Hilbert-Huang Transform Embedded Self-Attention Neural Network for EEG-based major depressive disorder vs. healthy controls classification

TL;DR

This paper introduces a new neural network model that uses EEG signals to accurately distinguish between people with major depressive disorder and healthy individuals.

Contribution

The novel integration of Hilbert-Huang Transform into a self-attention neural network improves time-frequency analysis for depression detection.

Findings

The proposed model achieved 98.78% accuracy in classifying MDD patients and healthy controls.

The model outperformed traditional methods with high sensitivity (99.23%) and specificity (98.27%).

The method enhances nonlinear processing and captures critical temporal-spectral patterns in EEG data.

Abstract

This paper proposes a novel approach for distinguishing Major Depressive Disorder (MDD) patients from healthy controls (HC), namely depression screening, using EEG signals, where the Hilbert-Huang Transform (HHT) is integrated into a Self-Attention neural network (HHT-SANN). The incorporation of the HHT enhances the model’s time-frequency analysis capabilities and allows for more effective nonlinear processing of the EEG data. By embedding the HHT within the self-attention module, the model captures intricate temporal and spectral patterns that are critical for accurate depression classification. We evaluated our method on a clinical EEG dataset comprising 34 MDD patients and 30 healthy controls from the Hospital of Universiti Sains Malaysia. Experimental results indicate that the proposed method achieves an accuracy of 98.78%, sensitivity of 99.23%, and specificity of 98.27%, outperforming traditional models and offering a more robust solution for depression detection. This work contributes to advancing the field of neuroinformatics by providing a more interpretable and effective model for mental health diagnostics based on EEG data.

Figures