An Effective Method for Detecting and Classifying Diabetic Retinopathy Lesions Based on Deep Learning

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) is a leading cause of vision loss in diabetic patients.
• Early diagnosis and treatment are crucial to prevent irreversible vision impairment.
• Current detection methods can be time-consuming and require specialized expertise.

Purpose of the Study:

• To develop and validate a deep learning-based method for automatic detection and classification of diabetic retinopathy lesions.
• To create a robust system independent of specific datasets for broader applicability.
• To improve the accuracy and efficiency of DR diagnosis.

Main Methods:

• A multi-dataset approach was used to create a comprehensive data pool.
• Faster R-CNN was employed for automatic detection and localization of DR lesions.
• Transfer learning and attention mechanisms were utilized for lesion classification.

Main Results:

• The proposed method achieved high performance on Kaggle and MESSIDOR datasets.
• Accuracy (ACC) reached 99.1% and 100%, while Area Under the Curve (AUC) reached 99.9% and 100%.
• The results demonstrate superior performance compared to existing methods in the literature.

Conclusions:

• The developed deep learning model effectively detects and classifies diabetic retinopathy lesions with high accuracy.
• This automated approach shows significant potential for early and reliable diagnosis of DR.
• The dataset-independent nature of the method enhances its clinical utility and scalability.