ESTRO 2021 Abstract Book

S1410

ESTRO 2021

Results The non-rigid registration process reached an accurate inter-individual registration. Mean Dice scores were 0,93 ± 0,01 for bones, 0,91 ± 0,02 for prostate, 0,96 ± 0,01 for bladder, 0,91 ± 0,02 for rectum and 0,98 ± 0,01 for body. Considering HU differences and statistical scores, it was shown how and where deep learning appeared widely superior to the bulk-density method. As shown in figure 2, the main source of HU errors was found in the cortical bones for the deep learning method and also in the soft tissue for the bulk-density. Dose errors appeared mainly at the external body contour interface for the GAN, and soft tissues errors remained for the bulk-density. Conclusion The proposed voxel-based workflow produced 3D maps showing not only the heterogeneity of errors for pCT generation from MRI, but also compared in a single coordinate system the planning dose distribution differences across the methods. Hence, it provides a way forward to quality control within the MRI-based planning radiotherapy. PO-1684 Correlations of tumour permeability parameters from DCE-MRI with ADC in nasopharyngeal carcinoma W.L. Mui 1 , A.W.M. Lee 2 , V.H.F. Lee 2 , W.T. Ng 2 , V. Vardhanabhuti 3 , S.Y. Man 1 , D.T.T. Chua 4 , X.Y. Guan 2 1 Hong Kong Sanatorium and Hospital, Department of Radiotherapy, Hong Kong, Hong Kong (SAR) China; 2 The University of Hong Kong, Department of Clinical Oncology, Hong Kong, Hong Kong (SAR) China; 3 The University of Hong Kong, Department of Diagnostic Radiology, Hong Kong, Hong Kong (SAR) China; 4 Hong Kong Sanatorium and Hospital, Department of Medicine, Hong Kong, Hong Kong (SAR) China

Purpose or Objective