Related Experiment Video
Updated: Dec 1, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.2K
CS2-Net: Deep learning segmentation of curvilinear structures in medical imaging
Lei Mou1, Yitian Zhao1, Huazhu Fu2
1Cixi Institute of Biomedical Engineering, Ningbo Institute of Materials Technology and Engineering, Chinese Academy of Sciences, Ningbo, China.
Medical Image Analysis
|November 9, 2020
Summary
A new deep learning model, the curvilinear structure segmentation network (CS 2-Net), accurately segments curvilinear structures in medical images. This automated method enhances disease diagnosis and treatment by improving the analysis of vessels and nerve fibers.
Area of Science:
- Medical image analysis
- Deep learning for biomedical imaging
- Computational anatomy
Background:
- Automated detection of curvilinear structures like blood vessels and nerve fibers is vital for medical image interpretation and disease management.
- Accurate segmentation and morphological analysis of these structures aid in diagnosing and treating various conditions, including cardiovascular, kidney, eye, lung, and neurological diseases.
Purpose of the Study:
- To propose a generic and unified convolutional neural network for segmenting curvilinear structures across diverse 2D/3D medical imaging modalities.
- To introduce the curvilinear structure segmentation network (CS 2-Net) with enhanced attention mechanisms for improved feature representation.
Main Methods:
- Developed the CS 2-Net, incorporating spatial and channel self-attention mechanisms in the encoder and decoder to capture hierarchical representations.
- Utilized 1x3 and 3x1 convolutional kernels for boundary feature extraction and extended 2D attention to 3D for depth information aggregation.
- Validated the network on six different imaging modalities using both 2D and 3D images across nine datasets.
Main Results:
- The CS 2-Net demonstrated superior performance in segmenting curvilinear structures compared to existing state-of-the-art algorithms.
- Experimental results across multiple datasets and modalities confirmed the method's effectiveness and generalizability.
- The attention mechanisms effectively enhanced feature discrimination and integration of local and global information.
Conclusions:
- The proposed CS 2-Net offers a robust and versatile solution for automated curvilinear structure segmentation in medical imaging.
- This advancement has the potential to significantly improve diagnostic accuracy and treatment planning for a wide range of diseases.
- The generic and unified approach of CS 2-Net makes it applicable across various imaging modalities and anatomical structures.
