Related Experiment Video
Updated: Jun 20, 2025
06:11
Brain-Computer Interface-controlled Upper Limb Robotic System for Enhancing Daily Activities in Stroke Patients
Published on: April 18, 2025
309
Classification of subtask types and skill levels in robot-assisted surgery using EEG, eye-tracking, and machine
Somayeh B Shafiei1, Saeed Shadpour2, James L Mohler3
1The Intelligent Cancer Care Laboratory, Department of Urology, Roswell Park Comprehensive Cancer Center, Buffalo, NY, 14263, USA. Somayeh.besharatshafiei@roswellpark.org.
Surgical Endoscopy
|July 22, 2024
Summary
Machine learning models using electroencephalogram (EEG) and eye-tracking data can accurately classify robot-assisted surgery (RAS) subtasks and skill levels. This approach enhances objective evaluation for surgical education and patient safety.
Area of Science:
- Robotics and Machine Learning in Surgery
- Biomedical Signal Processing
- Surgical Education Technology
Background:
- Objective evaluation of surgical skills in robot-assisted surgery (RAS) is crucial for training and patient safety.
- Current assessment methods may lack standardization and objectivity.
- Novel approaches are needed to enhance the precision of skill evaluation in RAS.
Purpose of the Study:
- To introduce and evaluate machine learning (ML) techniques for identifying surgical subtasks and classifying skill levels in RAS.
- To utilize electroencephalogram (EEG) and eye-tracking data for objective skill assessment.
- To compare the performance of different ML models in classifying RAS performance.
Main Methods:
- Collected EEG and eye-tracking data from surgeons performing nine distinct RAS subtasks on pigs.
- Employed four ML models: logistic regression, random forest, gradient boosting, and extreme gradient boosting (XGB) for multi-class classification.
- Utilized a 80/20 train-test split, optimized hyperparameters via grid search with fivefold cross-validation, and ensured reliability through 30 iterations of train-test splits.
Main Results:
- The ML approach demonstrated superior performance in classifying RAS subtasks and skill levels compared to existing methods.
- XGB and random forest models achieved high accuracy rates of 88.49% and 88.56%, respectively, with no significant difference between them.
- The models successfully differentiated between various surgical subtasks and skill levels based on neurophysiological and visual data.
Conclusions:
- ML models show significant potential to improve the objectivity and precision of RAS subtask and skill evaluation.
- This study represents a step towards more refined, objective, and standardized RAS training and competency assessment.
- Future research should focus on optimizing models for challenging subtask classes identified in this study.
