³¹P MRS Data Diagnosis Of Hepatocellular Carcinoma Based On Support Vector Machine

³¹Phosphorus Magnetic Resonance Spectroscopy (³¹P MRS) is an effective and non invasive technique for researching the biological metabolism and chemical compound changes of liver. It has significant clinical value and practical significance for analyzing the hepatic disease. This paper firstly pre-processes the hepatic ³¹P MRS pictures, and changes the curves of the pictures into numeric and available data, so the new subset of data sample is established. Then the classifier model which based on the Machine Learning algorithm-Support Vector Machine is designed, and it separately based on linear, polynomial and radial basis function kernel. In order to classify the hepatic ³¹P MRS data, and distinguish three categories of liver (normal liver, hepatocellular carcinoma and hepatocirrhosis), the classifier is used separately on the two data subsets, one subset includes 20 parameters which are often used in medical diagnosis, the other includes all spectroscopy data. In order to improve the performance of classifier, and find the features which are meaningful for medical diagnosis, the method of feature selection is used, which can dispose the irrelevant and interferential features. The optimal feature subset is established, and it is the input information of classifier. This paper presents the GA-SVM classification model,and cross-validation of k-fold is used to estimate the performance of the model.The iterative experiments show that the GA-SVM model has a good performance for hepatic ³¹P MRS data classification. Especially when the number of feature is 20, optimal feature subset is the input information of classifier and the performance of the model is obviously better than the feature subset which includes 20 features of medicinal. In this paper, Mahalanobis distance based on statistical numeric characteristic is applied to design the fitness function, and the selection which based on stochastic universal sampling is used to construct the GA algorithm, and the experiments prove the effectiveness of the algorithm. The optimal feature subset which is searched by algorithm provides an important referential value for liver disease.