VAMOS-OCTA: Vessel-Aware Multi-Axis Orthogonal Supervision for Inpainting Motion-Corrupted OCT Angiography Volumes

TL;DR

This work tackles motion-induced corruption in handheld OCTA volumes, where bulk motion can blank entire B-scans and degrade en face projections. It introduces VAMOS-OCTA, a 2.5D U-Net that reconstructs center B-scans from neighboring slices, guided by the Vessel-Aware Multi-Axis Orthogonal Supervision (VAMOS) loss. The loss blends a vessel-weighted reconstruction term with axial and lateral projection constraints on MIP and AIP to enforce vascular continuity across depth and across slices, with a projection weight of $\lambda_{proj}=3$. Across synthetic and real-world corruptions, VAMOS-OCTA achieves sharper B-scans and more coherent en face projections than prior methods, highlighting the value of multi-axis supervision for volumetric OCTA restoration. The authors also provide the code at https://github.com/MedICL-VU/VAMOS-OCTA, facilitating broader adoption and validation.

Abstract

Handheld Optical Coherence Tomography Angiography (OCTA) enables noninvasive retinal imaging in uncooperative or pediatric subjects, but is highly susceptible to motion artifacts that severely degrade volumetric image quality. Sudden motion during 3D acquisition can lead to unsampled retinal regions across entire B-scans (cross-sectional slices), resulting in blank bands in en face projections. We propose VAMOS-OCTA, a deep learning framework for inpainting motion-corrupted B-scans using vessel-aware multi-axis supervision. We employ a 2.5D U-Net architecture that takes a stack of neighboring B-scans as input to reconstruct a corrupted center B-scan, guided by a novel Vessel-Aware Multi-Axis Orthogonal Supervision (VAMOS) loss. This loss combines vessel-weighted intensity reconstruction with axial and lateral projection consistency, encouraging vascular continuity in native B-scans and across orthogonal planes. Unlike prior work that focuses primarily on restoring the en face MIP, VAMOS-OCTA jointly enhances both cross-sectional B-scan sharpness and volumetric projection accuracy, even under severe motion corruptions. We trained our model on both synthetic and real-world corrupted volumes and evaluated its performance using both perceptual quality and pixel-wise accuracy metrics. VAMOS-OCTA consistently outperforms prior methods, producing reconstructions with sharp capillaries, restored vessel continuity, and clean en face projections. These results demonstrate that multi-axis supervision offers a powerful constraint for restoring motion-degraded 3D OCTA data. Our source code is available at https://github.com/MedICL-VU/VAMOS-OCTA.

Figures (5)

• Figure 1: Overview of the VAMOS-OCTA framework. A 2.5D U-Net reconstructs a central target B-scan from a stack of neighboring B-scans. Training is guided by the proposed VAMOS loss, which combines vessel-weighted reconstruction with multi-axis orthogonal supervision. Axial and lateral projection constraints are enforced by collapsing each B-scan along depth and lateral dimensions, respectively, to form 1D maximum and average intensity profiles.
• Figure 2: Reconstructed B-scan for each method. Arrows highlight areas of interest: (iv@) oversmooths vessels and (v@) introduces horizontal banding, while VAMOS-OCTA (vi@) restores vessel contrast and spatial continuity.
• Figure 3: Depth-wise en face MIP reconstructions. (I) Ground Truth, (II) Corrupted, (III) Standard MSE, (IV) SOAD Weighted MSE, (V) wMSE + Axial, (VI) VAMOS-OCTA. Zoom panels highlight an area where (III), (IV), and (V) fail to remove artifacts, while VAMOS-OCTA (VI) restores a smooth and realistic projection with recovered vessels (arrows).
• Figure 4: VAMOS-OCTA restores volumes with real-world motion corruptions without known GT. (a, c) are real-world corruptions and (b, d) are their inpaintings, with red bars highlighting significant areas of motion.
• Figure 5: Mean Intensity Error (MIE) vs. corruption severity. Even with severely degraded OCTA volumes, VAMOS (green) maintains low error, while other methods’ errors rise sharply with more missing B-scans.