Identifying Mild Cognitive Impairment and Healthy Aging Using EEG Visual Working Memory Task with XGBoost

Mild Cognitive Impairment (MCI) is a stage that marks the transition from healthy aging to severe cognitive decline, often associated with impaired working memory. This study investigates the utility of EEG-based functional connectivity measures to distinguish MCI from healthy controls (HC) in a visual working memory task. Nineteen MCI and nineteen HC participants underwent two encoding durations (150 ms vs. 500 ms) with or without distractors to manipulate cognitive load. After artifact rejection via Independent Component Analysis (ICA), connectivity matrices were computed using Imaginary Coherence (ImCoh) and weighted Phase-Lag Index (wPLI) across five frequency bands (alpha, beta, delta, gamma, theta). Partial Least Squares (PLS) was applied to reduce the high-dimensional connectivity features, then used the PLS components as input to XGBoost. Our purposed method achieved an accuracy of 0.71 using ImCoh features in the alpha and beta bands at a 150-millisecond encoding time, and 0.71 using wPLI features in the alpha band at a 500-millisecond encoding time. The results suggest that the alpha and beta bands are particularly relevant in distinguishing MCI from HC when distractors are present. Additionally, we observed that ImCoh and wPLI capture complementary information at different encoding times, with ImCoh showing optimal performance at 150 ms and wPLI at 500 ms. These findings highlight the temporal and spectral sensitivity of connectivity measures, as well as the effect of distractor presence, in characterizing cognitive differences.