ESTRO 2024 - Abstract Book

S3071

Physics - Autosegmentation

ESTRO 2024

Conclusion:

Despite the small training dataset used to train the nnUNet network, both quantitative as well as qualitative measurements showed superior clinical acceptability of auto-contours using the 3D_H nnUNet model compared to commercial pre-trained RSDL model. Further optimisation of the nnUNet model is warranted to improve the robustness and clinical acceptability of auto-segmentation contours in order to realise the usefulness of nnUNet DL contours as a independent contour audit QA tool or to gain clinical efficiency if deployed as clinical tool for contouring.

Keywords: Deep learning, Contour QA

References:

1. Van Mourik, Anke M., et al. "Multi-institutional study on target volume delineation variation in breast radiotherapy in the presence of guidelines." Radiotherapy and Oncology 94. 3 (2010): 286-291

2. Isensee, F., Jaeger, P.F., Kohl, S.A.A. et al."nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation." Nat Methods (2020).

3.

Chlap

,

P.,

et

al.

PyDicer

(Version

0.1.0)[Computer

software].

(2023).

https://github.com/AustralianCancerDataNetwork/pydicer

4. Abadi, Sobhi, et al. "Feasibility of automatic assessment of four-chamber cardiac function with MDCT: initial clinical application and validation." European journal of radiology 74.1 (2010): 175-181.

1861

Poster Discussion

Performance of a heart substructures autocontouring tool after clinical implementation

Tom Marchant 1,2 , Gareth Price 2 , Alan McWilliam 2 , Kathryn Banfill 3,2 , Marcel van Herk 2 , Corinne Faivre-Finn 2,1

1 The Christie NHS Foundation Trust, Christie Medical Physics & Engineering, Manchester, United Kingdom. 2 The University of Manchester, Division of Cancer Sciences, Manchester, United Kingdom. 3 The Christie NHS Foundation Trust, Clinical Oncology, Manchester, United Kingdom

Purpose/Objective:

Several studies have identified an area at the base of the heart where excess dose is significantly correlated with overall survival [1], [2]. This area has recently been implemented as a new organ at risk (OAR) for lung radiotherapy planning at our centre, aiming to limit its maximum dose to 19.5Gy in 20 fractions. An autocontouring tool was developed to assist clinicians in segmenting the base of heart region (details were reported previously [3]). Ongoing QA of AI autocontouring tools is important to validate acceptable performance after initial commissioning and clinical