Calcification Point Detection In Mammogram Based On Neural Network

At present,breast cancer has seriously threatened women’s life and health to a large extent,and has become the chronic tumor disease with the highest female prevalence and cancer mortality in the world.The prevention and detection of breast cancer in the early stage can effectively increase its cure rate and prolong the survival time of patients.Medical imaging is an effective means to realize the pre-screening and diagnosis of breast cancer.Doctors can directly observe the microcalcification to diagnose the disease,but there will be misdiagnosis and missed diagnosis when manually detecting the microcalcification.With the popularization of Computer-Aided Diagnosis(CAD),the general survey and diagnosis of breast diseases have become more and more convenient.However,the accuracy of most microcalcification detection algorithms needs to be improved.In the research of automatic detection of microcalcification in mammograms,the precise segmentation of microcalcification is the key to realize the subsequent detection of microcalcification.This paper takes mammogram images as the research object,uses the fully connected Pulse Coupled Neural Network(PCNN)to segment the breast images,and further extracts the shape,grayscale and texture of the microcalcification based on the microcalcification segmentation map.Based on these features,Deep Neural Network(DNN)is used to classify benign and malignant microcalcification.The main work of this paper is as follows:1.Based on the fully connected PCNN,the segmentation of the mammogram microcalcification image is realized.By studying and adjusting the parameters of the PCNN model,a method for segmentation of mammogram microcalcification based on fully connected PCNN is proposed,and the optimal segmentation map of mammary microcalcification is obtained,and the connected domain filtering method is used to eliminate non-calcification.The point area is filtered out to ensure the accuracy and reliability of subsequent microcalcification detection.2.The clustering of microcalcification is realized based on the Fuzzy C-Means(FCM)algorithm,and 12 features of microcalcification are further extracted.By extracting features such as shape,grayscale,and texture on the clustered mammogram microcalcification cluster images,the number of microcalcification,the area and aspect ratio of microcalcification clusters,and the proportion of microcalcification in the area of microcalcification clusters are calculated.Shape features such as proportions,grayscale features such as maximum gray scale,minimum gray scale,grayscale mean value,grayscale variance of microcalcification,and texture characteristics such as energy,entropy,contrast,and correlation coefficient.3.Based on the Deep Neural Network,the classification of benign and malignant microcalcification is realized.Through a large number of experiments,the differential diagnosis effect of mammogram microcalcification images is analyzed.The experimental results show that the method in this paper is basically consistent with the artificially marked microcalcification diagnosis results,and it has a higher sensitivity and specificity in disease diagnosis.It also has a low false positive rate,which can help doctors to detect and diagnose breast cancer in a timely manner,which has certain practical significance in clinical medicine.