Automated feature-based grading and progression analysis of diabetic retinopathy

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Current diabetic retinopathy (DR) screening relies on human graders using feature-based guidelines.
• Deep learning models often use end-to-end learning, providing image-level grades without feature-specific insights.
• This limits interpretability and the ability to monitor disease progression based on specific retinal features.

Purpose of the Study:

• To develop and evaluate a feature-based retinal image analysis system for diabetic retinopathy (DR) grading.
• To provide a flexible system that supports grading and monitors disease progression.
• To overcome the limitations of end-to-end deep learning models in DR screening.

Main Methods:

• Development of an AI-powered feature-based retinal image analysis system.
• Evaluation of the system against two established grading systems: International Clinical Diabetic Retinopathy and Diabetic Macular Oedema Severity Scale, and UK National Screening Committee guidelines.
• External validation using large datasets from Kenya, Saudi Arabia, and China.

Main Results:

• The system demonstrated high sensitivity (91.2-94.2%) and excellent specificity (>93%) for DR detection across different grading schemes and international datasets.
• Crucially, no cases of severe non-proliferative DR, proliferative DR, or diabetic macular oedema (DMO) were missed.
• Performance remained consistent across diverse datasets, indicating robustness.

Conclusions:

• An AI feature-based system for DR grading shows significant potential.
• The system is not limited to specific grading schemes, offering flexibility.
• It can effectively support DR screening programs and disease progression monitoring.