ESTRO 2021 Abstract Book

S406

ESTRO 2021

Conclusion The current SECT calibration technique introduces high RSP, dose and range errors for paediatric patients, caused by the stoichiometric fit being unable to discern the differences in elemental compositions and densities between adults and children. The influence of high RSP and dose prediction errors on the outcomes of paediatric cases in terms of toxicity is yet to be investigated. We recommend DECT as a personalised pre- treatment imaging modality. OC-0521 A deep learning approach to generate synthetic CT in low field MR-guided radiotherapy for lung cases D. Cusumano 1 , J. Lenkowicz 2 , C. Votta 2 , M. Nardini 2 , L. Boldrini 1 , L. Placidi 2 , F. Catucci 1 , N. Dinapoli 2 , M.V. Antonelli 2 , A. Romano 2 , V. De Luca 2 , G. Chiloiro 2 , L. Indovina 2 , V. Valentini 2 1 Fondazione Policlinico Universitario Agostino Gemelli IRCCS, Radioterapia, Rome, Italy; 2 Fondazione Policlinico Universitario Agostino Gemelli IRCCS, Radioterapia, Roma, Italy Purpose or Objective Artificial Intelligence (AI) is playing a relevant role in Magnetic Resonance guided Radiotherapy (MRgRT), offering solutions able to automatize different manual operations, making more efficient and safer the different clinical workflows. One of the main applications of AI in MRgRT is represented by the generation of synthetic computed tomography (sCT) images from MR acquisitions. The aim of this study is to propose a Deep Learning (DL) approach able to generate sCT from low field MR images in the thoracic district, which represents the most complicated site due to the large heterogeneity of the lung tissue, which makes the sCT generation and the dose calculation processes extremely challenging. Materials and Methods A total of 178 patients treated on pelvic (40), abdominal (40) and thoracic (58) districts were enrolled and divided in training (120) and test sets (18). All the patients were treated on a low field MRgRT system, acquiring a 0.35 T T2/T1 MRI and a CT image during treatment simulation. A conditional Generative Adversarial Network (cGAN) was used for sCT generation: the test set was composed only by thoracic cases. The image accuracy was evaluated calculating the mean absolute error (MAE) and the mean error (ME) in terms of Hounsfield Units (HU) between synthetic and original CT. Two IMRT treatment plans were calculated for each patient, both considering the simulation MRI as reference image: the first plan was calculated using the original CT for Electron Density (ED) map, the second one using the sCT. Dose calculation was performed considering the presence of magnetic field, using a Montecarlo algorithm and a dose grid size of 2 mm 3 . A comparison between the two IMRT plans was performed in terms of gamma analysis and differences in Dose Volume Histogram (DVH). As regards the DVH comparison, four parameters (V95%, D2%, D98% and D50%) were considered for PTV and three parameters (D2%, D98% and D50%) for the organs at risk, which were homolateral lung, lungs and ribs. Results Figure 1 reports an example of generated sCT image, together with the relative dose distribution: on the 18 patients analysed the MAE was 15.1±8.8 HU while the ME was -10.88±11.9 HU. Mean gamma passing rates for the three tolerance criteria analysed (1%/1mm,2%/2mm and 3%/3mm) were respectively 70.8±8.4%, 90.7±5.2% and 96.0±3.3%. Table 1 reports the difference observed in terms of DVH parameters between the treatment plan calculated on sCT and those calculated on the original CT.

Conclusion