Recognition of Major Depressive Disorder Based on EEG Wavelet Coherence

Abstract

This study investigates whether features derived from wavelet coherence between EEG channels improve the recognition of Major Depressive Disorder (MDD) compared with baseline features obtained from wavelet transforms of individual channels. A comprehensive algorithm was developed, incorporating signal preprocessing, coherence estimation using the analytic Morlet mother wavelet, and the construction of derived inter-channel and network-level features. Experimental evaluation was performed on the TDBRAIN dataset with a subject-wise split into independent training and test cohorts. Baseline spectral–energetic features (relative wavelet energy and wavelet entropy) achieved a test accuracy of 76.2%, representing the highest performance among individual feature sets. Adding pair wise wavelet-coherence features substantially improved classification performance up to 83.3%. These results indicate that features reflecting functional interactions between electrodes are complementary to spectral descriptors and enhance the system’s ability to detect patterns associated with depressive disorders. Overall, the combination of wavelet-energy features with inter-channel coherence provides a stable and reproducible approach for automated MDD recognition from EEG signals.