A tensor PRISM algorithm for multi-energy CT reconstruction and comparative studies

Area of Science:

• Medical Imaging
• Computational Imaging
• Image Reconstruction

Background:

• Multi-energy CT (MECT) offers advantages over single-energy CT (SECT) due to its ability to acquire data across different energy spectra.
• MECT data exhibits unique sparsity characteristics, both within individual energy datasets and across different energies, which can be leveraged for improved image reconstruction.
• Existing methods may not fully exploit all available sparse characteristics and prior information inherent in MECT data.

Purpose of the Study:

• To develop a novel tensor PRISM model for multi-energy CT (MECT) image reconstruction.
• To consistently integrate prior knowledge, including low rank, intensity, and sparsity, using higher-dimensional tensor techniques.
• To improve the image quality of MECT compared to existing algorithms.

Main Methods:

• Proposed a new tensor PRISM model to consistently incorporate prior knowledge of low rank, intensity, and sparsity.
• Utilized higher-dimensional tensor techniques to represent and process MECT data.
• Solved the regularization and convex minimization problem using tensor unfolding and an extended tensor-based split-Bregman algorithm.
• Consistently treated and mixed different constraints within the new algorithm.

Main Results:

• The proposed tensor PRISM approach demonstrated significantly better performance than the popular l1 regularization algorithm.
• Numerical experiments confirmed superior image quality for MECT reconstruction using the new tensor PRISM model.
• The method effectively leverages low-rank properties (stationary background, energy similarity) and single-energy features (intensity, sparsity).

Conclusions:

• The developed tensor PRISM model offers a more effective way to reconstruct MECT images.
• This approach provides a consistent framework for utilizing diverse prior information in MECT.
• The proposed method represents a significant advancement in MECT image quality enhancement.