An EEG-based marker of functional connectivity: detection of major depressive disorder

• Published in: Cognitive neurodynamics Vol. 18; no. 4; pp. 1671 - 1687
• Main Authors: Li, Ling, Wang, Xianshuo, Li, Jiahui, Zhao, Yanping
• Format: Journal Article
• Language: English
• Published: Dordrecht Springer Netherlands 01.08.2024; Springer Nature B.V
• Subjects: Accuracy; Artificial Intelligence; Artificial neural networks; Biochemistry; Biomedical and Life Sciences; Biomedicine; Brain; Brain research; Cognitive Psychology; Coherence; Computer Science; Datasets; Deep learning; EEG; Electroencephalography; Frequency dependence; Information processing; Machine learning; Medical diagnosis; Mental depression; Mental disorders; Missing data; Neural networks; Neurosciences; Research Article; Wavelet transforms; Malaysia; Functional connectivity; CNN; Synchrosqueezed wavelet coherence; EEG; Major depressive disorder
• ISSN: 1871-4080; 1871-4099; 1871-4099
• DOI: 10.1007/s11571-023-10041-5

Major depressive disorder (MDD) is a prevalent psychiatric disorder globally. There are many assays for MDD, but rapid and reliable detection remains a pressing challenge. In this study, we present a fusion feature called P-MSWC, as a novel marker to construct brain functional connectivity matrices and utilize the convolutional neural network (CNN) to identify MDD based on electroencephalogram (EEG) signal. Firstly, we combine synchrosqueezed wavelet transform and coherence theory to get synchrosqueezed wavelet coherence. Then, we obtain the fusion feature by incorporating synchrosqueezed wavelet coherence value and phase-locking value, which outperforms conventional functional connectivity markers by comprehensively capturing the original EEG signal's information and demonstrating notable noise-resistance capabilities. Finally, we propose a lightweight CNN model that effectively utilizes the high-dimensional connectivity matrix of the brain, constructed using our novel marker, to enable more accurate and efficient detection of MDD. The proposed method achieves 99.92% accuracy on a single dataset and 97.86% accuracy on a combined dataset. Moreover, comparison experiments have shown that the performance of the proposed method is superior to traditional machine learning methods. Furthermore, visualization experiments reveal differences in the distribution of brain connectivity between MDD patients and healthy subjects, including decreased connectivity in the T7, O1, F8, and C3 channels of the gamma band. The results of the experiments indicate that the fusion feature can be utilized as a new marker for constructing functional brain connectivity, and the combination of deep learning and functional connectivity matrices can provide more help for the detection of MDD.