ESTRO 2024 - Abstract Book

S3913

Physics - Image acquisition and processing

ESTRO 2024

Figure 1 shows some sCT results from the second centre after three different types of training. Furthermore, as expected, the 2D method presents interslice artefacts especially for the bones structures.

Conclusion:

To conclude, the best image accuracy was for the supervised methods. The best MAE for the brain was obtained with the 2D network but the best general accuracy was obtained with the 3D supervised one. These brain results are similar to the literature. Nevertheless, 2D methods present interslice artifacts that could impact the dose and supervised deep learning approaches require a large amount of paired data. That is why, the unsupervised method which does not require multimodal registration step, also provides accurate sCTs, could be used in a clinical context. A dose evaluation will be performed to evaluate the accuracy of these sCTs for a RT clinical application.

Keywords: synthetic CT, brain, deep learning

References:

[1] Boulanger, M., Nunes, J. C., Chourak, H., Largent, A., Tahri, S., Acosta, O., ... & Barateau, A. (2021). Deep learning methods to generate synthetic CT from MRI in radiotherapy: A literature review. Physica Medica, 89, 265-281.

[2] Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., ... & Bengio, Y. (2020). Generative adversarial networks. Communications of the ACM, 63(11), 139-144.

[3] Johnson, J., Alahi, A., & Fei-Fei, L. (2016). Perceptual losses for real-time style transfer and super-resolution. In Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11-14, 2016, Proceedings, Part II 14 (pp. 694-711). Springer International Publishing.

[4] C. Hemon, V. Boussot, B. Texier, J.-L. Dillenseger & J-C. Nunes, « Guiding Unsupervised MRI-to-CT synthesis using Content and style Representation by an Enhanced Perceptual synthesis (CREPs) loss », MICCAI 2023, SynthRAD2023.