Resting-State Quantitative EEG Spectral Slowing and Machine Learning for Automated Screening of Alzheimer’s Disease

The current study assesses the diagnostic utility of resting-state quantitative electroencephalography (qEEG) spectral profiles together with supervised machine learning for the automated classification of Alzheimer’s disease (AD). We analyzed the dynamics of resting-state, ۱۹-channel eyes-closed EEG from a cohort of ۶۵ age-comparable participants (۳۶ AD and ۲۹ healthy controls). Power spectral densities were computed for canonical frequency bands using the Welch method, and a composite Brain Slowing Index (BSI) was calculated to capture pathological low-to-high frequency shift. Relative theta power (p=۰.۰۰۰۱) and global BSI (p=۰.۰۰۸۵) were significantly increased and relative alpha power (p=۰.۰۰۲۶) markedly decreased in the AD group as revealed by non-parametric analysis. Three classifiers were trained and tested using a leakage-free stratified CV-۵ fold procedure. The Support Vector Machine with a radial basis function kernel (RBF-SVM) outperformed the other models with a cross-validated diagnostic accuracy of ۸۶.۱۵%±۵.۳۴%, sensitivity of ۸۸.۵۷%±۶.۹۰%, and an Area Under the ROC Curve (ROC-AUC) of ۰.۹۱۵ ± ۰.۰۵۲. These findings show that automated qEEG spectral analysis is a highly sensitive, noninvasive, and cost-effective digital screening tool for clinical triage.

کلیدواژه ها:

Alzheimer’s disease ، Quantitative EEG ، Spectral slowing ، Brain Slowing Index ، Machine learning