Phase retrieval for X-ray differential phase contrast radiography with knowledge transfer learning from virtual...

Area of Science:

• Medical Imaging
• Computational Imaging
• Biomedical Engineering

Background:

• Grating-based X-ray phase contrast imaging offers advanced diagnostic potential.
• Quantitative analysis is limited by the ill-posed phase retrieval problem.
• Deep learning methods are hindered by the difficulty of obtaining ground truth phase data.

Purpose of the Study:

• To develop a robust phase retrieval method for grating-based X-ray phase contrast imaging.
• To address the challenge of acquiring ground truth phase data for deep learning training.
• To enhance the image quality of phase retrieval in 2D and 3D X-ray imaging.

Main Methods:

• Proposed a virtual differential absorption model to generate training datasets.
• Utilized transfer learning to apply knowledge from differential absorption images to phase retrieval.
• Validated the method through numerical simulations and experimental data.

Main Results:

• Successfully generated training data without requiring ground truth phase information.
• Demonstrated improved image quality in retrieved phase radiographs and CT slices compared to existing methods.
• Confirmed the feasibility of the proposed approach in both simulations and experiments.

Conclusions:

• The developed transfer learning method effectively solves the phase retrieval problem in X-ray phase contrast imaging.
• This technique enhances image quality for both 2D radiography and 3D computed tomography.
• The method shows significant potential for quantitative biomedical imaging applications.