Research On Breast Tumor Segmentation And Classification Method Of Medical Images

Purpose: Breast cancer is one of the malignant tumors which seriously threaten the women’s health. If it can’t be detected and treated in time, it may endanger the lives of patients. The computer aided diagnosis system of breast cancer is helpful to improve the efficiency and accuracy of breast cancer diagnosis, and is of great value in the research and clinical diagnosis of breast cancel. Segmentation and classification of breast tumor is one of the supporting technologies for computer aided diagnosis of breast cancer.In order to reduce the computational complexity of breast tumor segmentation algorithm and to improve the accuracy, this paper proposes a new breast tumor segmentation algorithm based on superpixels. On the medical image segmentation based on superpixels, selecting of superpixels segmentation algorithm reasonably is the key to improve the accuracy and versatility of breast tumor segmentation. To select the superpixels algorithm which is suitable for the segmentation of breast tumor, this paper proposes a new performance evaluation method for the medical imaging processing. The accuracy of breast cancer diagnosis is limited based on the ultrasound or mammography images, according to the characteristics of multi-modal breast image, a selective ensemble of breast tumor classification method is proposed in this paper.Method: This paper proposes a breast tumor segmentation algorithm based on the superpixels. It firstly uses the superpixels segmentation algorithm to segment the breast ultrasound image, and then the region growing algorithm is used to achieve the segmentation of breast tumor. In order to select the appropriate superpixels algorithm, a performance evaluation method combined with the speed of algorithm, boundary adherence and boundary accuracy, boundary recall and superpixels uniformity are proposed to systematically analyse the performance of superpixels algorithm based on the characteristics of medical image, while comparing the SLIC algorithm with Normalized cuts and Turbopixels algorithm. Based on the proposed method, SLIC algorithm is choosen to segment the breast ultrasounds images.A selective ensemble classifier method based on R is designed in this paper. The method first extracts the morphological features and texture feature of breast ultrasound image and texture features of mammography images, then using SVM classifier, Naive Bayes classifier and k-nearest neighbor classifier to classify the breast tumor. By selecting the suitable individual classifier by the index R, using the maximum vote strategy, we can finally get the tumor classification results.Conclusion: This paper put forward a superpixels based algorithm to segment breast ultrasound images which can obtain an accuracy of 95.28% and a sensitivity of 59.89%, and it provides a prerequisite for the following feature extraction and classification of breast tumor. In order to select the appropriate superpixels segmentation algorithm, this paper proposes a superpixels algorithm performance evaluation method, and it can propose a new idea for the design and application of superpixels algorithm in medical imaging processing. It provides some reference for the selection of superpixels algorithms in medical image processing.In this paper, a new selective ensemble method of breast tumor classification based on indicator R is proposed. Experimental results show that the fusion based on classifier level is better than that based on feature set of breast tumors segmentation. The indicator R proposed and used in this paper can effectively select the appropriate individual classifier, and produce a better performance of ensemble classifiers which achieve the accuracy of 88.73% and specificity of 97.06%. The method can be used and improved the efficiency and accuracy of the breast tumor diagnosis. It is needed in clinical application and large-scale screening of breast tumor.