ESTRO 2020 Abstract Book

S325 ESTRO 2020

Results Mean motion prediction error was below 2.1mm and absolute voxel-to-voxel dose differences between ideal and realistic tracking for all 5 cases were less than 10% or 5% in 93.2% or 80.1% of the VOI, respectively (Fig 2a - solid line: median, shade: range). Fig 2b shows DVHs for the two tracking scenarios compared to the static and the 4D non- mitigated results. Tracking substantially improves CTV coverage and dose homogeneity compared to non- mitigated delivery, and differences are small between ideal and realistic tracking, indicating the efficacy of the US-driven statistical motion model. However, tracking cannot completely restore dose homogeneity to that of the static case.

Hospital Zurich, Department of Radiation Oncology, Zurich, Switzerland

Purpose or Objective Pencil beam scanned proton therapy (PBS) naturally facilitates tumour tracking as long as the deformable 3D motion of the whole patient geometry is known in real- time – an impossible task with current online IGRT approaches. In this study, the feasibility of PBS tracking based on the reconstruction of tumour and lung motion using a statistical motion model and liver ultrasound (US) as a real-time motion surrogate has been investigated. Material and Methods Simultaneous free-breathing 4DMR and liver US images were acquired for five volunteers, resulting in 690-1056 variable 4DMR volumes per volunteer, with a temporal resolution of 0.4-0.6s and acquired over 7-11min. Deformation vector fields (DVF), extracted from each 4DMRI, were used to generate 5 synthetic 4DCT datasets from the same static lung patient CT (Fig 1a). Each dataset contained 99-159 full breathing cycles (Fig 1b) with the corresponding DVFs considered to represent the ground- truth motion for each 4DCT dataset. Using the simultaneously acquired liver US, a patient-specific motion model was created, based on principal component analysis and Gaussian process regression using the first 631-965 motion states per volunteer. Based on the corresponding US signal, this model was then used to predict DVFs of the lung for the last 35s of motion of each dataset not included in the model building (predicted motion). A 2-field PBS plan was optimised on the CTV of the reference CT (Fig 1a) and deformable 3D beam tracking simulated by adapting pencil beam positions laterally and in depth based on either the ground-truth (ideal) or predicted motions (realistic tracking). Resulting 4D dose distributions were compared in terms of absolute dose difference volume histograms (DDVH, VOI=CTV+20mm), and dose volume histograms (DVH) of the CTV.

Conclusion US-based motion model is a promising IGRT approach to guide 3D proton beam tracking in real-time if patient- specific models are first created based on the simultaneous pre-treatment acquisition of 4DMRI and liver US. However, to further mitigate residual patient/motion- specific effects, it will be necessary to combine tracking with other motion mitigation techniques (e.g. rescanning/gating or a combination) in order to fully restore dose homogeneity close to the static case. MK and AG contributed equally.

OC-0579 First clinical experience with proton therapy for HN cancer according to model-based selections

Abstract withdrawn from presentation

OC-0580 Bringing FLASH to the clinic: treatment planning considerations for ultrahigh dose-rate proton beams P. Van Marlen 1 , M. Dahele 1 , M. Folkerts 2 , E. Abel 2 , B. Slotman 1 , W. Verbakel 1 1 Amsterdam UMC, Radiation Oncology, Amsterdam, The Netherlands ; 2 Varian Medical Systems, Varian Medical Systems, Palo Alto, USA Purpose or Objective Pre-clinical research into ultrahigh dose-rate (e.g. ≥40Gy/s) “FLASH”-radiotherapy suggests a decrease in side-effects compared to conventional irradiation, while