Document resource
Introduction Neurovascular optical coherence tomography (nOCT) has begun clinical trials for FDA clearance. However, registration of these high-resolution data back to the anatomy in situ is currently not available. Accurate 3D registration with rotational angiography (3DRA) is therefore essential for downstream analysis and modeling. We propose a morphology-independent, physics-informed framework to reconstruct the true 3D OCT probe trajectory using single-plane fluoroscopy and OCT-derived geometric constraints, enabling accurate volumetric registration.Methods The OCT probe trajectory was reconstructed by optimizing a continuous representation against three inputs: (1) a 3D arterial centerline extracted from 3DRA, (2) a 2D fluoroscopic projection of the probe path, and (3) probe to lumen centroid distances derived from the OCT pullback. Given that the fluoroscopic projection strongly constrains the in-plane geometry, the reconstruction reduces to estimating the out-of-plane component of the trajectory. A compact multi-layer perceptron was used to model this remaining degree of freedom. The network was trained using a multi-term loss function enforcing geometric consistency, including OCT plane orthogonality to the centerline, agreement with OCT-derived radial measurements, and spatial smoothness of the catheter path. The reconstructed trajectory, together with the arterial centerline and the OCT centroid curve, was used to construct paired ruled surfaces that map the straight OCT pullback geometry to the true 3D probe path, enabling direct registration of the OCT volume into the 3DRA coordinate system without anatomical landmarks.Results The reconstructed probe trajectory demonstrated agreement with the 3D anatomy. The registered OCT volume aligned accurately with the vascular geometry, with correct localization of the aneurysm neck, artery bifurcations and device positions were obtained both in-vitro and in-vivo experiments.Conclusion With a single fluoroscopic projection, OCT-3DRA registration can be reduced to a one-dimensional reconstruction problem. This formulation enables efficient recovery of the 3D probe trajectory using a lightweight neural network. When biplane fluoroscopic projections are available, the probe path can be directly reconstructed without learning, and ruled surface methods can be extended to recover time-resolved trajectories.Disclosures M. Epshtein: None. M. Gounis: None. V. Anagnostakou: None.Abstract E-357 Figure 1(A) Left: OCT Slice Right: Longitudinal OCT pullback showing a cross-sectional frame along the apparent straight catheter path. (B) the in vitro model reconstructed directly from the OCT pullback (C) Final registration of the geometrically corrected OCT volume with the 3DRA arterial reconstruction. (D) 3D view of the probe projection surface (red) defined by the fluoroscopic imaging geometry, shown in correct spatial orientation relative to the 3DRA arterial reconstruction, the true catheter trajectory lies on this surface.