Beam Hardening and Visibility Hardening Correction In Dark-Field CT
Abstract
Purpose
X‑ray dark‑field imaging is sensitive to unresolved microstructure and has strong potential for lung tissue characterization, disease identification, and virtual histology. However, quantitative dark-field CT remains challenging due to uncertainties in the visibility spectrum and polychromatic non-linearity of the sample’s diffusion coefficient. We propose modeling the sample’s energy dependence to produce pseudo-monochromatic projections, reducing the effects of beam hardening and visibility hardening in dark-field CT.
Methods
We simulate polychromatic dark-field CT projections for a computational thorax phantom using energy-weighted and visibility-weighted integration of the attenuation and dark-field CT coefficients. We correct for polychromatic effects by modeling the energy-dependence of the dark-field CT projections as a power law, yielding pseudo-monochromatic projections. We compare CT reconstructions with no beam hardening corrections, with existing subtraction-based beam hardening corrections, and using the modeled energy dependence.
Results
CT reconstructions show that modeling the energy dependence of the dark-field CT coefficient yields superior results compared to reconstructions using uncorrected dark-field sinograms. By incorporating pseudo-monochromatic dark-field projections, the non-linearity of the diffusion coefficient is removed, reducing cupping artifacts and artificial dark-field signal due purely to beam hardening. The reconstruction is improved over existing beam hardening correction methods even when there is a model mismatch in the power-law energy dependence, highlighting the utility of this method.
Conclusion
Modeling the energy dependence of dark-field sinograms provides a practical route to quantitative dark-field CT by transforming polychromatic measurements into pseudo‑monochromatic projections. This approach reduces cupping and removes artificial dark-field signal and outperforms subtraction‑based corrections, even under model mismatch. These results support modeling the energy dependence of dark-field CT data for quantitative dark‑field CT. This will advance quantitative applications of dark-field lung imaging, such as tissue characterization, disease identification, and virtual histology.