TransAS-UNet:融合Swin Transformer和UNet的乳腺癌区域分割  

TransAS-UNet:regional segmentation of breast cancer Swin Transformer and of UNet algorithm

作　　者：徐旺旺 许良凤 李博凯[1,2] 周曦 律娜 詹曙[1,2] Xu Wangwang;Xu Liangfeng;Li Bokai;Zhou Xi;Lyu Na;Zhan Shu(Institute of Artificial Intelligence,Hefei Comprehensive National Science Center,Hefei 230601,China;School of Computer Science and Information Engineering,Hefei University of Technology,Hefei 230601,China;Anhui Water Conservancy and Electric Power Technical College,Hefei 231603,China;The First Affiliated Hospital of Anhui Medical University,Hefei 230022,China)

机构地区：[1]合肥综合性国家科学中心人工智能研究院,合肥230601 [2]合肥工业大学计算机与信息学院,合肥230601 [3]安徽水利电力职业技术学院,合肥231603 [4]安徽医科大学第一附属医院,合肥230022

出　　处：《中国图象图形学报》2024年第3期741-754,共14页Journal of Image and Graphics

基　　金：安徽省高校协同创新项目(GXXT-2022-041);合肥市自然科学基金项目(2021008)。

摘　　要：目的乳腺癌在女性中是致病严重且发病率较高的疾病,早期乳腺癌症检测是全世界需要解决的重要难题。如今乳腺癌的诊断方法有临床检查、影像学检查和组织病理学检查。在影像学检查中常用的方式是X光、CT(computed tomography)、磁共振等,其中乳房X光片已用于检测早期癌症,然而从本地乳房X线照片中手动分割肿块是一项非常耗时且容易出错的任务。因此,需要一个集成的计算机辅助诊断(computer aided diagnosis,CAD)系统来帮助放射科医生进行自动和精确的乳房肿块识别。方法基于深度学习图像分割框架,对比了不同图像分割模型,同时在UNet结构上采用了Swin架构来代替分割任务中的下采样和上采样过程,实现局部和全局特征的交互。利用Transformer来获取更多的全局信息和不同层次特征来取代短连接,实现多尺度特征融合,从而精准分割。在分割模型阶段也采用了Multi-Attention ResNet分类网络对癌症区域的等级识别,更好地对乳腺癌进行诊断医疗。结果本文模型在乳腺癌X光数据集INbreast上实现肿块的准确分割,IoU(intersection over union)值达到95.58%,Dice系数为93.45%,与其他的分割模型相比提高了4%~6%,将得到的二值化分割图像进行四分类,Accuracy值达到95.24%。结论本文提出的TransAS-UNet图像分割方法具有良好的性能和临床意义,该方法优于对比的二维图像医学分割方法。Objective Breast cancer is a serious and high-morbidity disease in women.Early detection of breast cancer is an important problem that needs to be solved all over the world.The current diagnostic methods for breast cancer include clinical,imaging,and histopathological examinations.The commonly used methods in imaging examination are X-ray,computed tomography(CT),and magnetic resonance imaging.etc.,among which mammograms have been used in early cancer to detect;however,manually segmenting the mass from the local mammogram is an very time-consuming and errorprone task.Therefore,an integrated computer aided diagnosis(CAD)is needed to help radiologists perform automatic and precise breast mass identification.Method In this work,we compared different image segmentation models based on the deep learning image segmentation framework.At the same time,on the based UNet structure,we adopt the Swin architec⁃ture to replace the downsampling and upsampling processes in the segmentation task,to realize the interaction between local and global features.At the same time we use a Transformer to obtain more global information and different hierarchi⁃cal features to replace short connections and realize multi-scale feature fusion to achieve accurate segmentation.In the seg⁃mentation model stage,we also use so as Multi-Attention ResNet classification network to identify the classification of can⁃cer regions Better diagnosis and treatment of breast cancer.During segmentation the Swin Transformer and atrous spatial pyramid pooling(ASPP)modules are used to replace the common convolution layer through analogy with the UNet struc⁃ture model.The shift window and multiple attention are used to achieve the integration of feature information inside the image slice and extract information complementarity between non-adjacent areas.At the same time,the ASPP structure can achieve self-attention of local information with an increasing receptive field.A Transformer structure is introduced to correlate information between different layer

关 键 词：乳腺癌 深度学习 医学图像分割 TransAS-UNet 图像分类 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]