Implementation of a Framework for Healthy and Diabetic Retinopathy Retinal Image Recognition

Area of Science:

• Biometrics
• Ophthalmology
• Computer Vision

Background:

• Biometric recognition systems heavily rely on feature extraction from traits like the retina.
• Existing retina recognition systems often perform poorly with diseased retinal images (e.g., diabetic retinopathy).
• Diseases like diabetic retinopathy, glaucoma, and cataract can significantly impact retina recognition accuracy.

Purpose of the Study:

• To develop a robust retina recognition framework that accommodates both healthy and diseased retinal images.
• To enhance the accuracy and reliability of retina-based biometric systems in clinical settings.
• To introduce a dual-approach system for improved feature extraction based on retinal image health.

Main Methods:

• A novel framework employing two distinct feature extraction approaches for retina image recognition.
• Approach 1: Utilizes structural features for healthy retinal image recognition.
• Approach 2: Employs vascular and lesion-based features specifically for diabetic retinopathy (DR) retinal image recognition, preceded by DR symptom detection.

Main Results:

• Achieved a 100% recognition rate for healthy retinal images.
• Obtained a 97.23% recognition rate for diabetic retinopathy (DR) retinal images.
• Reported a low false acceptance rate (FAR) of 0.0444 and a false rejection rate (FRR) of 0.0133.

Conclusions:

• The proposed dual-approach framework significantly improves retina recognition accuracy across healthy and DR-affected images.
• The system demonstrates robustness by adapting feature extraction techniques based on the presence of DR symptoms.
• This advancement offers a more reliable biometric solution for diverse ophthalmological conditions.