A Spatiotemporal Illumination Model for 3D Image Fusion in Optical Coherence Tomography

TL;DR

This work tackles illumination artifacts in Optical Coherence Tomography (OCT) caused by raster-scanned 3D volume acquisition by introducing a spatiotemporal illumination model that enforces continuity along both B-scans and volumes. The correction is parameterized in the log domain with per-B-scan and per-volume Hermite spline coefficients $c_i^V(j)$, and optimized via a 3D inverse problem that aligns illumination-corrected volumes against orthogonal scans. Quantitatively, the method reduces illumination artifacts in about $88\%$ of volumes (with $6\%$ exhibiting moderate residuals) and achieves a $22.5\%$ reduction in mean absolute differences between registered volumes, enabling forward-warped motion correction, denoising, and potential super-resolution in OCT. Overall, the approach improves cross-volume consistency while preserving clinically relevant morphology, supporting more reliable biomarkers and advanced 3D reconstruction in OCT.

Abstract

Optical coherence tomography (OCT) is a non-invasive, micrometer-scale imaging modality that has become a clinical standard in ophthalmology. By raster-scanning the retina, sequential cross-sectional image slices are acquired to generate volumetric data. In-vivo imaging suffers from discontinuities between slices that show up as motion and illumination artifacts. We present a new illumination model that exploits continuity in orthogonally raster-scanned volume data. Our novel spatiotemporal parametrization adheres to illumination continuity both temporally, along the imaged slices, as well as spatially, in the transverse directions. Yet, our formulation does not make inter-slice assumptions, which could have discontinuities. This is the first optimization of a 3D inverse model in an image reconstruction context in OCT. Evaluation in 68 volumes from eyes with pathology showed reduction of illumination artifacts in 88\% of the data, and only 6\% showed moderate residual illumination artifacts. The method enables the use of forward-warped motion corrected data, which is more accurate, and enables supersampling and advanced 3D image reconstruction in OCT.

Figures (3)

• Figure 1: Global illumination variation (ellipse) and banding artifacts (arrows) in enface images (depth-averaged volumes). Left: Input with vertical B-scan orientation. Right: forward-warped & merged.
• Figure 2: Compensation of illumination artifacts. Top left: Schematic of X-fast B-scans (blue) and Y-fast B-scans (red) after motion correction. Bottom: Visualization of the optimization goal in logarithmic scale, with legend / zoom in the top right. Blue curve: X-fast B-scan signal corresponding to the dark blue line in the top left. For clarity, we assume a sample with uniform (foreground) backscattering such that changes in measured signal only originate from illumination variance. Therefore, given the continuous illumination assumption in B-scans, the blue curve is continuous. The red circles represent the intensities of the orthogonal scan, which are interpolated and therefore written without indices. Because they originate from different B-scans, their illumination is not necessarily continuous. The dotted black curve, corresponding to the average of the uncorrected intensities, is discontinuous where $s^Y$ is discontinuous or has gaps. In forward warping, the averaging weights of the red intensities vary between points, which would distribute the black dots discontinuously somewhere between the blue and red inputs. The blue arrows represent the illumination correction of the X-fast intensities, which are continuous due to their spline parametrization. Added to the blue curve, they form the dashed green curve, the corrected X-fast intensities. Correction coefficients of the orthogonal scan are optimized such that the corrected intensities match the green line up to noise (in a different addend of the outer sum in Eq. 5).
• Figure 3: Representative enface images (depth-averaged volumes) from volumes merged without (row 1) and with illumination correction (row 2). (a) Good results after large and (b) small improvement. The zooms in (a) show a checkerboard overlay of the registered volumes before merging. Illumination difference makes the individual tiles in (1a) clearly distinguishable, while they are almost uniform after correction in (2a). (c) Minor residual banding artifacts (the black squares are gaps which originate from large saccadic motion). (d) Worst result (besides the residual banding artifact, slight lines with residual illumination deviation are visible throughout on a monitor). (e) The volume with most worsening (the other two worsened much less). In B-scans (which are not depth-averaged), the same change of illumination would be less visible due to higher noise levels. Contrast is adjusted jointly per column, corruptions in dark areas arise from the publication process.