Study On Axillary Lymph Nodes Segmentation Algorithm For Breast Cancer

With the transformation of industrialization,urbanization and residents’ living environment,China’s disease spectrum has undergone important changes,and breast cancer in women has surpassed lung cancer and has become the first cancer in the number of new cases of cancer,which seriously threatens women’s health and life safety.Axillary lymph node metastasis plays an important role in the diagnosis and prognosis of breast cancer,and accurate identification of axillary lymph node status is one of the key factors affecting early breast cancer screening.Clinically,the state of axillary lymph nodes is mainly diagnosed by mammography,ultrasound,MRI and other medical imaging examinations,among which ultrasound imaging is one of the important clinical detection methods,because of its non-invasive,non-radiation,less cost,convenient detection and other advantages,it has become one of the most common and cost-effective clinical methods for the diagnosis of axillary lymph node metastasis for breast cancer.However,the low contrast and artifacts in the axillary lymph node lesion area and the surrounding normal tissue area in the breast ultrasound image are time-consuming and laborious to rely only on the doctor’s manual reading for diagnosis,and there is poor consistency in the diagnosis of different doctors,and even the situation of missing and misdiagnosis is likely to occur.The axillary lymph node lesion area segmentation technology can mark the lymph node area for clinicians and provide reference for the diagnosis of lymph node metastasis,so as to reduce the workload of reading,improve the efficiency of diagnosis,and improve the efficiency and accuracy of clinical diagnosis of breast cancer.Therefore,study efforts for the establishment of an axillary lymph node segmentation algorithm has important scientific and clinical value.Based on artificial intelligence technologies such as deep learning,this paper studies the ultrasound image segmentation algorithm of axillary lymph nodes based on image features and pre-segmentation areas,and proposes two ultrasound image segmentation algorithms for axillary lymph nodes for breast cancer.(1)A symptomatic lymph node feature guidance segmentation network algorithm is proposed,which better solves the problem of poor edge feature extraction ability and low segmentation accuracy in the existing related work.According to the ultrasound image characteristics of axillary lymph nodes,this algorithm is trained to improve the accuracy of clinical data segmentation of axillary lymph nodes from the perspective of extracting detailed features of axillary lymph nodes.On the basis of encoder-decoder architecture,the feature guidance module is designed to realize efficient feature fusion and coefficient exploration in feature extraction,and on this basis,a feature guidance segmentation network of axillary lymph nodes is proposed to realize the accurate identification and segmentation of axillary lymph nodes in ultrasound images.The algorithm combines the characteristics of unclear outline and large noise of axillary lymph nodes in ultrasound images,and studies it from the perspective of depth extraction of edge detail features,and improves the overall segmentation accuracy of the network by strengthening the extraction of local features.Experiments show that the m-ACC of the model on the ultrasound image dataset of axillary lymph nodes is 0.977,the m-Io U score is 0.878,and the m-Dice can reach 0.932,which is better than the existing segmentation model,and the segmentation results can be used as a clinical diagnostic reference to achieve the accurate diagnosis of axillary lymph node metastasis in breast cancer.(2)A priori guidance algorithm for axillary lymphatic region is designed,which solves the problem of low segmentation accuracy due to lack of prior knowledge guidance in existing algorithms.Using the combination of traditional image feature analysis technology and deep learning technology,this paper designs a multi-task learning framework segmentation algorithm based on axillary lymphatic region prior guidance.The algorithm consists of a shared encoder backbone network for feature learning and two independent decoders for axillary lymphatic segmentation and axillary lymph node segmentation,respectively.The feature extraction part adds the axillary lymph node ultrasound image feature correlation analysis branch,and the feature map output by the encoder is fused into the context extraction module,and further input into the two segmentation branches.In addition,the prior guidance module of the decoder section improves the performance of axillary lymph node segmentation by fusing features from axillary lymphatic segmentation branches.The m-ACC,m-Io U,m-Dice,and Jaccard scores of this algorithm on the axillary lymph node ultrasound image segmentation dataset are 0.989,0.883,0.937 and 0.704,respectively,and the experimental results are better than the segmentation performance of similar algorithms.In addition,after a large number of data collection,data screening,data labeling and other work in the early stage,an ultrasound image dataset that can be used for axillary lymph node segmentation was sorted out,which laid a good foundation for subsequent axillary lymph node ultrasound image analysis related work.Based on the needs of clinical application,in the later stage of clinical image-assisted diagnosis of breast cancer diseases,studying and improving the accuracy of lesion segmentation will be one of the research hotspots in the field of medical image segmentation.