Reconstructing dynamic 3D anatomy from sparse X-ray projections
Background
Conventional CT and CBCT reconstruction requires many X-ray projections acquired over a full gantry rotation, imposing significant radiation dose and limiting temporal resolution. In image-guided radiotherapy and orthopaedic imaging, anatomy is often moving or deforming, and only sparse projection data may be available. Sparse radiographic views provide fundamentally incomplete observations of underlying structures, creating an ill-posed reconstruction problem that classical algorithms cannot reliably solve.
Aims
- Reconstruct dynamic 3D anatomy from sparse X-ray projections in clinically relevant settings including image-guided radiotherapy, CBCT, and orthopaedic imaging.
- Recover patient-specific anatomy and motion directly from limited projection data without requiring full angular sampling.
Methods
Physics-aware learning frameworks are developed that integrate continuous volumetric representations, patient-specific anatomical priors, and differentiable X-ray rendering. These components allow the model to reason about the underlying 3D anatomy consistent with the observed projections while respecting the physics of X-ray image formation.