Encrypted login | home

Program Information

A Bayesian Method to Derive Proton Stopping Powers of Human Tissues From Multi-Energy CT Data

no image available
A Lalonde

A Lalonde1*, E Baer2,3 , H Bouchard1,4 , (1) Universite de Montreal, Montreal, Canada, (2) UCL, London, UK, (3) National Physical Laboratory, Teddington, UK, (4) CRCHUM, Montreal, Canada


WE-G-605-3 (Wednesday, August 2, 2017) 4:30 PM - 6:00 PM Room: 605

Purpose: To develop a new approach to characterize human tissues from noisy multi-energy CT (MECT) data and evaluate its performance in estimating proton stopping powers (SPR).

Methods: A general method to decompose MECT data referred to as eigentissue decomposition (ETD) is adapted for noise. The new method, called Bayesian ETD, uses a prior function to extract the maximum a posteriori composition in each voxel. Simulated noisy dual-energy CT (DECT) data is used to compare the accuracy of SPR estimations made by the proposed method and two similar formalisms: maximum-likelihood ETD and a parametric approach (ρe–Z). Statistical and systematic deviations from tabulated data are introduced to mimic patient-to-patient variations. Patient DECT images with metal artifacts are used to evaluate the sensitivity of each method toward systematic errors on HU. The impact of using MECT over DECT is investigated by simulating CT data using two to five energy bins for equivalent noise levels. Potential reduction of range uncertainty using MECT over DECT is investigated using a probabilistic model.

Results: Using simulated DECT data for reference tissues, Bayesian ETD systematically gives the lowest root mean square (RMS) and mean errors. For a medium noise level, the RMS errors on SPR are 2.78%, 2.76% and 1.45% for ρe–Z, ETD and Bayesian ETD respectively. Values of 2.79%, 2.78% and 1.88% are obtained with randomly altered tissues composition. On patient images, Bayesian ETD noticeably reduces the effect of metal artifacts on SPR prediction maps. Finally, the proposed method is shown to reduce beam range uncertainties using more than two spectra. Indeed, Bayesian ETD with four energy bins reduces range uncertainty by a factor of up to 1.5 compared to ρe–Z.

Conclusion: The proposed method increases the robustness of ETD towards noise and artifacts in addition of yielding promising accuracy of SPR derived from MECT data.

Contact Email: