AI-Based Video Segmentation: Procedural Steps or Basic Maneuvers?

Area of Science:

• Surgical training and assessment
• Artificial intelligence in medicine
• Computer vision for surgical procedures

Background:

• Video review is crucial for surgical training, assessing skill, and decision-making.
• Current methods focus on procedural steps, overlooking fundamental maneuvers.
• Basic maneuvers (suturing, knot tying, cutting) offer insights into complexity, skill, and preference.

Purpose of the Study:

• To develop and evaluate an algorithm for identifying basic surgical maneuvers.
• To assess the algorithm's accuracy in differentiating maneuvers from surgical videos.

Main Methods:

• A deep learning model was trained to distinguish suture throws, knot ties, and suture cutting.
• Data comprised videos from 52 practicing clinicians performing simulated enterotomy repair.
• Qualitative analysis explored the utility of maneuver identification in open colon resection.

Main Results:

• The algorithm achieved 84% accuracy in differentiating maneuvers.
• Precision rates were 87.9% for suture throws, 60% for knot ties, and 90.9% for suture cutting.
• Qualitative feedback confirmed the potential benefits of maneuver identification in real-world scenarios.

Conclusions:

• Basic surgical maneuvers can signal error management and safety protocols, aiding skill assessment.
• The developed deep learning algorithm demonstrates reasonable accuracy in identifying these maneuvers.
• AI-assisted video review, incorporating maneuver identification, can augment traditional segmentation protocols.