Spectral decomposition of EEG microstates in post-traumatic stress disorder

Abstract

Microstates offer a promising framework to study fast-scale brain dynamics in the resting-state electroencephalogram (EEG). However, microstate dynamics have yet to be investigated in post-traumatic stress disorder (PTSD), despite research demonstrating resting-state alterations in PTSD. We performed microstate-based segmentation of resting-state EEG in a clinical population of participants with PTSD (N = 61) and a non-traumatized, healthy control group (N = 61). Microstate-based measures (i.e., occurrence, mean duration, time coverage) were compared group-wise using broadband (1–30 Hz) and frequency-specific (i.e., delta, theta, alpha, beta bands) decompositions. In the broadband comparisons, the centro-posterior maximum microstate (map E) occurred significantly less frequently (d = -0.64, pFWE = 0.03) and had a significantly shorter mean duration in participants with PTSD as compared to controls (d = -0.71, pFWE < 0.01). These differences were reflected in the narrow frequency bands as well, with lower frequency bands like delta (d = -0.78, pFWE < 0.01), theta (d = -0.74, pFWE = 0.01), and alpha (d = -0.65, pFWE = 0.02) repeating these group-level trends, only with larger effect sizes. Interestingly, a support vector machine classification analysis comparing broadband and frequency-specific measures revealed that models containing only alpha band features significantly out-perform broadband models. When classifying PTSD, the classification accuracy was 76 % and 65 % for the alpha band and the broadband model, respectively (p = 0.03). Taken together, we provide original evidence supporting the clinical utility of microstates as diagnostic markers of PTSD and demonstrate that filtering EEG into distinct frequency bands significantly improves microstate-based classification of a psychiatric disorder.

Keywords

• EEG
• Microstates
• Resting-state
• Machine learning
• PTSD