Offline identification of surgical deviations in laparoscopic rectopexy

Area of Science:

• Medical Informatics
• Surgical Safety
• Machine Learning in Healthcare

Background:

• Adverse events during surgery occur in 14.4% of patients, with a third being preventable.
• Deviations from standard surgical processes due to adverse events pose a significant challenge to patient safety.
• Automatic identification of these deviations is crucial for improving surgical outcomes.

Purpose of the Study:

• To propose and evaluate a method for automatically identifying surgeons' deviations from standard surgical processes.
• To focus on deviations related to surgical events rather than anatomical variations.
• To address the challenge of high variability in surgical procedure workflows.

Main Methods:

• Developed an approach using multi-dimensional non-linear temporal scaling with a hidden semi-Markov model.
• Utilized manual annotation of surgical processes for training and evaluation.
• Evaluated the method's performance using cross-validation techniques.

Main Results:

• Achieved over 90% accuracy in detecting deviations.
• Reported recall and precision for event deviations below 80% and 40%, respectively.
• Conducted a detailed error analysis of incorrectly detected observations.

Conclusions:

• The hidden semi-Markov model approach shows promise for automated surgical deviation detection.
• Error analysis provides insights for future method refinement.
• The method is feasible for skill analysis and developing computer-assisted surgical systems for situation awareness.