Benefits of Precise Dect-Based Elemental Composition Estimation In Carbon Ion Therapy
Abstract
Purpose
Dual-energy CT (DECT) enables material differentiation by exploiting the energy-dependent attenuation characteristics of tissues, which is particularly beneficial for carbon ion therapy. This study systematically evaluated a recently proposed machine-learning-based DECT (ML-DECT) elemental decomposition approach in terms of dose calculation accuracy, robustness to image noise, estimation of dose uncertainty, and dose monitoring capability in carbon ion therapy.
Methods
The ICRP110 human phantom served as the ground truth. Voxel-wise DECT numbers were used as inputs to the ML-DECT method to derive elemental compositions. To assess robustness, Gaussian noise of up to 5% was added to the DECT numbers before decomposition. The associated uncertainties were quantified and propagated through the whole process. Monte Carlo simulations of carbon ion pencil beam irradiations were performed to compute physical and biological dose distributions, dose uncertainties, and annihilation photon signals. Results were compared with those from a parameterized DECT (PA-DECT) method and the ground truth.
Results
Compared with PA-DECT, the ML-DECT method achieved up to 4% and 10% higher gamma passing rates for physical and biological doses, respectively, under the 1 mm/1% criterion. Under noisy conditions, the improvements increased to 11% and over 20%. The relative dose uncertainty was reduced by 5% for physical and 8% for biological doses. Furthermore, the mean relative error of annihilation photon signals decreased by 4% with ML-DECT.
Conclusion
The ML-DECT approach demonstrated superior performance in carbon ion therapy applications, which yields more accurate and noise-resilient dose calculations, reduced dose uncertainties, and enhanced sensitivity of dose monitoring signals.