Machine learning approaches for the study of AD with brain MRI data

Abstract

The use of automated or semi-automated approaches based on imaging data has been suggested to support the diagnoses of some diseases. In this context, Machine Learning (ML) appears as a useful emerging tool for this purpose, allowing from feature extraction to automatic classification. Alzheimer Disease (AD) and Frontotemporal Dementia (FTD) are two common and prevalent forms of early-onset dementia with different, but partly overlapping, symptoms and brain patterns of atrophy. Because of the similarities, there is a need to establish an accurate diagnosis and to obtain good markers for prognosis. This work combines both supervised and unsupervised ML algorithms to classify AD and FTD. The data used consisted of gray matter volumes and cortical thicknesses (CTh) extracted from 3TT1 MRI of 44 healthy controls (HC, age: 57.8±5.4 years), 53 Early-Onset Alzheimer Disease patients (EOAD, age: 59.4±4.4 years) and 64 FTD patients (FTD, age: 64.4±8.8 years). A principal component analysis (PCA) of all volumes and thicknesses was performed and a number of principal components (PC) that accumulated at least 80% of the data variance were entered into a Support Vector Machine (SVM). Overall performance was assessed using a 5-fold crossvalidation...