Related Experiment Videos
LHR-Net: a lightweight high-resolution network for retinal vessel segmentation
Jun Feng1, Haibin He1, Jinmei Guo1
1School of Mechanical and Electronic Engineering, Jiangxi College of Applied Technology, Jiangxi, Ganzhou 341000, People's Republic of China.
Biomedical Physics & Engineering Express
|July 22, 2026
Summary
We developed LHR-Net, a lightweight high-resolution network for retinal vessel segmentation. This advanced deep learning model achieves high accuracy in detecting blood vessels, aiding ophthalmic disease diagnosis.
Area of Science:
- Medical imaging analysis
- Computer vision
- Ophthalmology
Background:
- Retinal vessel segmentation is crucial for diagnosing eye diseases.
- Low contrast and complex vessel morphology pose challenges for automated segmentation.
Purpose of the Study:
- To propose LHR-Net, a novel lightweight high-resolution network for accurate retinal vessel segmentation.
- To improve upon existing methods by enhancing feature extraction and multi-scale analysis.
Main Methods:
- Developed LHR-Net based on HR-Net, featuring a high-resolution path, low-resolution branches, and multi-scale feature extraction.
- Incorporated Before-Activation Residual Blocks (BRBs) and a parallel channel attention mechanism.
- Utilized atrous convolution with varying dilation rates for multi-scale feature capture.
Main Results:
- LHR-Net achieved high accuracy (up to 97.40%) and AUC (up to 98.86%) across multiple datasets (DRIVE, STARE, CHASE_DB1, HRF).
- Demonstrated superior segmentation performance compared to state-of-the-art methods.
- The network is lightweight, with only 1.0 million parameters.
Conclusions:
- LHR-Net offers a highly effective and efficient solution for retinal vessel segmentation.
- The proposed architecture enhances the extraction of semantic and spatial information for improved microvessel prediction.
- This method shows significant potential for clinical applications in ophthalmic disease diagnosis.