Answers are grounded in this paper and cite their sources.
Interventional C-arm imaging enables minimally invasive cardiac procedures while reducing radiation exposure through sparse-view acquisition. However, reconstruction of three-dimensional vascular morphology from a limited number of projections remains a severely ill-posed inverse problem, particularly in the presence of Poisson quantum noise. This study evaluates two ray-tracing models, Siddon’s (weighted) and Bresenham’s (binary), combined with the Simultaneous Iterative Reconstruction Technique (SIRT) and Total Variation regularization (SIRT+TV). Synthetic phantoms derived from patient-specific coronary geometries were reconstructed from nine simulated noisy cone-beam projections acquired at different angles. Reconstructed volumes were binarized using Otsu and IoU-maximized thresholds. Automatic Otsu thresholding was ineffective for Bresenham-based reconstructions due to severe reconstruction artifacts. Nevertheless, IoU-maximized thresholding yielded a mean Dice-Sørensen coefficient of 0.795, indicating preservation of relevant vascular information. Siddon-based reconstruction produced fewer artifacts and achieved superior performance, reaching a Dice-Sørensen coefficient of 0.860 with SIRT. No clear advantage of TV regularization over SIRT alone was observed. The results demonstrate that key aspects of vascular morphology can be recovered from sparse-view projections. However, the observed reconstruction errors indicate that substantial methodological improvements are required before practical application can be considered.
No takes yet. Share an insight, caveat, or question.
Binesh et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: