Time-frequency domain prior constrained deep unfolding network for low-dose CT reconstruction

Area of Science:

• Medical Imaging
• Computer Vision
• Signal Processing

Background:

• Low-dose computed tomography (LDCT) reduces radiation exposure but introduces image noise and artifacts.
• Existing Convolutional Neural Networks (CNNs) show promise but have limitations in adaptability for CT reconstruction.
• Further improvements in LDCT reconstruction are needed to balance dose reduction with image quality.

Purpose of the Study:

• To develop an interpretable deep unfolding network for low-dose CT reconstruction.
• To enhance reconstruction performance by integrating iterative algorithm principles with deep learning.
• To achieve superior image quality in LDCT by effectively mitigating noise and artifacts.

Main Methods:

• Proposed an interpretable deep unfolding network leveraging time-frequency and image domain priors.
• Mapped the iterative optimization process into a deep unfolding network architecture.
• Introduced a Stage Information Memory Network (SIMN) to preserve information across network stages.

Main Results:

• The proposed model demonstrated superior performance on Mayo and Piglet datasets.
• Achieved state-of-the-art results in both quantitative metrics and visual quality.
• Validated the effectiveness of the interpretable network in complex CT reconstruction tasks.

Conclusions:

• The developed network effectively removes artifacts and noise from low-dose CT images.
• Achieved excellent reconstruction performance, enhancing diagnostic capabilities of LDCT.
• The interpretable deep unfolding approach offers a promising direction for future CT reconstruction research.