Abstract
Purpose: :
To develop and test a system that detects drusen and pigmentation abnormalities in Age-related maculopathy (ARM) patients, replicating the simplified severity scale for AMD given in AREDS Report #18, which results in an estimation of the 5-year risk of development of advanced stages of AMD.
Methods: :
We developed a system that detects the two characteristic features of ARM which were identified in the AREDS study: Drusen and pigmentation abnormalities. First, by using Amplitude Modulation-Frequency Modulation (AM-FM) methods, we obtain spatial, mathematical features that characterize drusen. Using morphology-based filters we extract the drusen features from the background and automatically detect their occurrence in a subject’s eyes. Similarly, we use a region of interest (ROI)-based algorithm to determine pigmentation abnormalities. These ROIs are based on the grid used for AREDS grading, analyzing first the macular region and second a disc-shaped area one to three disc diameters away from the macula. The hypothesis is that normal retinas will have less significant histogram changes than retinas with drusen or pigmentation abnormalities.
Results: :
The algorithm was tested on a set of 100 images from the AREDS database, using a single image of the central field per eye and per subject. For images with drusen, we obtained a rate of correct classification greater that 90%. For images with pigmentation abnormalities, we obtained a rate of detection of 75%. These results compare well with published results of human grading for ARM and AMD, which achieved sensitivities of 61% for ARM and 86% for AMD.
Conclusions: :
Age-related macular degeneration (AMD) is the most common cause of visual loss in the United States and is a growing public health problem. The ultimate goal of this system is to provide biomarkers to identify patients at risk of developing advanced stages of AMD and to improve outcomes by identifying more specific interventions. In this manner one will be able to select the best treatment earlier in the disease, which will lead to improved therapeutic outcomes.
Keywords: age-related macular degeneration • image processing