MBO-SVM-based exudate classification in fundus retinal images of diabetic patients

摘要

Exudates act as early indications of diabetic retinopathy which may lead to blindness among diabetic patients. The recognition of exudates with a high accuracy and sensitivity is an important and challenging diagnostic task. In the framework of computer assisted diagnosis of diabetic retinopathy, a novel hybrid algorithm using Migrating Bird Optimisation and Support Vector Machine (MB-SVM) classifiers has been proposed in this paper. Initially, the preprocessing is performed with Gaussian filter to eradicate the noisy intensities and background pixels from the given fundus retinal image. We have detected the optic disc by means of Circular Hough Transform and eliminated it using Otsu thresholding algorithm and Gray scale dilation with flat disc structuring. The blood vessels are extracted using thresholding algorithm and then edge detection is performed by using kirsch's template. After heuristic experimentation, it is evident that the performance of SVM classifier can be improved by concurrently optimising its parameters using MBO algorithm. The hybrid supervised model of MB-SVM has been employed to classify the fundus retinal images into normal and abnormal classes. The images that are classified as abnormal will be further categorised into soft, moderate and severe sub-classes. Experimental validation on a publicly available STARE data-set demonstrates the improved performance of the proposed method over existing methods.