Optical imaging for diabetic retinopathy diagnosis and detection using ensemble models.

Area of Science:

• Ophthalmology and Medical Imaging
• Artificial Intelligence in Healthcare
• Diabetology

Background:

• Diabetic Retinopathy (DR) is a leading cause of blindness in diabetic individuals, particularly affecting working-age populations.
• Early detection of DR is crucial but challenged by limitations in traditional methods.
• Elevated blood sugar levels in diabetes damage retinal blood vessels, leading to vision impairment.

Purpose of the Study:

• To develop and evaluate a novel Convolutional Neural Network (CNN) based model for the early diagnosis of Diabetic Retinopathy (DR).
• To leverage an ensemble approach combining multiple deep learning models for improved DR detection accuracy and resilience.

Main Methods:

• Utilized an ensemble of deep learning models including VGG19, Resnet50, and InceptionV3 for feature extraction from retinal fundus images.
• Concatenated features from individual models and fed them into a CNN for final classification.
• Evaluated the proposed model's performance on a publicly available dataset of fundus images.

Main Results:

• The ensemble approach combining VGG19, Resnet50, and InceptionV3 demonstrated high accuracy in classifying Diabetic Retinopathy.
• The integrated model effectively captured diverse features, from fine details (VGG19) to complex patterns (ResNet50) and multi-scale information (InceptionV3).
• The model showed promise in identifying subtle abnormalities like microaneurysms, neovascularization, and macular edema, crucial for early DR diagnosis.

Conclusions:

• Ensemble deep learning models offer a powerful and resilient tool for accurate Diabetic Retinopathy detection.
• This approach enhances the early identification of DR, potentially mitigating vision loss in diabetic patients.
• The combined strengths of VGG19, ResNet50, and InceptionV3 provide a comprehensive analysis of retinal images for DR screening.