A method for diagnosis support of mild cognitive impairment through EEG rhythms source location during working memory tasks. (April 2021)

• Record Type: Journal Article
• Authors: San-Martin, Rodrigo; Johns, Erin; Quispe Mamani, Godofredo; Tavares, Guilherme; Phillips, Natalie A.; Fraga, Francisco J.
• Is Part Of: Biomedical signal processing and control. Volume 66(2021)
• Publication Date: 2021-04
• Subjects: Mild Cognitive Impairment -- Alzheimer's disease -- Working memory -- Source localization (LORETA) -- Machine learning -- Support vector machine (SVM); Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2021.102499
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: EEG reveals Working Memory (WM) alterations in Mild Cognitive Impairment (MCI). MCI have more source-space EEG power than healthy elderly (HE) in gamma at precuneus. Source-space EEG features led to MCI vs. HE classification accuracy up to 96%. Abstract: Objective: We investigated group differences in current source density (CSD) patterns from EEG signals before and after a working memory (WM) task performed by mild cognitive impaired (MCI) subjects and healthy elderly (HE). Methods: EEG was recorded during N-back WM tasks in 41 age-, sex- and education-matched participants divided into MCI ( N = 19) and HE ( N = 22) groups. EEG epochs were divided into pre- and post-stimulus periods, named herein as working memory epochs (WME) and event-related epochs (ERE), respectively. Frequency–domain CSD was extracted for both WME and ERE on delta, theta, alpha, beta, and gamma bands using LORETA. Group comparisons were performed under Statistical non-Parametric Mapping. Moreover, after feature selection, we performed cross-validation with a Support Vector Machine (SVM) classifier. Results: MCI displayed increased spectral CSD on delta and theta (low-frequency) and decreased spectral CSD on (high-frequency) alpha and beta bands when compared to HE. Surprisingly, MCI patients presented an increase in gamma at precuneus and a decrease at occipital cortex. Group prediction through SVM achieved 96% accuracy, 98% specificity and 93% sensitivity when WME and ERE spectral CSD features were combined. Conclusions: Our findings confirmed the overall EEG slowing observed in classical MCI resting-state EEG literature as well as alpha desynchronization changes observed in task-related EEG literature. Furthermore, they also revealed MCI abnormalities in the gamma band. Significance: Our frequency-domain analysis of CSD patterns in task-related EEG, focusing both on pre- and post-stimulus periods, may be a clinically relevant tool to support MCI diagnosis.