Automated detection and quantification of retinal exudates

Abstract

Retinal exudates are a common manifestation of vascular damage in a variety of retinal diseases. We have used computerized image analysis to detect and measure the area of exudates from digitized colour fundus slides of patients with diabetic retinopathy and have assessed the repeatability, reproducibility, and accuracy of the technique. The analysis was entirely independent of the operator apart from choice of the region to be analysed. The coefficient of variation for repeatability was between 3% for large areas of exudate and 17% for small areas of exudate. The reproducibility was also within this range. Sensitivity was between 61 and 100% (mean 87%). False-positives were observed in 5 of 30 regions analysed, and these could have been eliminated by using more stringent criteria for selection of images for analysis. Time taken for the analysis was approximately 3 min.

Automated Detection of Hard Exudates in Fundus Images Using Improved OTSU Thresholding and SVM

“…It required the user to select the threshold manually according to the histogram. Phillips [9] detected the large EXs by a global threshold and segmented the smaller, lower intensity ones by local threshold. The thresholds were selected automatically, but the region of interest must be chosen manually.…”

An Effective Automated Technique for Retinal Disease Identification in Diabetic Retinopathy without Manually Labeled Kit

“…The methods for exudate segmentation can be approximately distributed into four different categories. Thresholding methods base the exudate reorganization on an overall or adaptive grey level analysis [2,3]. Region growing methods segment the images using the special contiguity of gray levels [4].Morphology methods employ grey scale morphological operators to identify all structures with estimated shape.…”

Automatic retina exudates segmentation without a manually labelled training set

“…Thresholding methods base the exudate identification on a global or adaptive grey level analysis. A first attempt was presented in [1] and recently a more sophisticated method based on image normalisation and distribution analysis was presented in [2]. Region growing methods segment the images using the spacial contiguity of grey levels; a standard region growing approach is used in [3], which is very computationally expensive by being employed to the whole image.…”