LC-Smart: A Deep Learning-Based Quality Control Model for Laparoscopic Cholecystectomy

LC-Smart: A Deep Learning-Based Quality Control Model for Laparoscopic Cholecystectomy

Objective: Critical view of safety (CVS) is a successful technique to reduce bile duct injury during laparoscopic cholecystectomy (LC). We aimed to create a deep learning-based quality control model for LC and reduce the learning curve for junior surgeons, which would automatically assess whether surgeons are CVS conscious during procedures.Methods: We retrospectively collected 308 LC videos from public datasets (Cholec80, Endoscapes) and Sun Yat-sen Memorial Hospital. Video frames were labeled using binary classification and feature optimization methods, such as black border clipping and sliding windows. Two neural networks, ResNet-50 and EfficientNetV2-S, were trained and evaluated based on F1 scores and accuracy. Additionally, We created an online CVS recognition system (LC-Smart), tested it using 171 films from two hospitals, and compared the results to two local senior doctors.

Inclusion Criteria:

• complete video data with no missing footage;
• surgical procedure identified as laparoscopic cholecystectomy;
• full visibility of the surgical area in the video;
• successful completion of the procedure;
• absence of significant anatomical variations
• video resolution no less than 720×560.

Exclusion Criteria:

• substantial intraoperative adhesions
• a history of previous abdominal or pelvic procedures
• a conversion to open surgery during the procedure
• significant bleeding that obscured structural identification.