Analysis of Dementia EEG Signals Using Empirical Mode Decomposition Variants and Deep Learning

Abstract

Dementia is among the most frequent neurological diseases that result in worsening cognitive functions and damaging consequences on quality of life. In this study, a novel method is proposed by using electroencephalogram (EEG) signals and signal decomposition methods to detect Alzheimer's disease (AD). Intrinsic Mode Functions (IMFs) are obtained by utilizing Empirical Mode Decomposition (EMD) and Ensemble Empirical Mode Decomposition (EEMD) methods. Five time-domain and five spectral-domain features are extracted from one-minute and five-second EEG segments, as well as from the corresponding IMF segments, using EEG recording. Topographic heat maps are generated for each feature. These feature maps give temporal and spatial information simultaneously on a single image. Feature map images are classified with two-dimensional convolutional neural network (2D-CNN). Three CNN architectures are used for classification and comparison between networks. The proposed method achieves promising results, with accuracy up to 96%.