Dynamic X-ray speckle-tracking imaging with high-accuracy phase retrieval based on deep learning

Area of Science:

• Advanced imaging techniques
• Materials science
• Biomedical imaging

Background:

• Speckle-tracking X-ray imaging offers flexible dynamic imaging with phase, transmission, and scattering data.
• Traditional methods exhibit phase distortion with abrupt density changes, limiting real-world sample analysis.
• Accurate phase retrieval is crucial for quantitative dynamic X-ray imaging.

Purpose of the Study:

• To develop a deep-learning based method for high-accuracy phase retrieval in dynamic X-ray speckle-tracking imaging.
• To overcome the phase distortion limitations of conventional speckle-tracking techniques.
• To enable precise quantitative analysis of dynamic processes using X-ray imaging.

Main Methods:

• Implementation of a deep-learning algorithm for X-ray speckle-tracking imaging.
• Utilizing simultaneous phase, transmission, and scattering image acquisition.
• Validation using phantom calibration and polyurethane foaming experiments.

Main Results:

• The deep-learning method achieved high-accuracy phase retrieval, consistent with theoretical profiles in phantom tests.
• Accurate visualization of complex bubble microstructure evolution during polyurethane foaming was demonstrated.
• The proposed technique effectively mitigates phase distortion issues inherent in traditional methods.

Conclusions:

• The developed deep-learning approach provides a promising solution for dynamic X-ray imaging with accurate phase retrieval.
• This method significantly enhances the application scope of speckle-tracking X-ray imaging.
• Potential for extensive applications in metrology and quantitative dynamic analysis across material science, physics, chemistry, and biomedicine.