Phase-Restoring Subpixel Image Registration: Enhancing Motion Detection Performance in Fourier-domain Optical Coherence Tomography

Abstract

Phase-sensitive Fourier-domain optical coherence tomography (FD-OCT) enables label-free imaging of cellular movements in vivo with detection sensitivity down to the nanometer scale. Due to this high sensitivity, it is widely employed in various emerging functional imaging modalities such as optoretinography (ORG), optical coherence elastography (OCE), and optical coherence tomography angiography (OCTA). However, achieving shot-noise limited detection sensitivity remains a major challenge for in-vivo measurements where the sample is constantly affected by vascular pulsation, breathing, eye and head motion, and other involuntary movements. Here, we propose a phase-restoring subpixel motion correction (PRSMC) method for post-hoc image registration in FD-OCT. Based on a generalized FD-OCT model, this method enables translational shifts of OCT images by arbitrary displacements while accurately restoring physically meaningful phase components, both with subpixel precision. With the sample movements estimated from averaged Doppler shift or normalized cross-correlation, we reconstructed the OCT images by correcting the axial displacement in the spectrum (k) domain and the lateral displacement in the spatial frequency domain, respectively. We validated our method in simulations, phantom experiments, and in-vivo optoretinogram imaging, where the advantages over conventional approaches for both amplitude stability and phase accuracy were demonstrated. Our approach significantly reduces the motion-induced phase error (MIPE) when imaging moving samples, achieving systematic phase sensitivities close to the shot-noise regime.

Light-evoked deformations in rod photoreceptors, pigment epithelium and subretinal space revealed by prolonged and multilayered optoretinography

“…1B, left column). To correct the negative influence of bulk tissue motion, we adopted our recently proposed phase-restoring sub-pixel motion correction method to register complex-valued OCT images down to subpixel precision (23). We noticed periodical oscillations on the order of a few micrometers in estimated lateral and axial shifts that matched the frequency of heart pulsation (∼4 Hz, see Fig.…”

“…To correct the motion-induced phase error in speckle patterns caused by bulk tissue motion, we estimated subpixel-level translational displacements between repeated B-scans using the single-step DFT algorithm (44). We registered the complex-valued OCT images using our recently developed phase-restoring subpixel motion correction approach (23), where the lateral and axial displacements were corrected by multiplying the corresponding exponential terms in the spatial frequency domain and the spectrum domain, respectively. Light-evoked dynamics of the outer retina was then extracted out by computing the temporal phase difference between the pixel pairs from two highly reflective outer retinal bands.…”

“…7C, center column) (36). We registered the complex-valued OCT images using our D R A F T recently developed phase-restoring subpixel motion correction algorithm (18), where the lateral and axial displacements were corrected by multiplying the corresponding exponential terms in the spatial frequency domain and the spectrum domain, respectively. After image registration (see Fig.…”

“…To achieve high 89 phase stability/sensitivity in-vivo, one needs to correct the 90 bulk tissue motion. For this purpose, we applied our recently developed phase-restoring motion correction method to register complex-valued OCT images with subpixel precision (18). Phase traces extracted from the prolonged ORG experiments (55 s in duration) were spatially averaged across pixels within each band.…”

Optimization method for Knox–Thompson paths in phase-based optoretinography

“…Subpixel-level registration procedures were implemented to ensure OCT image alignment at each temporal instance. 9…”