Attention-mechanism–based network characteristic analysis for major depressive disorder detection

¹ School of Electronics and Information Engineering, Liaoning Technical University - Huludao, 125105, China
² School of Electrical and Information Engineering, Tianjin University - Tianjin 300072, China

^(a) weidongdang@tju.edu.cn (corresponding author)

Received: 28 January 2022
Accepted: 23 June 2022

Abstract

Major depressive disorder (MDD) is a very serious mental illness that spreads all over the world and affects patients of all ages. Constructing an efficient and accurate MDD detection system is an urgent research task. In this paper, we develop an EEG-based multilayer brain network and an attention-mechanism–based convolutional neural network (AM-CNN) model to study MDD. In detail, based on mutual information theory, we first construct a multilayer brain network, in which each layer corresponds to a specific frequency band. The experimental results show that such a design can effectively reveal the brain physiological changes of MDD patients, from the perspective of network topology analysis. On this basis, multi-branch AM-CNN model is then designed, which uses multilayer brain network as input and can well achieve feature extraction and detection of MDD. On the publicly available MDD dataset, the proposed method achieves an identification accuracy of 97.22%. Our approach and analysis provide novel insights into the physiological changes of MDD patients and a reliable technical solution for MDD detection.