PURPOSE: To determine if fully automated computerized detection algorithms applied to ultrawide field (UWF) images can identify the risk of diabetic retinopathy (DR) progression associated with predominantly peripheral hemorrhages and microaneurysms (PPL-HMA). CONCLUSIONS: A fully automated HMA detection algorithm identified PPL-HMA in 20% of eyes and was associated with a 2-3 fold increased risk of DR progression, development of PDR and receipt of intravitreal injections independent of diabetes duration, baseline DR severity and HbA1c levels over 4 years. These findings suggest that the automated evaluation for PPL might assist in predicting future DR risks.
METHODS: Patients with UWF images and 4-year follow-up imaging were identified by electronic medical record review. All baseline UWF images were evaluated using a fully automated HMA detection algorithm to determine HMA count and location within, and peripheral to, the ETDRS 7-standard fields. Algorithm performance AUC was 0.902- 0.948 for mild and moderate nonproliferative DR (NPDR). PPL-HMA were defined as present when at least 1 field had a number of HMAs that were greater in the peripheral retina than within the ETDRS fields. 4-year individual and composite outcomes for DR progression were evaluated [≥2-step progression, development of proliferative diabetic retinopathy (PDR), intravitreal injections (IVT)].
RESULTS: Images from 1,712 eyes with baseline and 4-year follow-up UWF images were reviewed. Mean age was 52.6±16.6 years, diabetes duration 14.0±10.6 years, hemoglobin A1c 8.2%±1.8, 54.4% male, and 87.2% white. Baseline DR severity was: no DR 61.1%(1021), mild 22.4%(375), moderate 8.8%(147), severe 1.9%(32), PDR 4.5%(75), high-risk PDR 1.3%(21). PPL-HMA were present in 16.7%(279) of eyes with increasing frequency in more severe DR (No DR-9.1%, Mild-22.4%, Moderate-29.9%, Severe-46.9%, PDR-40.0%, High-risk-62.0%, trend p<0.0001). Progression rates for PPL-HMA (N=236) versus no PPL-HMA (N=1139) over 4-years in eyes with no, mild, moderate or severe NPDR were 14.4%(34) vs 5.53% (74), p<0.0001 for the composite outcome; 8.1%(19) vs 4.1%(55), p=0.0083 for ≥2-step progression; 10.2%(24) vs 2.3% (31), p<0.0001 for PDR development; and 4.7%(11) vs 1.9%(25), p=0.0081 for intravitreal injections. Findings remained significant after correcting for diabetes duration, hemoglobin A1c and DR severity.
