Multi-material decomposition using statistical image reconstruction for spectral CT.

Area of Science:

• Medical Imaging
• Computational Imaging
• Materials Science

Background:

• Spectral computed tomography (CT) enables material characterization by separating basis materials.
• Dual-energy (DE) CT provides two energy measurements, allowing decomposition of up to two materials.
• Clinical and industrial applications often require decomposition of three or more materials, necessitating additional constraints.

Purpose of the Study:

• To propose a novel penalized-likelihood (PL) method for reconstructing multi-material images from spectral CT data.
• To address the limitations of existing image-domain (ID) methods for triple-material decomposition.
• To improve the accuracy and reduce artifacts in multi-material decomposition using sinogram data.

Main Methods:

• Developed a penalized-likelihood (PL) method incorporating edge-preserving regularizers for multi-material reconstruction.
• Implemented an optimization transfer method utilizing pixel-wise separable quadratic surrogate (PWSQS) functions for efficient cost function minimization.
• The PWSQS algorithm allows simultaneous pixel updates while maintaining coupled basis materials for faster convergence.

Main Results:

• The proposed PL method significantly reduced noise, streak, and cross-talk artifacts compared to the image-domain (ID) method.
• Reconstructed basis component images showed substantial improvement in quality.
• Achieved significantly smaller root mean square errors in the decomposed material images.

Conclusions:

• The penalized-likelihood method with PWSQS is effective for multi-material decomposition in spectral CT.
• This approach offers superior performance over existing image-domain methods, particularly for complex material mixtures.
• The method holds promise for enhanced material characterization and quantification in various applications.