Computer-aided Diagnosis Algorithm Of Alzheimer’s Disease Based On Image Information Fusion

Alzheimer’s disease is usually insidious onset, duration of slow and irreversible,and the major clinical manifestation is intellectual impairment. Due to damage anddeath of brain cells are irreversible, while the medical profession has not yet beenfound effective method in treating this disease, so early diagnosis and prediction ofAlzheimer’s disease has become particularly important. With the development andmaturity of neuroimaging, imaging technology has played an increasingly importantrole in the diagnosis of brain lesions. In this paper, computer-aided diagnosisalgorithm of Alzheimer’s disease based on image information fusion is researched.A method which can accurate segment areas of the brain associated withAlzheimer’s disease is proposed, also, in view of single-mode images of MRI or PETfor classification of Alzheimer’s disease, the accuracy is not high，proposed a methodwhich can fuse MRI and PET3D image information. This scheme mainly includesobtaining the objects of study, image registration, image information fusion, featureextraction and feature classification ifve parts.All data used in this paper are selected from the ADNI database, including87subjects aging from60to90，with both their PET and MRI images.For image registration, since the peformance of registration is crucial to theintegrity and accuracy of feature extraction, choosing a proper registration algorithmis extremely important. In this paper, an affine registration is first applied as a linearmodel to remove global transformations between images, then, a non-linear method,named Log Domain Demons，is applied to remove local deformations. Once the linearand non-linear registration is done，the size，position, direction and textures of thealigned images become consistent. Image registration lays the foundation of thefollowing segmentation and feature extraction steps.For image information fusion, due to the high accuracy of registration, here wedirectly integrate MRI and PET3D images with the weighted average method in pixellevel, that is，weighted average gray value of each registered MRI images and theirscorresponding PET images.For feature extraction, voxel values of MRI, PET and the fused images areregarded as features in the thesis. Since MRI images present more morphometricinformation than PET does，an MRI template is created. Segment the template and use the segmented template to extract features of the87individuals, including the wholebrain, gray matter,34cerebral cortex areas and salient regions of two-sample t-test.For feature classification, supported vector machine (SVM) is introduced in thisthesis. First, the theoretical principle of SVM is described. Then，present thegeneralization ability of SVM and the type of kernel function. On this basis, use SVMto classify the extracted sample characteristics. Last, the idea of cross-validation isapplied to assesse the results of classification.With the above methods, theoretical analysis and experiments demonstrate thatthe proposed computer-aided diagnosis algorithm of Alzheimer’s disease based onimage information fusion has satisfying performance.