Comparison of five one-step reconstruction algorithms for spectral CT.

Area of Science:

• Medical Imaging
• Computational Imaging
• Photon Counting CT

Background:

• Dual-energy CT scanners are evolving, with spectral photon counting CT (SPCCT) emerging.
• SPCCT requires a reconstruction process involving material decomposition and tomographic reconstruction.
• One-step inversion methods perform these tasks simultaneously, but can be slower than two-step methods.

Purpose of the Study:

• To compare the convergence speeds of five different one-step inversion algorithms for SPCCT reconstruction.
• To evaluate the impact of μ-preconditioning on the convergence of these algorithms.
• To assess practical aspects like memory footprint and parameter tuning.

Main Methods:

• Adapted five one-step algorithms (Bayesian, three quadratic surrogate-based, one primal-dual) for SPCCT reconstruction.
• Utilized a three-material decomposition basis and spatial regularization across five energy bins.
• Investigated the effect of three different μ-preconditioners on algorithm convergence.

Main Results:

• Significant differences in convergence speeds were observed across the evaluated algorithms.
• The Mechlem et al. algorithm required only ten iterations, while others needed hundreds or thousands.
• μ-preconditioning was found to impact convergence speed, with specific preconditioners showing varying effectiveness.

Conclusions:

• The Mechlem et al. one-step inversion algorithm offers superior convergence speed for SPCCT reconstruction compared to the other methods studied.
• Algorithm choice and μ-preconditioning strategy are critical for optimizing reconstruction time in SPCCT.
• Further practical considerations such as memory usage and parameter tuning are important for clinical implementation.