Purchase this article with an account.
S.D. Mitra, P.S. King, M. Wilson, S. Nemeth, Y.H. Kwon, y.I. Kim, m.L. Pereira, P. Soliz; Automated Cup to Disc Ratios from 3-D Surface Recovery of the Optic Disc from Stereo Image Pairs: A Comparison to Multiple Readers . Invest. Ophthalmol. Vis. Sci. 2003;44(13):3370.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Purpose: Stereo images of the optic disc remain the mainstay of researchers and clinics for diagnosing and monitoring the health of optic nerve head. Cup to disc ratios are often derived from these photographs with varying degrees of inter- and intra-reader variability. The objective of this study was to demonstrate a totally automatic computer-based technique for calculating cup to disc ratios based on depth from stereo image pairs by comparing topological features drawn by two or more ophthalmologists of optic disc and cup margins. Methods: 159 stereo optic disc image pairs of subjects presenting with various stages of glaucoma were analyzed by at least two ophthalmologists. The ophthalmologist-derived analysis was used as ground truth for initial comparisons with computer generated measures. A cross correlation technique for registering corresponding features in the stereo pairs was combined with a coarse-to-fine disparity search algorithm that extracted depth information from disparity between the features. The resulting sparse disparity matrix elements are encoded as gray levels and processed through a cubic B-spline operation to reduce intrinsic noise and generate a smooth representation of the optic disc surface. The topological maps were compared with the disc and cup margins provided by the ophthalmologists, who used stereo visual cues for determining the cup and disc. Results: The stereo pairs of subjects were taken spanning over 20 years. The computer-generated volume cup/disc ratios showed very high agreement with the linear cup/disc ratios computed from the ophthalmologists’ analysis. The average correlation coefficient between the computer-generated and ground truth data was found to be 0.9355 for 159 stereo pairs. Conclusions: The 3-D surface recovery methodology demonstrated significant merit in generating cup and disc measurements that compare well with those provided by the ophthalmologists. These results suggest that an automatic disc analysis system, demonstrating high correlation with ophthalmologists’ analysis, could eliminate the inter- and intra-observer variability and would lead to a quantitative, more sensitive and precise means for making accurate cup to disc measures necessary in longitudinal studies.
This PDF is available to Subscribers Only