Related Experiment Video
Updated: Jun 16, 2026
09:17
Using Retinal Imaging to Study Dementia
Published on: November 6, 2017
From discarding to leveraging: quality-aware collaborative learning for robust diabetic retinopathy grading.
Yuan Pan1, Xiangwen Cai2, Pan Xiong1
1School of Medical Information Engineering, Guangzhou University of Chinese Medicine, Guangzhou, China.
Frontiers in Medicine
|June 15, 2026
Summary
Diabetic retinopathy screening struggles with low-quality images showing severe disease. A new quality-guided framework (QGDR) uses image quality as a continuous signal, improving diagnostic accuracy for all image types.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Diabetic retinopathy (DR) screening is challenged by poor image quality in severe cases due to disease-related opacities.
- Conventional methods often discard critical low-quality images or misclassify them, hindering effective screening.
- Image quality assessment (IQA) is typically a binary preprocessing step, failing to leverage quality information dynamically.
Purpose of the Study:
- To develop and validate a novel framework, QGDR, that utilizes image quality as a continuous guidance signal for DR diagnosis.
- To address the paradox of low-quality images often containing the most severe DR indicators.
- To improve the reliability and coverage of automated DR screening systems.
Main Methods:
- Proposed QGDR, a quality-guided dynamic routing framework incorporating multi-level IQA, quality-conditioned context gating, and adaptive gated fusion.
- Developed a multi-level IQA module to extract hierarchical quality features.
- Implemented a dynamic routing mechanism that routes inputs to scale-specialized experts based on predicted image quality.
Main Results:
- QGDR achieved high accuracy (78.32% on EyeQ, 80.85% on DDR) and outperformed baseline models including CNNs, transformers, and foundation models.
- Performance was maintained or improved on low-quality images (e.g., 82.21% on DDR-Reject), exceeding good-quality accuracy.
- Validated cross-dataset generalization on IDRiD and DeepDRiD cohorts, confirming robustness.
Conclusions:
- Treating image quality as a continuous guidance signal, rather than a filter, enhances DR diagnostic reliability.
- The QGDR framework effectively leverages semantic quality information to improve screening for both high- and low-quality images.
- QGDR offers a promising solution for preserving screening coverage without compromising diagnostic accuracy in real-world DR screening.
