Similarity - based analysis of FDG-PET images of alzheimer's disesase patients :|a method for automated diagnosis and severity prediction with the aim of therapy response monitoring

Abstract

This study aimed to evaluate 18-Fluorodeoxyglucose positron emission tomog raphy (18F-FDG-PET) images of the brain for the computer-aided characterization detection of Alzheimer’s disease (AD) intuitive image similarity-measure-based ap proach. The first objective was to diagnose AD automatically. The second objective was to determine the association between the similarity measure and neuropsycholog ical assessments. Therefore, we aimed to develop a new AD evaluation algorithm that can give an early diagnosis of the disease and define an objective severity index that correlates with well-known neuropsychological tests. 125 patients with AD, 132 Cog nitively Normal (CN), and a total of 257, FDG-PET data were obtained from ADNI. We then found a distance value indicative of the similarity between any 3D image to available CN and AD patient images in the database using the mutual information method. The diagnosis was based on a threshold value for the distance value. Then, the Mini Mental State Examination (MMSE) and Clinical Dementia Rating (CDR) results of all patients and the distance values obtained from FDG-PET were analyzed using an analysis of variance. The algorithm achieved an AUC ROC of 0,969 using a leave-one-out method for the original dataset (n=197) and 0,873 using the independent testing dataset (n=60). The correlation was 0,642 between MMSE scores and imaging scores, and for CDR global test correlation between imaging and testing was 0,677. A simple and intuitive similarity-based algorithm can be used for the early detection of AD using molecular imaging as well as determining an objective severity index. No ROI and feature computations should be performed.|Keywords : Alzheimer’s disease, 18F-FDG-PET, Similarity Index, Neuropsychological Assessments, Severity Index.