Classification of positron emission tomography brain images using first and second derivative features

Abstract

Computer-Aided Diagnosis (CAD) for Positron Emission Tomography (PET) brain images is of importance for better quantifying and diagnosing neurodegenerative diseases like Alzheimer Disease (AD). This paper presents new features based on first and second derivatives, computed on brain PET images and aiming at better image classification in the case of AD. Brain images are first segmented into Volumes Of Interest (VOIs) using an atlas. To quantify the ability of features to separate AD from Healthy Control (HC), the orientation field for each VOI is studied. First, 3D gradient images are computed. First and second derivatives over each VOI is then computed. Inputting the mean, then the first and second derivatives features within VOIs into a Support Vector Machine (SVM) classifier, yields better classification accuracy rate than when inputting only the mean value as a feature.