Feature fusion and WOA-GWO optimization for Alzheimer's disease detection with sparse EEG channels.

Alzheimer's Disease (AD) is a neurodegenerative disorder with insidious onset, making early diagnosis challenging. Electroencephalogram (EEG) is a promising noninvasive tool for AD diagnosis, but high-density EEG configurations cause computational burdens and hinder clinical translation. Thus, developing an efficient sparse EEG channel selection method with high classification accuracy is urgent for AD auxiliary diagnosis. This study proposes a multi-strategy enhanced Whale Optimization Algorithm-Grey Wolf Optimizer (WOA-GWO) hybrid model for EEG channel selection, combined with a nonlinear dynamic feature fusion framework. We extracted geometric features from second-order difference plot (SODP) and complexity features (sample entropy, fuzzy entropy) of EEG signals, then adopted the ReliefF algorithm for feature fusion and key feature selection. The WOA-GWO model was improved via chaotic initialization, nonlinear convergence factors, spiral-hierarchical position update, and random perturbation to avoid local optima. Experimental results show that the proposed framework achieves a classification accuracy of 96.97% for AD detection, with significantly reduced EEG channel dimensions (four optimal channels identified: T5, FP1, T4, F4). The WOA-GWO model outperforms the original WOA and GWO in convergence speed and optimization accuracy, and the fused features exhibit strong discriminability for AD-related EEG abnormalities. This work provides a reliable computational framework for developing lightweight, portable AD diagnostic systems, and the identified optimal EEG channels offer neurophysiological evidence for AD electrophysiological biomarkers.