Application Of MRF Model In Computer Aided Diagnosis Of Breast Cancer

At present,breast cancer disease has gradually developed into one of the diseases with the highest incidence of endangering women's health.By actively examining and obtaining initial treatment,breast cancer mortality can be effectively reduced.In the past,manual diagnosis of the diagnosis of benign and malignant breast tumors,the diagnosis depends on the doctor's experience to a large extent,at the same time due to visual fatigue and other reasons resulting in the diagnosis of the doctor during the diagnosis of errors,omissions occur.With the development of computer science and technology,medical computer-aided diagnosis systems have emerged,and to a great extent reduce the dependence of diagnosis results on physician experience,can make objective analysis of medical images,and greatly improve the efficiency and accuracy of physician diagnosis.The rate is of great significance for improving the detection rate of breast cancer so that patients can obtain timely and effective treatment.Extracting the mass in the image is the basis for the classification of benign and malignant breast cancer computer-aided diagnosis systems.For the problem of the over-segmentation caused by pairwise markov random field(Markov Random Field,MRF)segmentation algorithm using simple prior model when used the algorithm in breast mass segmentation,an improved MRF segmentation algorithm based on simple linear iterative cluster(Simple Linear Iterative Cluster,SLIC)algorithm was proposed.In the first step of the algorithm,the SLIC algorithm is used to pre-segment the image into super-pixel blocks with high local area consistency.According to the characteristics of the super-pixel area,a corresponding neighborhood system is constructed and an MRF is constructed,replace the pixels with superpixels as the process unit to accomplished the segmentation of breast mass.The experimental results show that the MRF segmentation algorithm based on SLIC acquire a satisfactory segmentation effect in breast mass segmentation efficiently.Before the diagnosis of the mass is achieved,the image feature vector needs to be determined.Since feature vectors are the basis for the final classifier classification,it is critical to extract the appropriate features.For this reason,this paper proposes an improved sparse representation Local Binary Pattern(LBP)algorithm to effectively extract the texture features of the mass.Firstly,the LBP values of each point in the mass area were extracted in blocks,and the LBP histograms of each patch area were extracted to complete the normalization.These histograms were combined to obtain the texture features with larger dimensions.Then,the LBP feature vector is dictionary-learned and the feature vector is sparse.This paper uses the K-SVD algorithm to continuously update the dictionary matrix until the result meets the sparse requirement,so that the final texture feature vector is sparse and the complexity of the classifier is reduced.The experimental results show that this method is effective not only to reduce the false positive result of diagnosis,but also to improve the overall performance of CAD system of breast cancer.