Automatic Detection of Abnormalities and Grading of Diabetic Retinopathy in 6-Field Retinal Images: Integration of Segmentation Into Classification.

Purpose: Classification of diabetic retinopathy (DR) is traditionally based on severity grading, given by the most advanced lesion, but potentially leaving out relevant information for risk stratification. In this study, we aimed to develop a deep learning model able to individually segment seven different DR-lesions, in order to test if this would improve a subsequently developed classification model. Methods: First, manual segmentation of 34,075 different DR-lesions was used to construct a segmentation model, with performance subsequently compared to another retinal specialist. Second, we constructed a 5-step classification model using a data set of 31,325 expert-annotated retinal 6-field images and evaluated if performance was improved with the integration of presegmentation given by the segmentation model. Results: The segmentation model had higher average sensitivity across all abnormalities compared to the retinal expert (0.68 and 0.62) at a comparable average F1-score (0.60 and 0.62). Model sensitivity for microaneurysms, retinal hemorrhages and intraretinal microvascular abnormalities was higher by 42.5%, 8.8%, and 67.5% and F1-scores by 15.8%, 6.5%, and 12.5%, respectively. When presegmentation was included, grading performance increased by 29.7%, 6.0%, and 4.5% for average per class accuracy, quadratic weighted kappa, and multiclass macro area under the curve, with values of 70.4%, 0.90, and 0.92, respectively. Conclusions: The segmentation model matched an expert in detecting retinal abnormalities, and presegmentation substantially improved accuracy of the automated classification model. Translational Relevance: Presegmentation may yield more accurate automated DR grading models and increase interpretability and trust in model decisions.

Inverse Cross-sectional and Longitudinal Relationships between Diabetic Retinopathy and Obstructive Sleep Apnea in Type 2 Diabetes: Results from a National Screening Program.

Purpose: In previous smaller studies, associations were demonstrated between diabetic retinopathy (DR) and obstructive sleep apnea (OSA), but longitudinal relationships have not been evaluated in larger cohorts. The aim of the present study was to assess the cross-sectional and prospective associations between DR and OSA in a national cohort of patients with type 2 diabetes. Design: Cross-sectional and 5-year longitudinal registry-based cohort study. Participants: For cases, we included 153 238 patients with type 2 diabetes who had attended diabetic eye screening and were registered in the Danish Registry of Diabetic Retinopathy (DiaBase). Each of these were matched by 5 control participants without diabetes of the same age and gender (n = 746 148). Methods: Exposure and outcome data as well as systemic morbidity and use of medications were identified in national registers, including the DiaBase, the Danish National Patient Register, the Danish National Prescription Registry, and the Danish Civil Registration System. The index date was defined as the date of the first DR screening registered in DiaBase. Main Outcome Measures: Exposure was defined as present and level-specific DR, and main outcomes were crude, age- and gender-adjusted, and multivariable adjusted odds ratios (ORs) for prevalent OSA as well as hazard ratios (HR) for 5-year incident OSA and DR. Results: Patients with type 2 diabetes independently were more likely to have prevalent OSA (OR, 2.01; 95% confidence interval [CI], 1.95-2.08) and to develop OSA within 5 years (HR, 1.55; 95% CI, 1.46-1.64). Patients with type 2 diabetes and DR at baseline were less likely to have prevalent OSA (OR, 0.57; 95% CI, 0.52-0.62) or to demonstrate incident OSA (HR, 0.86; 95% CI, 0.74-0.99). Likewise, patients with OSA had a lower risk to develop DR (HR, 0.83; 95% CI, 0.74-0.92). Conclusions: In a registry-based national cohort study, patients with type 2 diabetes had a higher risk of OSA. However, a 43% decreased risk of prevalent OSA was demonstrated in patients with DR, and prospectively, OSA and DR both were related inversely with each other.

Polymorphisms in the CTSH gene may influence the progression of diabetic retinopathy: a candidate-gene study in the Danish Cohort of Pediatric Diabetes 1987 (DCPD1987).

BACKGROUND: The incidence of type 1 diabetes mellitus (T1DM) is increasing globally, and as a consequence, more patients are affected by microvascular complications such as diabetic retinopathy (DR). The aim of this study was to elucidate possible associations between diabetes-related single-nucleotide polymorphisms (SNP) and the development of DR. METHODS: Three hundred and thirty-nine patients with T1DM from the Danish Cohort of Pediatric Diabetes 1987 (DCPD1987) went through an ophthalmic examination in 1995; 185 of these were reexamined in 2011. The development of DR was assessed by comparison of overall DR level between baseline and follow-up in the worst eye at baseline. Patients were graded on a modified version of the Early Treatment Diabetic Retinopathy Study (ETDRS) scale, and 20 SNPs were genotyped in 130 of the 185 patients. RESULTS: We found the CTSH/rs3825932 variant (C > T) was associated with reduced risk of progression to proliferative diabetic retinopathy (PDR) (OR [95 % CI] = 0.20 [0.07-0.56], p = 2.4 × 10(-3), padjust = 0.048) and ERBB3/rs2292239 variant (G > T) associated with increased risk of two-step progression (OR [95 % CI] = 2.76 [1.31-5.80], p = 7.5 × 10(-3), padjust = 0.15). The associations were independent of other known risk factors, such as HbA1c, sex, and diastolic blood pressure. CONCLUSION: In conclusion, CTSH/rs3825932 and ERBB3/rs2292239 SNPs were associated with reduced risk of progression to PDR and two-step progression of DR on the ETDRS scale accordingly. The variant CTSH remained statistically significant after adjusting for multiple testing. Our results suggest an overlap between genetic variants that confer risk of T1DM and progression of DR.

