NeRF-CA: Dynamic Reconstruction of X-ray Coronary Angiography with Extremely Sparse-views

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Dynamic 4D coronary angiography (CA) reconstruction from 2D X-ray images is clinically challenging.
• Current methods often need significant user input or large datasets.
• Neural Radiance Fields (NeRF) show promise but struggle with sparse views, motion, and complex anatomy in CA data.

Purpose of the Study:

• To develop a fully automatic 4D CA reconstruction method using sparse X-ray angiograms.
• To address the limitations of NeRF in medical imaging, specifically for CA.
• To enable high-fidelity 4D CA reconstruction from minimal clinical data.

Main Methods:

• Introduced NeRF-CA, a novel approach for 4D CA reconstruction.
• Decoupled the dynamic coronary artery from the static background to manage cardiac motion.
• Adapted NeRF for sparse-view X-ray data, handling complex vessel morphology.

Main Results:

• Achieved adequate 4D CA reconstructions from as few as four sparse angiograms.
• Significantly outperformed state-of-the-art sparse-view X-ray NeRF methods.
• Validated quantitatively and qualitatively using 4D phantom datasets.

Conclusions:

• NeRF-CA represents a significant advancement in automatic 4D CA reconstruction from sparse views.
• The method effectively overcomes challenges of motion and data sparsity in CA.
• The publicly available codebase aims to accelerate research in 4D medical image reconstruction.