Abstract

Automated classification of FDG-PET images combining voxels of interest and neuropsychological assessments

Author(s): Wenlu Yang, Xiaoman Zhang, Fangyu He, Xudong Huang

To performthe automated classification ofAD orMCI subjects vs. healthy control (HC) subjects fromADNI PET images database, the study presents a novel systematic method of combining voxels of interest in positron emission tomography (PET) images and the neuropsychological assessments of subjects. It aimes to find the appropriate technology for the early detection of AlzheimerÂ’s disease (AD) or mild cognitive impairment(MCI). The method includes four steps: pre-processing, extracting independent components using ICA, selecting voxels of interest, and classifying them using a Support Vector Machine (SVM) classifier. PET image data were obtained fromtheADNI database including 91 HC, 50 patientswith baseline diagnosis ofAD and 105 patients with a baseline diagnosis of MCI. As a result, we achieved an excellent discrimination between AD patients and HC (accuracy 97.5\%, sensitivity 93.5\%, specificity 99.7\%), and a good discrimination betweenMCI patients and HC (accuracy 94.5\%, sensitivity 92.7\%, specificity 96.5\%). The experimental results showed that the proposed method can successfully distinguish AD or MCI from HC and that it is suitable for the automated classification of PET images