ARRA IMPACT REPORT:
Automated Screening Diabetic Retinopathy


Public Health Burden
As the rate of diabetes has climbed in recent years, so has diabetic retinopathy (DR), a leading cause of blindness in the United States, which affects nearly one third of adults with diabetes.1 DR disproportionately affects minorities, affecting 47 percent of Hispanics,2 and 39 percent of non-Hispanic blacks, who also have a three-fold higher rate of vision threating disease than whites.3 A landmark NEI clinical trial in 2010 demonstrated the effectiveness of new therapeutic options to improve vision in patients with diabetic retinopathy. However, disparities in access to eye care limit the benefit of new therapies to certain populations. Wide-scale screening for eye diseases will be economically prohibitive without computer-assisted diagnostic tools to reduce the need for trained specialists to analyze retinal images. ARRA funding propelled development of automated disease screening and diagnostic technologies that will catch the disease earlier in a greater number of patients.

Detection Algorithms
In the retina, diabetes mainly affects the blood capillaries and other vessels resulting in microvascular structural changes, which may be diagnosed by trained eye specialists analyzing photographs of the retina. Funds from ARRA accelerated the development of automatic detection algorithms, which process digital images of normal and diseased retinas.4 One DR lesion detection algorithm first matches patterns of retinal vessels to a template to quickly localize the optic disk and other landmarks. The algorithm successfully localized features in 1189 of 1200 images (99 percent), adapting for different image resolutions and contrast.5 The algorithm then characterizes features based on pixel intensity, size, and geometry metrics to detect potential pathologic structures such as a microaneurysm or hemorrhage in a retinal vessel, abnormal angiogenesis (sprouting of new vessels), or the number and size of fatty deposits called drusen.6

Multidisciplinary Strategies to Detect DR
Another ARRA-funded team of engineers, computational, biochemical, and clinical biologists7 applied simultaneous multimodal retinal functional imaging8 to detect the earliest abnormalities in DR that are undetectable in the fundus image, such as thickening of the basement membrane and loss of pericytes, the elongated contractile cells that wrap around the blood capillaries, and imaged deposits of substances that play protective (melanin) and activating (lipofuscin) roles in eye disease.9 They also modeled metabolic and gene expression changes. Working with rodent models of diabetes, the team searched for early biomarkers of disease, demonstrating that a deficiency in a natural inhibitor of angiogenesis, thrombospondin-1 (TSP1), leads to abnormal blood vessel growth10 but delivering TSP1 therapeutically could rescue the angiogenesis.11 They also showed that loss of retinal pericytes (due to high levels of blood sugar in diabetes) can trigger retinal inflammation which leads to diabetic retinopathy.12

High Specificity Automated Screening
VisionQuest, LLC received an ARRA grant from the small business Biomedical Research, Development, and Growth to Spur the Acceleration of New Technologies Program to develop a tool that automatically detects diabetic eye disease from digital photographs.13 Testing the system in populations from predominantly Hispanic regions of south Texas and New Mexico, where an estimated 200,000 diabetics live, 50 percent of whom do not receive annual examinations, they compared their automated screening tool against the gold-standard opinions of eye specialists. The automated tool was successful in detecting DR with sensitivity of 92–97 percent.14 Developers expect the automated system will ultimately be able to screen patients five times faster than human graders and can be used in conjunction with telemedicine to increase access to eye care in underserved areas. Furthermore, while the system was designed to detect DR, it was found to be highly accurate in also diagnosing other diseases such as age-related macular degeneration, the leading cause of blindness in the U.S.

Contributing NIH Institutes & Centers

  • National Eye Institute (NEI)