Pattern Classification Of MRI With Cognitive Impairment

As a new discipline, brain science ripes with the development of brain imaging technology, especially after the magnetic resonance imaging(MRI) technology springs up, people can carry out analysis and research on the brain structure through the more precise brain imaging data. After the pattern recognition method was introduced in the brain science research, its applications continue to expand with the huge advantage of machine learning, a typical example is to find the pathogenic brain region of the neurological diseases by finding the difference between diseases patients and normal subjects on the brain magnetic image in medicine science. At the same time, through the study and train on the difference between the patients and the normal subjects, we can help the diagnosis of the disease.In this paper, we performed the pattern classification study of the Alzheimer's disease(AD) based on the brain magnetic resonance imaging. We first conducted the pattern classification analysis of gray matter and white matter based on Voxel-Based Morphometry(VBM), we used Principal Component Analysis(PCA) and Locally Linear Embedding(LLE) as the feature selection algorithm, respectively.And we used linear support vector machine(SVM) as the classification algorithm. Then we used Histogram Oriented Gradient(HOG) algorithm as the feature description algorithm to the converse the feature, and after that we conducted the classification experiments again.We determined the optimal classification algorithm in this experiment through a series of comparison between different combinations of each algorithm. Among them, the optimal classification accuracy of AD patients and normal control is 0.90 of gray matter with the algorithm combination of VBM+HOG+PCA. Finally we find the difference in brain regions between each classification pairs using statistical differences research method and characteristic analysis method.Finally, through the analysis on the classification performance, we had a relatively complete understanding of the ideas and methods of using pattern recognition to diagnosis neural disease and to find pathogenic brain regions based on brain magnetic image, and our work also had certain positive effect on promoting the application of pattern recognition method in this field.