A convolutional neural network-based diagnostic method using resting-state electroencephalograph signals for major depressive and bipolar disorders. (February 2022)

• Record Type: Journal Article
• Authors: Lei, Yu; Belkacem, Abdelkader Nasreddine; Wang, Xiaotian; Sha, Sha; Wang, Changming; Chen, Chao
• Is Part Of: Biomedical signal processing and control. Volume 72(2022)Part B
• Publication Date: 2022-02
• Subjects: Neural networks -- Electroencephalography -- Major depressive disorder -- Bipolar disorder; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.103370
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: A deep learning method to recognize bipolar disorder and major depressive disorder. The learned features are consistent with neurobiological studies. Different modules in the network have been analyzed. Abstract: Background: Early and accurate diagnosis of bipolar and major depressive disorders is important in clinical practice. However, no diagnostic biomarkers can discriminate bipolar from major depressive disorder with high accuracy at present. Methods: We propose a novel convolutional neural network architecture using multichannel raw resting-state electroencephalograph signals to differentiate bipolar disorder from major depressive disorder. This method has great potential in diagnosing mental disorders. In total, 101 patients with major depressive disorder, 82 patients with bipolar disorder, and 81 healthy controls were assessed. Clinical diagnosis was performed by psychiatrists based on the Diagnostic and Statistical Manual of Mental Disorders, fifth edition. Participants were instructed to fix their eyes on a cross on the monitor during the collection of resting-state electroencephalograph signals. Results: A classification mean accuracy of 96.88% was achieved using the resting-state electroencephalograph signals with the 10-fold cross-validation method. The results indicate that deep neural networks could learn efficient feature patterns automatically without manual feature selections. Both statistical analysis and feature visualization on different brain regions showed that the classification performance and the learned features of the proposed model were consistent with that obtained in neurobiological studies on bipolar and major depressive disorders. Conclusions: The combined use of deep neural networks and electroencephalograph signals is an effective approach for the computer-aided diagnosis of bipolar and major depressive disorders.