Estimation of Proton Stopping Power from Optimized Combinations of Virtual Monoenergetic Images Generated By Dual-Energy CT
Abstract
Purpose
To develop and evaluate a data-driven strategy that optimally combines the full set of virtual monoenergetic images (VMIs) from dual-energy CT to construct an optimal image pair for estimating electron density and proton stopping power ratio (SPR), and to compare the results with quantitative maps provided by a commercial dual-layer DECT system.
Methods
Thirteen tissue-equivalent inserts are scanned using a clinical dual-layer DECT system (Spectral CT 7500, Philips Healthcare, Best, The Netherlands). Hounsfield unit (HU) values are extracted from 16 VMIs reconstructed at energies ranging from 50 to 200 keV. Principal component analysis (PCA) is applied to the VMIs’ HU dataset to derive two principal component images, which served as inputs to an eigentissue decomposition (ETD) method for estimating electron density and SPR. A leave-one-out cross-validation scheme is implemented in which one insert is excluded at a time from the calibration and PCA processes. PCA and calibration are performed using the remaining 12 inserts, and the resulting model is then applied to the excluded insert to estimate electron on density and SPR. This procedure is repeated for all 13 inserts. The estimated values are compared with scanner-derived electron density and SPR maps for 200 MeV protons.
Results
Across the 13 tissue-equivalent inserts, the PCA-based method reduced RMS errors in electron density estimation from 1.53% to 0.86% and in SPR estimation from 1.09% to 0.89% compared with scanner-derived values. Mean errors improved from 1.22% to −0.17% for electron density and from 0.74% to 0.07% for SPR.
Conclusion
The proposed PCA-based approach optimally combines information from all available VMIs and provides an optimal input image pair for improved estimation of electron density and proton stopping power ratio compared with commercial dual-layer scanner outputs, even under leave-one-out cross-validation. This method shows strong potential for proton therapy treatment planning workflows.