ESTRO 2024 - Abstract Book

S3083

Physics - Autosegmentation

ESTRO 2024

Conclusion:

Our findings demonstrate that DL autocontouring models trained using heterogeneous data with regards to multiple sources of domain shift can generalize well to an unseen external validation dataset when evaluated using DSCs. Furthermore, the model can even be robust to sources of domain shift that were not present in the training data. For example, the external validation data were acquired with different image acquisition and annotation protocols to any of the training data. This is an important finding because clinical datasets can feature multiple sources of domain shift and it will not always be feasible to include all of them in the training data of the model. Therefore, the model can be used in cases where center-specific labelled data are either unavailable or of limited availability. Our results did not demonstrate a race bias. However, one limitation of this study is that the public dataset’s race information was not provided, so the racial heterogeneity of the training data was not known. Therefore, further investigation into possible demographic bias using more controlled experiments with more validation data is recommended.

Keywords: Deep Learning, Prostate, MRI

References:

[1] F. Isensee, P. F. Jaeger, S. A. A. Kohl, J. Petersen, and K. H. Maier-Hein, “nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation,” Nat Methods, vol. 18, no. 2, pp. 203–211, 2021, doi: 10.1038/s41592-020-01008-z.

[2] G. Litjens et al., “Evaluation of prostate segmentation algorithms for MRI: The PROMISE12 challenge,” Med Image Anal, vol. 18, no. 2, pp. 359–373, Feb. 2014, doi: 10.1016/j.media.2013.12.002.

[3] G. Lemaître, R. Martí, J. Freixenet, J. C. Vilanova, P. M. Walker, and F. Meriaudeau, “Computer-Aided Detection and diagnosis for prostate cancer based on mono and multi-parametric MRI: A review,” Comput Biol Med, vol. 60, pp. 8–31, May 2015, doi: 10.1016/J.COMPBIOMED.2015.02.009.