Abstract
Abstract: :
Purpose: To compare the variation in thickness and reflectivity of the RNFL circumferentially and on repeated testing in normal eyes and in eyes with different forms of optic neuropathy. Methods: 20 normal eyes and 23 eyes with different optic neuropathies were analyzed with a third party OCT software (OCT-Pro) which defines the RNFL border based on the reflectivity pattern across the retina and also provides the mean reflectivity and total retinal thickness in addition to the RNFL thickness at 100 locations (circular scan 1.73 mm radius; 5 scans per eye). A new Gaussian filtering method was implemented so that the weight of the applied filter was varied for each individual scan until an optimum weight filter was determined by minimizing the residuals of the 100 points from the median of the 5 repetitions (Matlab, Natick, MA). The optimal weight of the Gaussian filter was determined independently for each of the 5 scans and also for parameters of RNFL thickness, total retinal thickness and reflectivity. Results: The weight of the applied Gaussian filter provided a new parameter, which characterized the measurement variation in substructure of RNFL thickness, RNFL reflectivity, and total retinal thickness. The filtered scans were the best estimate of the structure of the RNFL. When the optimization procedure produced a higher weight filter (heavier filter), this corresponded to greater measurement variability of that particular scan compared to the median of the 5 scans. The scans requiring the highest weight filter could also be identified as "outlier" scans that contained measurement or alignment artifacts. When scans required very little filtering (low weight) the borders of the RNFL were very reproducible, defining a substructure or pattern of "jaggedness" that was characteristic for that eye. There was no difference in the weight of filtering needed for normal and abnormal eyes (no difference in measurement variability). The RNFL thickness, reflectivity, and retinal thickness were highly correlated with one another in the normal and abnormal eyes. Conclusion: A new Gaussian filtering routine was devised that not only defined the most reproducible substructure of the RNFL for a given patient's eye, but also provided a new method for quantifying measurement variability and identification of scans with measurement or alignment artifacts. Support: Swiss National Science Foundation, Freiwillige Akademische Gesellschaft, Roche Research Foundation, Veterans Administration Merit Review Grant, and Research to Prevent Blindness
Keywords: 432 imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound) • 430 imaging/image analysis: clinical • 484 nerve fiber layer