was categorized into three surgical skill levels (expert, intermediate, and novice groups) based on predefined criteria. We evaluated the applicability of our model to automatically categorize the three surgical skill levels using parameters derived from the model, including time spent in each phase and the confidence level for phase recognition (AI confidence score [AICS]).

The overall accuracy of the surgical phase recognition model was 82.3%. Manual analysis showed that the time for dissection in the gallbladder neck in the novice group was significantly longer than in the expert group (P < 0.01), and a similar trend was observed in the model-based analysis. The time for clipping and cutting the cystic duct and artery in the novice group was significantly longer than in the expert group in both manual and model-based analyses. AICS was higher in the expert group than in the intermediate group (P = 0.02).

We developed an automated surgical phase recognition model in Lap-C with AI, which was applicable for surgical skill assessment by measuring the time required for dissection of the gallbladder neck, clipping and cutting the cystic duct and artery, and calculating the AICS. Our model is expected to contribute to the efficiency of surgical education.