Detection of Mild Cognitive Impairment Using Image Differences and Clinical Features

In this study, we present a systematic method for early detection of mild cognitive impairment (MCI) from magnetic resonance images (MRI) using image differences and clinical features. Early detection of MCI has pivotal importance to delay or prevent the onset of Alzheimer's disease (AD). Subjects were selected from the Open Access Series of Imaging Studies (OASIS)database and included 89 MCI subjects and 80 controls. T1 weighted MRI scans were analyzed to identify their voxel-by-voxel differences in gray matter (GM) intensity between MCI group and control group. Based on the differences, a threshold-based unseeded region growing algorithm was designed to determine multiple regions which atrophy is characteristic of MCI. A feature ranking method was then adopted to select a small number of regions that presented relatively more pronounced atrophy. Next, support vector machine (SVM) based classification was applied by using the clinical features of subjects and the features of selected regions. Our method was tested by leave-one-out cross-validation and it demonstrated high classification accuracy (90%).