Author: Giavanna Luisa Jadick, Patrick J La Riviere, Cailey Riggs 👨🔬
Affiliation: University of Chicago 🌍
Purpose: We explore the accuracy of physics models for propagation-based x-ray phase-contrast imaging (XPCI) with an eye to future clinical implementation, providing a practical assessment of detection tasks in which the commonly applied projection approximation (PA) appears to be insufficient.
Methods: XPCI simulations were performed with a 20-keV source and three propagation distances (0, 5, and 10 cm) using two models: PA, which assumes a sample is thin enough to ignore internal refraction; and multislice (MS), a computationally intensive but more realistic approach. An image-quality phantom was designed to emulate high-resolution mammographic imaging with various detection tasks: a 5-cm thick PMMA slab with rows of 1-mm thick nylon fibers with widths 25, 50, and 100 microns; 100-micron wide glass specks of thicknesses 10, 20, and 40 microns; and ICRU breast tissue spheres with radii 100, 150, and 200 microns. A photon-counting detector was modeled with 10-micron pixels, a Lorentzian point-spread function, and Poisson noise.
Results: In propagation-based imaging, phase effects manifest as high- and low-intensity fringes at structure edges. These increase in amplitude, number, and width as the propagation distance R increases. In all simulations, PA underestimated fringe amplitude. The effect was most prominent when R = 0 cm, where PA predicts no phase effect. For the fibers and spheres, MS clearly showed one to two single-pixel fringes per edge. For the specks, fringes were harder to resolve relative to background noise; however, PA underestimated contrast, possibly due to partial volume averaging. As R increased and PA began to show phase effects, its fringe amplitudes grew closer to those of MS.
Conclusion: Accurate forward modeling is essential for quantitative phase retrieval. PA is commonly used for convenience and analytical invertibility. As XPCI approaches clinical feasibility, the approximation should be revisited, and MS-based methods should be developed to optimize patient imaging.