Classification among healthy, mild cognitive impairment and Alzheimer's disease subjects based on wavelet entropy and relative beta and theta power

Diagnosis of Alzheimer's disease (AD), mild cognitive impairment (MCI), and healthy subjects (Healthy) is currently lacking an automated tool. It requires experience of neuropsychologists and has sensibilities of 80% when separating between Healthy and MCI. The aim of this work is to evaluate the performance of a method for classification among the three groups using a database of 17 Healthy, 9 MCI and 15 AD. The method uses wavelet decomposition of the EEG signal (Haar mother wavelet and 5 decomposition levels) to calculate the wavelet entropy and theta and beta relative power of the EEG signal. These features are used as inputs to a three-way classifier consisting in a support vector machine with polynomial kernel and a two-layer neural network. The last implements a vote procedure. Wavelet entropy was evaluated together with the sample entropy and approximated entropy to choose the one that best detected changes in the complexity of the EEG signal. The results show that it is possible to automatically classify a subject of a particular group with an overall accuracy of 92.6%, close to the best result found in the literature that is 97.9%. The method could be the basis for the implementation of a diagnosis-support quantitative tool oriented to aid in clinical diagnosis, especially when the classification between the three groups is not one of the more represented researches in the consulted literature.

Automated summary

By using wavelet decomposition and EEG signals, the study successfully achieved automatic classification of Alzheimer's disease, mild cognitive impairment, and healthy subjects with an accuracy of 92.6%. This method could be the basis for implementing a diagnosis-support tool.