Single-Shot Iterative Reconstruction of X-Ray Grating Interferometry
Abstract
Purpose
X-ray grating interferometry enables the simultaneous acquisition of attenuation, dark-field, and differential phase-contrast images. Dark-field provides sensitivity to sub-resolution microstructural variations inaccessible to conventional radiography. This capability has demonstrated clinical promise, particularly in lung disease assessment and breast cancer imaging. Current implementations, however, rely on phase-stepping, in which a grating is translated to acquire 5–20 exposures per projection. Phase-stepping substantially increases radiation dose and acquisition time, requires high-precision mechanical motion prone to instability, and is sensitive to patient motion, posing challenges to clinical translation. Single-shot interferometric imaging, analogous to a conventional X-ray, provides a lower-dose alternative, but existing Fourier approaches for parameter recovery are vulnerable to Moiré artifacts and mis-estimation.
Methods
We investigate a model-based iterative reconstruction (MBIR) framework for single-shot recovery of attenuation and dark-field signals from one set of interferometric reference and with-object projections. Then, amplitude, bias, and phase-shift parameters are estimated using a physics-based forward model and iterative optimization, eliminating the need for phase-stepping.
Results
The method was evaluated using experimental measurements acquired with a Modulated Phase Grating Interferometry system. Porous materials were imaged to assess accuracy and robustness. Recovery using 10 phase steps was compared with single-shot recovery using just one of the 10 steps. Compared with Fourier-based single-shot reconstruction, the proposed MBIR approach substantially improved attenuation and dark-field recovery. Attenuation image quality was comparable between the single-shot and ten-step methods. Residual moiré patterns in attenuation images were attributable to micro-motion between reference and object projections. Incorporating a sub-pixel translation parameter within reconstruction eliminated these artifacts. The dark-field signal-to-noise ratio was similar and, in fact, slightly higher for single-shot recovery (12.7) compared with 10 phase-steps (7.1).
Conclusion
A robust single-shot, model-based reconstruction has been developed that reduces acquisition time, eliminates mechanical grating motion, and lowers dose. Future work will strengthen joint attenuation/dark-field estimation.