ESTRO 2024 - Abstract Book

S3045

Physics - Autosegmentation

ESTRO 2024

Table 1: Performance scores for all structures grouped for comparison with inter-observer variability. Performance non-inferior to inter-observer scores is marked in bold. *= created with model C*.

Model A

Model B

Model C/C*

Inter-observer

Median

DSC 0.75 0.77 0.79 0.34

HD95

DSC 0.76 0.78 0.80 0.38

HD95

DSC 0.89 0.88 0.84 0.72

HD95

DSC 0.87 0.90 0.83 0.72 0.40 0.16

HD95

Chambers

12 27 10 17

11 23 10 15

7 8 9

3.7 1.5 6.5 8.0 8.3

Aorta

Pulmonary artery

SVC + IVC

10

CA

n.a.

0

n.a.

0.30* 0.29*

37* 14*

Valves

0.07

20

0.09

18

16.5

Conclusion:

Cardiac auto-contouring models trained on oncological patients perform inconsistently on VT patients due to the presence of implanted devices. Developing a new auto-contouring model with combined oncological and VT patient data, led to substantial improvement model performance, especially for patients with severe device-induced artefacts. The full model auto-contouring performance for the four chambers, major vessels and valves is non-inferior compared to manual contouring.

Keywords: STAR, auto-contouring, cardiac substructures

References:

1. Balgobind, B.V. et al. Refining critical structure contouring in STereotactic Arrhythmia Radioablation (STAR): Benchmark results and consensus guidelines from the STOPSTORM.eu consortium, Radiotherapy and Oncology, Volume 189, 2023, ISSN 0167-8140, https://doi.org/10.1016/j.radonc.2023.109949. 2. Isensee, F., Jaeger, P.F., Kohl, S.A., Petersen, J., & Maier-Hein, K.H. (2021). nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation. Nature methods, 18(2), 203-211. https://doi.org/10.1038/s41592-020-01008-z

1307

Digital Poster

Development and Clinical Validation of Guideline-Based Autosegmentation for Pelvic Lymph Node Areas

Weir Chiang You, Chien Chih Chen, Jia Fu Lin

Taichung Veterans General Hospital, Department of Radiation Oncology, Taichung City, Taiwan

Purpose/Objective: