ESTRO 2020 Abstract book

S192 ESTRO 2020

framework for patient volunteer data acquired on an MR- Linac at multiple days during their treatment. Material and Methods A lung cancer patient undergoing radical radiotherapy treatment on a conventional linac consented to being imaged on an MR-Linac on day 1, 6, and 7 of their treatment with a slice-selective fast gradient echo sequence interleaving surrogate and motion slices. Sagittal surrogate images from a fixed location including the tumour were used to generate surrogate signals, whereas motion images provided anatomical and motion information from varying locations in sagittal and axial orientations covering the anatomy of interest. All images were acquired with a resolution of 2x2x10mm³ and the motion images were offset by increments of 2mm in the out-of-plane direction to facilitate super-resolution reconstruction. Surrogate signals were generated by principal component analysis on the deformation vector field after registration to a reference surrogate image (Figure 1a). For each acquisition day 990 motion images were reconstructed to a 2mm isotropic MCSR image and a motion model was fitted. To quantify the model fitting accuracy, the model was applied to the MCSR and the image acquisition was simulated for each motion image time point. Each simulated motion image was then 2D registered to the original motion image to quantify the residual error between the model estimate and the original input data. Results A motion image and corresponding model estimated image are shown in Figure 1 (b) and (c) respectively. The volumetric MCSR image is shown in Figure 1 (d-f) and the quantitative residual fitting errors within the patient anatomy are given in Table 1. The mean absolute (95th percentile) residual fitting error ranged between 0.52mm and 0.80mm (1.42mm and 2.20mm).

Conclusion The first patient scanned with an adaptive 4DCBCT imaging protocol has been acquired showing large reductions in scan time (69%) and imaging dose (85%). OC-0338 High-resolution image reconstruction and motion modelling for a lung cancer patient on an MR- Linac B. Eiben 1 , E.H. Tran 1 , A. Wetscherek 2 , A. Shiarli 3 , J. Bertholet 2 , U. Oelfke 2 , J.R. McClelland 1 1 University College London, Medical Physics and Biomedical Engineering, London, United Kingdom ; 2 Institute of Cancer Research and Royal Marsden NHS Foundation Trust, Joint Department of Physics, London, United Kingdom ; 3 Institute of Cancer Research and Royal Marsden NHS Foundation Trust, Radiotherapy and Imaging Department, London, United Kingdom Purpose or Objective Real-time tumour monitoring and high-resolution volumetric motion information on the patient anatomy during radiotherapy delivery are both essential to utilise the full potential of an MR-Linac. Motion-including dose- reconstruction and downstream treatment adaptation or toxicity analysis is only possible with volumetric information. However, spatial and temporal resolution are competing parameters in volumetric MR acquisitions which in the thorax are further complicated by respiratory motion. We propose to use a unified image registration and motion modelling framework to overcome these inherent MR limitations. The framework takes surrogate signals and unsorted motion images as an input to reconstruct a motion-compensated super-resolution (MCSR) image and fit a motion model directly to the input data. Here, for the first time, we present the results of our motion modelling