American Journal for Young Scientists · ISSN 3068-9015
Using EEG to make Biomarkers for Alzheimer's Research Project
Authors: Ranganathan Srinivasan, Siddharth Nori
Published: September 26, 2025
Category: Computational Neuroscience
Abstract
Alzheimer's disease (AD) is associated with disrupted neural synchrony and altered network connectivity, which can be assessed using electroencephalography (EEG). In this paper, we employed a generalized Kuramoto model to quantify directed functional coupling between EEG channels in the alpha frequency range (8–12 Hz) for both AD patients and cognitively normal (CN) controls. The model extends the classic Kuramoto framework by estimating interaction strengths and preferred phase lags for each directed channel pair via regularized linear regression, yielding coupling matrices and phase offsets. From these matrices, we computed node-level graph measures of out-strength and in-strength to characterize the influence each channel exerts on and receives from others. Comparative analysis revealed distinct coupling patterns between AD and CN groups, with AD networks exhibiting altered connectivity profiles. To evaluate the diagnostic utility of these network features, we conducted a classification experiment using Support Vector Machine (SVM) with RBF kernel, XGBoost, and Multi-Layer Perceptron (MLP) models on 14 electrodes after data preprocessing and channel selection. An 80/20 train-test split was applied. MLP achieved the highest average test accuracy (0.8824), outperforming XGBoost (0.8235) and SVM (0.7647). Electrode-specific performance indicated that occipital (O1, O2) and frontal (F3, F7) regions contributed most to classification accuracy, with O2 reaching 0.9412 accuracy using MLP. These findings suggest that generalized Kuramoto-based coupling measures combined with machine learning classification can effectively discriminate AD from CN EEG recordings, and highlight specific brain regions as potential EEG biomarkers for early detection.