Brain region ranking for 18FDG-PET computer-aided diagnosis of Alzheimer's disease. (May 2016)

• Record Type: Journal Article
• Authors: Garali, I.; Adel, M.; Bourennane, S.; Guedj, E.
• Is Part Of: Biomedical signal processing and control. Volume 27(2016)
• Page Start: 15
• Page End: 23
• Publication Date: 2016-05
• Subjects: Computer-aided diagnosis (CAD) -- Feature selection (FS) -- Positron emission tomography (PET) -- Support vector machine (SVM) -- Receiver operating characteristic (ROC) -- Alzheimer's disease (AD); Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2016.01.009
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Region-based approach for brain PET images classification in Alzheimer disease. Computer-aided diagnosis in neurodegenerative diseases. Rank brain region of interest ability for classification. Abstract: Positron emission tomography (PET) is a functional molecular imaging, which helps to diagnose neurodegenerative diseases, such as Alzheimer's disease (AD), by evaluating cerebral metabolic rate of glucose after administration of (18)F-fluoro-deoxy-glucose ((18)FDG). A quantitative evaluation, using computer aided methods, is of importance to improve medical care. In this paper a novel ranking method of the effectiveness of brain region of interest to classify healthy and AD brain is developed. Brain images are first segmented into 116 regions according to an anatomical atlas. A spatial normalization and four gray level normalization methods are used for comparison. Each extracted region is then characterized by a feature set based on gray level histogram moments, as well as age and gender of a subject. Using a receiver operating characteristic curve for each region, it was possible to define a Separating Power Factor (SPF) to rank region's ability to separate healthy from AD brain images. Using a set of selected regions, according to their rank, and when inputting them to a support vector machine classifier, it was possible to show that classification results were similar or slightly better than those obtained when using the whole gray matter voxels of the brain or the 116 regions as input features to the classifier. Computational time was reduced compared to the other methods to which our approach was compared.