DRSegNet: A cutting-edge approach to Diabetic Retinopathy segmentation and classification using parameter-aware...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) is a leading cause of global blindness, presenting diagnostic challenges due to its complex development and the eye's intricate structure.
• Accurate and timely diagnosis of DR is crucial for preventing vision loss and improving patient outcomes.

Purpose of the Study:

• To propose and evaluate a novel, AI-driven approach for the accurate identification of diabetic retinopathy.
• To leverage synthetic data generation and advanced machine learning techniques for enhanced DR diagnosis.

Main Methods:

• Utilized Generative Adversarial Networks (GANs) for high-quality synthetic data generation.
• Employed K-Means Clustering-Based Binary Grey Wolf Optimizer (KCBGWO) and Fully Convolutional Encoder-Decoder Networks (FCEDN).
• Integrated transfer learning with Extreme Learning Machines (ELM) for feature extraction and classification.

Main Results:

• Achieved exceptional performance on the IDRiD dataset.
• Reported 99.87% accuracy, 99.33% sensitivity, and 99.78% specificity for DR detection.
• Demonstrated the efficacy of the proposed model in a substantial evaluation.

Conclusions:

• The proposed approach shows significant promise for advancing diabetic retinopathy diagnosis.
• This study establishes a new benchmark in medical image analysis for DR.
• The findings support the development of more effective and timely treatments for DR.