Automated Detection of Hard Exudates in Fundus Images Using Improved Otsu Thresholding and Svm

One common cause of visual impairment among people of working age in the industrialized countries is Diabetic Retinopathy (DR). Automatic recognition of hard exudates (EXs) which is one of DR lesions in fundus images can contribute to the diagnosis and screening of DR.The aim of this paper was to automatically detect those lesions from fundus images. At first,green channel of each original fundus image was segmented by improved Otsu thresholding based on minimum inner-cluster variance, and candidate regions of EXs were obtained. Then, we extracted features of candidate regions and selected a subset which best discriminates EXs from the retinal background by means of logistic regression (LR). The selected features were subsequently used as inputs to a SVM to get a final segmentation result of EXs in the image. Our database was composed of 120 images with variable color, brightness, and quality. 70 of them were used to train the SVM and the remaining 50 to assess the performance of the method. Using a lesion based criterion, we achieved a mean sensitivity of 95.05% and a mean positive predictive value of 95.37%. With an image-based criterion, our approach reached a 100% mean sensitivity, 90.9% mean specificity and 96.0% mean accuracy. Furthermore, the average time cost in processing an image is 8.31 seconds. These results suggest that the proposed method could be a diagnostic aid for ophthalmologists in the screening for DR.