Microstate Analysis of Resting-State EEG Signals for Classifying Tinnitus from Healthy Subjects

Abstract

Background: Electroencephalography (EEG) is a noninvasive technique for studying brain electrophysiology with high temporal resolution. Microstate analysis examines EEG recordings as a succession of quasi-stable microstates, allowing evaluation of extensive brain network activity linked to neuropsychiatric disorders like tinnitus. Objective: This study distinguishes tinnitus patients from healthy controls by using features acquired by microstate analysis. Methods: This study investigated EEG microstate differences between 16 healthy controls and 10 tinnitus patients. Four microstates were extracted and analyzed using Multivariate Analysis of Variance (MANOVA), revealing significant differences in duration, coverage, and occurrence between groups. Machine learning algorithms, including support vector machine (SVM) and K-Nearest Neighbors (KNN), and others were employed to classify tinnitus patients based on microstate features, achieving high accuracy, precision, specificity, recall, and F1-score. Results: MANOVA analysis revealed a significant difference in the duration of microstate A, which is associated with phonological processing and auditory perception, between the two groups. Additionally, significant differences in the coverage and occurrence of microstate B, related to visual networks, were observed. The SVM classifier achieved the highest accuracy of 96.44% in differentiating tinnitus patients from healthy controls, with impressive precision (97.64%), specificity (95.62%), and F1-score (97.24%). KNN also performed well, achieving a maximum recall of 97.24%. Conclusion: This study reveals the potential of EEG microstate analysis, incorporating time-related features, to improve tinnitus diagnosis and classification. Using SVM and KNN, we achieve high accuracy in identifying tinnitus-associated brain patterns, highlighting the clinical utility of EEG for neurological disease management.