Related Experiment Video
Updated: Mar 7, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.6K
LCNet: lightweight segmentation network for blood vessel segmentation in retinal imaging
Minshan Jiang1, Cuicui Xie1, Shuai Huang1
1Shanghai Key Laboratory of Contemporary Optics System, College of Optical-Electrical and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, People's Republic of China.
Medical Engineering & Physics
|March 6, 2026
Summary
LCNet, a new deep learning model, improves retinal vessel segmentation accuracy. This lightweight network efficiently handles complex cases, offering a valuable tool for computer-aided diagnosis.
Area of Science:
- Medical Imaging
- Computer Vision
- Deep Learning
Background:
- Accurate retinal vessel segmentation is vital for diagnosing eye conditions using computer-aided methods.
- Current deep learning models face challenges with thin, fuzzy boundaries and high computational costs.
Purpose of the Study:
- To introduce LCNet, a lightweight U-shaped network designed for efficient and accurate retinal vessel segmentation.
- To address limitations of existing models regarding parameter count, computational demands, and boundary detection.
Main Methods:
- Developed LCNet using depth-separable convolutions to reduce parameters and computational load.
- Integrated a synergistic coordinate attention module for enhanced feature learning and an atrous spatial pyramid pooling module for multiscale feature capture.
- Incorporated four side-output layers for additional supervision during training.
Main Results:
- Achieved high global accuracies on benchmark datasets: 96.02% (DRIVE), 97.95% (STARE), 97.95% (CHASEDB1), and 97.77% (IOSTAR).
- Demonstrated efficiency with only 2.65 million parameters and 21.2 GFLOPs on the DRIVE dataset.
- Showcased effectiveness on fundus images with lesions and optical coherence tomography angiography images.
Conclusions:
- LCNet is a highly efficient and accurate lightweight model for retinal vessel segmentation.
- The proposed architecture effectively overcomes challenges with complex boundaries and resource constraints.
- LCNet shows promise for clinical applications in computer-aided diagnosis of retinal diseases.
