An EEG-based framework for Alzheimer's disease diagnosis via single-channel multi-entropy fusion and ensemble learning.

Alzheimer's Disease (AD) is a prevalent neurodegenerative disorder characterized by progressive cognitive decline, and its early diagnosis is vital for timely intervention. However, most existing EEG-based diagnostic methods rely on multi-channel devices confined to hospital settings, limiting their accessibility for community screening. To address this gap, this study aims to develop a lightweight processing approach for portable, low-cost devices to support early AD diagnosis using single-channel EEG data, integrating signal purification with complementary entropy-based feature extraction and ensemble learning. Specifically, EEG signals are first decomposed into five brain rhythms (δ, θ, α, β, γ) using Discrete Wavelet Transform (DWT). Empirical Mode Decomposition (EMD) is then applied to extract Intrinsic Mode Functions (IMFs) from the selected rhythm. Four entropy measures, including Differential Entropy (DE), Singular Spectral Entropy (SSE), Fuzzy Entropy (FE), and Sample Entropy (SE), are subsequently extracted from each IMF to construct the Multi-IMF Entropy Matrix (MIMFEM). After that, an ensemble learning framework combining five classifiers is designed to discriminate AD patients from healthy controls (HCs). The proposed method is evaluated on two public datasets using Leave-One-Out Cross-Validation (LOOCV). Using single-channel EEG data, our approach achieves classification accuracies of 97.92% on the Florida-based dataset and 81.54% on the AHEPA dataset, with corresponding F1-scores of 97.87% and 82.86%, respectively. These results, achieved with limited data sources, show that our approach provides an effective, practical solution for single-channel wearable AD screening and addresses the critical need for low-cost, accessible early diagnostic tools, advancing Artificial Intelligence (AI) in healthcare through a lightweight model.