Severity Grading and Early Retinopathy Lesion Detection through Hybrid Inception-ResNet Architecture

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) is a leading cause of vision impairment and blindness globally.
• Early detection of DR is crucial for preventing vision loss, but often challenging due to subtle initial symptoms.
• Current diagnostic methods can be limited in detecting early-stage DR and grading its severity accurately.

Purpose of the Study:

• To propose a novel framework for the early detection and severity grading of diabetic retinopathy.
• To leverage a hybrid deep learning model, Inception-ResNet, for enhanced DR diagnosis.
• To improve the accuracy and efficiency of DR assessment through advanced data preprocessing techniques.

Main Methods:

• Retinal images were preprocessed using augmentation and intensity normalization.
• A hybrid Inception-ResNet architecture was employed for extracting image features.
• A classification step was implemented to identify DR and determine its stage (mild, moderate, severe, proliferative).

Main Results:

• The proposed framework demonstrated effective outcomes in identifying DR and grading its severity.
• Performance was found to be competitive when compared to existing approaches.
• The study successfully extracted vector image features for categorization of different DR stages.

Conclusions:

• The hybrid deep learning framework offers a promising approach for early diabetic retinopathy detection and severity grading.
• Smart data preprocessing combined with Inception-ResNet enhances diagnostic capabilities.
• Further research is suggested to address study constraints and improve future iterations.