Theoretical Prediction of Bias in Model-Based Material Decomposition

Area of Science:

• Medical Imaging
• Computational Physics
• Materials Science

Background:

• Spectral imaging enhances computed tomography (CT) quantitative capabilities over single-energy methods.
• Spectral CT is prone to bias, potentially leading to material misclassification and reduced accuracy.
• Current bias quantification relies on empirical methods, offering limited theoretical understanding.

Purpose of the Study:

• To develop a theoretical framework for predicting bias in model-based material decomposition for spectral CT.
• To differentiate and quantify statistical bias and model mismatch bias.
• To validate the theoretical predictions against empirical measurements under diverse imaging parameters.

Main Methods:

• Proposed a theoretical model to estimate statistical bias (noise propagation) and model mismatch bias (forward model discrepancies).
• Conducted empirical validation across varying spectral separation, mAs, and spectral response mismatches.
• Analyzed sensitivity to perturbations in tube potential and effective energy.

Main Results:

• The theoretical framework accurately predicted bias across a broad range of spectral CT imaging conditions.
• Statistical bias prediction showed strong agreement with empirical data, with minor deviations at low spectral separation and mAs.
• Model mismatch bias correlated with spectral response mismatches and was more sensitive to tube potential variations.

Conclusions:

• The developed theoretical model provides accurate bias prediction for spectral CT systems.
• This framework facilitates the design of spectral CT systems that are more tolerant to bias.
• Improved bias prediction enhances material decomposition accuracy and quantitative potential in spectral CT.