ESTRO 37 Abstract book

ESTRO 37

S536

PO-0971 Monte Carlo validation of a new dual-energy CT method for proton therapy in a patient-like geometry A. Lalonde 1 , C. Remy 2 , E. Baer 3 , H. Bouchard 2 1 Centre Hospitalier de l'Université de Montréal, Radio- Oncologie, Montréal, Canada 2 Universite de Montreal, Physique, Montreal, Canada 3 University College London, Medical Physics and Biomedical Engineering, London, United Kingdom Purpose or Objective To evaluate the performance of a novel method to extract tissue parameters for Monte Carlo (MC) dose calculation using dual-energy CT (DECT) in the context of proton therapy treatment planning. Material and Methods A novel and highly realistic ground truth numerical phantom is created using a real patient’s pelvis scan. Contours made by an expert are used to define tissues and assign known chemical compositions of human reference tissues. The density of each voxel is overwritten following the original distribution of CT numbers. From this numerical phantom, simulated images are created for single-energy CT (SECT) and DECT with different levels of random noise. The conversion from simulated CT data to MC inputs is done using the method of Schneider et al. (2000) for SECT, and a recently published method called eigentissue decomposition for DECT. Dose distributions for single proton beams are calculated using the MC code TOPAS with SECT and DECT determined inputs, and compared to ground truth dose distributions in order to quantify the error on range associated for both approaches. Results For noiseless images, the DECT-based tissue segmentation outperforms the SECT approach in the context of MC treatment planning with a range error of almost 0 mm, compared to 1.5 mm for the SECT approach (figure 1), as previously predicted by several studies. In the presence of a realistic level of random noise, errors on the range of 0.5 mm and 1.8 mm are obtained for DECT and SECT respectively (figure 2), indicating that DECT used with the eigentissue decomposition predicts proton ranges more accurately than SECT for MC dose calculation, even in the presence of noise. In terms of dose distribution, the root-mean squared error on the absorbed dose on the whole patient volume for noiseless images is 0.17% and 0.05% for SECT and DECT respectively. With a realistic level of noise, these values grow to 1.22% and 0.8%, validating the improved reliability provided by DECT in the context of MC dose calculation.

Figure 2. a) Dose distribution calculated with TOPAS on the ground truth numerical phantom. Relative error on absorbed dose using b) SECT tissue segmentation and c) DECT with realistic noise level. Conclusion A recently published method for tissue characterization with DECT is successfully implemented in TOPAS. Dose distributions calculated with MC for single proton beams in a highly realistic phantom shows a noticeably improved proton range prediction using DECT over SECT for simulated images affected by noise.

Poster: Physics track: (Quantitative) functional and biological imaging

PO-0972 Geometric image biomarker changes at the third treatment week predict late xerostomia T.A. Vedelaar 1 , L.V. Van Dijk 1 , T. Zhai 1 , C.L. Brouwer 1 , J.A. Langendijk 1 , R.J.H.M. Steenbakkers 1 , N.M. Sijtsema 1 1 University Medical Center Groningen, Department of Radiation oncology, Groningen, The Netherlands Purpose or Objective Radiation-induced xerostomia is a common toxicity of the treatment which has a major impact on the quality of life of head and neck cancer (HNC) patients. Parotid gland response to radiation dose during and after treatment varies between different patients. Mid-treatment delta image biomarkers (ΔIBMs) of the parotid gland that predict the risk on xerostomia could contribute to the selection of patients in order to adapt treatment and potentially reduce the probability of developing late xerostomia. The aim of this study is to predict xerostomia 12 months after treatment (XER12m) with ΔIBMs at the beginning of the 3rd treatment week in combination with dosimetric, clinical and treatment parameters. Material and Methods Geometric IBMs (20) were calculated of 57 patients on the CT images acquired before treatment (planning CT) and on a repeat CT acquired at the beginning of 3rd treatment week. The difference between both IBMs resulted in geometric ΔIBMs. Additionally, xerostomia scores before start (XERbaseline) and at the beginning of the 3rd treatment week (XERw3), and mean dose to the contralateral parotid gland were considered as possible predictors of XER12m. To select the best predictors, a stepwise forward selection based on the likelihood test (p<0.05) was performed. This process has been bootstrapped 1000 times. The final logistic multivariable models were compared with a reference model based on parotid gland dose and baseline xerostomia scores. All models were corrected for baseline xerostomia scores and internally validated. Results Twenty-four (42%) patients developed moderate-to- severe XER12m. The contralateral parotid glands surface change (ΔSurface) was the most frequently selected ΔIBM. ΔSurface was significantly associated to XER12m (p<0.001) and added significantly to an univariable model with XERbaseline (likelihood-ratio test; p =0.001). The performance of the ΔIBM model with ΔSurface and XERbaseline (AUC=0.87) was high and improved compared to the reference model (AUC=0.84) (Model 1 Table). Furthermore, the addition of XERw3 to this ΔIBM model (XERbaseline and ΔSurface) was significant (likelihood ratio test p=0.005). The addition of XERw3 further

Figure 1. a) Dose distribution calculated with TOPAS on the ground truth numerical phantom. Relative error on absorbed dose using b) SECT tissue segmentation and c) DECT without noise.