This proposed project is motivated by two observations. First, broad-scale screening of diabetics for retinopathy is economically prohibitive without the introduction of computer-assisted diagnosis of retinal images. Second, screening by family physicians or other non-ophthalmologists does not result in sufficiently high sensitivity or specificity. There are over 20 million people in the US with diabetes and it is estimated that less than half of those are screened periodically for diabetic retinopathy. Access to this type of healthcare is an obstacle to the individual, while having an affordable solution to provide this service to the large volume of diabetics presents a significant challenge. Basing a comprehensive screening program for US citizens on human """"""""readers"""""""" to grade each case would prohibitively expensive. Like other medical applications, such mammograms and Pap smears, computer-assisted technology could provide the foundation for the solution to comprehensive, periodic screening of our at risk population. Numerous investigators have developed specific algorithms, each to detect one type of lesion, such as dark lesions, white lesions, or vessel characteristics. These algorithms have been tested using a single modality (pixel format, SLO, standard funduscope, color, red free, etc). Each new camera requires significant re-tuning of the algorithms. The goal of this project is to demonstrate then validate an entirely new approach for computer-assisted grading of retinal images. This algorithm is based on the human vision system and is """"""""tuned"""""""" to each type of lesion, modality, grading system, etc. through the presentation of examples of each to a single algorithm. Sensitivity and specificity will be calculated. The goal is to achieve 99% sensitivity and 90% specificity. The significance of this proposed research is two-fold. First, by providing a validated, robust computer-assisted grading system, all existing reading centers would benefit by the added efficiency of our system. Our system does not replace current readers, it simply allows increases by factors of 4-5 throughput of cases without sacrifice of sensitivity and specificity. Our Product Development Plan expands on the economics of our approach. Second, by increasing the productivity of reading centers, a much larger population of at risk diabetics can be screened, leading to improved quality of life.

Public Health Relevance

Today there are about 10 million diabetics that are not receiving annual eye examinations. Without these examinations early detection of vision threatening retinopathy is not possible. The result is early loss of vision for many of these diabetics. There is an insufficient number of healthcare specialists to perform eye examinations for this population. Without the computer-based screening, a broad-scale screening of the population will not be possible.

National Institute of Health (NIH)
National Eye Institute (NEI)
Small Business Innovation Research Grants (SBIR) - Phase II (R44)
Project #
Application #
Study Section
Special Emphasis Panel (ZRG1-BDCN-F (12))
Program Officer
Wujek, Jerome R
Project Start
Project End
Budget Start
Budget End
Support Year
Fiscal Year
Total Cost
Indirect Cost
Visionquest Biomedical
United States
Zip Code
Yu, H; Barriga, E S; Agurto, C et al. (2012) Fast localization and segmentation of optic disk in retinal images using directional matched filtering and level sets. IEEE Trans Inf Technol Biomed 16:644-57
Agurto, Carla; Barriga, E Simon; Murray, Victor et al. (2011) Automatic detection of diabetic retinopathy and age-related macular degeneration in digital fundus images. Invest Ophthalmol Vis Sci 52:5862-71
Agurto, Carla; Murray, Victor; Barriga, Eduardo et al. (2010) Multiscale AM-FM methods for diabetic retinopathy lesion detection. IEEE Trans Med Imaging 29:502-12