Spectral CT imaging method based on blind separation of polychromatic projections with Poisson prior

Area of Science:

• Medical Imaging
• Computational Imaging
• Signal Processing

Background:

• Conventional computed tomography (CT) suffers from hardening artifacts.
• Narrow-energy-width projections can mitigate these artifacts but require known incident spectra.
• Obtaining these projections in a blind scenario (unknown spectra) is challenging.

Purpose of the Study:

• To develop a spectral CT blind separation algorithm for acquiring narrow-energy-width projections without prior spectral information.
• To improve the accuracy of narrow-energy-width projection reconstruction in computed tomography.

Main Methods:

• An X-ray multispectral forward model was utilized.
• A constrained optimization problem was formulated based on Poisson statistics.
• A block coordinate descent algorithm, alternating between nonnegative matrix factorization and Gauss-Newton, was employed.

Main Results:

• The developed algorithm successfully decomposes projections consistent with narrow-energy-width characteristics.
• Experimental results demonstrate improved accuracy in obtaining these projections.
• The method addresses the challenge of unknown incident spectra in spectral CT.

Conclusions:

• The proposed blind separation algorithm is effective for generating narrow-energy-width projections in spectral CT.
• This advancement enhances the potential for artifact reduction and accuracy improvement in CT imaging.
• The algorithm offers a viable solution for scenarios with unknown incident X-ray spectra.