Detecção e classificação automáticas de alterações estruturais cerebrais em imagens de ressonância magnética para o auxílio ao diagnóstico do Alzheimer

Abstract

Alzheimer's disease is a progressive and irreversible neurodegenerative condition with development characterized by asymmetrical brain atrophies. Across life, due to neuronal aging, a cognitively healthy brain presents structural changes. However, with the development of the Alzheimer's, these changes are more accentuated, being less intensive in the prodromal stage of Alzheimer's, known as mild cognitive impairment, and more pronounced as the disease progresses. In this research, we developed three approaches to aid in diagnosing patients with mild cognitive impairment and mild Alzheimer's disease, each of which analyzed distinct and complementary properties related to the disease, namely asymmetry, atrophy, and biological brain aging. In addition, we combined and tested the approaches to measure impact on outcomes. In the analysis of structural asymmetries, we extracted multi-scale attributes of the hippocampal regions and created an asymmetry index. We obtained classification results comparable to other studies focused on the hippocampal regions, and the created asymmetry index showed statistically significant results consistent with the medical literature. In the analysis of structural atrophies, we automatically selected and classify discriminative landmark points between populations. Next, we classify the images based on a quantitative assessment obtained from the results of the landmark points. The obtained image classification results surpassed many studies published in the literature and are comparable to others. In the analysis of the biological aging of the brain, we developed an age estimation model using deep learning and images of cognitively normal patients with the same age range as patients with mild cognitive impairment and mild Alzheimer's disease. The age estimation error between the groups showed statistically significant results and presented a significant correlation with the Mini-Mental State Examination clinical test. The estimation results were competitive among existing studies, but the classification results were below expectations. Finally, combining the first two approaches brought positive contributions to the results, with up to 3% gains in AUC.