Explainable machine learning using EMG and accelerometer sensor data quantifies surgical skill and identifies...

Area of Science:

• Robotics and Artificial Intelligence in Medicine
• Biomedical Engineering and Sensor Technology
• Surgical Education and Skill Assessment

Background:

• Traditional surgical skill evaluation relies on subjective methods, lacking precision and scalability.
• Objective, interpretable, and scalable assessment tools are crucial for modern robotic surgery training.
• Emerging wearable sensor technology offers potential for objective motor skill analysis.

Purpose of the Study:

• To develop an explainable machine learning (XAI) framework for classifying surgeon skill levels using sEMG and accelerometer data.
• To identify objective neuromuscular biomarkers indicative of surgical expertise in robotic procedures.
• To enhance transparency and provide actionable feedback for surgical training.

Main Methods:

• Collected surface electromyography (sEMG) and accelerometer data from 26 participants performing standardized robotic tasks.
• Extracted time-domain, frequency-domain, and nonlinear dynamical features from sensor data.
• Utilized supervised machine learning classifiers (SVM, Random Forest, XGBoost, Naïve Bayes) and XAI techniques (SHAP, LIME) for classification and interpretation.

Main Results:

• Ensemble models achieved over 72% classification accuracy for surgeon skill levels.
• Nonlinear features, including Lyapunov exponents and entropy, were key predictors of skill.
• XAI revealed distinct feature sets characterizing novice versus expert performance, enabling interpretable feedback.

Conclusions:

• The XAI framework effectively combines wearable sensor data with machine learning for objective robotic surgical skill assessment.
• Identified neuromuscular biomarkers capture nuanced motor control aspects differentiating skill levels.
• This approach provides a transparent, interpretable pathway for personalized surgical feedback and training enhancement.