Abstract
Aim
Diabetic retinopathy is a leading cause of preventable blindness and is expected to worsen in the coming decades. There is limited research around the localisation of diabetic retinopathy; current studies have mapped a subset of retinopathy lesion locations by hand[1]. Better understanding of the location of diabetic retinopathy may inform the development of a visual search strategy for retinopathy screening.
Methods
757 retinal photographs (45-degree field of view) of people with diabetic retinopathy were obtained. These included annotations for microaneurysms, haemorrhages (intraretinal and pre-retinal), exudates, and cotton wool spots[2]. Additional annotations were added by consensus for venous beading, intraretinal microvascular abnormalities (IRMA), and neovascularisation. Pre-retinal haemorrhage annotations were removed to allow analysis of intraretinal haemorrhages only. Images were standardised by aligning all macula and optic disc positions to correct for rotation, centration, and field of view, with custom software available at https://github.com/tim-murphy/dr_lesion_location. Retinopathy annotations were extracted into frequency matrices, and a frequency distribution heatmap generated for lesion subtypes.
Results
Diabetic retinopathy occurs diffusely around the posterior pole, with cotton wool spots correlating with radial peripapillary capillaries, and exudates clustering in the temporal macula. Colocation of cotton wool spots and exudates is rare. Venous beading was more frequent closer to the optic disc. Microaneurysms and intraretinal haemorrhages were distributed diffusely. Neovascularisation occurred more frequently around the optic disc. IRMA were rare in this data set, but were diffusely distributed.
Conclusions
Diabetic retinopathy lesion subtypes display characteristic distribution patterns in 45-degree fundus photographs. The increased frequency of exudates around the temporal macular region is inconsistent with previous findings[1] and warrants further study. These patterns may inform visual search strategies for diabetic retinopathy screening. Future work includes separating by severity grade to determine if distribution patterns change with disease progression, and including an analysis of ultra-widefield images.
References
1. Munuera-Gifre E, Saez M, Rodríguez-Poncelas A, et al (2020) Analysis of the location of retinal lesions in central retinographies of patients with Type 2 diabetes. Acta Ophthalmol 98(1):e13–e21. https://doi.org/10.1111/aos.14223
2. Li T, Gao Y, Wang K, Guo S, Liu H, Kang H (2019) Diagnostic assessment of deep learning algorithms for diabetic retinopathy screening. Inf Sci (Ny) 501:511–522
Diabetic retinopathy is a leading cause of preventable blindness and is expected to worsen in the coming decades. There is limited research around the localisation of diabetic retinopathy; current studies have mapped a subset of retinopathy lesion locations by hand[1]. Better understanding of the location of diabetic retinopathy may inform the development of a visual search strategy for retinopathy screening.
Methods
757 retinal photographs (45-degree field of view) of people with diabetic retinopathy were obtained. These included annotations for microaneurysms, haemorrhages (intraretinal and pre-retinal), exudates, and cotton wool spots[2]. Additional annotations were added by consensus for venous beading, intraretinal microvascular abnormalities (IRMA), and neovascularisation. Pre-retinal haemorrhage annotations were removed to allow analysis of intraretinal haemorrhages only. Images were standardised by aligning all macula and optic disc positions to correct for rotation, centration, and field of view, with custom software available at https://github.com/tim-murphy/dr_lesion_location. Retinopathy annotations were extracted into frequency matrices, and a frequency distribution heatmap generated for lesion subtypes.
Results
Diabetic retinopathy occurs diffusely around the posterior pole, with cotton wool spots correlating with radial peripapillary capillaries, and exudates clustering in the temporal macula. Colocation of cotton wool spots and exudates is rare. Venous beading was more frequent closer to the optic disc. Microaneurysms and intraretinal haemorrhages were distributed diffusely. Neovascularisation occurred more frequently around the optic disc. IRMA were rare in this data set, but were diffusely distributed.
Conclusions
Diabetic retinopathy lesion subtypes display characteristic distribution patterns in 45-degree fundus photographs. The increased frequency of exudates around the temporal macular region is inconsistent with previous findings[1] and warrants further study. These patterns may inform visual search strategies for diabetic retinopathy screening. Future work includes separating by severity grade to determine if distribution patterns change with disease progression, and including an analysis of ultra-widefield images.
References
1. Munuera-Gifre E, Saez M, Rodríguez-Poncelas A, et al (2020) Analysis of the location of retinal lesions in central retinographies of patients with Type 2 diabetes. Acta Ophthalmol 98(1):e13–e21. https://doi.org/10.1111/aos.14223
2. Li T, Gao Y, Wang K, Guo S, Liu H, Kang H (2019) Diagnostic assessment of deep learning algorithms for diabetic retinopathy screening. Inf Sci (Ny) 501:511–522
| Original language | English |
|---|---|
| Pages | 1-38 |
| Number of pages | 38 |
| Publication status | Unpublished - 10 Sept 2023 |
| Externally published | Yes |
| Event | O=MEGA23 Clinical Conference / 4th World Congress of Optometry - Melbourne, Australia Duration: 8 Sept 2023 → 10 Sept 2023 |
Conference
| Conference | O=MEGA23 Clinical Conference / 4th World Congress of Optometry |
|---|---|
| Abbreviated title | O=MEGA23/WCO4 |
| Country/Territory | Australia |
| City | Melbourne |
| Period | 8/09/23 → 10/09/23 |
