The Application And Research Of Feature Extraction And Classification Method Based On MRI

With the development of medical imaging technology and the improvement of computer technology, it produces a large number of multi-angle, high-resolution medical images.It is possible to extract more details from images with computer-aided diagnosis, also reduce the burden on doctors and improve the diagnostic accuracy. In order to explore a new method for the diagnosis of Alzheimer’s disease, by extracting morphology and textural features based on brain MRI, with dynamic chaotic particle swarm algorithm, present a new inertia weight calculation function to optimize the parameters of SVM. Finally classify the images into two groups, patients with Alzheimer’s disease and normal.Based on the characteristics of brain MRI, at first the thesis pretreated images including head movement correction, denoise, segmentation, registration, etc. As the traditional method is directly based on a particular layer of the image or the whole brain to extract brain MRI feature, some brain details may be lost, the advantages of MRI such as high resolution, multi-level, multi-angle are not used. So the thesis fuses morphological features and texture features to retain more image details.The thesis is based on voxel theory to extract morphological features, firstly locate the focal regions of Alzheimer’s disease using two-sample T-test, then extract the proportion of gray matter from focal regions are used as the morphological features of the image, finally extract the key slices with statistical analysis of all image slices from whole brain for next step. For image’s texture feature extraction, the thesis convert the image from spatial domain to frequency domain using non-subsampled contourlet transform, then extract the grey level co-occurrence matrix from low-frequency region, and calculate mean value, variance, entropy, contrast and other parameters of matrix. Then extract the matrix’s mean value, variance, energy and other parameters from high-frequency region. Finally two groups of feature are extracted after analysis and optimization.Based on the morphology and texture features, the thesis classificates the MRI images with support vector machine algorithm. In order to improve classification accuracy and reduce classification time, the thesis optimizes the method to choose the parameters such as kernel parameter of SVM, the penalty factor with the improved chaotic particle swarm algorithm, at the same time propose a new method of inertia weight calculation to optimize the PSO algorithm. The experiment proved that the imporved chaos PSO improved the accuracy and speed of classification.The thesis fuses morphological features and texture features to extract multi features from brain MRI, and locate the focal regions of Alzheimer’s disease, it could effectively improve diagnosis accuracy. Then proposes a new method of inertia weight calculation to optimize the PSO algorithm, combine with SVM to further improve disease diagnosis accuracy.