Joint Reconstruction and Spectrum Refinement for Photon-Counting-Detector Spectral CT

Area of Science:

• Medical Imaging
• Physics
• Computer Science

Background:

• Photon-counting detector CT (PCD-CT) offers significant advantages in noise reduction, dose efficiency, and material-specific imaging.
• Accurate spectral reconstruction is crucial for PCD-CT applications but is hindered by spectrum estimation errors, particularly from energy threshold calibration inaccuracies.
• Current methods struggle with ill-conditioned problems arising from pixel-wise spectral variations.

Purpose of the Study:

• To develop a robust method for accurate spectral reconstruction in PCD-CT.
• To address the challenges of spectrum estimation errors caused by energy threshold calibration bias and intrinsic detector variations.
• To improve the well-posedness and stability of spectral reconstruction in PCD-CT.

Main Methods:

• Proposed a joint reconstruction and spectrum refinement algorithm (JoSR) utilizing a novel non-negative matrix factorization (NMF) based spectrum model.
• Formulated an optimization objective combining polychromatic spectral imaging and basis material decomposition.
• Employed a proximal regularized block coordinate descent algorithm to solve the non-convex optimization problem.

Main Results:

• The JoSR method effectively reduces the dimensionality of spectral unknowns, making the reconstruction problem well-conditioned.
• Simulation studies and experimental validation on a laboratory PCD-CT system demonstrated the method's efficacy.
• Achieved superior image quality and quantitative accuracy compared to existing state-of-the-art methods.

Conclusions:

• The proposed JoSR algorithm provides a stable and accurate solution for spectral reconstruction in PCD-CT.
• The NMF-based spectrum model effectively handles energy threshold calibration errors and intrinsic detector variations.
• JoSR significantly advances the capabilities of PCD-CT for clinical applications requiring precise material decomposition and spectral information.