Purchase this article with an account.
K. W. Tobin, E. Chaum, M. Abdelrahman, V. P. Govindasamy, T. P. Karnowski; Preliminary Results for the Statistical Diagnosis of Retinal Pathology by Image Content. Invest. Ophthalmol. Vis. Sci. 2007;48(13):5022.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Diabetic retinopathy is the leading cause of blindness in the working age population around the world. Computer assisted analysis has the potential to assist in the early detection of diabetes by regular screening of large populations. The widespread availability of digital fundus cameras today is resulting in the accumulation of large image archives of diagnosed patient data that captures historical knowledge of retinal pathology. Through this research we are developing a content-based image retrieval (CBIR) method to verify our hypothesis that retinal pathology can be identified and quantified from visually similar retinal images in an image archive. In this talk we will present preliminary results based on an anonymized dataset of 395 red-free retinal images corresponding to 269 patients exhibiting 18 different pathologies taken through dilated pupils using a Topcon TRC 50IA retina camera.
Our CBIR method for diagnosing retinal pathology is based upon the indexing of image content. For our purposes, image content is derived from the statistical characteristics of the population of dark and bright lesions that may exist in the macular region of the fundus. Our approach uses the query-by-example method to retrieve a population of images from a database that exhibit similar features. Our assumption is that the population of images returned through the query process represents a Bayesian posterior probability, PQ(ω|v), of the state of pathology, ω, given the observation of image features, v. The state of pathology, ω, that we are predicting for an unknown query is associated with the particular pathology, severity of disease, and manifestations present in the returned population.
We will present results for this limited population of 395 images that demonstrate our ability to predict the state of pathology in a patient at a performance level exceeding 90% accuracy.
Although we are still in the early stages of this research, we have made significant progress in the characterization and indexing of retinal imagery representing many pathologies and associated disease manifestations. Our future work includes the expansion of our current data repository to incorporate data from additional DR studies, and the eventual determination of the efficacy of this CBIR approach through a clinical study with the University of Tennessee Health Science Center.
This PDF is available to Subscribers Only