Programmatically Localizing Diabetic Retinopathy Features in 45-Degree Retinal Photographs Using Anatomical...

Area of Science:

• Ophthalmology
• Medical Imaging
• Diabetic Retinopathy Research

Background:

• Diabetic retinopathy (DR) is a leading cause of vision loss.
• Understanding the precise spatial distribution of DR features is crucial for diagnosis and management.
• Previous studies have not comprehensively mapped the localization of various DR pathologies.

Purpose of the Study:

• To investigate and map the specific locations of diabetic retinopathy features within the posterior pole of the eye.
• To analyze the frequency distribution and spatial patterns of different DR-related lesions.
• To compare the localization of various DR features and vascular structures.

Main Methods:

• Utilized a dataset of 757 macula-centered fundus photographs from the DDR dataset.
• Extracted locations of diabetic retinopathy features, arterioles, and venules from DDR, RITE, and IOSTAR datasets.
• Normalized images and collated feature positions to create a frequency distribution matrix; calculated Sørensen-Dice coefficients for location comparison.

Main Results:

• Arterioles exhibited distinct arcuate patterns, while venules were more diffusely distributed.
• Microaneurysms, hemorrhages, and exudates were predominantly found in the temporal macula.
• Cotton wool spots localized near radial peripapillary capillaries; neovascularization was observed throughout the posterior pole.

Conclusions:

• Diabetic retinopathy manifestations are widespread in the posterior pole but concentrated temporally in the macula.
• The spatial mapping of DR features provides valuable insights for optimizing visual search patterns during screening.
• This detailed localization can enhance the efficiency and accuracy of diabetic retinopathy detection.