Efficient and Practical Framework for Bias Estimation in Spectral CT.

Area of Science:

• Medical Imaging
• Computational Physics
• Statistical Modeling

Background:

• Spectral computed tomography (CT) is vital for quantitative imaging but faces challenges in accurately predicting spectral quantitative bias.
• Bias in CT arises from complex factors including model inaccuracies, hardware imperfections, and nonlinear processing steps.
• Accurate bias prediction is computationally expensive with conventional methods, hindering system design and optimization.

Purpose of the Study:

• To develop a practical, projection-based statistical framework for efficient estimation of noise-induced spectral bias.
• To offer a computationally inexpensive alternative to Monte Carlo (MC) simulations for bias prediction.
• To enable rapid, bias-aware optimization of spectral CT acquisition parameters.

Main Methods:

• Modeled X-ray attenuation through a patient-equivalent path with iodine inserts.
• Simulated ideal and realistic photon-counting detector responses across multiple energy bins.
• Employed Bayesian statistics to compute material decomposition probabilities and estimate bias, referencing MC simulations.

Main Results:

• The proposed bias estimator closely matched MC-derived bias, with an average relative iodine bias difference of only 0.44%.
• The estimator's runtime was 0.5% of that required for MC simulations, demonstrating significant computational efficiency.
• Identified critical noise-bias tradeoffs, showing optimal thresholds for noise differ from those for bias.

Conclusions:

• Efficient spectral bias and noise estimation are crucial for designing accurate quantitative CT systems.
• The developed modular framework allows for rapid, bias-aware optimization of spectral acquisition parameters.
• This approach is adaptable to various spectral CT technologies, including photon-counting detectors.