Dose-volume-based evaluation of convolutional neural network-based auto-segmentation of thoracic organs at risk

Noémie Johnston; Jeffrey De Rycke; Yolande Lievens; Marc van Eijkeren; Jan Aelterman; Eva Vandersmissen; Stephan Ponte; Barbara Vanderstraeten

doi:10.1016/j.phro.2022.07.004

Dose-volume-based evaluation of convolutional neural network-based auto-segmentation of thoracic organs at risk

Phys Imaging Radiat Oncol. 2022 Jul 25:23:109-117. doi: 10.1016/j.phro.2022.07.004. eCollection 2022 Jul.

Authors

Noémie Johnston¹, Jeffrey De Rycke², Yolande Lievens^{2

3}, Marc van Eijkeren^{2

3}, Jan Aelterman^{4

5}, Eva Vandersmissen⁶, Stephan Ponte¹, Barbara Vanderstraeten^{2

3}

Affiliations

¹ Centre Hospitalier Universitaire de Liège, Service de Radiothérapie, Liège, Belgium.
² Ghent University, Faculty of Medicine and Health Sciences, Department of Human Structure and Repair, Gent, Belgium.
³ Ghent University Hospital, Department of Radiotherapy-Oncology, Gent, Belgium.
⁴ Ghent University, Department of Physics and Astronomy, Ghent University Centre for X-ray Tomography, Gent, Belgium.
⁵ Ghent University, Department TELIN / IMEC, Image Processing Interpretation Group, Gent, Belgium.
⁶ Agfa NV, Radiology Solutions R&D, Mortsel, Belgium.

Abstract

Background and purpose: The geometrical accuracy of auto-segmentation using convolutional neural networks (CNNs) has been demonstrated. This study aimed to investigate the dose-volume impact of differences between automatic and manual OARs for locally advanced (LA) and peripherally located early-stage (ES) non-small cell lung cancer (NSCLC).

Material and methods: A single CNN was created for automatic delineation of the heart, lungs, main left and right bronchus, esophagus, spinal cord and trachea using 55/10/40 patients for training/validation/testing. Dice score coefficient (DSC) and 95th percentile Hausdorff distance (HD95) were used for geometrical analysis. A new treatment plan based on the auto-segmented OARs was created for each test patient using 3D for ES-NSCLC (SBRT, 3-8 fractions) and IMRT for LA-NSCLC (24-35 fractions). The correlation between geometrical metrics and dose-volume differences was investigated.

Results: The average (±1 SD) DSC and HD95 were 0.82 ± 0.07 and 16.2 ± 22.4 mm, while the average dose-volume differences were 0.5 ± 1.5 Gy (ES) and 1.5 ± 2.8 Gy (LA). The geometrical metrics did not correlate with the observed dose-volume differences (average Pearson for DSC: -0.27 ± 0.18 (ES) and -0.09 ± 0.12 (LA); HD95: 0.1 ± 0.3 mm (ES) and 0.2 ± 0.2 mm (LA)).

Conclusions: After post-processing, manual adjustments of automatic contours are only needed for clinically relevant OARs situated close to the tumor or within an entry or exit beam e.g., the heart and the esophagus for LA-NSCLC and the bronchi for ES-NSCLC. The lungs do not need to be checked further in detail.

Keywords: Dice; Dose; Lung cancer; Radiotherapy; Treatment planning; Volume.