ESTRO 2024 - Abstract Book

S4548

Physics - Machine learning models and clinical applications

ESTRO 2024

Conclusion:

The proposed probabilistic methods allow for the prediction of an uncertainty that is correlated with the real error. Furthermore, the use of these uncertainty maps allows assessing the quality of sCT. The increase in uncertainty is higher when the CBCT data comes from a different center to the one used for training. Incorporating this uncertainty into the clinical RT workflow could hold promise, not only for segmentation tasks, but also for estimating the delivered dose in the context of ART. Evaluating the influence of these HU uncertainties on dose uncertainty would be of particular interest.

Keywords: Uncertainty, sCT, Deep-learning

References:

[1] M. Mirza et S. Osindero, « Conditional Generative Adversarial Nets ». arXiv, http://arxiv.org/abs/1411.1784

[2] M. W.-K. Law et al., « A study of Bayesian deep network uncertainty and its application to synthetic CT generation for MR-only radiotherapy treatment planning », Medical Physics, 2023 Sep 4, doi: 10.1002/mp.16666.

[3] M. Hemsley et al., « Deep Generative Model for Synthetic-CT Generation with Uncertainty Predictions », in Medical Image Computing and Computer Assisted Intervention – MICCAI 2020, p. 834-844. doi: 10.1007/978-3-030-59710 8_81.

[4] A. Thummerer et al., « SynthRAD2023 Grand Challenge dataset: Generating synthetic CT for radiotherapy », Medical Physics, vol. 50, no 7, p. 4664-4674, 2023, doi: 10.1002/mp.16529.