Surgical Hand Gesture Recognition Utilizing Electroencephalogram as Input to the Machine Learning and Network...

Area of Science:

• Neurosurgery
• Robotic Surgery
• Machine Learning

Background:

• Automated surgical skill assessment and feedback are crucial for robot-assisted surgery (RAS) training.
• Existing methods utilize surgical videos, kinematics, or electromyograms (EMG).
• Electroencephalogram (EEG) data offers a potential, less-explored avenue for gesture detection.

Purpose of the Study:

• To extract features from EEG data for classifying robot-assisted surgical gestures.
• To evaluate the efficacy of machine learning algorithms in conjunction with EEG-derived features for surgical gesture recognition.

Main Methods:

• EEG data was collected from five RAS surgeons during 34 robot-assisted radical prostatectomies.
• Eight dominant and six non-dominant hand gesture types were identified and synchronized with EEG.
• Network neuroscience algorithms extracted functional brain network and power spectral density features.
• Feature selection used analysis of variance (ANOVA) F-value; validation employed 10-fold cross-validation.

Main Results:

• The extra trees (ET) algorithm, using 60 selected features, achieved 90% accuracy, 90% precision, and 88% sensitivity for dominant hand gestures.
• For non-dominant hand gestures, the system achieved 93% accuracy, 94% precision, and 94% sensitivity.
• The proposed method demonstrates robust performance in classifying distinct surgical gestures.

Conclusions:

• EEG-based feature extraction combined with machine learning provides an effective method for automated surgical gesture detection in RAS.
• This technique holds promise for developing objective and automated surgical skill assessment tools.
• Further research can refine these methods for real-time feedback during surgical training.