ESTRO 2020 Abstract book

S285 ESTRO 2020

Undersampled data with a mean SSIM of 0.48 ±0.05 and a PSNR value of 19.84 ±1.65dB could be reconstructed to an SSIM of 0.96±0.02 and a PSNR of 35.06 ±1.5dB, while the mean SSIM value of images with motion artefacts could be improved from 0.87 ±0.14 to 0.97 ±0.02. The classifier network reached a detection accuracy of 100% in all folds and on the test set.

image for Organs At Risk (OAR) delineation, synthetic CT (sCT) generation and on-treatment verification. The sCT was produced using MriPlanner (Spectronic Medical, Sweden) and the VMAT plan created using RayStation (RaySearch Laboratories, Sweden). For on-treatment verification the CBCT was matched to the large FOV MR, which was relabelled as a “CT” in the DICOM header so it could be used on the TrueBeam linear accelerator (Varian, Palo Alto, USA). The prostate and seminal vesicles were independently delineated using the back-up CT fused with the small FOV MR. The volume ratio between MR-only prostate and seminal vesicles volume and the MR-CT volume was calculated. The sCT dose calculation accuracy was evaluated by registered the CT to the sCT and recalculating the MR-only plan on the CT. The dose difference to the PTV D50 was calculated and the doses compared using a gamma analysis. The CBCT matching accuracy was assessed by two radiographers independently matching the first fraction CBCT to the backup CT and comparing to the clinical MR-CBCT match. Results The prostate and seminal vesicle volumes were smaller for the MR-only pathway than using the MR-CT fusion, with the volume ratio being V = 86 ± 2 % (mean ± sem). The sCT dose calculation was accurate, with a mean dose difference in the PTV D50 of ∆D=0.2±0.1% (range - 0.5%,1.0%) . The mean gamma pass rate within the body contour with gamma criteria 1%/1mm was (range 95.6%, 99.4%). See fig 1 for example sCT dose distribution and MR- CBCT match.

Conclusion The suggested multi-model AI framework shows great potential of reducing motion artefacts in MR images. Furthermore, undersampled images were reconstructed to nearly the same SSIM as the original images. With that, the framework could allow moderate patient movement during scans and enables rapid image acquisition, useful for e.g. accelerating MRI-only workflows in ERT. OC-0473 Evaluation of first 20 patients treated with MR-Only prostate radiotherapy with MR-CBCT matching J. Wyatt 1,2 , R.A. Pearson 1,2 , J. Frew 1 , C. Walker 1 , N. Richmond 1 , M. Wilkinson 1 , K. Wilkes 1 , S. Driver 1 , S. West 1 , K. Pilling 1 , R.L. Brooks 1 , D. Ainslie 1 , E. Wilkins 1 , H.M. McCallum 1,2 1 Newcastle upon Tyne Hospitals, Northern Centre for Cancer Care, Newcastle upon Tyne, United Kingdom ; 2 Newcastle University, Northern Institute for Cancer Research, Newcastle upon Tyne, United Kingdom Purpose or Objective MR-only radiotherapy removes the need for a planning CT, enabling the superior MR soft-tissue to be used for delineation without the MR-CT registration uncertainty. MR-CBCT soft-tissue matching removes the need for implanted fiducial markers, which is essential for extending MR-only treatments to other clinical sites but has not been previously reported. This paper presents an evaluation of the first clinical implementation of MR-only radiotherapy using MR-CBCT soft-tissue matching. Material and Methods 20 patients received planning MR and CT scans on a radiotherapy couch top. All patients were delineated, planned and treated using the MR alone, with the CT scan used for QA purposes only. Two MR sequences were acquired: a small Field Of View (FOV) MR image for target delineation and a large FOV MR

Fig 1 The differences between the clinical MR-CBCT match and the mean of the backup CT-CBCT matches were small (fig 2), with mean differences 0.1 ± 0.4 mm, 0.1 ± 0.4 mm, - 0.2 ± 0.2 mm for the vertical, longitudinal and lateral axes respectively.

Fig 2

Conclusion