Detection of Optic Disc in Fundus Images Using K-Mean clustering with Outlier Removal Method

Diabetic retinopathy is a retinal disorder characterized by the pathological involvement of the blood vessels situated in the inner region of the retina. DR initially causes no symptoms or mild symptoms of vision problems, if left unnoticed, will lead to blindness. Ophthalmologists examine the fundus images showing the evolution of DR to look for any anomalies. Multiple strategies for supplementing the ophthalmologist with automated DR identification have been proposed in recent years. An important step towards early DR prediction is the optic disc (OD) segmentation, which can be challenging for any computer aided retinopathy prediction system. This study presents an improved clustering technique using k-means for finding outliers in clustered data. The algorithm is implemented on a benchmark dataset as well as a collection of real images obtained from hospitals. The efficacy of the algorithm was evaluated through experimental comparisons with alternative methods. The results presented show a considerable increase in the accuracy for segmenting the OD with less segmentation error.