A Deep Learning Model for Classification of MDD Patients Studied with EEG Microstate Features

The quasi-steady electric fields in electroencephalogram (EEG) are usually called microstates. In the field of neuroscience, methodologies like fMRI, EEG, and MEG have uncovered essential insights into the dynamics of extensive brain networks. Additionally, these techniques have quantified network characteristics using computational neuroscience approaches, which are vital for comprehending both brain information processing and pathological alterations. This study explored the role of abnormal resting-state microstate features in the recognition of MDD patients by deep learning. Through factor analysis of microstate parameters, a new feature set that could explain the original microstate features was created and applied to the construction of a convolutional neural network (CNN) classification model. At the same time, a backpropagation learning mechanism was introduced, and the model's performance in distinguishing patients with MDD from healthy individuals was comprehensively evaluated using the k-fold cross-validation approach, alongside metrics such as accuracy, precision, and AUC. This provides strong support for medical research and clinical decision-making. The findings indicate that the accuracy of CNN classification is 97.8%, the precision is 98.2%, and the AUC is 0.980. The performance of the model shows that features of resting-state EEG microstate segmentation can be obtained to effectively assist in the identification of MDD patients.