Prediction of surgical skill using machine learning with optimal EEG locations in two different laparoscopic surgery training task

Introduction: The aim of this study was to evaluate whether surgical skill can be reliably classified using low-density, mobile EEG recordings during laparoscopic simulation tasks with different cognitive–motor demands. Specifically, we compared Peg Transfer, a primarily motor-driven task, with String Pass (Threading), which involves higher cognitive–motor integration, to assess task-dependent neural and behavioral discriminability.Materials and Methods: Surgeons and non-experienced participants performed two standardized laparoscopic simulation tasks: Peg Transfer and String Pass. EEG data were acquired using a mobile EEG system with a sparse electrode configuration to ensure feasibility in realistic training environments. Task-related EEG features were extracted, focusing on band-specific spectral power to characterize local oscillatory activity. These features were used to train a Support Vector Classifier (SVC) for differentiating levels of surgical expertise. Behavioral performance was assessed using task completion time and reaction time, while subjective workload was evaluated using the NASA Task Load Index (NASA-TLX).Results: Classification performance revealed a clear task-dependent pattern. Consistent with these findings, EEG power analyses showed clearer and more widespread brain activation differences between surgeons and students during the more complex String Pass task, whereas neural differences during Peg Transfer were weaker and more localized. SVM-based classification using task-related EEG features achieved higher discriminative performance for the String Pass task, with the strongest separation observed in the theta band (AUC = 0.88). In contrast, Peg Transfer exhibited weak discriminative power at the single-band level. Behavioral results were consistent with machine learning findings: expert surgeons and medical students were more clearly differentiated during String Pass based on completion time and reaction time, whereas no significant group differences were observed for Peg Transfer. Subjective workload ratings measured by NASA-TLX did not differ significantly between groups for either task. Conclusion: These findings demonstrate that robust classification of surgical expertise is achievable using a reduced, low-density EEG setup, supporting its potential for real-time and objective skill assessment. This strategy has the potential to support objective surgeon certification processes and competency-based surgical training frameworks.