Extracting Salient Features for EEG-based Diagnosis of Alzheimer's Disease Using Support Vector Machine Classifier

Alzheimer's disease (AD) is one of the most common and fastest growing neurodegenerative diseases in the western countries. Development of different biomarkers tools are key issues for the diagnosis of AD and its progression. Prediction of cognitive performance of subjects from electroencephalography (EEG) and identification of relevant biomarkers are some of the research problems. Although EEG is a powerful and relatively cheap tool for the diagnosis of AD and dementia, it does not achieve the standards of clinical performance in terms of sensitivity and specificity to accept as a reliable technique for the screening of AD. Hence, there is an immense need to develop an efficient system and algorithms for diagnosis. Accordingly, the objective of this research paper is to analyze different features for the diagnosis of AD to serve as a possible biomarker to distinguish between AD subject and normal subject. The research is carried out on an experimental database. Three different features such as spectral-, wavelet-, and complexity- based features are computed and compared on the basis of classification accuracy obtained. The classification is carried out using support vector machine classifier giving 96% accuracy on complexity- based features and increased performance in terms of sensitivity and specificity. The results show the improved performance in the diagnosis of AD. It is observed that the severity of AD is depicted in EEG complexity. These features used in research work can be considered as the benchmark for AD diagnosis.