Resting-State EEG Microstate Features for Major Depressive Disorder Classification

Author

George M. V. Quinn, The Graduate Center, City University of New YorkFollow

Date of Degree

9-2024

Document Type

Thesis

Degree Name

M.S.

Program

Cognitive Neuroscience

Advisor

Lucas Parra

Subject Categories

Bioelectrical and Neuroengineering | Biological Psychology | Clinical Psychology | Cognitive Neuroscience | Cognitive Science | Psychiatric and Mental Health

Keywords

Microstate, Depression, Machine Learning

Abstract

Neuroimaging studies have revealed consistent abnormalities in functional connectivity within specific neural networks that may serve as biomarkers for major depressive disorder (MDD). It is important to find inexpensive, non-invasive techniques that target these biomarkers to make diagnosis easier and more objective. EEG microstates are quasi-stable potential topographies that are thought to reflect the quasi-stable network activity of the underlying neural generators. MDD has been shown to alter features of the four canonical EEG microstates (A, B, C, D) with some conflicting results. The most consistent network abnormalities in MDD are found in the anterior default mode network, and this network has been linked to the class F microstate using source localization and combined EEG-fMRI techniques. Here, analysis is extended to seven microstate classes (A through G) recently reviewed for their functional significance to find resting-state EEG microstate features that aid in MDD classification. The MODMA dataset was used in this study. This resting-state EEG dataset was obtained from 24 inpatients and outpatients diagnosed with MDD by at least one clinical psychiatrist (30.88 ± 10.37 years, female = 11) and 29 healthy controls (HC) (31.45 ± 9.15 years, female = 9). Seven microstate classes were extracted from all EEG data. Standard microstate analysis was conducted to obtain the temporal features (duration, coverage, occurrence, and transition probabilities) for each participant. The same seven microstate classes were extracted from each participant individually to analyze group topography differences. Individual temporal features and normalized class topographies were used as features to classify MDD using a support vector machine (SVM). There were no significant differences between the MDD and HC groups for the temporal features of microstate classes A through G. Analysis of the difference between group-level topographies of the seven extracted microstate classes revealed a significant difference in class F topography (p = 0.015). Microstate class topographies outperformed temporal features in SVM classification, with class F topography yielding the highest accuracy (87.1% ± 13.6). These results imply that the altered functional connectivity in cases of MDD is more likely to manifest as altered microstate class topographies, specifically class F, rather than temporal features when using resting-state EEG data. Resting-state EEG microstate class topographies, specifically class F, may be useful in targeting the network-based biomarkers for MDD.

Recommended Citation

Quinn, George M. V., "Resting-State EEG Microstate Features for Major Depressive Disorder Classification" (2024). CUNY Academic Works.
https://academicworks.cuny.edu/gc_etds/6037