Supervised Contrastive Embedding for Medical Image Segmentation

被引:8
作者
Lee, Sangwoo [1 ]
Lee, Yejin [1 ]
Lee, Geongyu [1 ]
Hwang, Sangheum [1 ,2 ,3 ]
机构
[1] Seoul Natl Univ Sci & Technol, Dept Data Sci, Seoul 01811, South Korea
[2] Seoul Natl Univ Sci & Technol, Dept Ind & Informat Syst Engn, Seoul 01811, South Korea
[3] Seoul Natl Univ Sci & Technol, Res Ctr Elect & Informat Technol, Seoul 01811, South Korea
基金
新加坡国家研究基金会;
关键词
Image segmentation; Semantics; Feature extraction; Robustness; Decoding; Task analysis; Training; Medical image segmentation; contrastive learning; boundary-aware sampling; domain robustness;
D O I
10.1109/ACCESS.2021.3118694
中图分类号
TP [自动化技术、计算机技术];
学科分类号
0812 ;
摘要
Deep segmentation networks generally consist of an encoder to extract features from an input image and a decoder to restore them to the original input size to produce segmentation results. In an ideal setting, the trained encoder should possess the semantic embedding capability, which maps a pair of features close to each other when they belong to the same class, and maps them distantly if they correspond to different classes. Recent deep segmentation networks do not directly deal with the embedding behavior of the encoder. Accordingly, we cannot expect that the features embedded by the encoder will have the semantic embedding property. If the model can be trained to have the embedding ability, it will further enhance the performance as restoring from those features is much easier for the decoder. To this end, we propose supervised contrastive embedding, which employs feature-wise contrastive loss for the feature map to enhance the segmentation performance on medical images. We also introduce a boundary-aware sampling strategy, which focuses on the features corresponding to image patches located at the boundary area of the ground-truth annotations. Through extensive experiments on lung segmentation in chest radiographs, liver segmentation in computed tomography, and brain tumor and spinal cord gray matter segmentation in magnetic resonance images, it is demonstrated that the proposed method helps to improve the segmentation performance of popular U-Net, U-Net++, and DeepLabV3+ architectures. Furthermore, it is confirmed that the robustness on domain shifts can be enhanced for segmentation models by the proposed contrastive embedding.
引用
收藏
页码:138403 / 138414
页数:12
相关论文
共 46 条
[11]  
Chen T, 2020, PR MACH LEARN RES, V119
[12]  
Chen Xinyun, 2020, arXiv preprint arXiv:2003.04297
[13]  
Dai J., ARXIV210111939
[14]   Dermatologist-level classification of skin cancer with deep neural networks [J].
Esteva, Andre ;
Kuprel, Brett ;
Novoa, Roberto A. ;
Ko, Justin ;
Swetter, Susan M. ;
Blau, Helen M. ;
Thrun, Sebastian .
NATURE, 2017, 542 (7639) :115-+
[15]  
Hadsell R, 2006, 2006 IEEE COMPUTER S, P1735
[16]   Boundary-aware Instance Segmentation [J].
Hayder, Zeeshan ;
He, Xuming ;
Salzmann, Mathieu .
30TH IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2017), 2017, :587-595
[17]   Momentum Contrast for Unsupervised Visual Representation Learning [J].
He, Kaiming ;
Fan, Haoqi ;
Wu, Yuxin ;
Xie, Saining ;
Girshick, Ross .
2020 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2020), 2020, :9726-9735
[18]   Deep Residual Learning for Image Recognition [J].
He, Kaiming ;
Zhang, Xiangyu ;
Ren, Shaoqing ;
Sun, Jian .
2016 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), 2016, :770-778
[19]   MultiResUNet : Rethinking the U-Net architecture for multimodal biomedical image segmentation [J].
Ibtehaz, Nabil ;
Rahman, M. Sohel .
NEURAL NETWORKS, 2020, 121 :74-87
[20]   Two public chest X-ray datasets for computer-aided screening of pulmonary diseases [J].
Jaeger, Stefan ;
Candemir, Sema ;
Antani, Sameer ;
Wang, Yi-Xiang J. ;
Lu, Pu-Xuan ;
Thoma, George .
QUANTITATIVE IMAGING IN MEDICINE AND SURGERY, 2014, 4 (06) :475-477