Clustering and modeling of EEG coherence features of Alzheimer's and mild cognitive impairment patients

Using multiple discriminant analysis (MDA) and k-means clustering, coherence features extracted from the EEGs of a group of 56 subjects were analyzed to assess how feasible an automated coherence-based pattern recognition system that detects Alzheimer's disease (AD) would be. Sixteen of the subjects were AD patients, 24 were mild cognitive impairment (MCI) patients while 16 were age-matched controls. With MDA, an overall classification rate (CR) of 84% was obtained for AD vs. MCI vs. Controls classifications. The high CR implies that it is possible to distinguish between the three groups. The coherence features were also statistically analyzed to derive a neural model of AD and MCI, which indicated that patients with AD may have a greater number of damaged cortical fibers than their MCI counterparts, and furthermore, that MCI may be an intermediary step in the development of AD.

I. Introduction

Coherence features calculated from the electroencephalogram (EEG) have been successfully used to distinguish Alzheimer's disease (AD) and/or mild cognitive impairment (MCI) patients from age-matched controls [1]–[3]. Specifically, AD patients show lower coherence than MCI patients and controls [3]–[5]. And, MCI patients also show reduced coherence relative to controls [1], [2].