Journal of Electronic Measurement and Instrument (电子测量与仪器学报) 2013/27:9 PP.850-858
A method which can accurate segment areas of the brain associated with Alzheimer's disease is proposed. Firstly, a template which fits for all the samples is designed. Secondly, during the acquisition of MR features, we make full use of deformation field information from the registration of images. Then, in the processing of PET images, we pre-register them to the corresponding MR images. Finally, according to features of each cerebral area, the features are classified using support vector machine. The accuracy from the whole brain is: MR: 0.8736, PET: 0.9195, 0.3 MR+0.7PET: 0.8621; from the gray matter is: MR: 0.8736, PET: 0.9195, 0.3MR+0.7PET: 0.8621; and from the two sample t-test is: MR: 0.8391, PET: 0.9195, 0.3MR+0.7PET: 0.8966. When the classification is carried out using cerebral cortical areas, the highest image accuracy in entorhinal cortex of MR is 0.8931, the highest image accuracy in precuneus of PET is 0.9195, and the highest image accuracy in entorhinal cortex of 0.3MR+0.7PET is 0.9425. The experimental results show that compared with existing methods, this method can distinguish mild AD patient and normal elder more accurately, and which will help to prevent and early diagnose AD.