Our study focuses on EEG-based analysis of Alzheimer’s disease (AD) and related disorders using a Green-Computing AI framework. The method relies on coherence matrices to quantify functional connectivity between 19 scalp electrodes across five standard frequency bands (delta, theta, alpha, beta, gamma). Each coherence matrix reflects the degree of synchronization between pairs of brain regions, providing a compact representation of neural interaction patterns.
After computing the coherence matrices, we apply the Discriminant Feature Test (DFT) to identify the most informative features for distinguishing disease groups. DFT ranks all coherence features according to their discriminative power, measured by entropy-based separability across classes. The top-ranked (K_n) features from the upper triangle of the matrix are retained as raw discriminative features.
For each electrode pair with at least (m) (1 < m ≤ 5) selected band features, we further derive two complementary representations:
- Linear Normal Transform (LNT) features, which map the selected coherence values into a linearly separable subspace; and
- Support Vector Machine (SVM) features, which capture nonlinear decision boundaries between classes.
The final feature vector combines the selected raw, LNT, and SVM features. Three binary classifiers—AD vs CN, AD vs FTD, and FTD vs CN—are trained using a leave-one-out cross-validation scheme. For each test subject, the outputs from the relevant binary classifiers are averaged to form the final multi-class prediction.
