Automated evaluation of retinal hyperreflective foci changes in diabetic macular edema patients before and after intravitreal injection

OBJECTIVE To examine the global prevalence and major risk factors for diabetic retinopathy (DR) and vision-threatening diabetic retinopathy (VTDR) among people with diabetes. RESEARCH DESIGN AND METHODS A pooled analysis using individual participant data from population-based studies around the world was performed. A systematic literature review was conducted to identify all population-based studies in general populations or individuals with diabetes who had ascertained DR from retinal photographs. Studies provided data for DR end points, including any DR, proliferative DR, diabetic macular edema, and VTDR, and also major systemic risk factors. Pooled prevalence estimates were directly age-standardized to the 2010 World Diabetes Population aged 20–79 years. RESULTS A total of 35 studies (1980–2008) provided data from 22,896 individuals with diabetes. The overall prevalence was 34.6% (95% CI 34.5–34.8) for any DR, 6.96% (6.87–7.04) for proliferative DR, 6.81% (6.74–6.89) for diabetic macular edema, and 10.2% (10.1–10.3) for VTDR. All DR prevalence end points increased with diabetes duration, hemoglobin A1c, and blood pressure levels and were higher in people with type 1 compared with type 2 diabetes. CONCLUSIONS There are approximately 93 million people with DR, 17 million with proliferative DR, 21 million with diabetic macular edema, and 28 million with VTDR worldwide. Longer diabetes duration and poorer glycemic and blood pressure control are strongly associated with DR. These data highlight the substantial worldwide public health burden of DR and the importance of modifiable risk factors in its occurrence. This study is limited by data pooled from studies at different time points, with different methodologies and population characteristics.

Biomedical imaging is a driver of scientific discovery and a core component of medical care and is being stimulated by the field of deep learning. While semantic segmentation algorithms enable image analysis and quantification in many applications, the design of respective specialized solutions is non-trivial and highly dependent on dataset properties and hardware conditions. We developed nnU-Net, a deep learning-based segmentation method that automatically configures itself, including preprocessing, network architecture, training and post-processing for any new task. The key design choices in this process are modeled as a set of fixed parameters, interdependent rules and empirical decisions. Without manual intervention, nnU-Net surpasses most existing approaches, including highly specialized solutions on 23 public datasets used in international biomedical segmentation competitions. We make nnU-Net publicly available as an out-of-the-box tool, rendering state-of-the-art segmentation accessible to a broad audience by requiring neither expert knowledge nor computing resources beyond standard network training.