Surgical motion analysis using discriminative interpretable patterns

Area of Science:

• Robotics and Automation in Surgery
• Medical Education Technology
• Biomechanics and Movement Analysis

Background:

• Surgical motion analysis is crucial for advancing automated surgical training systems.
• Quantitative evaluation of surgical skills is essential for improving patient care.
• Automatic capture of surgical motions is increasingly feasible with new technologies.

Purpose of the Study:

• To develop and present an approach for discovering and ranking discriminative patterns in surgical motion data.
• To enable automated assessment of surgical trainee skills through pattern recognition.
• To provide interpretable feedback on surgical performance.

Main Methods:

• Decomposition of continuous kinematic data into overlapping gestures represented as strings (bag of words).
• Application of term frequency-inverse document frequency (tf-idf) for comparative statistical analysis.
• Discovery of discriminative gestures based on relative occurrence frequency.

Main Results:

• Experimental validation on three surgical motion datasets.
• Accurate classification of individual gestures, skill levels, and surgical interfaces using identified patterns.
• Demonstration of how patterns provide detailed feedback for trainee skill assessment.

Conclusions:

• The proposed approach offers a valuable addition to surgical learning tools.
• It enables feedback on specific exercise components contributing to performance classification.
• This method facilitates objective and quantitative assessment of surgical proficiency.