Empirically Informed Blood Vessel Network Reconstruction for Multimodality Image-Based Vascular Systems Biology

Reconstruction of vascular topology from imperfect 3D imaging data is essential for accurate hemodynamic simulations. Therefore, we developed a "gap-filling" method predicated on vascular biology and empirical patterns to restore incomplete vessels, independent of the imaging modality and spatial resolution employed.

Our gap-filling algorithm reconstructs a vascular network based on distributions of its extant morphological features and employs different rules for healthy and tumor tissue. Algorithm performance was evaluated on (i) incomplete synthetic vascular trees with simulated CT noise, (ii) micro-CT-acquired brain tumor vasculature, (iii) micro-MRI-derived vasculature from a healthy mouse brain, and (iv) vasculature from the murine hippocampus imaged with light-sheet microscopy (LSM).

For synthetic vascular trees, our algorithm restored more gaps when the volume fraction was between 80% and 95%. Peak vessel segment recovery occurred for intermediate volume fractions (86%-89%). Gap-filling performance was impacted when image noise was introduced. Across CT, MRI, and LSM, our method successfully reconnected missing vessel segments while preserving the overall features of the vascular network. Gap-filling of tumor vasculature, with its abnormal topology, was better when tumor-specific reconstruction rules were employed.

Our method showed efficient restoration of vascular network connectivity for 3D data derived from common imaging modalities at different resolutions. We also demonstrated the significance of applying distinct biologically-informed rules when reconstructing healthy and tumor vascular networks.