Diagnosis of Alzheimer's Disease Using Atlas-Based Volume Measurement Method on 3D T1 Weighted MR Images

Abstract

Alzheimer's Disease is a brain disease that begins with aging. Diagnosis of the disease, its follow-up and measurements of the related brain regions can be performed with high-resolution three-dimensional structural magnetic resonance images. In this study, an atlas-based volume measurement and classification model were designed that can perform volumetric measurement of 116 subcortical regions on 70 Alzheimer 70 Normal 3D T1-weighted MR images taken from the OASIS database. The measured volume values were normalized by dividing gray matter, parenchyma, and total brain volume in each subject. Thus, 4 different datasets with 140x116 matrix size, including raw measured values, were obtained. Datasets were ranked from the most meaningful feature to the most meaningless feature with entropy, t-test, roc, Bhattacharyya, Wilcoxon feature ranking methods. The ranked data were combined in each cycle, respectively, and the classification process was performed by giving linear and rbf kernel support vector machines with 10-fold cross validations. Data cluster, feature ranking method and classification method that give the best results with the least feature were determined by analyzing all scenario. The effect of normalization and feature ranking methods on the classification results were examined. As a result of experimental operations, the roc feature ranking based linear support vector machine gives the highest rates with 95.71% sensitivity, 94.29% specificity, 95.00% accuracy, 0.95 area under curve values using 107 features with total brain volume normalization.