3-D OCT Motion Correction Efficiently Enhanced with OCT Angiography

Abstract

Purpose : Ophthalmic optical coherence tomography (OCT) and OCT angiography (OCTA) suffer from motion artifacts caused by involuntary eye motion during the sequential OCT scanning process. Motion correction (MoCo) can achieve more accurate and reproducible disease metrics. We extend an existing MoCo approach by incorporating OCTA data and perform a quantitative evaluation to assess its reliability and accuracy.

Methods : An existing 3-D OCT MoCo algorithm that closely models the scanning process of a volume pair scan with orthogonal fast-scan directions is extended to not only use the structural, but also the angiographic OCT signal to guide the co-registration of both scans. Operating on the entire OCTA volume increases runtime significantly. Instead, a vessel contrast preserving en-face projection is used which is computed in a fast way without time-consuming layer segmentations.
The method was evaluated on 18 eyes (10 healthy, 8 with various pathologies) from different subjects at both 3x3 mm and 6x6 mm field sizes. In each case, 3 repeated scan-pairs were acquired to assess reproducibility, adding up to 230 OCT volumes.

Results : Figure 1 compares evaluation results for the OCT only registration and our combined approach. The proposed method improves both registration and reproducibility performance. The representative en-face images in Figure 2 visualize differences in the registered volumes.

Conclusions : A novel approach of fully 3-D OCT MoCo using a combination of OCT and OCTA data was introduced. The method showed an improvement over a registration using only OCT data with respect to transverse registration accuracy and success rate at a low additional computational cost.

This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.

 

Quantitative evaluation results. Registration performance describes the agreement between the registered volumes of each volume pair, reproducibility performance compares independently acquired and merged volume pairs. The vessel probability was computed on the en-face OCTA projection of the superficial layer using a vesselness filter. The pixel spacing along depth is 4.5 µm.

 

En-face images of the X/Y-fast OCT(A) and merged OCTA volumes. In A, the additional angiographic contrast helped to better align the scans, leading to less blurring in the merged image. In the OCT only registration in B, the vessels marked with arrows were poorly aligned between both scans and thus appear twice in the merged volume. The combined registration result did not show this problem.