End-to-end differentiable spectral ct imaging
AssigneesGE Precision Healthcare LLC, The Board of Trustees of the Leland Stanford Junior University
InventorsSen Wang, Yirong Yang, Adam Wang, Grant M. Stevens
Life of this patent
Estimating the term…
Loading patent…
AssigneesGE Precision Healthcare LLC, The Board of Trustees of the Leland Stanford Junior University
InventorsSen Wang, Yirong Yang, Adam Wang, Grant M. Stevens
Estimating the term…
An end-to-end differentiable image-processing pipeline for spectral Computed Tomography (CT) systems, including both Photon Counting (PCCT) and dual-energy (DECT) configurations, is provided. In the disclosed framework, spectral data acquired from a detector across two or more distinct X-ray energy spectra or energy bins are processed through a sequence of differentiable operators, including one or more differentiable preprocessing operators, a differentiable material-decomposition operator, and a differentiable image-reconstruction operator. The material-decomposition operator can be implemented using an unrolled iterative solver, an implicit-differentiation formulation, or a trained empirical proxy model. An image-domain loss between reconstructed and reference images, or other differentiable regularization criteria, is backpropagated through the pipeline to update parameters of the preprocessing operators, the material-decomposition operator, and/or a forward detector model. This enables automatic physics-informed calibration that improves quantitative image fidelity, reduces artifacts, and eliminates dependence on manual recalibration or Monte Carlo reference data.
A method, comprising:
A spectral CT system, comprising:
A computer program product comprising a non-transitory computer-readable memory having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to:
Dependencies are inferred from the claim text.