Diabetic retinopathy screening through artificial intelligence algorithms: A systematic review.

Area of Science:

• Ophthalmology and Medical Imaging
• Artificial Intelligence in Healthcare
• Computer Vision for Disease Detection

Background:

• Diabetic retinopathy (DR) is a leading cause of vision loss, often asymptomatic in early stages.
• Effective and accessible screening methods are crucial for timely diabetes management.
• Artificial intelligence (AI) presents a significant opportunity to improve DR screening.

Purpose of the Study:

• To review current state-of-the-art AI techniques for diabetic retinopathy screening.
• To identify research gaps and guide future investigations in automated DR screening.
• To synthesize existing research and highlight areas for further development.

Main Methods:

• Analysis of deep learning (DL) methods applied to diabetic retinopathy screening, focusing on techniques like classification, detection, and segmentation.
• Review of studies utilizing datasets such as the Indian Diabetic Retinopathy Image Dataset (IDRiD).
• Examination of preprocessing techniques and specific DL models including YOLO, ViT, and U-Net.

Main Results:

• A surge in DL-based DR screening research, particularly post-2021, with a focus on exudate detection.
• Classification and segmentation are dominant AI approaches (46.5% and 41.9% respectively), with preprocessing significantly improving results.
• Integration of AI models, like CNNs and U-Net, shows potential for improved DR stage classification.

Conclusions:

• AI, especially integrated DL models, holds promise for transforming diabetic retinopathy screening.
• Significant challenges remain, including the need for high-quality labeled data and model interpretability.
• Addressing data scarcity and model complexity is essential for reliable real-world clinical application and improved patient outcomes.