ESTRO 2021 Abstract Book

S263

ESTRO 2021

Conclusion While keeping the number of acquired kV images the same, the proposed Mixed Model Tracking, which allows explicit modelling of time dependence in correlation models, is clearly superior to currently used Linear Model Tracking, with error reductions up to 2.7mm, depending on the patient’s respiratory motion pattern. OC-0360 A surrogate-driven motion model for incorporating motion irregularity into 4D proton treatment Y. Zhang 1 , L. Huang 1 , G. Fattori 1 , A. Duetschler 1 , D. Weber 1 , A. Lomax 1 1 Paul Scherrer Institut, Center for Proton Therapy, Villigen-PSI, Switzerland Purpose or Objective Due to the recent availability of temporally resolved log-file information for proton therapy, the retrospective reconstruction of fraction specific 4D dose distributions has become a possibility (A Meijers et al Med Phys 2019 6(3):1140). For this, an estimation of the deformable patient motion during breathing typically relies on a single 4DCT acquired at the time of treatment planning. However, motion during delivery will likely be different in both frequency and magnitude. Here we aim to develop and evaluate a patient specific motion model which can estimate 4D volumetric images using logged surface motion during dose delivery, in order to improve the accuracy of 4D dose reconstruction under conditions of breathing variability. Materials and Methods A motion modelling-estimation framework based on principal component analysis (PCA) and partial data reconstruction has been developed. Surface marker motion and deformable motion fields (DVFs), both extracted from 4DCT, were used to generate a correlation model. Marker position was monitored during treatment delivery in real-time via an Optical-Tracking-System (OTS), from which time resolved DVFs could be reconstructed via this pre-built motion model. Validation was performed using 4D data from an abdomen patient under free breathing treatment (motion up to 20mm). Firstly, 8-fold, leave-one-phase-out cross- validation (LOOCV) was applied for each 4DCT phase, where the predicted 4DCT and its derived 4D plan were directly compared to the ground truth 4DCT scenario. As comparison, the best modelling scenario where all phases were considered by the model was deduced. Prediction accuracy for models using either external surface or internal diaphragm surrogates was analyzed. Secondly, OTS motion data of the planning and delivery day were used to generate time resolved 4DCTs in the presence of irregular motion patterns, with which 4D dose distributions were calculated. These results were compared to 4D doses calculated based on the single pre-treatment 4DCT. Results As the validation results shown in fig 1, comparable 4D dose distributions were obtained based on estimated and ground-truth 4DCT, independent on the use of internal or external motion surrogates, with V dosediff>20% <2% and V dosediff>10% <10% in the PTV. Conversely, by using 4DCTs warped by the predicted motion fields from OTS motion, rather than that of the planning 4DCT, both intra- and inter- fractional motion variability has been