Purchase this article with an account.
Martin F Kraus, Jonathan Jaoshin Liu, Julia Schottenhamml, Lauren Branchini, Tony Ko, Gadi Wollstein, Joel S Schuman, Jay S Duker, James G Fujimoto, Johachim Hornegger; Improved Reproducibility in 3D-OCT Quantitative Measurements through use of Advanced 3D Motion Correction using Orthogonal Scan Patterns. Invest. Ophthalmol. Vis. Sci. 2014;55(13):4812.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To evaluate whether the reproducibility of quantitative measurements extracted from 3D-OCT can be improved using post processing based motion correction methods.
Use of a baseline motion correction algorithm was compared with an advanced algorithm and to not performing motion correction. The advanced algorithm features several additions designed to improve results on previously hard to correct data sets. These include a two-stage optimization approach, tilt correction, illumination correction as well as a robust intensity distance measure and regularization term. 3D-OCT volumes from 73 healthy and diseased human eyes were acquired with a software modified Optovue RTVue OCT system. Three independent pairs of orthogonal volumes were acquired for each location (Optic Nerve Head, Macula) and each subject eye. The volumes were then processed with the three methods. For each resulting 3D volume, 2D maps were generated using retinal layer and blood vessel segmentation. All possible pairs of independent result volumes from one method and location were rigidly registered with each other and the corresponding segmentation maps were mapped into a common space. The per-pixel absolute difference or error and its mean and median were calculated per pair. Finally, the overall means of these errors over different subgroups of the data were calculated and different algorithms and subgroups compared for differences. Statistical significance testing was performed using the Wilcoxon signed rank test or the Mann-Whitney-U test, depending on whether the data in question were paired.
The evaluation consistently showed that errors were lowest using the advanced correction, followed by the baseline algorithm, followed by performing no correction. For example, the mean of the mean retinal thickness absolute error over all data was 3.2 pixels without correction, 2.3 pixels using baseline correction and 1.6 pixels using advanced correction. These differences in mean error were found to be statistically significant (p < 0.05).
These results show that the reproducibility of quantitative measurements derived from 3D-OCT data can be substantially improved, especially when using advanced motion correction. This enables more precise quantification, which can enable earlier diagnosis of disease and more sensitive tracking of disease progression.
This PDF is available to Subscribers Only