Detecção de Alzheimer por meio de eletroencefalogramas utilizando aprendizado profundo

Abstract

Alzheimer’s disease (AD) is a progressive neurodegenerative disorder that impairs memory, cognition, and behavior, affecting millions of people worldwide. Diagnosing this condition can be challenging, as neuroimaging exams are often expensive or difficult to access. Electroencephalography (EEG), which records real-time brain electrical activity, emerges as a more accessible and low-cost alternative, capable of identifying alterations in brain waves in the frontotemporal region associated with the disease. This study proposes the use of deep learning models for the automatic classification of EEG signals. The dataset includes recordings from 88 individuals, comprising patients with Alzheimer’s disease, Frontotemporal Dementia, and healthy controls. Signal preprocessing involves the selection of frontotemporal channels, segmentation of EEG signals into 10-second windows, and application of the Fourier Transform to obtain the power spectrum. These data are used as input for the convolutional neural networks developed, aiming to classify individuals according to their diagnosis. The results indicated that the models performed well in distinguishing Alzheimer’s patients from healthy controls, achieving accuracies between 85% and 93%, and showed moderate performance in differentiating Alzheimer’s from Frontotemporal Dementia, achieving accuracies between 62% and 73%.