Purchase this article with an account.
A. Laude, C.-K. Lu, T. B. Tang, A. F. Murray, R. D. Henderson, I. J. Deary, B. Dhillon; Boundary Detection of Optic Disc and Parapapillary Atrophy From Color Fundus Images Using Dual-Channel Color Morphology and Snakes. Invest. Ophthalmol. Vis. Sci. 2010;51(13):1800.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
The presence of parapapillary atrophy (PPA) representing chorio-retinal atrophy around the optic disc (OD) has been associated with certain relatively common eye conditions (eg. Glaucoma and myopia). However, the significance of its development and extent has not been fully established. Although often detected in color fundus photography, there is to date no computer-aided measuring tool available for its accurate detection and measurement. We describe a novel approach to automatically segment the OD and PPA and compare this against the performance by an ophthalmologist.
Pre-processing techniques were initially applied to color fundus images on the red and blue channels separately in order to segment the OD and the OD-plus-PPA respectively. Average filtering was performed within an initial mask to create an enclosed homogeneous area. The OD and OD-plus-PPA boundaries were then further segmented by using a free-form deformable model called 'snakes without edges', based on techniques of curve evolution, level sets and 'Mumford-Shah functional'. We carefully selected the step size of the energy function to ensure that our snakes stopped at the desired boundaries. PPA was then derived from the subtraction of the OD from the OD-plus-PPA. We applied this technique on fundus images taken from a database of a well-characterized cohort and compared the accuracy of boundary detection against the manually-labeled ground truth information drawn by an ophthalmologist.
Of the 33 randomly selected images of 25 subjects with PPA, 27 were of sufficient quality for analysis. Our proposed algorithm achieved a mean accuracy level of 86.6% (S.D.=5.9) in detecting OD, 87.1% (S.D.=6.5) in detecting OD-plus-PPA and 73.5% (S.D.=12.8) in detecting PPA.
Our proposed algorithm achieved good accuracy compared to the gold standard of a human expert. Further work to test out this algorithm in a larger sample is indicated. Possible application includes semi-automated screening systems for diagnosis of eye conditions associated with PPA in the community.
This PDF is available to Subscribers Only