ESTRO 2020 Abstract book

S97 ESTRO 2020

PD-0178 NTCP model development and comparison for brain image changes after IMPT for head and neck cancer. G.M. Engeseth 1,2 , R. He 1 , D. Mirkovic 3 , P. Yepes 3 , C.H. Stokkevag 2,4 , H.E. Pettersen 2 , K.A. Wahid 1 , A. Adair 3 , R. Wu 3 , X. Zhang 3 , A.A. Mohamed 1 , C.D. Fuller 1 , S. Frank 1 , R. Mohan 3 , G.B. Gunn 1 1 University of Texas MD Anderson Cancer Center, Department of Radiation Oncology, Houston, USA ; 2 Haukeland University Hospital, Department of Oncology and Medical Physics, Bergen, Norway ; 3 University of Texas MD Anderson Cancer Center, Department of Radiation Physics, Houston, USA ; 4 University of Bergen, Department of Physics and Technology, Bergen, Norway Purpose or Objective Recent research suggest that Linear Energy Transfer (LET (keV/µ)) may be associated with the development of radiation induced MR image changes (MR‐IC) of the brain tissue for patients treated with passive scattered proton therapy. While increasing proportion of head and neck cancer (HNC) patients are treated with IMPT, LET correlations with MR‐IC have not yet been investigated for this group. Therefore, the purpose of this study was to explore the potential influence of LET on the development of MR‐IC in HNC patients treated with IMPT. Material and Methods 13 HNC patients, all diagnosed with MR‐IC, were included in the analysis. Lesions contoured on post contrast T1‐ weighted MRIs were propagated to the patient planning CT through rigid registration. Treatment plans were re‐ calculated using an in‐house fast Monte Carlo system to obtain dose‐ and LET distributions for further voxel based analysis. Voxels inside and outside the lesions were defined as events and non‐events, respectively. Four logistic regression models were developed; one multivariate model with LET and dose as predictors (M1) and three univariate models with dose (M2), variable RBE weighted dose (M3) and LET (M4) as predictors. For three of the models (M1‐M3) information from all voxels within the lesions as well as voxels outside with dose >40 Gy(RBE) were used for modelling. The fourth model (M4), was developed on a data set in which the voxels inside and outside the lesions were matched by dose with a ratio of 1:2, thus creating an similar distribution of dose levels between the event and non‐event voxels. Model performance were assessed using the AUROC, calculated by leave one out cross validation (LOOCV), where 13 training and test data sets were created by successively leaving a single patient out as test data, while the remaining patient data were used for training the model. Results All model coefficients were statistically significant with a p‐value< 0.001. Both the multivariate model (M1), and the univariate model with LET as predictor (M4), showed increasing probability of MR‐IC with increasing LET values (Figure 1a and 1c). The calculated dose at 5% probability of MR‐IC based on M2 and M3, were 64.6 Gy(RBE) and 68.8 Gy(RBE) , respectively (Figure 1b). For the multivariable model, the linear relationship between the LET and the dose was described by a slope of ‐2.4 Gy / (keV/µ) (Figure 1d). From our LOOCV analysis we found that the AUC was approximately similar for both the multivariable model and the two models that utilized dose or variable RBE dose as predictors, all with AUC above 0.82. The performance of the model with LET as predictor was however lower, with a mean AUC of 0.671 (Table I).

Figure 1. Fusion of axial (A,C) and coronal (B,D) Ct‐slices from the same level in DIBH planning CT (pink) and : the first (A,B) and the last (C,D) CBCT (green), showing inter‐ fractional heart motion in one patient.

Figure 2. Box plot showing mean heart displacement (mm) of the fifteen patients from the chest wall on each axis (X;Y;Z). Conclusion Few studies examined the 3D movement of the heart during a course of DIBH‐RT using daily CBCT. During the course of DIBH, no significant inter‐fractional heart displacement has been observed. However, we found a high intra‐ and inter‐individual heterogeneity of the heart position. RT technologist should evaluate the heart position on the daily CBCT and give good feedback after each RT session in order to continuously optimize their DIBH.