Complexity of olfactory-evoked EEG as an evidence-based marker of Alzheimer's disease

Olfactory impairment is an early symptom of Alzheimer's disease (AD). However, currently used olfactory task-based functional magnetic resonance imaging (fMRI), functional near-infrared spectroscopy (fNIRS), and electroencephalogram features are not powerful enough to detect the impairment. To address this issue, we propose an explainable Artificial Intelligence (XAI) framework that comprises discriminant analysis/naive Bayes/thresholding classifiers driven by the sample entropy (SE) of olfactory event-related potentials (OERPs) at the Fz/Pz electrodes. The proposed XAI framework exhibits a higher accuracy (92.14%) than methods in the literature, namely, support vector machine (88.20%), logistic regression (67.42%), thresholding (82.5%), and light gradient boosting (80.68%) classifiers fed respectively by inter-electrode β-γ magnitude squared coherence of OERPs, P2 latency of OERPs, fMRI activation pattern of primary olfactory cortex, and NIRS oxygenation difference in the orbito-frontal cortex. Reduction in SE in AD patients is caused by low dynamicity of OERPs as a consequence of diminished sensitivity to the olfactory stimulus.