Automated surgical phase recognition and analysis in single-incision laparoscopic cholecystectomy using artificial...

Area of Science:

• Artificial Intelligence in Surgery
• Surgical Phase Recognition
• Deep Learning Models

Background:

• Single-incision laparoscopic cholecystectomy (SILC) presents unique technical challenges and a steeper learning curve than conventional methods.
• Developing AI systems for surgical phase recognition and prediction in SILC is crucial for enhancing training, assessing quality, and improving safety.
• This study focused on creating a multicenter AI system for SILC phase recognition and prediction.

Purpose of the Study:

• To develop and evaluate a multicenter deep learning-based system for surgical phase recognition and prediction specifically for SILC.
• To assess the model's performance in classifying and predicting temporal transitions of surgical phases.
• To investigate the impact of training data size on model performance.

Main Methods:

• A deep learning model (Trans-SVNet) was developed using 122 SILC videos from two medical centers.
• The model was trained to identify distinct surgical phases within SILC procedures.
• Performance was evaluated on 26 independent SILC videos using metrics like accuracy, precision, recall, Jaccard Index, and Mean Absolute Error (MAE) variants.

Main Results:

• The Trans-SVNet model demonstrated high performance in phase classification, achieving an overall accuracy of 0.933, precision of 0.939, and recall of 0.939.
• For phase transition prediction, the model achieved an overall MAE of 37s, eMAE of 34s, and pMAE of 50s.
• Increasing the number of training videos significantly improved the model's performance.

Conclusions:

• The Trans-SVNet model successfully enabled automatic classification and temporal prediction of surgical phases in complete SILC videos.
• AI holds significant potential for analyzing large surgical datasets, leading to clinically relevant applications in the future.
• Continued refinement of AI models can further enhance surgical analysis and support safer surgical practices.