ESTRO 2024 - Abstract Book

S3056

Physics - Autosegmentation

ESTRO 2024

Conclusion:

Delineation uncertainty varied in different directions for each structure for both manual observers and AI contours, highlighting the need for anisotropic margins. The uncertainties also varied between cases, suggesting using standardised population-based margins may not be appropriate. Significant variation in contouring occurred with manual contouring despite the use of training and a protocol. This may be due to uncertainty within the imaging and our inability to truly define ground-truth. Auto-contouring of pelvic and para-aortic lymph node regions had lower uncertainties than the manual contours, suggesting that automated nodal delineation was more consistent and requires smaller PTV margins. However, this was not the case for other structures. Furthermore, auto-contouring systems make different errors to manual contouring systems, as reflected by the automated systems having very large uncertainty for the superior border of the parametrium and anorectum. Delineation uncertainty calculations produce numerical values that enable comparison between auto-contour performance and human contours in terms of inter-observer variation. These are clinically useful assessment methods to inform the required CTV-PTV margins and should be carried out as part of auto-contouring system evaluation.

Keywords: Auto-segmentation, Delineation Uncertainty

References:

[1] Tudor G, Bernstein D, Riley S, Rimmer Y, Thomas S, van Herk M, et al. Geometric Uncertainties in Daily Online IGRT: Refining the CTV-PTV Margin for Contemporary Photon Radiotherapy. British Institute of Radiology. 2020. 10.1259/geo-unc-igrt.

[2] van Herk M. Errors and margins in radiotherapy. Semin Radiat Oncol. 2004 Jan;14(1):52-64. doi: 10.1053/j.semradonc.2003.10.003

[3] Isensee F, Jaeger PF, Kohl SAA, Petersen J, Maier-Hein KH. nnU-Net: a self-configuring method for deep learning based biomedical image segmentation. Nat Methods. 2021 Feb;18(2):203-211. doi: 10.1038/s41592-020-01008-z. Epub 2020 Dec 7.

1466

Digital Poster

The validation of Limbus AI based autocontouring for radical lung radiotherapy planning

Jacqueline Poxon 1 , Alison Starke 1 , John Conibear 2 , Dean Jones 1 , Niall MacDougall 1

1 Barts Health NHS Trust, Radiotherapy Physics, London, United Kingdom. 2 Barts Health NHS Trust, Clinical Oncology, London, United Kingdom