Identifying Exudates in Diabetic MaculopathyA. Osareh, M. Mirmehdi, B. Thomas, R. Markham, Identifying Exudates in Diabetic Maculopathy. 2nd International Workshop on Computer Assisted Fundus Image Analysis. Bjarne Ersboll, (eds.), pp. 17–17. October 2001. No electronic version available.
Aim: to develop an automatic method to analyse colour retinal images to detect and classify exudates (EXs) in diabetic maculopathy. Method: We propose a method comprising four different stages. Firstly, colour retinal images are normalised due to wide variation in the colour of retina from different patients. Then, to improve both the contrasting attributes of EXs and the overall colour saturation in the image, a local contrast enhancement technique is applied. In the third stage, a colour segmentation method based on Fuzzy C-means (FCM) clustering is performed. In our experiments, FCM could distinguish more than 98\% of EXs successfully. In the final stage, ten relevant features are employed to divide the feature space into two disjoint classes and then the FCM segmented regions are classified as EX/non-EX classes by a Neural Network (NN) classifier using the feature measurements. The network training and testing was performed on 42 retinal images, which contained 4037 objects, each of which was labeled by an ophthalmologist. Results: The NN could achieve 92\% sensitivity and 82\% specificity. However, alternative results can be obtained by varying the threshold on the network output, e.g. 83\% sensitivity and 94\% specificity. In addition, we investigated other classifiers and compare the results. The overall performances for the NN, K-Nearest Neighbors, Radial Basis Function and Quadratic Gaussian classifiers were 90.1\%, 86.32\% (K=4), 87.39\% and 78.33\% respectively. Conclusion: The results are very promising and show that automated identification of EX lesions on the basis of colour information is of practical use to ophthalmologists. We are presently obtaining more data and expect the performance to be improved continually through richer training and testing information.