Creative Commons

This dissertation was written as a part of the MSc in Data Science at the International Hellenic University. Alzheimer's disease (AD) is one of the most common neurodegenerative diseases in modern society and is the main cause of cognitive impairment in elderly people, affecting nearly 70% of more than 55 million dementia cases worldwide. Although there is no cure for AD, early diagnosis of the disease can improve the quality of life of AD patients. Electroencephalography (EEG) is one of the most used methods for the diagnosis of Alzheimer’s disease. This dissertation presents a novel method, by developing a Time Delay Neural Network (TDNN) model, aiming to correctly classify a dataset of people to predict who is healthy, who is having Mild Cognitive Impairment and who is suffering from Alzheimer’s disease, with the use of EEG. The EEG data were split into 2-second no-overlapping segments and preprocessing methods, segments like filtering and Fast Fourier Transform (FFT), were applied to them. Two training methods for splitting the data were used: by segment or by patient, with the later method being something newly introduced to the current literature. Overall, the approach of this thesis is something different from what can be found in the literature, because TDNNs have never been used before, for the prediction of Alzheimer’s disease. However, due to the time-series nature of our data, some experiments managed to achieve up to 99% accuracy.