Purchase this article with an account.
Stefan B Ploner, Julia Schottenhamml, Eric Moult, Lennart Husvogt, A. Yasin Alibhai, Nadia K Waheed, Jay S Duker, James G Fujimoto, Andreas K Maier; Correction of artifacts from misregistered B-scans in orthogonally scanned and registered OCT angiography. Invest. Ophthalmol. Vis. Sci. 2019;60(9):3097. doi: https://doi.org/.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
While individual volumetric optical coherence tomography (OCT) scans suffer from distortion especially along the slowly scanned transverse axis, registration enables merging of two orthogonally oriented B-scans into one averaged volume with corrected distortion and reduced noise. However, registration errors can occur, which appear as blurred or duplicated vessels in en-face OCT angiography (OCTA) images. We present a fully automatic method that excludes artifacts and locally smooths resulting en-face images to achieve a uniform noise level. Noise non-uniformity emerges from differing numbers of averaged samples. This occurs when misregistrations or saccadic OCTA B-scans are excluded or due to varying depth of projected layers in en-face images. Uniform noise is critical for subsequent analysis algorithms to produce consistent results throughout the image.
Two orthogonally scanned OCT volumes are registered and individual en-face OCTA images are computed. By design, errors in registration occur along B-scans, thus they extend along horizontal or vertical lines in the en-face images. As vessels do not coincide within misregistrations, these lines are detected via their high difference between the en-face images. The scan with B-scan orientation along the line can be identified as the one containing the misregistration and only that scan's B-scan is excluded, while the valid data in the other scan is preserved.To compensate for locally increased noise due to smaller numbers of averaged samples, adaptive Gaussian smoothing is applied. The standard deviation is controlled locally by the number of excluded scans and the remaining scans' sampling densities and projected layers' thickness, such that the total contribution of samples per pixel is kept uniform. The method was evaluated on 8 healthy eyes and 6 eyes with various pathologies.
Figure 1 shows the classification performance, Figure 2 compares merged images.
The number of artifacts in orthogonal scan based OCTA volume registration was reduced while preventing noise non-uniformity due to varying numbers of averaged samples per pixel.
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.
Leave-one-out cross-validation normalized confusion matrix.
Vertically scanned OCTA image (A) and merged images w/o (B) and w/ automatic exclusion (C) of B-scans. Misregistrations at arrows. (D) Adaptively smoothed result. (E/F) Zooms into (C/D). Increased noise between lines.
This PDF is available to Subscribers Only