CS2-Net: Curvilinear Structure Segmentation Network for Medical Images

Lei Mou, Huazhu Fu, Yitian Zhao, YONGHUAI LIU, Jun Cheng, Yalin Zheng, Pan Su, Jianlong Yang, Li Chen, Alejandro F Frangi, Jiang Liu

Research output: Contribution to journalArticle

Abstract

The automated detection of curvilinear structures, e.g., blood vessels or nerve fibers, in
medical images is becoming a paramount step in facilitating the management of many
diseases. Precise measurement of the morphological changes of these curvilinear structures
provides indicative information to clinicians for understanding the mechanism,
diagnosis, and treatment of many diseases. In this work, we propose a general unified
convolutionary neural network for curvilinear structure segmentation of images in
various 2D/3D medical imaging modalities. We introduce a novel curvilinear structure
segmentation network (CS2-Net) based on Dual Attention Network, which includes a
self-attention mechanism in the encoder and decoder to learn rich hierarchical representations
of curvilinear structures. Two types of attention modules - spatial attention
and channel attention - are utilized to enhance the inter-class discriminating power and
intra-class responsiveness, so as to further integrate local features with their global dependencies
and normalization, adaptively. To further facilitate the curvilinear structure
segmentation in medical images, we employ a 1x3 and a 3x1 convolutional kernel to
capture more boundary feature. In addition, we extend the 2D attention to the 3D field
to enhance the network’s ability to aggregate depth information of different layers. The
proposed curvilinear structure segmentation network is rigorously validated using both
2D and 3D images in six different imaging modalities. Experimental results on nine
datasets show that the proposed method outperforms on the whole state-of-the-art algorithms
in different metrics.
Original languageEnglish
JournalMedical Image Analysis
Publication statusAccepted/In press - 5 Oct 2020

Keywords

  • Curvilinear structure
  • blood vessel
  • nerve fiber
  • segmentation
  • attention
  • deep neural network

Fingerprint Dive into the research topics of 'CS2-Net: Curvilinear Structure Segmentation Network for Medical Images'. Together they form a unique fingerprint.

  • Cite this

    Mou, L., Fu, H., Zhao, Y., LIU, YONGHUAI., Cheng, J., Zheng, Y., Su, P., Yang, J., Chen, L., Frangi, A. F., & Liu, J. (Accepted/In press). CS2-Net: Curvilinear Structure Segmentation Network for Medical Images. Medical Image Analysis.