Related Experiment Video
Updated: Dec 25, 2025
12:28
Quantification of Vascular Parameters in Whole Mount Retinas of Mice with Non-Proliferative and Proliferative Retinopathies
Published on: March 12, 2022
4.1K
NFN＋: A novel network followed network for retinal vessel segmentation
Yicheng Wu1, Yong Xia1, Yang Song2
1National Engineering Laboratory for Integrated Aero-Space-Ground-Ocean Big Data Application Technology, School of Computer Science and Engineering, Northwestern Polytechnical University, Xi'an 710072, China.
Summary
Accurate segmentation of retinal blood vessels is crucial for early diabetic retinopathy diagnosis. The novel NFN+ deep learning model significantly improves retinal vessel segmentation accuracy on fundus images.
Area of Science:
- Medical Imaging
- Computer-Aided Diagnosis
- Deep Learning
Background:
- Accurate segmentation of retinal blood vessels is essential for the early diagnosis of diabetic retinopathy.
- Existing methods face challenges with limited annotated data, inter-vessel variations, and complex structured prediction for retinal vessel segmentation, especially capillaries.
Purpose of the Study:
- To propose a novel deep learning model, NFN+, for accurate retinal vessel segmentation on color fundus images.
- To effectively extract multi-scale information and leverage deep feature maps for improved segmentation.
Main Methods:
- Developed the NFN+ model featuring a cascaded design with two identical multi-scale backbones connected by inter-network skip connections.
- The front network generates a probabilistic retinal vessel map, refined by a subsequent network for implicit vessel structure representation.
- Utilized augmented images and averaged refined maps for final segmentation results.
Main Results:
- The NFN+ model achieved state-of-the-art retinal vessel segmentation accuracy on DRIVE, STARE, and CHASE datasets.
- Reported Area Under the Curve (AUC) values of 98.30%, 98.75%, and 98.94% respectively.
- Demonstrated that the cascaded design enhances performance gain in retinal vessel segmentation.
Conclusions:
- The proposed NFN+ model effectively addresses challenges in retinal vessel segmentation, particularly for capillaries.
- The deep learning approach with multi-scale feature extraction and refinement offers superior performance for computer-aided diagnosis systems.
- NFN+ represents a significant advancement in achieving high accuracy for retinal vessel segmentation on color fundus images.
