Synchrotron microtomography image restoration via regularization representation and deep CNN prior

Area of Science:

• Medical Imaging
• Image Processing
• Scientific Instrumentation

Background:

• Synchrotron-based X-ray microtomography (S-µCT) is vital in medical science but suffers from artifacts like ring artifacts and noise.
• These artifacts, including quantum and electronic noise, often appear together, degrading image quality and hindering research.
• Existing methods struggle to address the complex, simultaneous nature of these image corruptions.

Purpose of the Study:

• To develop a novel algorithm for the simultaneous removal of mixed artifacts and noise in S-µCT images.
• To improve the quality of reconstructed CT images obtained from S-µCT systems.
• To enhance the reliability and applicability of S-µCT in scientific research.

Main Methods:

• A hybrid approach combining regularization-based methods (low-rank tensor decomposition, total variation) for ring artifacts.
• Integration of a convolutional neural network (CNN) for random noise modeling.
• A plug-and-play framework merging traditional regularization with deep learning for efficient image restoration.

Main Results:

• The proposed algorithm effectively removed both ring artifacts and random noise from S-µCT images.
• Both simulation and real-world data experiments confirmed the method's efficacy.
• Quantitative analysis showed superior performance in PSNR, SSIM, and MAE compared to existing methods.

Conclusions:

• The developed algorithm is an effective tool for restoring corrupted S-µCT images.
• This method has the potential to significantly advance the application and impact of S-µCT technology.
• Improved image quality will facilitate more accurate research and diagnostics using S-µCT.