Automatic Diagnosis of Mild Cognitive Impairment Using Electroencephalogram Spectral Features

Abstract

Alzheimer's disease (AD) is one of the most expensive and fatal diseases in the elderly population. Up to now, no cure have been found for AD, so early stage diagnosis is the only way to control it. Mild cognitive impairment (MCI) usually is the early stage of AD which is defined as decreasing in mental abilities such a cognition, memory, and speech not too severe to interfere daily activities. MCI diagnosis is rather hard and usually assumed as normal consequences of aging. This study proposes an accurate, mobile, and nonexpensive diagnostic approach based on electroencephalogram (EEG) signal. EEG signals were recorded using 19 electrodes positioned according to the 10-20 International system at resting eyes closed state from 16 normal and 11 MCI participants. Nineteen Spectral features are computed for each channel and examined using a correlation based algorithm to select the best discriminative features. Selected features are classified using a combination of neurofuzzy system and k-nearest neighbor classifier. Final results reach 88.89%, 100%, and 83.33% for accuracy, sensitivity, and specificity, respectively, which shows the potential of proposed method to be used as an MCI diagnostic tool, especially for screening a large population.

Keywords: Early Alzheimer's disease; electroencephalogram spectral features; k-nearest neighbor; mild cognitive impairment; neurofuzzy.

Figure 1

Electrode placement and scalp zones: 1 = frontal, 2 = left temporal, 3 = central, 4 = right temporal, 5 = occipital

Figure 2

Typical structure for Takagi-Sugeno inference system

Figure 3

Block diagram of proposed method

Figure 4

Classification results for zone individually selected features, before (left) and after cascading k-nearest neighbor (right)