Artificial intelligence-based image segmentation can accelerate treatment-planning workflows, but geometric segmentation errors may propagate to downstream dosimetric calculations. This study presents a secondary computational analysis of published patient-level nnU-Net results for 16 Glioblastoma Multiforme cases evaluated in Boron Neutron Capture Therapy planning. Manual and neural-network-generated outputs were compared for tumour volume, irradiation time, and D98, D50, and D2 dose indicators. Mean absolute error, mean absolute percentage error, bias, Pearson correlation, and non-parametric comparisons were calculated. An exploratory retrospective audit additionally stratified cases using a Dice coefficient threshold of 0.80. The mean absolute percentage error was 73.71% for tumour volume, 1.88% for irradiation time, 7.21% for D98, 2.83% for D50, and 1.27% for D2. In cases with Dice coefficient >= 0.80, D98 percentage error was 1.64%, compared with 10.55% in lower-Dice cases. Higher segmentation overlap was associated with lower downstream computational error in the analysed cases. The Dice-informed stratification is proposed as a retrospective reliability-audit method rather than a prospective clinical decision rule, because Dice calculation requires a reference contour.
L. Xing, M. L. Giger, and J. K. Min, Artificial Intelligence in Medicine: Technical Basis and Clinical Applications. London, UK: Academic Press, 2020.
A. Zhang, L. Xing, J. Zou, and J. C. Wu, “Shifting machine learning for healthcare from development to deployment and from models to data,” Nature Biomedical Engineering, vol. 6, pp. 1330-1345, 2022.
B. Ibragimov and L. Xing, “Segmentation of organs-at-risk in head and neck CT images using convolutional neural networks,” Medical Physics, vol. 44, pp. 547-557, 2017.
H. Seo, C. Huang, M. Bassenne, R. Xiao, and L. Xing, “Modified U-Net (mU-Net) with incorporation of object-dependent high-level features for improved liver and liver-tumor segmentation in CT images,” IEEE Transactions on Medical Imaging, vol. 39, pp. 1316-1325, 2020.
C. Pezzi, F. Morosato, B. Marcaccio, et al., “Dosimetric comparison of the BNCT treatment planning performances when using a nnU-Net to automatically segment Glioblastoma Multiforme,” Health and Technology, vol. 15, pp. 811-822, 2025, [Online]. Available: https://doi.org/10.1007/s12553-025-00996-2.
T. J. Netherton, C. E. Cardenas, D. J. Rhee, L. E. Court, and B. M. Beadle, “The emergence of artificial intelligence within radiation oncology treatment planning,” Oncology, vol. 99, pp. 124-134, 2021.
M. Islam and L. Xing, “A data-driven dimensionality-reduction algorithm for the exploration of patterns in biomedical data,” Nature Biomedical Engineering, vol. 5, pp. 624-635, 2021.
R. Challen et al., “Artificial intelligence, bias and clinical safety,” BMJ Quality & Safety, vol. 28, pp. 231-237, 2019.
M. Roberts et al., “Common pitfalls and recommendations for using machine learning to detect and prognosticate for COVID-19 using chest radiographs and CT scans,” Nature Machine Intelligence, vol. 3, pp. 199-217, 2021.
R. Salmorbekova and M. Talipov, “Digital risk matrix and safety management workflow for airport infrastructure in developing countries: a data-driven prioritization approach,” Vibroengineering Procedia, vol. 62, pp. 653-661, Jun. 2026, [Online]. Available: https://doi.org/10.21595/vp.2026.26121.
