Purchase this article with an account.
Stewart R Lake, Keryn Williams, Murk Bottema, Karen Reynolds; Merging retinal optical coherence tomography images in three dimensions. Invest. Ophthalmol. Vis. Sci. 2021;62(8):1783.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Optical coherence tomography (OCT) provides high resolution cross-sectional two dimensional retinal images, with the ability to reconstruct local retinal shape from these sections in three dimensions. OCT is capable of sampling a limited area of retina in a single image. Wider field images have become available with developing technologies including swept source (SS) OCT, as well as through the merging in series of individual B scans in two dimensions. Merging SS OCT images in three dimensions (3D) has not previously been reported. Here, overlapping SS OCT cubes were reconstructed in 3D and merged via two different methods to explore the utility and limitations of composite images for retinal shape analysis.
From three eyes of three human participants, overlapping pairs of 9 x 9 mm SS OCT cubes were taken. Twenty five B scans were sampled from each cube, and used to reconstruct local retinal shape within the area covered by the scanning laser ophthalmoscope (SLO). For each eye, three pairs of corresponding points were identified in the SLO images from each cube. The reconstructed 3D images were merged via two different methods, using a rigid transformation to preserve shape features: sequential rotations using the axis-angle method via calculation of, then rotation with a quaternion.
Composite images from one eye are shown in Figure 1. The two methods produced similar images, with no overall difference in alignment (mean (standard deviation) axial difference within en face overlapping area was 30.8 (23) pixels (quaternion rotation) versus 30.9 (23) pixels (axis angle rotation), where 1536 pixels = 3 mm, two sample t-test, p<0.005).
Merged OCT images provided a good qualitative description of wide field retinal shape in three dimensions. Each rotation induced small errors, although the single rotation quaternion method performed no better than the sequential rotation required for the axis angle method. Although the magnitude of induced error was low with both rotation methods, care should be taken in performing quantitative analyses of composite images.
This is a 2021 ARVO Annual Meeting abstract.
Figure 1. Composite images generated by axis angle rotation (above), and quaternion rotation (below).
This PDF is available to Subscribers Only