Research On Medical Image Classification Based On And SVM

With a variety of image equipment widely used in medicine, the computer-aideddiagnosis based on medical image has developed rapidly in recent years, which can helpthe radiologists make more accurate judgments and identification to medical images.Pattern classification based on the feature extraction is an important step incomputer-aided diagnosis of medical image. It increase the difficulty that the medicalimage classification is a small sample problem which may cause overfitting orunderfitting. So the classifier should fit the small sample and be able to overcome theoverfitting. Support vector machine (SVM) has good classification capability and a goodpromotional in small sample, nonlinear and high-dimensional space by replacing thetraditional empirical risk minimization with structural risk minimization. So SVM is agood selection to medical image classification. Research in this area is in the ascendant,and it has important theoretical significance and broad application prospects. Thisdissertation includes the following main research results:(1) Feature selection will also affect the classification accuracy in classification. Butclassical SVM can not distinguish the importance of attributes between training samples.An improved algorithm for attributes reduction, which is based on the rough sets andinformation gain, is put forward in this paper. It is used to distinguish the importance ofthe samples attributes and reduct the attributes. Both the feature dimension of the inputsamples and the training time of the algorithm is reduced.(2) High-quality samples is required as the classification accuracy of SVM isdirectly dependent on the selected training samples. Extract samples by clusteringtechnology. The importance of each attribute is considered as the same in traditionalcluster analysis but it is different in practical applications. Therefore, a new concept ofweighted similarity degree is presented which accords with the reality much more.(3) It increase the difficulty that the medical image classification is a small sampleproblem which may over or less learn. And SVM is a good choice for medical image classification. The proposed method is applied to medical image classification whichshows that method presented can get 96.77%。...