Related Experiment Video
Updated: Aug 23, 2025
09:17
Using Retinal Imaging to Study Dementia
Published on: November 6, 2017
21.7K
[Research on grading algorithm of diabetic retinopathy based on cross-layer bilinear pooling]
Liming Liang1, Renjie Peng1, Jun Feng1
1School of Electrical Engineering and Automation, Jiangxi University of Science and Technology, Ganzhou, Jiangxi 341000, P. R. China.
Summary
A novel algorithm improves diabetic retinopathy (DR) grading by using cross-layer bilinear pooling and advanced training techniques. This method achieves high accuracy in classifying DR severity, aiding early detection and treatment.
Area of Science:
- Ophthalmology
- Computer Science
- Artificial Intelligence
Context:
- Diabetic retinopathy (DR) grading is challenging due to subtle differences between severity levels.
- Accurate DR grading is crucial for timely intervention and preventing vision loss.
Purpose:
- To develop an advanced retinopathy grading algorithm that addresses the challenges of subtle DR classification differences.
- To enhance the accuracy and reliability of automated diabetic retinopathy grading systems.
Summary:
- A new algorithm utilizes Hough circle transform for image preprocessing and SEResNeXt as the backbone.
- It incorporates cross-layer bilinear pooling for classification, enhanced by random puzzle generation, center loss (CL), and focal loss (FL) during training.
- The model achieved 90.84% QWK on the IDRiD dataset and 88.54% AUC on the Messidor-2 dataset.
Impact:
- The proposed algorithm demonstrates significant application value in the field of diabetic retinopathy grading.
- This advancement can lead to more efficient and accurate screening tools for diabetic eye disease.
- Improved DR grading accuracy supports better patient outcomes and healthcare resource allocation.
