Long-range temporal dependencies in EEG as a biomarker of depression

Signal complexity measures computed from electroencephalogram (EEG) are widely used as feature descriptors for the diagnosing major depressive disorder (MDD). Because of their intricate formulations, it is not so straightforward to back track and infer the changes in the pattern of EEG that have caused the deflections. We define aperiodic nature of EEG in MDD patients with the help of mean square of autocorrelation function (ACF) coefficients (MSACFC) as a feature descriptor. MSACFC quantifies changes in the long-range dependencies in a signal via an easily explainable formulation compared to the traditional signal complexity measures. We analysed the MSACFC computed from EEG recordings of 5minute duration within 1 Hz to 30 Hz band in eyes closed state obtained from an age and gender-matched cohort of 28 healthy controls (HCs) and 30 MDD patients at 19 scalp electrodes. From Wilcoxon rank-sum test MSACFC at F8 electrode is found to be significantly higher in HCs (0.0027±0.0035) than in MDD patients (2.9313 × 10^-4± 2.7236 × 10^-4) with a p-value of 3.7 × 10^-6. At the F8 electrode, MSACFC has a large effect size (Cohen's d) of 0.99563 > 0.8 for 95% confidence interval signifying a substantially strong difference between the groups. The EEG in MDD patients is more aperiodic and the ACF has lower magnitude compared to HCs.