Related Experiment Video
Updated: Aug 5, 2026
12:28
Quantification of Vascular Parameters in Whole Mount Retinas of Mice with Non-Proliferative and Proliferative Retinopathies
Published on: March 12, 2022
VesselMetaKAN: vessel-guided meta-learned interpretable classification for diabetic retinopathy grading
TianQi Yang1, GuoYong Chen2, Lu Liu3
1The First Affiliated Hospital of Guangxi Medical University, Nanning, Guangxi, China.
Frontiers in Medicine
|July 31, 2026
Summary
VesselMetaKAN improves diabetic retinopathy grading using vessel guidance and meta-learning. This AI approach enhances public health screening for preventable blindness.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Diabetic retinopathy (DR) is a major cause of preventable blindness globally, impacting 93 million people.
- Accurate automated DR grading is essential for public health screening but faces challenges like sparse evidence and domain shift.
Purpose of the Study:
- To develop an automated system for robust diabetic retinopathy grading.
- To address limitations in current deep learning methods for DR detection.
Main Methods:
- Proposed VesselMetaKAN, a two-stage framework with explicit vessel guidance and meta-learning.
- Stage 1: GMF-SwinUnet for vessel segmentation. Stage 2: KAN-MAML for classification, using vessel maps for guidance.
Main Results:
- VesselMetaKAN achieved 74.9% accuracy and 0.838 QWK on APTOS 2019, outperforming EfficientNet-B4.
- Ablation studies validated the contributions of vessel guidance, KAN interpretability, and meta-learning.
Conclusions:
- VesselMetaKAN offers a principled, structure-aware, and interpretable solution for DR grading.
- The framework is suitable for robust public health screening programs.
