Related Experiment Video
Updated: Apr 29, 2026
07:23
Retinal Vascular Reactivity as Assessed by Optical Coherence Tomography Angiography
Published on: March 26, 2020
8.1K
nnLoGoNet: a hybrid local-global network for retinal vessel segmentation with Skeleton Recall Loss
Haixia Bai1,2, Fuquan Wu3, Yushuai Zhou4
1Eye Center of Second Affi liated Hospital, School of Medicine, Zhejiang University, Hangzhou, Zhejiang, China.
Scientific Reports
|April 27, 2026
Summary
nnLoGoNet enhances retinal vessel segmentation by combining local and global feature extraction. This novel approach improves accuracy and connectivity in ophthalmic imaging for better disease detection.
Area of Science:
- Ophthalmic image analysis
- Medical image processing
- Deep learning for healthcare
Background:
- Retinal vessel segmentation is crucial for diagnosing eye diseases.
- Challenges include complex structures, low contrast, and fragmented vessels.
- Existing methods struggle with fine details and global topology.
Purpose of the Study:
- To develop an advanced deep learning model for accurate retinal vessel segmentation.
- To improve the delineation of micro-vessels and preserve vascular connectivity.
- To enhance disease screening and clinical diagnosis in ophthalmology.
Main Methods:
- Proposed nnLoGoNet, a hybrid local-global network based on the nnUNet framework.
- Integrated Convolutional Neural Networks (CNNs) for local features and Transformers for global context.
- Introduced a Skeleton Recall Loss for topology-aware supervision with minimal computational cost.
Main Results:
- nnLoGoNet achieved superior performance on Optical Coherence Tomography Angiography (OCTA) datasets.
- Reached state-of-the-art results on Color Fundus Photography (CFP) benchmarks.
- Significantly improved vessel connectivity and reduced model complexity compared to existing methods.
Conclusions:
- nnLoGoNet offers a robust and efficient solution for retinal vessel segmentation.
- The hybrid architecture and novel loss function effectively address segmentation challenges.
- The method holds promise for advancing ophthalmic image analysis and diagnostics.
