SP-XTIN: A single projection grating-based X-ray tri-contrast imaging network

Abstract

Background and Objective:

Grating-based X-ray imaging (GBXI) enables the acquisition of tri-contrast signals—absorption, phase, and dark- field—making it highly promising for applications in clinical diagnostics. However, traditional GBXI requires phase stepping of gratings, leading to high radiation doses. In this study, a single projection grating-based X-ray tri-contrast imaging network (SP-XTIN) is proposed.

Methods:

A Pix2pixHD-based architecture is adopted, and a multi-task learning strategy is employed to transform the generator into a multi-output model that can simultaneously generate tri-contrast images. Additionally, an edge loss term is integrated into the loss function to enhance edge preservation in the tri-contrast images.

Results:

The proposed SP-XTIN is validated on two experimental datasets: one acquired with synchrotron radiation (SR) and another using a laboratory X-ray tube source. For the SR dataset, the feature similarity index measure (FSIM) values for absorption, phase, and dark-field signals achieved were 0.9871, 0.9863, and 0.9786, respectively. Using the laboratory X-ray tube source dataset, the FSIM values were 0.9883, 0.9670, and 0.9631.

Conclusion:

The proposed SP-XTIN is effective in advancing GBXI technology. These results highlight its effectiveness and are expected to contribute to the further development of this field.