Analysis Of Left Ventricular Movement Of Tagged Mri Based On Svm

The Support Vector Machine (SVM) is a novel machine learning method proposed by Vapnik et al. It has becoming a hotspot in the field of machine learning due to its excellent performance. It also has been successfully applied in many different fields, e.g. face recognition, handwritten numerals recognition, text auto-categorization. The purpose of the research on tagged MRI of left ventricle (LV) deformation analysis is to attain the track of each material point motivation in LV. In this paper, SVM is imported into the deformation analysis of LV in a creative way , and a new model based on SVR is proposed. The work including: (1) Systematic introduction of Support Vector Regression (SVR) We presented the knowledge of Support Vector Regression (SVR) systematically since there have being few papers about it now. In detail, we presented the basic theory on the studying level and the algorithms of SVR such as ε -SVR and v -SVR on the algorithm level.(2) The improvement of training method on SVMBased on the deeply analysis of SVM learning, we proposed an improved training method based-on v -SVR. Simulation experiments are implemented to show the performance also.(3) Discussion of SVM Kernel and the parameters of KernelThe most common problem we will meet in SVM learning is how to select SVM Kernel and Kernel parameters, which has been researched deeply by many researchers. This paper also studys this field and proposes a improved strategy of Kernel selection, which was proved to have a good performance by simulation experiments.(4) Design the computational model of LV deformationSince the cardiac movement includes many different kinds of deformation e.g. displacement, torsion, systole and diastole. There are few rules we can follow, it seems difficult to regress or predict the trace of cardiac movement according to the sparse data observed. The model proposed in this paper has many merits, such as simple algorithms, no needs of hypothesis on cardiac muscle, precise calculation, fast implementation, etc.