Decision support system for major depression detection using spectrogram and convolution neural network with EEG signals

• Published in: Expert systems Vol. 39; no. 3
• Main Authors: Loh, Hui Wen, Ooi, Chui Ping, Aydemir, Emrah, Tuncer, Turker, Dogan, Sengul, Acharya, U. Rajendra
• Format: Journal Article
• Language: English
• Published: Oxford Blackwell Publishing Ltd 01.03.2022
• Subjects: Artificial neural networks; Automation; classification; CNN; COVID-19; Decision support systems; Deep learning; Electrocardiography; electroencephalogram (EEG); Electroencephalography; Fourier transforms; Machine learning; major depressive disorder (MDD); Medical imaging; Mental depression; Neural networks; spectrograms; STFT
• ISSN: 0266-4720; 1468-0394
• DOI: 10.1111/exsy.12773

The number of Major Depressive Disorder (MDD) patients is rising rapidly these days following the incidence of COVID‐19 pandemic. It is challenging to detect MDD through personal interviews and by observing electroencephalogram (EEG) signals. Hence, an automated MDD detection system developed using deep learning techniques can help reduce the workload of clinicians by diagnosing MDD accurately. In this study, we have proposed a novel deep learning model based on Convolutional Neural Network (CNN) and spectrogram images. In this work, Short‐Time Fourier Transform (STFT) is first applied to the EEG signals to obtain spectrogram images of MDD patients and healthy subjects. These spectrogram images are then fed to the CNN model for automated detection of MDD patients and healthy subjects. The EEG signals used in this study were obtained from public database with 34 MDD patients and 30 healthy subjects. The highest classification accuracy, precision, sensitivity, specificity, and F1‐score of 99.58%, 99.40%, 99.70%, 99.48%, and 99.55% respectively were obtained with hold‐out validation. Our MDD detection model is highly accurate and needs to be validated with more diverse MDD database before it can be used in clinical settings. Also, we plan to use our developed prototype to detect depression using other physiological signals like electrocardiogram (ECG) and speech signals for accurate and faster diagnosis.