ESTRO 2024 - Abstract Book

S3758

Physics - Image acquisition and processing

ESTRO 2024

10.1002/mp.15840. https://onlinelibrary.wiley.com/doi/abs/10.1002/mp.15840. [2] Jun-Yan Zhu et al. “Unpaired Image-to-Image Translation Using Cycle-Consistent Adversarial Networks”. In: 2017 IEEE International Conference on Computer Vision (ICCV). 2017 IEEE International Conference on Computer Vision (ICCV). Venice: IEEE, Oct. 2017, pp. 2242–2251. isbn: 978-1-5386-1032-9. doi: 10.1109/ICCV.2017.244. url: http://ieeexplore.ieee.org/document/8237506/. url:

394

Digital Poster

Dosimetric evaluation of synthetic 3DCTs reconstructed from a single x-ray radiograph using CNN

Estelle Loÿen, Damien Dasnoy, Benoît Macq

UCLouvain, ICTEAM, Louvain-la-Neuve, Belgium

Purpose/Objective:

Every proton therapy treatment room is equipped with x-ray imaging devices to align the patient accurately in front of the treatment beam. However, the classical fluoroscopy deployed in the treatment room is not well adapted for proton dose calculation. This can be overcome by reconstructing a 3D computed tomography (3DCT) using convolutional neural network (CNN). In this study, the suitability of these images for proton therapy of lung cancer patients was evaluated in terms of dosimetric accuracy.

Material/Methods:

Five patients treated for a lung cancer in Cliniques Universitaires Saint-Luc (Brussels) are studied. For each patient, a data augmentation tool is used to create 1500 synthetic 3DCTs (sCTs) from the planning 4DCT [1]. 1500 digitally reconstructed radiographs (DRRs) are generated from each sCT and given as input to a convolutional neural network that learns to reconstruct the associated 3DCTs. A test set composed of 50 ground-truth (GT) 3DCTs and their corresponding DRRs is used to evaluate the accuracy of the predicted (P) 3DCTs. In proton therapy treatment planning, the stopping power ratio (SPR) is used for calculating the water equivalent path length (WEPL), which is successively used for the computation of the required proton energy in the beam axis. In this study, SPR values are obtained using the CT calibration implemented in MCSquare [2]. WEPLs are computed along 3 beams whose angles are specific to the position of the tumor. SPR and WEPL difference maps are obtained by using the equations below:

where x̄ represents the mean of x and the subscripts GT and P denote the estimations based on the ground-truth and predicted 3DCTs, respectively.