Differentiable SV-FBP framework makes cone-beam CT 10x faster and more flexible

A team led by Chengze Ye and Andreas Maier has published a robustness and stability analysis of the differentiable shift-variant filtered backprojection (SV-FBP) framework for cone-beam CT reconstruction. Unlike traditional approaches that require handcrafted weighting schemes, this data-driven method learns redundancy weights directly from the acquisition geometry. The study demonstrates that SV-FBP remains stable even under highly irregular and discontinuous source trajectories, with performance depending more on spatial sampling distribution than trajectory ordering or continuity.

Under sparse-view conditions, differentiable SV-FBP delivers competitive reconstruction quality while providing an order-of-magnitude reduction in computation time compared to iterative methods at moderate sampling densities. However, under severe undersampling, the lack of iterative data consistency leads to performance degradation. Notably, the framework generalizes to non-planar multi-isocenter geometries—such as Lissajous-saddle trajectories—without requiring architectural modifications. These findings position differentiable SV-FBP as a flexible, efficient solution for robotic and non-standard CBCT acquisition scenarios.