A Deep Learning Reconstruction Framework for Differential Phase-Contrast Computed Tomography With Incomplete Data.

Area of Science:

• Medical Imaging
• Computational Imaging
• Soft-Tissue Analysis

Background:

• Differential phase-contrast computed tomography (DPC-CT) is vital for soft-tissue and low-atomic-number sample analysis.
• Incomplete projection data is a common challenge in DPC-CT, hindering conventional reconstruction algorithms.
• Existing methods often require complex parameter tuning, are sensitive to noise, and are time-consuming.

Purpose of the Study:

• To develop a novel deep learning reconstruction framework for DPC-CT with incomplete projection data.
• To address the limitations of traditional algorithms in handling sparse-view, limited-view, and missing-view DPC-CT scenarios.
• To improve the efficiency and accuracy of DPC-CT image reconstruction.

Main Methods:

• A deep learning neural network is tightly coupled with the DPC-CT reconstruction algorithm.
• The framework operates within the domain of DPC projection sinograms to estimate complete sinograms from incomplete data.
• The trained framework reconstructs final DPC-CT images directly from incomplete projection sinograms.

Main Results:

• The proposed framework successfully generates complete phase-contrast projection sinograms, avoiding artifacts from incomplete data.
• Validation with synthetic and experimental datasets (sparse-view, limited-view, missing-view) demonstrates superior performance.
• The deep learning approach achieves superior imaging quality at a faster speed and with fewer parameters compared to existing methods.

Conclusions:

• The developed deep learning framework provides an effective solution for DPC-CT reconstruction with incomplete data.
• This approach significantly enhances imaging quality and processing speed, making DPC-CT more practical.
• The study highlights the potential of advanced deep learning techniques in advancing DPC-CT applications.