Automated surgical action recognition and competency assessment in laparoscopic cholecystectomy: a proof-of-concept study

Hung-Hsuan Yen ^1,2, Yi-Hsiang Hsiao ^1,2, Meng-Han Yang ³

⁰Department of Surgery, National Taiwan University Hospital Hsin-Chu Branch, No. 2, Sec. 1, Shengyi Rd., Zhubei City, Hsinchu County, 302058, Taiwan.

Surgical Endoscopy +

|

March 21, 2025

View abstract on PubMed

Summary

Automated models can now assess surgical skills during laparoscopic cholecystectomy by analyzing surgical actions. This technology offers objective feedback for improving surgical education and competency.

Area Of Science

Medical Education
Surgical Technology
Artificial Intelligence in Medicine

Background

Laparoscopic cholecystectomy (LC) is a common surgical procedure.
Current competency assessment methods for LC lack focus on surgical actions.
Automated surgical action recognition models are needed.

Purpose Of The Study

To analyze surgical actions during the Calot's Triangle Dissection (CTD) phase of LC.
To develop and evaluate automated models for surgical competency assessment and action recognition.

Main Methods

Analysis of 80 LC videos from the Cholec80 dataset.
Evaluation of Strasberg's critical view of safety (CVS) score and surgical actions.
Development of a Random Forest model for competency prediction and a Video-Masked Autoencoders (VideoMAE) model for action recognition.

Main Results

The Random Forest model achieved 93% accuracy in predicting competency.
Key features for prediction included CVS score, CTD duration, and action percentages.
The VideoMAE model attained 89.11% accuracy in surgical action recognition.

Conclusions

Surgical actions are crucial for competency assessment in LC.
Automated models provide objective, data-driven feedback for surgical training.
These tools can significantly enhance surgical education and skill development.

Keywords:

Artificial intelligence Cholecystectomy Competency assessment Machine learning Surgical action recognition