Resting-state brain oscillations predict cognitive function in psychiatric disorders: A transdiagnostic machine learning approach.
NEUROIMAGE-CLINICAL 2021;
30:102617. [PMID:
33752077 PMCID:
PMC7985402 DOI:
10.1016/j.nicl.2021.102617]
[Citation(s) in RCA: 9] [Impact Index Per Article: 3.0] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Subscribe] [Scholar Register] [Received: 11/03/2020] [Revised: 02/03/2021] [Accepted: 02/28/2021] [Indexed: 11/23/2022]
Abstract
Resting EEG activity associated with cognitive function in psychiatric disorders.
Using EEG, random forest modeling predicts cognitive performance but not diagnosis.
High alpha oscillations associated with better episodic memory and processing speed.
Beta oscillations associated with worse performance in several cognitive domains.
EEG power changes in psychiatric disorders may be related to cognitive dysfunction.
Background
Cognitive dysfunction is widespread in psychiatric disorders and can significantly impact quality of life. Deficits cut across traditional diagnostic boundaries, necessitating new approaches to understand how cognitive function relates to large-scale brain activity and psychiatric symptoms across the diagnostic spectrum.
Objective
Using random forest regression, we aimed to identify transdiagnostic patterns linking cognitive function to resting-state EEG oscillations.
Methods
216 participants recruited through an outpatient psychiatric clinic completed the Cambridge Neuropsychological Test Automated Battery and underwent a 5-minute eyes-closed resting state EEG recording. We built random forest regression models to predict performance on each cognitive test using the resting-state EEG power spectrum as input, and we compared model performance to a sampling distribution constructed with random permutations. For models that performed significantly better than chance, we used feature importance estimates to identify features of the EEG power spectrum that are predictive of cognitive functioning.
Results
Random forest models successfully predicted performance on measures of episodic memory and associative learning (Paired Associates Learning, PAL), information processing speed (Choice Reaction Time, CRT), and attentional set-shifting and executive function (Intra-Extra Dimensional Set Shift, IED). Oscillatory power in the upper alpha range was associated with better performance on PAL and CRT, while low alpha power was associated with worse CRT performance. Beta power predicted poor performance on all three tests. Theta power was associated with good performance on PAL, and delta and theta oscillations were identified as predictors of good performance on IED. No differences in cognitive performance were found between diagnostic categories.
Conclusion
Resting oscillations are predictive of certain dimensions of cognitive function across various psychiatric disorders. These findings may inform treatment development to improve cognition.
Collapse