Advanced volume rebuilding overcomes quilting, stretching, and banding image artifacts in orthogonally-scanned OCT

Abstract

Purpose : Achieving diagnostic quality images is especially challenging in elderly patients: unstable fixation causes increased image distortion and frequent vitreous opacities (floaters) increase image brightness variation. These image artifacts can impact diagnosis and reproducibility, and persist when applying generic image enhancement methods which are typically limited to noise reduction. Herein, we pinpoint the origin of these artifacts and tackle them with specialized algorithms that model the OCT scanning process to achieve appropriate compensation.

Methods : We evaluated 34 swept source OCT angiography scans of 8 elderly patients retrospectively. Artifacts included quilting, stretching, and banding. Quilting and stretching result from model insufficiencies of standard backward-warped displacement fields. We demonstrate superior capability for describing the scan distortion using forward-warped displacements. For additional compensation of banding image artifacts, we equalize the illumination strength of coregistered A-scans. In contrast to standard methods, the scanning process is taken into account to prevent overcorrection and processing artifacts: Morphological features are preserved along the (orthogonal) fast scan directions of each scan. Yet, even abrupt illumination changes are corrected in slow scan direction. We perform inverse model optimization, which is enabled by forward-warped displacements, to correct each input scan individually, such that consistency of all coregistered data is maximized.

Results : The figures demonstrate image artifacts, compensation, and comparison with previous methods in challenging cases. Quilting (triangles) is caused by residual distortion near saccadic motion. Stretching results from inappropriate filling of gaps caused by saccades or at the image boundary. Banding (arrows) can be caused by quick illumination changes in slow scan direction within a volumetric scan, generating stripes of varying brightness. While merged volumes typically show the average brightness of the merged scans, a similar artifact appears if one scan has a gap, and the brightness instantaneously changes to that of the remaining scan.

Conclusions : We demonstrate correction of quilting, stretching and banding OCT image artifacts commonly observed in scans of elderly patients, by using forward-warped 3D motion correction and inverse model-based 3D image reconstruction.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.