Short-Trajectory artifact-reversal diffusion for CCTA calcium deblooming.

BACKGROUND

Calcium blooming in coronary CT angiography (CCTA) enlarges calcified plaques, obscures the coronary lumen, and complicates stenosis assessment.

OBJECTIVE

To develop an efficient restoration method that suppresses blooming while preserving calcium and recovering adjacent lumen structure.

METHODS

We propose Short-Trajectory Artifact-Reversal (STAR) Diffusion, which progressively transforms a blooming-corrupted patch toward a debloomed target. Soft Annular Boundary Attention (SABA) reduces attention leakage from the blooming-prone annulus into the calcium core.

RESULTS

On 232 held-out synthetic 3D test cases, each sampled from a distinct anatomical location and unique plaque, background, calcium concentration, and exposure/noise configuration, STAR Diffusion achieved Dice 0.9248, Core RMSE 134.99 HU, and Annulus RMSE 90.05 HU. Metrics were computed after slice-wise inference and volumetric reassembly, with one matched case-level measurement per test volume for direct comparison across methods. The method improved calcium-region agreement and annular error relative to classical, GAN-based, and latent diffusion baselines, while ablations showed complementary contributions from spatial-prior conditioning and SABA. STAR Diffusion also reduced inference time compared with long-trajectory latent diffusion.

CONCLUSIONS

In synthetic paired-data experiments, short-trajectory, spatially guided diffusion with boundary-aware attention reduced blooming while preserving plaque morphology and apparent lumen visibility. Clinical examples showed qualitatively similar behavior across plaque shapes and backgrounds.