3DGR-CAR: Coronary artery reconstruction from ultra-sparse 2D X-ray views with a 3D Gaussians representation

TL;DR

This work addresses 3D coronary artery reconstruction from ultra-sparse 2D X-ray views by introducing a 3D Gaussian representation (3DGR) and a Gaussian Center Predictor (GCP). The GCP, implemented as a U-Net-based network, predicts rough 3D center positions from a single projection to initialize Gaussians, which are then refined via a projection-based loss that combines $\mathcal{L}_{L2}$ and $\mathcal{L}_{clL2}$, along with a centerline constraint. A dedicated training objective for the GCP combines Chamfer distance, Soft-ClDice, and depth losses including $\mathcal{L}_{SILog}$, $\mathcal{L}_{GradL1}$, and $\mathcal{L}_{MaskL1}$, with empirically tuned weights. Empirical results on ImageCAS and ASOCA show that 3DGR-CAR significantly outperforms FBP, NeRP, and vanilla 3DGR for 2-view sparse reconstructions, delivering higher projection and volume quality with substantially faster convergence. The approach enables accurate, low-dose coronary artery reconstruction suitable for clinical workflows, illustrating the practical impact of combining 3D Gaussian representations with learned center initialization in ultra-sparse imaging scenarios.

Abstract

Reconstructing 3D coronary arteries is important for coronary artery disease diagnosis, treatment planning and operation navigation. Traditional reconstruction techniques often require many projections, while reconstruction from sparse-view X-ray projections is a potential way of reducing radiation dose. However, the extreme sparsity of coronary arteries in a 3D volume and ultra-limited number of projections pose significant challenges for efficient and accurate 3D reconstruction. To this end, we propose 3DGR-CAR, a 3D Gaussian Representation for Coronary Artery Reconstruction from ultra-sparse X-ray projections. We leverage 3D Gaussian representation to avoid the inefficiency caused by the extreme sparsity of coronary artery data and propose a Gaussian center predictor to overcome the noisy Gaussian initialization from ultra-sparse view projections. The proposed scheme enables fast and accurate 3D coronary artery reconstruction with only 2 views. Experimental results on two datasets indicate that the proposed approach significantly outperforms other methods in terms of voxel accuracy and visual quality of coronary arteries. The code will be available in https://github.com/windrise/3DGR-CAR.

Figures (5)

• Figure 1: Coronary arteries occupy only approximately 0.1% of a typical cardiac volume. (a-c): Gaussian centroid initial positions derived from our Gaussian center predictor and FBP results. (d): Ground truth. (e-g) Coronary reconstruction results from FBP fbp1976reconstruction, NeRP shen2022nerp and 3DGR.
• Figure 2: Overview of 3DGR-CAR. The green box illustrates the pipeline for training GCP; the yellow box denotes the Gaussian position are initialized by GCP; the blue box represents Gaussian parameters optimization with sparse-view projections.
• Figure 3: Comparison of the new projection quality evaluation with increasing numbers of projections on ImageCAS.
• Figure 4: Comparison of volume quality evaluation with increasing numbers of projections on ImageCAS.
• Figure 5: Comparison of 3D vascular and new projections from different methods. The blue circles marked the most differences between 3DGR-FBP and 3DGR-GCP. (left: 3D coronary reconstruction results; right: new projections of coronary arteries).