Improved OCT motion correction in pathology using advanced eye motion modeling

Abstract

Purpose : Patients of higher age or with reduced acuity exhibit increased involuntary eye motion during fixation. This causes distortion and gaps in optical coherence tomography (OCT) and OCT angiography (OCTA), which can substantially complicate qualitative and quantitative image analysis. We evaluated whether motion correction in late-stage age-related macular degeneration (AMD) can be improved by advanced motion modeling.
While a multitude of correction approaches exist, their accuracy and reliability in pathology is insufficient for the high resolution of OCT due to (1) toleration of implausible displacement estimates that limit distortion correction, (2) independent displacement estimation of A- or B-scans that is unreliable in areas with few features, sequential correction procedures where (3) initial steps lack correction of following ones and require (4) superfluous resamplings that cause interpolation artifacts, and (5) use of inaccurate features.

Methods : Our from-scratch redesigned motion model estimates (1) plausible, (2) closed motion trajectories throughout each volume-scan that map A-scans to their correct position in a single mapping (3, 4) based on the similarity of the OCT data itself (5). 3-D displacements and tilt along the transverse directions were parameterized with a Hermite B-spline along acquisition time. Futher key features include regularization for continuity, gradient descent optimization, OCTA white line removal, and preliminary illumination normalization.
Evaluation is ongoing. To date, we evaluated on data of 14 eyes (Table 1) that covers a 3x3 mm (1) or 6x6 mm (13) field around the fovea with a 500x500 orthogonal raster scan (5 B-scan repeats, 3.7 s acquisition) from a prototype 400 kHz swept source OCT device.

Results : Compared to our previous motion correction scheme, inter-scan consistency improved further, and, if present, tilt inaccuracies and motion-induced blurring and double vessels were significantly reduced. However, lack of proper illumination correction compromised the benefits in 2 volumes.

Conclusions : Correct modeling of eye motion can significantly reduce motion-induced artifacts. While broader evaluation is indicated to prove generalization, already using only OCT data improved reliability beyond our combined OCT + OCTA approach based on the previous motion model.

This is a 2021 ARVO Annual Meeting abstract.

 

Pathology distribution and change in motion-correction.

 

Example en-face images that demonstrate motion correction.