Automatic parapapillary atrophy shape detection and quantification in colour fundus images

Abstract

Parapapillary atrophy (PPA) in the retina has been associated with eye diseases (e.g. glaucoma) and certain eye conditions (e.g. myopia). However, no computer-aided measuring tool thus far is available to quantify the extent of the PPA. In this paper, a novel approach to automatically segment and quantify the optic disc (OD) and PPA is proposed. The methodology exploits both the red and blue channels of the colour image to maximise information extraction of features (PPA) whilst keeping interference (blood vessels) to a minimum. A combination of several techniques, including scanning filter, thresholding, region growing as well as modified Chan-Vese (C-V) model with a shape constraint is used to segment and quantify the OD and PPA. Our proposed approach is evaluated against the reference boundary drawn by an ophthalmologist. Experimental results show that our method can repeatedly detect both the sizes of the OD and PPA region automatically, and achieved a mean accuracy level of 91.3% and 92.5% in defining the size of the OD and PPA, respectively. Moreover, the correlation coefficient of the ground truth and the results from proposed method is 0.98 for both the PPA and OD.