ESTRO 2022 - Abstract Book

S19

Abstract book

ESTRO 2022

model and the best LR model are shown in Fig. 1 for the 500 ms forecast, which is necessary for MLC-tracking based on 4 Hz cine MRIs.

Conclusion LSTM networks show great potential as respiratory motion predictors and could be used to compensate for the system latencies in SI direction for MR-guided radiotherapy with MLC-tracking. Independent testing with data from a different institution is planned.

OC-0044 Rapid distortion correction enables accurate real-time adaptive radiotherapy on an MRI-Linac

P. Liu 1,2 , S. Shan 1,2 , D. Waddington 1,2 , B. Dong 2 , G. Liney 3,4,5 , P. Keall 1,2

1 The University of Sydney, ACRF Image X Institute, Sydney, Australia; 2 Ingham Institute For Applied Medical Research, Ingham Institute For Applied Medical Research, Liverpool, Australia; 3 Liverpool and Macarthur Cancer Therapy Centre, Department of Medical Physics, Liverpool, Australia; 4 University of Wollongong, Centre for Medical Radiation Physics, Wollongong, Australia; 5 University of New South Wales, Faculty of Medicine, Sydney, Australia Purpose or Objective Real-time tumour tracking during radiotherapy relies on precise target localization to maintain target alignment with the radiation beam. MRI-Linacs are well-suited for this purpose, but MRI is affected by image distortion caused by magnetic field gradient non-linearities. In this work, a real-time distortion correction method was developed and integrated with an MLC tracking system. This method streams 2D cine-MRIs from the MRI-Linac during irradiation and mathematically corrects for distortion. The corrected image is then used to modulate the MLC leaves to compensate for target motion. Materials and Methods The gradient field of the Australian MRI-Linac, a 1T open bore prototype system, was characterized using spherical harmonics (SH) (Fig1A). A grid phantom with 3718 markers was imaged and the distorted marker positions were compared to known marker positions to calculate SH distortion correction coefficients. The SH correction algorithm was then modified to enable real-time implementation. Prior to tracking, a deformation vector field (DVF) was calculated from the SH coefficients and the tracking slice location (Fig1B). Subsequent images during irradiation have the same slice location and are corrected for distortion by applying this DVF (Fig1C). To verify this method, a 1D motion phantom underwent sinusoidal motion with an amplitude of 20 mm and a period of 9 s. The phantom was tested at three locations 11, 15 and 17 cm from the isocentre, with the magnitude of distortion increasing with distance. Cine-MRIs were acquired at 128 × 128 resolution at 3 Hz and corrected by applying the pre- calculated DVF. The target position was found using template matching and sent to the MLC tracking algorithm to obtain new leaf positions. The DVF method was compared to tracking using uncorrected images and images corrected with SH.