Study Of Key Technology On Label Distribution Embedded Active Contour Model For Breast Tumor Segmentation

Breast cancer is one of the diseases with high mortality for women.Early detection and early treatment is the key to cure breast cancer.Because of its advantages of noninvasive,low price and simple operation,ultrasound imaging technology has become one of the main tools for early diagnosis of breast cancer.Tumor segmentation in ultrasound image is very important for early diagnosis of breast cancer.However,there are different gray levels and quality problems in breast ultrasound image,which will seriously affect the segmentation accuracy of the algorithm.In order to solve this problem,this paper studies an active contour segmentation model based on label distribution embeddingFirstly,in the framework of encoder decoder network,a deep label distribution learning model is constructed to learn the label distribution of each pixel;secondly,a new active contour energy function is constructed.A new label distribution fitting energy term is introduced under the traditional active contour framework.After that,the tumor is segmented by the optimized energy function.The active contour model based on label distribution embedding proposed in this paper has the following two characteristics:1.The deep label distribution learning model can obtain the label distribution map,which has nothing to do with the change of gray level and has a certain robustness to the different quality of gray level;2.The label distribution fitting energy term can embed the information of the label distribution map into the active contour.In order to improve the accuracy of segmentation,it is necessary to keep the label distribution consistent and the result of segmentation is less affected by the quality of grayIn order to verify the performance of this method,a systematic experiment was carried out on the breast image annotation data set of 135 benign and 51 malignant cases.The four main indexes of ACC,Jaccard,TP and FP of this method were 0.9908,0.8917,0.6273 and 0.034,respectively.The overall performance of this method is better than that of the existing method,and it shows a certain robustness to the different gray levelsThis work enriches the technical means of breast ultrasound image tumor segmentation,which is helpful to improve the auxiliary diagnosis effect of breast cancer,and can provide some reference for other similar image segmentation tasks.