Pose-aware diffusion improves extremely sparse-view CT reconstructions
Extremely Sparse-View Computed Tomography from 2D Projections via Pose-Aware Diffusion Priors
Computer Vision and Pattern RecognitionArtificial Intelligence
Summary
Sometimes, doctors need to take CT scans with very few X-ray views because of time, movement, or radiation safety. Existing methods either rely on many training examples or produce images with artifacts. The authors developed a way to teach a computer to understand each viewpoint’s position from just single 2D images, so it can better reconstruct 3D scans using very few views. This method works well even when training data has similarly limited views and outperforms other techniques on public CT datasets.
What this means in practice
- •For medical imaging teams: Produce higher-quality 3D CT scans from fewer X-ray views to reduce patient exposure and scanning time.
- •For industrial nondestructive testing teams: Generate accurate 3D images from minimal 2D projections for faster inspection of materials with limited scan angles.
Authors
Linrui Dai, Jinxiu Liang, Chu Zhou, Mae Yamaguchi, Ryuya Takahashi, Imari Sato
Abstract
Extremely sparse-view computed tomography (xSV-CT) is critical when scanning time, motion, or radiation tolerance limits each scan to only a few projections. In many such scenarios, the same acquisition limits also constrain dataset construction, making pre-reconstructed 3D volumes from dense scans unavailable. This exposes a gap in current sparse-view approaches: Optimization-based methods, which use the sparse measurements directly but lack data-driven priors to suppress artifacts, and learning-based methods, which use strong priors but usually require densely scanned training volumes acquired outside the xSV-CT regime. We study this train-sparse / infer-sparse setting by learning a pose-aware diffusion prior from one 2D projection per training sample, then guiding diffusion sampling with measured-view fidelity, shared-volume reprojection, and projection-physics consistency for tomographic reconstruction. On two public CT datasets across 1-, 4-, and 8-view settings, the proposed method achieves the best novel-view synthesis scores and best or competitive reconstruction quality compared with baselines that use dense-volume supervision.