eXplainable Artificial Intelligence Improves EEG-Based Cognitive Workload Assessment Induced by Fine Motor Activity in Neurosurgeons

Cognitve workload associated with fine motor activity in neurosurgeons was monitored by using a wearable electroen-cephalographic (EEG) device. The most informative EEG features were selected by means of an explainable Artificial Intelligence (XAI) algorithm. XAI represents a promising novel approach in this application field and offers new opportunities for extracting information from EEG data beyond traditional statistical and Machine Learning-based methods. Six neurosurgeons performed the Purdue Pegboard Test (PPT) at two difficulty levels related to low or high cognitive load. EEG signals were acquired with an eight dry electrode device. Absolute powers in six different frequency bands of interest were explored. Three most involved EEG features resulted from SHapley Additive exPlanations (SHAP) methods, namely absolute power in delta band on C3 and Fz channels and the absolute power in theta band on Fz. Summary plots showed a decrease of the three identified EEG features in the high cognitive load task. These findings demonstrate the potential of Artificial Intelligence-supported wearable EEG solutions to monitor cognitive load over time, to track the cognitive load of trainee neurosurgeons and to design adaptive training courses.