MEFD-JTFR-Based Classification of Alzheimer's Disease, Frontotemporal Dementia, and Healthy Controls From EEG

Abstract

Neurodegenerative diseases such as Alzheimer's disease (AD) and frontotemporal dementia (FTD) are difficult to diagnose at an early stage because of overlapping clinical symptoms. In this letter, an electroencephalogram (EEG)-based classification framework is presented that combines multivariate empirical Fourier decomposition (MEFD) with joint time-frequency representation (JTFR) for the classification of AD, FTD, and healthy control subjects. The proposed framework decomposes multichannel EEG signals into oscillatory mode components and constructs a JTFR from the multivariate instantaneous amplitude and frequency of the decomposed components. Three statistical features are extracted from the JTFR to characterize temporal spectral dynamics and are classified using k-nearest neighbor, decision tree, and ensemble bagged tree classifiers. Experimental results demonstrate a classification accuracy of 98.27% under subject-dependent evaluation, while achieving 96.63% under leave-one-subject-out cross-validation. These findings indicate that the proposed framework effectively captures discriminative EEG characteristics for multiclass dementia classification. © 2017 IEEE.