Mental Fatigue Classification with High-Density Diffuse Optical Tomography: A Feasibility Study.

High-Density Diffuse Optical Tomography (HD-DOT) presents as a promising tool for not only clinical use but also daily monitoring of mental states. This study employed wearable HD-DOT to evaluate mental fatigue, specifically examining the differences in functional near-infrared spectroscopy (fNIRS) data between states of low and high fatigue among healthy participants for data collection. Data processing involved filtering, channel selection, and dimensionality reduction through Uniform Manifold Approximation (UMAP) and Projection, followed by classification using Support Vector Machines (SVM). We developed two models to assess the accuracy and generalizability of our findings: one based on individually tailored models and another employing a leave-one-participant-out cross-validation strategy. We evaluated different kernel functions, resulting in various accuracy, F1 score, and Area Under the Curve (AUC) metrics. The study achieved an average accuracy of approximately 90% for participant-specific classifiers, underscoring the effectiveness of our approach to differentiate between low and high states of mental fatigue. Our analyses led to a robust model demonstrating high classification accuracy, proving its suitability and potential for real-time Brain-Computer Interface (BCI) applications.