Research On Segmentation Method Of Mammography Target X-ray Image Based On Neural Network

Breast cancer has a higher the incidence rate and mortality rate than other disease.Early treatment can effectively reduce the mortality patients for breast cancer.Medical imaging can detect breast cancer earlier,and is simple,noninvasive and safe.It is the first choice for the clinical diagnosis of breast cancer.Breast mass and calcification are the important manifestations of breast cancer.However,the breast tissues and lesions have low contrast due to low pixel resolution.Related physicans probably bring the error diagnosis when reading a large number of related images.This paper takes the mammography images as the research object,focusing on the segmentation of breast mass and the determination of benign and malignant calcifications.It also uses pulse coupled neural network models（PCNN）+an attention mechanism model+Graph-Based Visual Saliency（GBVS）to improve the segmentation accuracy of breast masses and the recognition accuracy of benign and malignant calcifications.Related works are introduced as follows:（1）To solve the problems about the less images and unequal benign and malignant calcification images,we add some experimental images for Mammographic Image Analysis Society（MIAS）database and（Digital Database for Screening Mammography,DDSM）database.The above two database contain 2400 experimental images,including 1600 training images and 800 testing images.（2）As the traditional PCNN models have low computer complexity and the mammography images are easy to bring much noise in most cases,we propose an improved FC-MSPCNN model（MFC-MSPCNN）,which optimizes the weight matrix NW_ijkl and linking strength parameterβ,resets the decay factorαand the parameter V,and only needs two iterations to obtain the experimental results.（3）As the complex information of mammography images,mass segmentation is easily affected from high-density glands and tissues.In this paper,a multi-step breast mass segmentation method based on PCNN and GBVS is proposed,and gets the finer mass segmentation results.The GBVS model can well recognize the significant region and get the coarse segmentation result of breast images.The MFC-MSPCNN model can accurately segment the mass regions of breast images.Morphological algorithm is used to obtain the final segmentation results of breast images.It is tested in MIAS database and DDSM database to verify the effectiveness and robustness of the method.（4）According to the characteristics of mammography images,a debugged FC-MSPCNN model（DFC-MSPCNN）+Swin Transformer model are proposed to determine the benign and malignant calcifications,remove the previous linking strengthβand weight matrix W_ijkl,reset the parameters,and define the key parameters.The swin transformer model is designed by using window moving operation.After each feature extraction,the down sampling operation is carried out,and a better judgment result of benign and malignant calcifications is obtained.