Purchase this article with an account.
K.W. Tobin, E. Chaum, P. Govindasamy, T.P. Karnowski; Screening for Diabetic Retinopathy by Image Content . Invest. Ophthalmol. Vis. Sci. 2006;47(13):5700.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
The World Health Organization estimates that 135 million people have diabetes mellitus worldwide and that the number of people with diabetes will increase to 300 million by the year 2025. It is estimated that as much as $167 million dollars and 71,000–85,000 sight–years could be saved annually in the U.S. alone with improved retinal screening methods. To address the issues of providing low–cost, high–throughput screening, we are developing a technology that uses a content–based image retrieval (CBIR) method to perform rapid analysis and diagnosis of diabetic retinopathy (DR) in a telemedicine environment. CBIR methods are being developed for many applications including biology and medicine. To date there have been few successful applications of CBIR to large historical repositories of medical image data; the goal being to provide useful datamining capabilities to the researcher, clinician, or student. The "semantic gap" between image meaning (human perception) and image description (machine learning) has been a primary factor. One approach to reduce this gap and produce a useful method is to focus the CBIR application on a specific problem, such as diagnosis of DR.
With our method, an electronic fundus photograph is submitted to the CBIR software engine where important retinal structures are extracted and described. The machine description of these structures is then used to automatically retrieve a population of visually similar, previously diagnosed patient images.
A statistical analysis is performed on the retrieved population to estimate a general characterization of the disease state.
Current telemedicine methods for diagnostics using retinal imagery depend upon trained human readers through a national reading center approach. Our method augments the reading center model by providing a rapid, consistent, and potentially low–cost diagnosis while the patient is in the doctor's office, increasing the chance that the patient will receive timely follow–up medical care and treatment. Through this presentation, we will review our method and present results achieved to date.
This PDF is available to Subscribers Only