The Research On Segmentation Of Breast Tumors Based On Multi-scale Dense Encoder-decoder And Two-level Context Enhanced Residual Attention

The incidence of breast cancer in women has gradually increased,and it is also has been one of the serious diseases leading to female death nowadays.Aiming to effective early detection of breast cancer,and improve the quality of life of patients,more and more researchers are devoted to breast cancer image detection and analysis technology.Among many medical imaging detection technologies,MRI is an effective breast cancer detection tool.Therefore,automatic tumor segmentation based on breast magnetic resonance imaging(MRI)is one of the research hotspots today.With the increasing application of deep learning networks in the image field,there have also been many breakthroughs in breast cancer detection,but they are still insufficient.The difficulties of breast MRI segmentation include the serious imbalance between small lesions and normal tissues,as well as the overlap among different lesions.In order to deal with these challenges,this paper proposes a Multiscale Dense Encoder-Decoder Network based on Two-Level Context Enhanced Residual Attention Mechanism(TLCRAM-MDED).With respect to TLCRAM-MDED,we design the encoding structure combining two-level residual attention structure(including channel level and pixel level attention)with dense block to extract and refine the features of different layers.Meanwhile,a Dense Multi-scale Atrous Convolution is used at the end of the encoder to obtain a larger receptive field and enrich the extracted semantic information.Moreover,residual attention model(TLCRAM)is also used for the refinement during decoding stage,while a long connection formed with the encoder TLCRAM output is applied to supplement the features and to gradually recover the segmentation details.In addition,for the implementation of Dense Net,a new Checkpoint technology is also used,which can also effectively reduce the consumption of video memory.Finally,the segmentation results are also applied to the prediction of sentinel lymph node metastasis in breast tumors.We validated prosed strategy improvement and model in the DCE sequence of challenging breast cancer MRI dataset.The average Dice coefficient is up to 81.04%,which outperforms compared state-of-the-arts.