Classificação diagnóstica de Doença de Alzheimer e demência fronto-temporal a partir da análise de eletroencefalograma por algoritmos de aprendizado de máquina

Abstract

Alzheimer’s disease (AD) is the most prevalent dementia among the elderly. It is a multifactorial pathology designated as a neurocognitive disorder that develops gradually. One of the first manifestations of the disease is memory loss, and as it progresses, it tends to affect other cognitive domains such as language, praxis, attention, executive functions and spatial orientation. The diagnosis of AD is primarily clinical, and treatment is essential with acetylcholinesterase inhibitors and NMDA (N-methyl-D-aspartate) receptor antagonists. In order to provide an objective tool to aid diagnosis, Alzheimer’s disease and frontotemporal dementia classifiers were developed based on analysis of electroencephalographic data using machine learning algorithms. The electroencephalogram data was obtained pre-processed from a public database, and the classifiers were developed using the following machine learning algorithms: decision tree, MLP (multi-layer perceptron), KNN (k-nearest neighbour), Random Forest and SVM (support vector machine). The Random Forest classification algorithm obtained the best average accuracy (0.6033 ± 0.0985) and the highest average area under the ROC curve (0.7189 ± 0.0586) after cross-validation (5 folds). The overall performance was not good, and in part, this may be due to the low sample size (88), affecting all the algorithms tested. In addition, only two features were extracted: the mean and the entropy. Perhaps extracting more features could result in greater accuracy for the algorithms.