Related Experiment Video
Updated: Jan 18, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.3K
Automated segmentation of retinal vessel using HarDNet fully convolutional networks
Yuanpei Zhu1, Yong Liu2,3, Xuezhi Zhou3
1School of Physics and Electronic Engineering, Xinxiang University, Xinxiang, China.
Plos One
|September 8, 2025
Summary
This study introduces an improved HarDNet model for enhanced retinal vessel segmentation in fundus images. The model excels at identifying small vessels and complex structures, improving early disease detection.
Area of Science:
- Medical image analysis
- Computer-aided diagnosis
- Deep learning for medical imaging
Background:
- Deep learning advances automatic segmentation for fundus images, crucial for detecting diseases like diabetes and hypertension.
- Challenges persist, including limited data and structural variations, hindering accurate segmentation of small vessels.
Purpose of the Study:
- To develop an enhanced HarDNet-based model for improved retinal vessel segmentation.
- To address limitations in current deep learning models for handling small vessels and complex vascular structures.
Main Methods:
- An enhanced HarDNet model integrating HarDNet, Receptive Field Block (RFB), and Dense Aggregation modules was proposed.
- The model was designed to effectively extract multi-scale features for improved segmentation accuracy.
Main Results:
- The model achieved high accuracies: 0.9685 (±0.0035) on DRIVE and 0.9744 (±0.0029) on CHASE_DB1.
- It surpassed state-of-the-art models like U-Net, ResU-Net, and R2U-Net in retinal vessel segmentation.
- Exceptional performance was noted in segmenting tiny vessels and branch regions, closely matching gold standards.
Conclusions:
- The proposed model demonstrates robust and accurate performance in retinal vessel segmentation.
- Its effectiveness in handling intricate vascular structures offers significant advantages for medical image analysis.
- This provides valuable technical support for research and applications in early disease detection.
