A fundus image dataset for intelligent diabetic retinopathy system

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Diabetic retinopathy (DR) is a leading cause of irreversible vision loss, affecting the working-age population globally.
• Deep learning (DL) in ultra-wide-field (UWF) imaging enhances DR grading and peripheral lesion detection, surpassing traditional methods.
• A critical barrier to AI in UWF-DR diagnosis is the scarcity of standardized, high-quality, and accessible datasets.

Purpose of the Study:

• To construct a comprehensive and publicly available dataset of UWF fundus images for diabetic retinopathy research.
• To facilitate the development and validation of more robust and generalizable AI-assisted diagnostic systems for DR.
• To overcome limitations in current AI models by providing a standardized resource for training and testing.

Main Methods:

• Compilation of 1,630 UWF fundus images from 809 patients diagnosed with or without diabetic retinopathy.
• Annotation and classification of images by three senior ophthalmologists to ensure high-quality, expert-validated labels.
• Dataset designed for the development and validation of artificial intelligence systems for UWF-based DR diagnosis.

Main Results:

• Creation of a novel, standardized dataset of UWF fundus images specifically for diabetic retinopathy analysis.
• The dataset provides a valuable resource for training AI models to improve diagnostic consistency and accuracy.
• Addresses the need for reliable data to enhance the clinical applicability of AI in detecting peripheral retinal lesions.

Conclusions:

• The developed UWF-DR dataset is crucial for advancing AI-driven diagnostic tools in ophthalmology.
• Availability of this dataset will empower researchers to build more efficient and accurate AI systems for DR detection.
• Facilitates the wider clinical adoption and real-world application of AI in managing diabetic retinopathy.