Theoretical assessment of DECT noise on physical and biological dose accuracy in carbon-ion radiation therapy

Accurate characterization of tissue parameters is essential for precise dose calculation in Carbon Ion Radiation Therapy (CIRT). Recent advances in Dual-Energy CT (DECT) have improved the estimation of tissue parameters, yet DECT-based methods are susceptible to image noise. The influence on physical and biological dose accuracy has not been thoroughly investigated, undermining the evidence-based clinical application and protocol optimization.

To systematically examine how image noise in DECT influences the estimated tissue parameters and its consequent impact on the accuracy of physical and biological dose distributions in CIRT.

Four DECT-based elemental decomposition methods were evaluated. A machine-learning (ML) approach was compared with three parameterization (PA) methods, that is, the Hünemohr model using input parameters ( ${\rho }_e$ , $Z_{eff}$ ) obtained from the Saito-Hünemohr, Saito-Landry, and Bourque schemes, respectively. Clinically relevant noise levels of 0%, 2%, and 5% were chosen to assess the four methods for calculating the carbon-ion range deviations in 85 reference tissues and estimate the elemental composition of the ICRP110 human phantom. Physical and biological dose distributions in CIRT were calculated using Monte Carlo simulations. The biological doses were modeled using the Linear Quadratic Model (LQM), Microdosimetric Kinetic Model (MKM), and Local Effect Model (LEM). Gamma analysis was applied to evaluate the dose deviations.

Across the investigated noise levels, the ML approach consistently outperformed the three PA methods. Compared with PA-based methods, the ML approach reduced the average water-equivalent range deviations by 0.4 mm-1.4 mm and improved gamma passing rates by 0.8%-3.9% (physical dose) and 6.4%-24.1% (biological dose) under the 1 mm/1% criteria.

The ML method provides superior robustness and accuracy in physical and biological dose calculation based on DECT of various noise levels. The LEM demonstrated superior noise robustness compared to the LQM and the MKM.