Model-driven phase retrieval network for single-shot x-ray Talbot-Lau interferometer imaging.

Area of Science:

• Medical Imaging
• Computational Imaging
• X-ray Optics

Background:

• Differential Phase Contrast (DPC) imaging using X-ray Talbot-Lau interferometry offers accelerated data acquisition.
• However, single-shot DPC imaging is prone to significant image artifacts, limiting its clinical utility.

Purpose of the Study:

• To develop a novel method for artifact reduction in single-shot X-ray DPC imaging.
• To enable high-quality DPC image reconstruction from a single projection using advanced computational techniques.

Main Methods:

• A deep convolutional neural network (CNN) was developed, incorporating the physical imaging model of X-ray Talbot-Lau interferometry.
• The CNN was trained to estimate DPC images directly from single-shot projection data, bypassing the need for extensive labeled datasets.
• Validation was performed using both phantom and biological specimens.

Main Results:

• The proposed CNN-based method successfully generated high-quality DPC images from single-shot projections.
• Significant reduction in image artifacts was observed compared to conventional methods.
• The approach demonstrated effective performance on both phantom and biological samples.

Conclusions:

• This study introduces a new paradigm for single-shot X-ray DPC imaging.
• The integration of CNNs with physical models offers a powerful approach for artifact reduction and image quality enhancement in accelerated X-ray imaging.
• This method holds promise for advancing rapid and high-fidelity X-ray phase contrast imaging applications.