Automatic detection of Alzheimer’s disease from EEG signals using hybrid PSO-GWO algorithm

Abstract

Early diagnosis of Alzheimer’s disease (AD) is vital. EEG is effective; however, its multi-channel property leads to redundancy and affects classification performance. Current studies frequently neglect the synergy of multi-feature extraction and Intelligent optimisation algorithm for overall performance in EEG channel screening.

This study innovatively combines Particle Swarm Optimization (PSO) and Grey Wolf Optimization (GWO) to develop a PSO-GWO hybrid model that aims to overcome the limitations of PSO, in particular its tendency to converge to local optima. The model significantly improves the performance of multi-channel EEG signal screening for AD. First, geometric features are extracted from AD EEG signals using a second order difference plot (SODP), resulting in two sets of uncorrelated features. Key features SAV, SCC and CTM are selected using XGBoost feature importance ranking and statistical analysis. The Relief algorithm then merges these key features into a fused vector for each channelNext, the PSO-GWO method is used to determine the optimal channel combination (Fp1, T3, T5, P3, and O2), which is input into the XGBoost classifier. 5-fold Cross-validation and LOSO validation accuracy of 96.35% and 91.08%, respectively, are achieved between patients and the normal control group. Finally, SHAP analysis highlights the positive contributions of the selected channels, confirming the effectiveness of the framework in accelerating channel selection and improving AD detection efficiency.

This study fills the void of collaborative optimisation of multi-features and intelligent algorithms in EEG channel screening, provides an efficient framework for AD detection, and enhances the understanding of neurological disease mechanisms.