Research And Application Of Support Vector Machine Technology In Computer-aided Medical Diagnosing System For Breast Cancer

In the context of Data Mining Technology Based on Medical Image, a project by the National Natural Science Foundation of China, this paper researches key technologies and main algorithms for the application of theories, which include Support Vector Machine (SVM) and Rough Set (RS) in the field of medicine, and put forward classification algorithm of improved Support Vector Machine, which applies in a computer-aided medical diagnosing system based on the mammograms database for breast cancer. The main research work include the following aspects.1. Image Preprocessing and Feature ExtractionAccording to the characteristics of mammograms, histogram equalization algorithm is used. For feature extraction, Gray-Level Co-Occurrence Matrices (GCMs) are constructed and 26 texture feature independent of directions are extracted, and 4 statistical features are added, and these features constructed inputs of the normal and abnormal classifier mammograms. The implemented image segmentation algorithm effectively cut up the lump of potential region, on the basis of which these features can describe the shape are extracted, and construct the inputs of the Benign and malignant classifier mammograms. The experiments show the two methods of feature extraction enhance the performanance of the classifier mammograms.2. Proximal Support Vector Machine Algorithm Based on Modified (MPSVM)Deeply analysing SVM theory, a classification idea of the proximal support vectormachine (PSVM) is studied. The algorithm is faster, easier to achieve. MPSVM, which effectively enhanced the accuracy of unbalanced dataset classification, is proposed in this paper for that there is defect of an overall unsatisfing performance result from fitting better the class with more data points when PSVM is applied to non-equilibrium sample sets.3. Classifier Based on Combining Rough Set with MPSVM (RS-MPSVM)Lucubrating in RS theory, the method combining Rough Set with MPSVM is proposed.Attribute discretization algorithm is used to discretize continuous texture feature, using RS theory to determine the importance of attributes and remove redundant features in order to reduce attributes, which can reduce dimensions of the input space, and RS-MPSVM classifier, which can be trained, is designed. Experiments show that the method effectively improved classification performance.4. Based on the features, such as mammograms's texture, shape, and support classification technology, using MyEclipse, a computer-aided medical diagnosing prototype system based on the support vector machine, which integrate the above-mentioned researches , is developed for breast cancer.