ESTRO 2020 Abstract book

S999 ESTRO 2020

locally varying effects. Predicting dosimetric impact prior to dose planning is important for assessment of automatically generated or deformed contours, especially in online adaptive RT. We propose a parametric model incorporating distance and angular relationship of OAR contour errors to the PTV boundary for dosimetric impact prediction, without prior knowledge of the planned RT dose. Material and Methods Clinical OAR and PTV contours were obtained retrospectively for 5 prostate patients, treated with 60 Gy in 20 VMAT fractions. A set (n~2000) of 3D Gaussian perturbations, distributed randomly around the volumetric OAR surface, were applied to bladder and rectum contours for each patient. Dosimetric impact was determined from the clinical planned dose distribution for each perturbation, via dose volume histogram (DVH) analysis. A non-linear model, parameterised in terms of perturbation magnitude ( m ), distance ( dPTV ), and angle ( φ ) from perturbation centre to the nearest point on the PTV surface was fit to these data for each clinical DVH statistic (Fig. 1). Patient specific and cohort-wide fits were obtained to measure patient-to-patient variance.

modalities, like PET or MR, may be useful for physicians to better discriminate soft tissues or to assess the local spread of a disease. The association of PET/CT-MR imaging, known as tri-modality technique, might be a powerful tool for radiotherapy to delineate precisely the volumes of interest. As the greatest precision is required in radiotherapy, keeping patient in the same position during the entire tri-modality workflow would be benefit to improve the multi-modal registration. The purpose of this study is to evaluate intrinsic positioning uncertainties associated with a PET/CT–MR solution. Material and Methods The PET/CT–MR solution is the association of a PET/CT (Discovery 710 GE) and a MR Imager (Optima MR450w GE) coupled to a transfer table, Zephyr XL DIACOR, allowing the patient to be transferred from one imager to another in the same position. The multi-modal registration is performed by means of the Integrated Registration software implemented on AW Server 3.2 (GE). To evaluate the registration software’s accuracy a digital phantom was used. Known rotations and translations were introduced and the corresponding modified images were registered with the non-transformed image. The uncertainties associated to the procedure for moving from PET/CT to MR were estimated using the phantom cartesian 3d (Alara Expertise – Spin Up !). Multi-modal images of the phantom were registered and the rigid transformation matrix was determined giving positioning uncertainties. The same workflow was used to determine positioning uncertainties on a retrospective cohort of 20 head and neck cancer patients positioned according to radiotherapy modalities with a thermoplastic mask and a knee wedge. The uncertainty for each rotation and translation was defined as the inter-quartile range obtained for all registrations. Results The relative uncertainty obtained from the registration algorithm was estimated to 5%. For the study on physical phantom, uncertainties for translations of 3.3 mm, 0.3 mm and 0.4 mm were obtained respectively for the left-right (L-R), antero-posterior (A-P) and superior-inferior (S-I) directions. For roll, pitch and yaw rotations, uncertainties were under 1°. Finally, for the clinical study, we obtained higher uncertainties, in each direction, corresponding to the images offsets before registration: 10.3 mm in A-P, 8.9 mm in S-I and 4.1 mm in L-R. In the case of rotations, the uncertainties obtained were under 1°. Conclusion This study shows that uncertainties of patient positioning are affected by a set of factors intervening during the whole workflow: registration software’s accuracy, patient's movements and isocenter difference between the two devices. Finally, despite the tri-modality solution, image registration remains essential before treatment planning. PO-1715 Predicting dosimetric impact of (auto)contouring errors for OARs in prostate VMAT radiotherapy D. Sandys 1 , M. Tyyger 1 , B. Al-Qaisieh 1 , N. Michael 1 1 Leeds Cancer Centre, Medical Physics & Engineering Department, Leeds, United Kingdom Purpose or Objective Contouring errors and uncertainty are well established as a major cause of dosimetric uncertainty in RT. For clinical OAR dose statistics there are no established relationships between geometric and dosimetric error. This is attributed to dose distribution inhomogeneity and relative error locations. Previous work attempted to demonstrate relationships between global contour similarity metrics (DICE, MDA, etc.) and dosimetric or treatment outcome measures using linear regression, with little success. Global metrics and linear regression are ill-suited to detect

Results Dosimetric impact was found to be linear in m , whilst showing a sigmoid-like fall off with dPTV and an angular dependence which could be modelled as a third order polynomial. Patient specific predictive models (Fig. 2a) were fit alongside a cohort-wide model. Residuals from patient-specific and cohort models were found to be indistinguishable with an equivalence bound equal to the individual patient data standard deviation (p<0.0001). This indicates that inter-patient variations in dosimetric impact were insignificant compared to intra-patient variation.

Conclusion