Classification of Brain MR Images for the Diagnosis of Alzheimer’s Disease Based on Features Extracted from the Three Main Brain Tissues

Abstract

Alzheimer’s disease (AD) is a progressive brain disorder that causes neurons to degenerate and die as the disease progress. AD is the most common cause of dementia, accounting for 60% to 80% of all cases, and has been recognized as a public health problem by the World Health Organization. In this study, we propose a method to aid in the diagnosis of AD that automatically extracts and classifies image features of the white matter (WM), gray matter (GM), and cerebrospinal fluid (CSF) tissues from the hippocampal regions. Our method uses the features as input to support vector machine (SVM) classifiers to perform the MR image classification in CN × AD and CN × MCI cases. For that, we preprocess all ADNI images and define the regions of interest for analysis. Then, we extract the GM, WM, and CSF tissues using an automated brain tissue segmentation method. Considering the intensities inside both hippocampal regions and each segmented tissue, we extract five statistical metrics from the voxel intensities inside each hippocampal region to use as features. Then, we train SVM classifiers with distinct kernels using a ten-fold nested cross-validation to perform the classification. From the classification experiments, the highest obtained AUC values for the CN × MCI and CN × AD classification cases were 0.814 and 0.922, respectively. It is important to emphasize that we obtained these results using an automated pipeline, with no human intervention, and a relatively small set of features.KeywordsAlzheimer’s diagnosisBrain tissue segmentationMR image classification