Music-evoked EEG signals for mental health assessment using two- and three-class classification

Electroencephalography (EEG), a non-invasive method of measuring brain activity, has become a valuable tool in mental health research for examining neural activity associated with various psychological disorders and neurofeedback therapies, including music therapy. However, the use of EEG under music and no-music conditions for mental health screening remains insufficiently validated.

In this preliminary study, we implemented an experimental protocol in which 48 participants underwent EEG recordings during both music and no-music sessions with identical relaxation instructions. The Patient Health Questionnaire-9 (PHQ-9) and the Generalized Anxiety Disorder 7-item (GAD-7) scale were also included. Five complementary EEG features were extracted for each mental fatigue class: power spectral density (PSD) across different frequency bands, band powers, alpha-theta ratio, theta-beta ratio, and channel coherence. Binary classification (mental fatigue) and multi-output binary classification (anxiety, depression, and stress components) were implemented using a deep residual multi-layer perceptron (MLP), a feedforward neural network with residual skip connections. Leave-Subject-Out cross-validation ensured robust generalization, with all metrics computed at the subject level rather than epoch level.

For binary classification, the model achieved 72.92% accuracy in the music condition and 58.33% accuracy in the no-music condition, with notably high recall (96.67%) in the music condition. Multi-output classification in the music condition showed stress with high accuracy (93.75%) but low AUC (35.56%), reflecting the high prevalence of the stressed class (85.7%), while anxiety and depression presented greater challenges (54.17% and 62.50% accuracy, respectively).

These preliminary results suggest that music may modulate brain activity patterns in ways that enhance the discriminability of mental fatigue states, although the fixed condition order means order and habituation effects cannot be ruled out.