Penalized maximum likelihood reconstruction for x-ray differential phase-contrast tomography.

Area of Science:

• Medical Imaging
• Image Reconstruction
• Differential Phase Contrast (DPC)

Background:

• Differential Phase Contrast (DPC) imaging offers valuable information but often requires phase retrieval for reconstruction.
• Existing non-iterative methods struggle with artifacts, particularly smearing in tomographic acquisitions.

Purpose of the Study:

• To propose a novel cost function with regularization for simultaneous iterative reconstruction of attenuation, phase, and scatter images from DPC data.
• To evaluate this method's performance, especially for sliding window acquisitions in DPC tomography, without needing phase retrieval.

Main Methods:

• A penalized maximum likelihood reconstruction algorithm was derived for direct image reconstruction from DPC detector values.
• Independent smoothing priors were applied to attenuation, phase, and scatter images.
• The algorithm was tested on sliding window DPC data from synchrotron acquisitions and compared to phase retrieval methods.

Main Results:

• The proposed iterative method significantly improved image quality compared to phase retrieval-based reconstructions.
• Independent priors showed no significant mutual influence.
• Reconstruction of sliding window acquisitions demonstrated substantially reduced smearing artifacts.

Conclusions:

• The proposed cost function, despite being non-convex, effectively reduces aliasing and streak artifacts compared to phase retrieval methods.
• This iterative approach successfully reconstructs sliding window DPC data with negligible smearing.