A Bayesian approach to solve proton stopping powers from noisy multi-energy CT data

Area of Science:

• Medical Physics
• Radiological Imaging
• Computational Biology

Background:

• Accurate characterization of human tissues is crucial for proton therapy planning.
• Multi-energy computed tomography (MECT) offers potential for improved tissue differentiation.
• Noise in MECT data poses a significant challenge for accurate parameter estimation.

Purpose of the Study:

• To develop a novel formalism for characterizing human tissues from noisy MECT data.
• To evaluate the performance of the proposed method in estimating proton stopping powers (SPR).
• To assess the robustness of the method against varying noise levels and tissue compositions.

Main Methods:

• Adapted a principal component analysis-based formalism, eigentissue decomposition (ETD), using a Bayesian estimator, termed Bayesian ETD.
• Utilized maximum a posteriori fractions of eigentissues to determine physical parameters for proton beam dose calculation.
• Simulated dual-energy CT (DECT) and MECT data with varying noise levels and elemental compositions to evaluate SPR estimation accuracy against existing methods.

Main Results:

• Bayesian ETD demonstrated systematically lower root-mean-square (RMS) errors and negligible bias in SPR estimation compared to maximum-likelihood ETD and a state-of-the-art ρe-Z formalism, especially under noisy conditions.
• For medium noise levels, Bayesian ETD achieved an RMS error of 1.53% on SPR, significantly outperforming other methods (2.76-2.78%).
• Bayesian ETD with increased energy bins (up to five) further reduced SPR estimation errors and improved proton beam range prediction accuracy by up to 1.5 times compared to DECT-based methods.

Conclusions:

• Bayesian ETD is a robust and promising approach for extracting SPR from noisy MECT data, outperforming existing methods.
• The method's resilience to noise and improved accuracy are expected to enhance precision in proton therapy.
• With the advent of advanced CT scanners, Bayesian ETD is poised to extract more information and improve treatment outcomes beyond DECT capabilities.