SVM-based Diagnosis of the Alzheimer’s Disease using 18FFDG PET with Fisher Discriminant Rate

Alzheimer's disease (AD) is characterized by impaired glucose metabolism. It can be detected using 18F-FDG inPositron Emission Tomography (PET) medical imaging modality. In this work an automatic method for diagnosis of AD based on region of interest (ROI) is presented. Brain image of subject is automatically parcellated into 116 pre-defined ROIs using Montreal Neurological Imaging (MNI) atlas. Discovering the most discriminative regions in atlas-based approach of AD is very important. Because of the t-test, feature selection scheme widely used in medical science, is not a sensitive measure, in this study Fisher linear discriminant ratio (FDR) is evaluated. Base on features extracted from most discriminative regions, a support vector machine is adapted to discriminant normal control (NC) from AD (or mild cognitive impairment (MCI)). For classifying AD from NC, our proposed method achieves 88.1% of classification accuracy, while the accuracy of voxel-wise and t-test methods are only 79.2% and 84.4% respectively. Also proposed method yields a higher diagnostic accuracy in discriminate NC and MCI.