Diagnosing Alzheimer’s Disease and Frontotemporal Dementia Using Machine Learning and EEG

Abstract

Alzheimer’s disease (AD) and frontotemporal dementia (FTD) are types of neurodegenerative dementias characterized by progressive cognitive decline. Electroencephalography (EEG) signal analysis is becoming a promising, inexpensive method to early diagnose AD and FTD. Prior research has applied machine learning to classify AD and healthy patients (HC) based on EEG readings, but not distinguishing between AD, FTD, and healthy patients. In the present paper, power spectral features will be extracted from raw EEG recordings for random forest classifiers and artificial neural networks to differentiate among AD, FTD, and HC patients. The first two minutes of 88 EEG recordings were used from the 2nd Department of Neurology in Thessaloniki. The models achieved 96%, 92%, and 95% accuracy when dealing with binary classification problems (AD and HC, AD and FTD, HC and FTD) and 90.4% for all three classes. The accuracies improve upon the results of previous literature.