《计算机应用》唯一官方网站 ›› 2023, Vol. 43 ›› Issue (12): 3918-3926.DOI: 10.11772/j.issn.1001-9081.2023010045

• 多媒体计算与计算机仿真 • 上一篇    下一篇

基于邻近切片注意力融合的直肠癌分割网络

兰冬雷1,2(), 王晓东1,2, 姚宇1,2, 王辛3, 周继陶3   

  1. 1.中国科学院 成都计算机应用研究所, 成都 610213
    2.中国科学院大学, 北京 100049
    3.四川大学华西医院 腹部肿瘤科, 成都 610041
  • 收稿日期:2023-01-17 修回日期:2023-03-15 接受日期:2023-03-16 发布日期:2023-06-06 出版日期:2023-12-10
  • 通讯作者: 兰冬雷
  • 作者简介:王晓东(1973—),男,四川乐山人,研究员,主要研究方向:网络工程
    姚宇(1980—),男,四川宜宾人,研究员,博士,主要研究方向:机器学习、模式识别
    王辛(1977—),女,四川成都人,教授,博士,主要研究方向:直肠肿瘤的诊断与治疗
    周继陶(1985—),女,四川成都人,主治医师,博士,主要研究方向:直肠肿瘤的诊断与治疗。
  • 基金资助:
    国家自然科学基金资助项目(82073338);四川省科技计划项目重点研发项目(2022YFS0217)

Rectal cancer segmentation network based on adjacent slice attention fusion

Donglei LAN1,2(), Xiaodong WANG1,2, Yu YAO1,2, Xin WANG3, Jitao ZHOU3   

  1. 1.Chengdu Institute of Computer Application,Chinese Academy of Sciences,Chengdu Sichuan 610213,China
    2.University of Chinese Academy of Sciences,Beijing 100049,China
    3.Department of Abdominal Oncology,West China Hospital,Sichuan University,Chengdu Sichuan 610041,China
  • Received:2023-01-17 Revised:2023-03-15 Accepted:2023-03-16 Online:2023-06-06 Published:2023-12-10
  • Contact: Donglei LAN
  • About author:WANG Xiaodong, born in 1973, research fellow. His research interests include network engineering.
    YAO Yu, born in 1980, Ph. D., research fellow. His research interests include machine learning, pattern recognition.
    WANG Xin, born in 1977, Ph. D., professor. Her research interests include rectal tumor diagnosis and treatment.
    ZHOU Jitao, born in 1985, Ph. D., chief physician. Her research interests include rectal tumor diagnosis and treatment.
  • Supported by:
    National Natural Science Foundation of China(82073338);Key Research and Development Project of Sichuan Provincial Science and Technology Program(2022YFS0217)

摘要:

针对直肠癌目标靶区在磁共振成像(MRI)图像的大小、形状、纹理和边界清晰程度不同等问题,为了克服患者之间的个体差异性并提高分割精度,提出一种基于邻近切片注意力融合的直肠癌分割网络(ASAF-Net)。首先,使用高分辨率网络(HRNet)作为主干网络,并在特征提取过程始终保持高分辨率特征表示,以减少语义信息和空间位置信息的损失;其次,通过邻近切片注意力融合(ASAF)模块融合并增强相邻切片之间的多尺度上下文语义信息,使网络能够学习相邻切片之间的空间特征;最后,在解码网络使用全卷积网络(FCN)和空洞空间金字塔池化(ASPP)分割头协同训练,并通过添加相邻切片间的一致性约束作为辅助损失缓解训练过程中出现的相邻切片差异过大的问题。实验结果表明,与HRNet相比,ASAF-Net在平均交并比(IoU)、平均Dice相似系数(DSC)指标上分别提升了1.68和1.26个百分点,平均95%豪斯多夫距离(HD)降低了0.91 mm。同时,ASAF-Net在直肠癌MRI图像多目标靶区的内部填充和边界预测方面均能实现更好的分割效果,有助于提升医生在临床辅助诊断中的效率。

关键词: 直肠癌, 图像分割, 注意力机制, 特征融合, 深度学习

Abstract:

Aiming at the problem that the target regions of rectal cancer show different sizes, shapes, textures, and boundary clarity on Magnetic Resonance Imaging (MRI) images, to overcome the individual variability among patients and improve the segmentation accuracy, an Adjacent Slice Attention Fusion Network for rectal cancer segmentation (ASAF-Net) was proposed. Firstly, using High Resolution Network (HRNet) as the backbone network, the high-resolution feature representation was maintained during the feature extraction process, thereby reducing the loss of semantic information and spatial location information. Secondly, the multi-scale contextual semantic information between adjacent slices was fused and enhanced by the Adjacent Slice Attention Fusion (ASAF) module, so that the network was able to learn the spatial features between adjacent slices. Finally, in the decoder, the co-training of Fully Convolutional Network (FCN) and Atrous Spatial Pyramid Pooling (ASPP) segmentation heads was carried out, and the large differences between adjacent slices during training was reduced by adding consistency constraints between adjacent slices as an auxiliary loss. Experimental results show that compared with HRNet, ASAF-Net improves the mean Intersection over Union (IoU) and mean Dice Similarity Coefficient (DSC) by 1.68 and 1.26 percentage points, respectively, and reduces the 95% mean Hausdorff Distance (HD) by 0.91 mm. At the same time, ASAF-Net can achieve better segmentation results in both internal filling and edge prediction of multi-objective target regions in rectal cancer MRI image, and helps to improve physician efficiency in clinical auxiliary diagnosis.

Key words: rectal cancer, image segmentation, attention mechanism, feature fusion, deep learning

中图分类号: