Abstract
Purpose :
Diabetes affects 9.3% of the adult population globally and is a leading cause of preventable blindness. The burden of diabetic complications will increase rapidly in the coming decades. Effective detection of diabetic retinopathy offers the best strategy for timely intervention. As has been shown for glaucoma, systematic search strategies may enable more efficient grading, however there are no formalised strategies for grading diabetic retinopathy in clinical practice. Better understanding of the localisation of features of diabetic retinopathy may inform the development of a search strategy to aid the training of eye health care providers.
Methods :
Three publicly available datasets of annotated retinal photographs were obtained, with a combined total of 757 images with diabetic retinopathy and 70 images of normal retinal vasculature. Annotations for venous beading, intraretinal microvascular abnormalities (IRMA) and neovascularisation were added to the existing annotations. Images were standardised to correct for centration, rotation and field of view. Vessel and lesion locations were extracted and collated into a frequency matrix. Heatmaps were generated showing frequency distributions of the location of arterioles, venules, microaneurysms, haemorrhages, exudates, cotton wool spots (CWS), IRMA, venous beading and neovascular changes.
Results :
Retinal arterioles typically conform to two distinct temporal arcades, whereas the distribution of the temporal venular arcades is more heterogeneous (Figure 1). Microaneurysms and haemorrhages are diffusely distributed in the posterior pole. Exudates appear to cluster in the temporal macular region. CWS and IRMA occur most frequently in the peripapillary area. Neovascular changes are diffusely distributed in the posterior pole. Venous beading is clustered around the venular arcades in the peripapillary area (Figure 2).
Conclusions :
Our study finds that diabetic retinopathy lesions are unevenly distributed in the posterior pole. The cause of exudate clustering at the temporal macular is not well understood and requires further study. These lesion distribution patterns will be used to derive a clinical algorithm to facilitate accurate diabetic retinopathy grading by trainee eye health care providers.
This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.