Purpose:
Optical coherence tomography (OCT) with adaptive optics (AO) provides unprecedented 3D resolution of the microscopic retina in vivo. The high resolution and high magnification, however, result in substantial eye motion artifacts that are difficult to remove using conventional OCT registration algorithms. Here we develop a new feature-based algorithm that is tailored for AO-OCT imaging and consists of a two step process of layer segmentation and rigid registration.
Methods:
A multi-resolution scheme followed by multi-scale directional edge detection and dynamic programming estimates the location of the major retinal layers for segmentation. Individual B-scans are next registered axially and rotationally using similarity transformation, assuming a locally flat retinal pigment epithelium (RPE). To evaluate the method, retinal volumes were acquired on several subjects with no pathology using the AO-OCT instrument described in Cense et al.[1] 12 volumes, each 900×600×100 voxels (width×height×depth with voxel size: 0.9×1.1×9.0 µm3), were collected within the central retina, segmented, and then registered. Effectiveness of the algorithm was qualitatively accessed by viewing fly-through videos of the processed volumes and projected B-scans in the slow scan direction, and quantified by measuring residual axial deviations of the segmented RPE layer.
Results:
The algorithm was found to correct eye motion artifacts substantially in all volumes processed. The visibly apparent axial motion in the original projected B-scan (left figure) is noticeably reduced (center figure). Across the 12 volumes, the residual mean-square error of the RPE layer before and after processing is 81.6±57.7 and 3.2±4.7 pixels respectively. Segmentation of a B-scan from the same volume is shown in the rightmost image.
Conclusions:
A method to remove motion artifacts from AO-OCT volumes is presented, and demonstrated effective on normal retinas. The proposed method is implemented as a plug-in for ImageJ and will be publically available.[1] B. Cense et al., Opt. Express 17, 4095-4111 (2009).
Keywords: image processing • imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound) • imaging/image analysis: non-clinical