Introduction
Blindness is a common outcome of diabetic-related eye diseases. When background changes occur in the central retina, the condition is termed diabetic maculopathy, and visual acuity is at risk. Much of the blindness can be prevented if the condition is detected early enough for laser treatment. Unfortunately, because visual loss is often a late symptom of advanced diabetic maculopathy, many patients remain undiagnosed even as their disease is causing severe retinal damage. Hence, there is an urgent need for mass-screening retinal examination for the early detection and treatment of such diseases.Current methods of detection and assessment of diabetic maculopathy is manual, expensive, potentially inconsistent, and require highly trained personnel to facilitate the process by searching large numbers of fundus images.
In contrast, a good, automatic method based on modern digital image processing techniques will be faster, will need less, perhaps no human intervention, and will yield consistent results. The aim of our work is to extend the capabilities and productivity of the ophthalmologist and to provide decision support to physicians. We hope to develop a system that will perform our overall aims and objectives including identifying the proportion of the colour retinal image that contains exudates (EXs), and separating them from the other retinal anomalies and pathologies. In this paper, we report a method that first normalises the colours of the retinal image, since this can vary between different races. It then performs local contrast enhancement followed by Fuzzy C-Means (FCM) clustering to highlight salient regions, extracts relevant features, and finally classifies those regions using a multi-layer perceptron neural network.Most of the work carried out so far in this area consider either Fluorescein Angiogram images or gray level images. The former is time consuming for physicians, inconvenient for patients, costly, and cause nonuniform illumination across the image due to varying amounts of background fluorescence. In the latter,monochrome images of the retina do not always capture all the available information for a more accurate segmentation. Other semi-automated methods for measuring EXs have been developed that need human intervention for defining a threshold, thus reducing the objectivity of the technique. Gardner et al used artificial neural networks for identification of EXs by classifying whole regions of size 20x20 pixels.
Blindness is a common outcome of diabetic-related eye diseases. When background changes occur in the central retina, the condition is termed diabetic maculopathy, and visual acuity is at risk. Much of the blindness can be prevented if the condition is detected early enough for laser treatment. Unfortunately, because visual loss is often a late symptom of advanced diabetic maculopathy, many patients remain undiagnosed even as their disease is causing severe retinal damage. Hence, there is an urgent need for mass-screening retinal examination for the early detection and treatment of such diseases.Current methods of detection and assessment of diabetic maculopathy is manual, expensive, potentially inconsistent, and require highly trained personnel to facilitate the process by searching large numbers of fundus images.
In contrast, a good, automatic method based on modern digital image processing techniques will be faster, will need less, perhaps no human intervention, and will yield consistent results. The aim of our work is to extend the capabilities and productivity of the ophthalmologist and to provide decision support to physicians. We hope to develop a system that will perform our overall aims and objectives including identifying the proportion of the colour retinal image that contains exudates (EXs), and separating them from the other retinal anomalies and pathologies. In this paper, we report a method that first normalises the colours of the retinal image, since this can vary between different races. It then performs local contrast enhancement followed by Fuzzy C-Means (FCM) clustering to highlight salient regions, extracts relevant features, and finally classifies those regions using a multi-layer perceptron neural network.Most of the work carried out so far in this area consider either Fluorescein Angiogram images or gray level images. The former is time consuming for physicians, inconvenient for patients, costly, and cause nonuniform illumination across the image due to varying amounts of background fluorescence. In the latter,monochrome images of the retina do not always capture all the available information for a more accurate segmentation. Other semi-automated methods for measuring EXs have been developed that need human intervention for defining a threshold, thus reducing the objectivity of the technique. Gardner et al used artificial neural networks for identification of EXs by classifying whole regions of size 20x20 pixels.
VIDEO DEMO
No comments:
Post a Comment