Comparison between Early Treatment Diabetic Retinopathy Study 7-field retinal photos and non-mydriatic, mydriatic and mydriatic steered widefield scanning laser ophthalmoscopy for assessment of diabetic retinopathy.

AIMS: To compare non-mydriatic, mydriatic and steered mydriatic widefield retinal images with mydriatic 7-field Early Treatment Diabetic Retinopathy Study (ETDRS)-standards in grading diabetic retinopathy (DR). METHODS: We examined 95 patients (190 eyes) with type 1 diabetes. A non-mydriatic, a mydriatic and four steered mydriatic 200° widefield retinal images were captured (Optos 200Tx, Optos plc, Dunfermline, Scotland) and compared to mydriatic 7-field 45° ETDRS images (Topcon 3D OCT-2000, Topcon, Tokyo, Japan). Images were graded for DR according to ETDRS-protocol by a trained and certified grader masked to the results of the corresponding grading. For agreement kappa-statistics were used. RESULTS: Exact level agreement with 7-field images was found in 76.3%, 76.1% and 70.7% for non-mydriatic, mydriatic and steered mydriatic widefield images, respectively. Corresponding values for one-level agreement were 99.0%, 98.9% and 99.5%, respectively. Non-mydriatic matched mydriatic widefield images almost fully with exact and one-level agreement of 96.8% and 100.0%, respectively. Mydriatic steered images resulted in higher grading in 24 eyes. CONCLUSIONS: Widefield images matched 7-field images favorably. Widefield images can be captured without pupil-dilation and only one image is needed. However, because of overlapping eyelashes and distortion some lesion might be missed. Mydriatic steered images in selected cases may solve some of these problems.

Retinal vascular fractals predict long-term microvascular complications in type 1 diabetes mellitus: the Danish Cohort of Pediatric Diabetes 1987 (DCPD1987).

AIMS/HYPOTHESIS: Fractal analysis of the retinal vasculature provides a global measure of the complexity and density of retinal vessels summarised as a single variable: the fractal dimension. We investigated fractal dimensions as long-term predictors of microvasculopathy in type 1 diabetes. METHODS: We included 180 patients with type 1 diabetes in a 16 year follow-up study. In baseline retinal photographs (from 1995), all vessels in a zone 0.5-2.0 disc diameters from the disc margin were traced using Singapore Institute Vessel Assessment-Fractal image analysis software. Artefacts were removed by a certified grader, and fractal dimensions were calculated using the box-counting method. At follow-up (in 2011), diabetic neuropathy, nephropathy and proliferative retinopathy were assessed and related to baseline fractal dimensions in multiple regressions adjusted for sex and baseline age, diabetes duration, HbA1c, BP, BMI, vibration perception threshold, albuminuria, retinopathy and vessel diameters. RESULTS: Mean baseline age and diabetes duration were 21.0 and 13.4 years, respectively, and of patients 50.0% were males. The mean fractal dimension was 1.3817. The 16 year incidences of neuropathy, nephropathy and proliferative retinopathy were 10.8%, 8.0% and 27.9%, respectively. Multiple regression analyses showed a lower fractal dimension to significantly predict incident neuropathy (OR 1.17 per 0.01 fractal dimension decrease [95% CI 1.01, 1.36]), nephropathy (OR 1.40 per 0.01 fractal dimension decrease [95% CI 1.10, 1.79]) and proliferative retinopathy (OR 1.22 per 0.01 fractal dimension decrease [95% CI 1.09, 1.37]). CONCLUSIONS/INTERPRETATION: The retinal vascular fractal dimension is a shared biomarker of diabetic microvasculopathy, thus indicating a possible common pathogenic pathway. Retinal fractal analysis therefore is a potential tool for risk stratification in type 1 diabetes.

Retinal vessel calibers predict long-term microvascular complications in type 1 diabetes: the Danish Cohort of Pediatric Diabetes 1987 (DCPD1987).

Diabetic neuropathy, nephropathy, and retinopathy cause significant morbidity in patients with type 1 diabetes, even though improvements in treatment modalities delay the appearance and reduce the severity of these complications. To prevent or further delay the onset, it is necessary to better understand common underlying pathogenesis and to discover preclinical biomarkers of these complications. Retinal vessel calibers have been associated with the presence of microvascular complications, but their long-term predictive value has only been sparsely investigated. We examined retinal vessel calibers as 16-year predictors of diabetic nephropathy, neuropathy, and proliferative retinopathy in a young population-based Danish cohort with type 1 diabetes. We used semiautomated computer software to analyze vessel diameters on baseline retinal photos. Calibers of all vessels coursing through a zone 0.5-1 disc diameter from the disc margin were measured and summarized as the central artery and vein equivalents. In multiple regression analyses, we found wider venular diameters and smaller arteriolar diameters were both predictive of the 16-year development of nephropathy, neuropathy, and proliferative retinopathy. Early retinal vessel caliber changes are seemingly early markers of microvascular processes, precede the development of microvascular complications, and are a potential noninvasive predictive test on future risk of diabetic retinopathy, neuropathy, and nephropathy.