Fractality and a wavelet-chaos-methodology for EEG-based diagnosis of Alzheimer disease

Abstract

Recently the senior author and his associates developed a spatiotemporal wavelet-chaos methodology for the analysis of electroencephalograms (EEGs) and their subbands for discovering potential markers of abnormality in Alzheimer disease (AD). In this study, fractal dimension (FD) is used for the evaluation of the dynamical changes in the AD brain. The approach presented in this study is based on the research ideology that nonlinear features, such as FD, may not show significant differences between the AD and the control groups in the band-limited EEG, but may manifest in certain subbands. First, 2 different FD algorithms for computing the fractality of EEGs are investigated and their efficacy for yielding potential mathematical markers of AD is compared. They are Katz FD (KFD) and Higuchi FD. Significant features in different loci and different EEG subbands or band-limited EEG for discrimination of the AD and the control groups are determined by analysis of variation. The most discriminative FD and the corresponding loci and EEG subbands for discriminating between AD and healthy EEGs are discovered. As KFD of all loci in the β subband showed very high ability (P value <0.001) in discriminating between the groups, all KFDs are abstracted in 1 global KFD by averaging across loci in each of the 2 eyes-closed and eyes-open conditions. This leads to a more robust classification in terms of common variation of electrode positions than a classification based on separate KFDs of certain loci. Finally, based on the 2 global features separately and together, linear discriminant analysis is used to classify EEGs of AD and elderly normal patients. A high accuracy of 99.3% was obtained for the diagnosis of the AD based on the global KFD in the β-band of the eyes-closed condition with a sensitivity of 100% and a specificity of 97.8%.