Classificação de imagens de ressonância magnética nas classes Normal, Comprometimento Cognitivo Leve e Alzheimer usando projeções PCA e Kernel PCA, e máquinas de vetores suporte

Abstract

Given the aging of the population in Brazil and the world, with a possible inversion of the age pyramid, moving towards an increasing number of older adults (≥ 60 years), projected 1.4 billion, 16.4% of the world population, and 42.5 million, 18.63% of the population in Brazil, by 2030. This transition and an estimated world population with dementia syndromes of 150 million – Alzheimer’s Disease (AD) corresponding from 60 to 80% of these cases – and as a result of these numbers, an estimated cost of 1 trillion U.S dollars with a projection to double by 2030. There is also the problem of time spent and the accuracy of specialists for the diagnosis, due to the way of differentiating before the affected people and the gradual symptomatic process. These symptoms include difficulty remembering new information, difficulty in solving problems, and completing familiar tasks at home, confusion about time or place, and problems in interpreting visual stimuli. Symptoms are reflections of a characteristic pathology that consists of progressive atrophy of the brain, mainly in the cortical and subcortical structures, including its biomarker and the hippocampal region. An imaging method that presents high quality in the visualization of brain structures is magnetic resonance imaging, which can be used in the diagnosis of AD. In this context, the objective of this work is to apply a set of algorithmic techniques that can identify, from the magnetic resonance images of brains, if a patient is healthy, has AD or has a mild cognitive impairment (MCI). In order to reduce the computation and the volume of data of the magnetic resonance images, techniques of extraction of characteristics were used. Projecting a set of magnetic resonance images on the bases, whether linear related to the Principal Component Analysis (PCA) or non-linear related to the Kernel PCA and applying to the models trained with better performance in face of the collected statistical metrics (precision, balanced accuracy, sensitivity, specificity and Area Under the Curve (AUC)), thus creating a computer-aided diagnostic system to assist the specialist in identifying this neurocognitive disorder (AD), enabling an early diagnosis to take action against the disease’s progress.