The Research And Application Of MCI Imaging Classification Based On Spectral Clustering

Mild cognitive impairment (MCI) is an intermediate state between normal aging and Alzheimer’s disease (AD). It is easily convert to dementia and give social and family life to create enormous pressure. Therefore, early research is the key to reducing the MCI dementia,It has become a focus of current research and has important significance.Blood oxygen level-dependent functional magnetic resonance imaging (BOLD-fMRI) of task state as a non-invasive technology that make the concentration change of oxygen as a contrast agent to avoid brain damage, and the brain can be reliable and accurate positioning. BOLD-fMRI have high complexity and high dimensional, spectral clustering can clustering on any cluster sample space and get a good clustering results. In this paper, we adopt an improved spectral clustering algorithm for BOLD-fMRI data dimensionality reduction, acquire some significant difference mode, get the brain activated voxels, The final structure classification matrix, get good classification model, It provide a reference value for the detection and diagnosis of MCI, The main work is as follows.First, this paper detail introduce traditional spectral clustering algorithm and Nystrom adaptive spectral clustering algorithm, This paper make three areas improvement based on the traditional spectral clustering algorithm according to problems of two algorithm, Three aspects respectively structure similar matrix, determine the number of packets K and memory overflow problems on large data sets.Second, Experiment collected the BOLD-fMRI data sets, to process the data, DPARSF do data preprocessing, several tests select the appropriate template and baseline values to extract BOLD rate of change.Third, traditional spectral clustering algorithm, Nystrom adaptive spectral clustering and improved spectral clustering algorithm are used for clustering brain activated voxel, Extract significant differences BOLD change pattern. In view of the BOLD-fMRI data gathered the effect of the class there are no definite evaluation standard, since this paper determine a comprehensive evaluation corresponding to BOLD significant differences patterns of activation and the location of extraction voxels in brain regions,and select the parameters, Finally, activated voxels structure classification matrix to obtain the correct classification rate, in this article obtained classification accuracy has improved compared to the previous algorithm, and extracted voxels activated have been reported in previous studies.