Prediction of working memory and neuropsychiatric symptom variability with oscillatory and non-oscillatory EEG measures

Poster Session D - Monday, April 15, 2024, 8:00 – 10:00 am EDT, Sheraton Hall ABC

Fleming Peck¹ (fpeck@ucla.edu), Jean-Baptiste Pochon¹, Sandra Loo¹, Catherine Sugar¹, Carrie Beardon¹, Robert Bilder¹, Jesse Rissman¹, Agatha Lenartowicz¹; ¹UCLA

Working memory (WM) is the capacity-limited process supporting the transient maintenance of goal-relevant information. The aim of the present study was to relate several complementary indicators of neural dynamics underlying WM to across-subject variance in measures of WM task performance, trait-level WM capacity, and psychiatric symptomatology. Electroencephalography (EEG) data were analyzed from a heterogenous sample of 100 adults (a combination of care-seeking and non-care-seeking individuals) as they performed WM tasks, including a spatial capacity task (SCAP) and dot-pattern expectancy task (DPX). Machine learning models were trained to predict: (1) WM task performance (accuracy), (2) trait WM capacity (WAIS and WMS WM subtests), and (3) psychiatric symptomatology (Brief Psychiatric Rating Scale). Features included EEG-based measures of power, oscillation symmetry, non-oscillatory power spectrum properties, and complexity, which were extracted from the encoding, maintenance, probe, and inter-trial stages of each task. We found that measures of non-oscillatory power spectrum properties during SCAP significantly predicted variance in WM capacity, and oscillation symmetry during DPX (a goal-maintenance task) significantly predicted psychiatric symptomatology. These models heavily weighted features derived from frontal and occipital regions-of-interest. We observed variability in prediction and feature importance among tasks, with no single measure dominating prediction at a given stage. However, we generally found better prediction during task than during the inter-trial fixation period, suggesting task-based brain activity is more predictive of behavioral outcomes than task-free activity. Ultimately, this project expands on the utility of integrating both oscillatory and non-oscillatory measures of EEG signal in prediction of trait- and state-like outcome measures.

Topic Area: EXECUTIVE PROCESSES: Working memory