Multi-stage AI reconstructs 3D coronary arteries from limited X-rays
Multi-Stage NeRF for Efficient 3D Coronary Artery Reconstruction from Two Narrow-Angle Angiographic Projections
Computational GeometryComputer Vision and Pattern Recognition
Summary
X-ray coronary angiography shows heart vessels in 2D during procedures, but doctors need 3D views for better understanding. The researchers created a two-step AI method called NeCA++ that builds a rough 3D model first and then refines details only in likely vessel areas. This approach works well even when only two narrow-angle X-ray views are available, which is common in clinics. It reconstructs accurate 3D coronary artery shapes quickly and more reliably than previous methods.
What this means in practice
- •For cardiac intervention teams: Obtain detailed 3D coronary artery models from routine clinical X-ray images taken at narrow angles to support treatment planning.
- •For medical imaging device developers: Integrate multi-stage NeRF-based reconstruction algorithms to enhance software for coronary artery visualization in X-ray systems with limited-angle views.$Commercial implications: Enables marketable imaging software upgrades for catheter labs delivering improved 3D vessel models despite constrained imaging angles.
Authors
Deyu Meng, Mojtaba Lashgari, Yiying Wang, Abhirup Banerjee
Abstract
X-ray coronary angiography is the clinical gold standard for coronary artery disease during real-time cardiac interventions, but provides only 2D projections of inherently 3D vessels. Existing learning-based 2D-to-3D reconstruction methods typically require wide angular coverage or multiple views, assumptions that are rarely satisfied in routine practice where only two projections with narrow angular separation are available. To address these challenges, we propose NeCA++, a multi-stage self-supervised neural radiance field (NeRF) framework tailored to clinically realistic acquisition constraints. The framework decomposes reconstruction into two stages that progressively refine spatial support and representation capacity. In the first stage, a coarse 3D representation of the vasculature is reconstructed, restricting the subsequent optimisation to regions with a higher likelihood of vessel presence, termed an active region. Afterward reconstruction is restricted to this region while higher-resolution representations are progressively activated to recover fine vascular details. This multi-stage strategy focuses learning on anatomically plausible regions, mitigates gradient dilution under extreme sparsity, and stabilises global topology before recovering fine vascular branches. Furthermore, two vessel-specific regularisations are introduced: a ray-aligned constraint to reduce projection-induced ambiguity, and a bimodal density penalty to enable early vessel-background separation. Extensive experiments across three datasets (ImageCAS, ASOCA, and Synthetic RCA) and four angular configurations demonstrate consistent superiority over state-of-the-art baselines, particularly under clinically realistic narrow-angle settings, while achieving reconstruction within 58 seconds per case.