Attention GCN-LSTM Model for Dementia Identification Using Functional Brain Networks

초록

Timely detection of dementia is essential for effective intervention and management of the diseases. EEG-based assessments have shown promise in capturing brain dynamics related to cognitive decline, yet conventional methods often overlook the integration of spatial and temporal features in the analysis. Current approaches lack the ability to comprehensively evaluate brain connectivity across various stages of dementia, leading to suboptimal classification performance. This study proposes a novel ensemble model that integrates Graph Convolutional Networks (GCNs) and Long Short-Term Memory (LSTM) networks with attention mechanisms. We constructed multiplex brain functional networks using Phase Locking Value (PLV) connectivity across five EEG frequency bands to facilitate a detailed analysis of connectivity patterns in normal cognition/controls (NC), mild cognitive impairment (MCI), and Alzheimer's disease (AD) groups. The proposed attention GCN-LSTM model outperformed individual attention-based GCN and LSTM models, achieving an accuracy of 83.12%, precision of 84.23%, recall of 83.14%, and an F1-score of 83.68%. The findings reveal significant connectivity differences among the NC, MCI, and AD groups, indicating alterations in brain network organization corresponding to cognitive impairment severity. These results enhance the effectiveness of integrating spatial and temporal information for improved dementia classification. The hybrid model presents a promising approach for early detection and intervention strategies, contributing to the ongoing efforts to enhance diagnostic accuracy in neurodegenerative diseases.

키워드