Surgical skill level classification model development using EEG and eye-gaze data and machine learning algorithms.

Area of Science:

• Robotics
• Machine Learning
• Surgical Education

Background:

• Robot-assisted surgery (RAS) requires specialized skills.
• Objective assessment of surgical expertise is crucial for training.
• Current methods for evaluating surgical skill can be subjective.

Purpose of the Study:

• To develop machine learning models for classifying surgical expertise in RAS.
• To utilize electroencephalogram (EEG) and eye-gaze data for skill prediction.
• To compare the performance of different classification algorithms.

Main Methods:

• Collected EEG and eye-gaze data from 11 participants performing RAS procedures.
• Used modified Global Evaluative Assessment of Robotic Skills (GEARS) for skill evaluation.
• Applied multinomial logistic regression (MLR), random forest (RF), and gradient boosting (GB) models.

Main Results:

• Gradient boosting (GB) model with EEG features achieved high accuracy (e.g., 85% for retraction).
• Combining EEG and eye-gaze data with GB further improved classification accuracy (e.g., 93% for retraction).
• The GB model demonstrated superior performance in classifying skill levels across subtasks.

Conclusions:

• Objective skill classification models using biosignals are feasible for RAS.
• These models can provide valuable, objective feedback for surgical training.
• Implementation in clinical settings can enhance the RAS surgical training process.