Radiation Physics and Engineering
A peer-reviewed journal published by K. N. Toosi University of Technology

‎Improving head and neck organs at risk segmentation in CT using residual U-Net with slice-based preprocessing‎

Document Type : Research Article

Authors

Khashayar Heshmati Jannat Magham 1
Laleh Rafat-Motavalli 1
Hashem Miri-Hakimabad 1
Mahdieh Dayyani 2, 3

1 Department of Physics‎, ‎Faculty of Science‎, ‎Ferdowsi University of Mashhad‎, ‎Mashhad‎, ‎Iran

2 Research and Education Department‎, ‎Reza Radiotherapy and Oncology Center‎, ‎Mashhad‎, ‎Iran

3 Radiation Oncology Department‎, ‎Reza Radiotherapy and Oncology Center‎, ‎Mashhad‎, ‎Iran

https://doi.org/10.22034/rpe.2026.559226.1321
Abstract
Accurate segmentation of organs at risk (OARs) in head and neck CT scans is essential for effective radiotherapy planning. Although U-Net based deep learning models have achieved strong performance, small or anatomically complex OARs remain challenging to segment accurately. This study investigates how image slicing (cropping) influences segmentation accuracy and efficiency when using a Residual U-Net for head and neck OAR segmentation. A total of 63 CT scans from a public dataset and an institutional dataset were used. Slice-based preprocessing was applied by cropping regions surrounding target masks. Residual U-Net models were trained and tested using both full-size and sliced (cropped) images for 41 OARs. Segmentation accuracy was evaluated using Dice Similarity Coefficient (DSC) and Intersection over Union (IoU). Additional experiments incorporating dropout layers and longer training epochs were performed to improve optic chiasm and optic nerves segmentation. Slice-based networks achieved a 4.1% increase in IoU and a 3.2% increase in Dice score compared with full-size networks. For 11 complex structures, IoU and Dice scores improved by 10.9% and 9.0%, respectively. The standard deviation of both metrics decreased, indicating more consistent performance. Slice-based networks also demonstrated 130% faster training and about 8× faster prediction times. Adding dropout layers and extending training epochs further improved segmentation of the optic chiasm and optic nerves. Slice-based preprocessing combined with a Residual U-Net architecture improves segmentation accuracy and computational efficiency in head and neck CT imaging. This approach shows strong potential for practical use in radiotherapy planning.

Highlights

  • Slice-based preprocessing improves Dice (3.2%) and IoU (4.1%) scores for OAR segmentation.
  • Small and complex organs show the most significant performance gains with this method (IoU: 10.9%).
  • The approach reduces performance variability and ensures more consistent results.
  • Training is over 2 times faster and inference speed increases by 8 times.

Keywords

Auto-Segmentation
Residual U-Net
Head and Neck CT
Organs at Risk

Copyright
RPE is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0).

Conflict of Interest
The authors declare no potential conflict of interest regarding the publication of this work‎.

Funding
‎The authors declare that no funds‎, ‎grants‎, ‎or other financial support were received during the preparation of this manuscript‎.

Adams, R. and Bischof, L. (1994). Seeded region growing. IEEE Transactions on Pattern Analysis and Machine Intelligence, 16(6):641–647.
Breiman, L. (2001). Random forests. Machine learning, 45(1):5–32.
Brouwer, C. L., Steenbakkers, R. J., Gort, E., et al. (2014). Differences in delineation guidelines for head and neck cancer result in inconsistent reported dose and corresponding NTCP. Radiotherapy and Oncology, 111(1):148–152.
Brudfors, M., Balbastre, Y., Ashburner, J., et al. (2022). Fitting segmentation networks on varying image resolutions using splatting. In Annual Conference on Medical Image Understanding and Analysis, pages 271–282. Springer.
Chan, J. W., Kearney, V., Haaf, S., et al. (2019). A convolutional neural network algorithm for automatic segmentation of head and neck organs at risk using deep lifelong learning. Medical Physics, 46(5):2204–2213.
C ¸i¸ cek,Ö., Abdulkadir, A., Lienkamp, S. S., et al. (2016). 3D U-Net: learning dense volumetric segmentation from sparse annotation. In International Conference on Medical Image Computing and Computer-Assisted Intervention, pages 424–432. Springer.
Cortes, C. and Vapnik, V. (1995). Support-vector networks. Machine Learning, 20(3):273–297.
Gibbons, E., Hoffmann, M., Westhuyzen, J., et al. (2023). Clinical evaluation of deep learning and atlas-based auto-segmentation for critical organs at risk in radiation therapy. Journal of Medical Radiation Sciences, 70:15–25.
Han, X., Hoogeman, M. S., Levendag, P. C., et al. (2008). Atlas-based auto-segmentation of head and neck CT images. In International Conference on Medical Image Computing and Computer-assisted Intervention, pages 434–441. Springer.
Haque, I. R. I. and Neubert, J. (2020). Deep learning approaches to biomedical image segmentation. Informatics in Medicine Unlocked, 18:100297.
Harari, P. M., Song, S., and Tomé, W. A. (2010). Emphasizing conformal avoidance versus target definition for IMRT planning in head-and-neck cancer. International Journal of Radiation Oncology* Biology* Physics, 77(3):950–958.
He, K., Zhang, X., Ren, S., et al. (2016). Deep residual learning for image recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 770–778.
Howard, A. G., Zhu, M., Chen, B., et al. (2017). Mobilenets: Efficient convolutional neural networks for mobile vision applications. arXiv preprint arXiv:1704.04861.
Hu, J., Shen, L., and Sun, G. (2018). Squeeze-and-excitation networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7132–7141.
Huang, G., Liu, Z., Van Der Maaten, L., et al. (2017). Densely connected convolutional networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 4700–4708.
Iakubovskii, P. (2019). Segmentation models py-torch. GitHub. GitHub repository https://github.com/qubvel/segmentation models. pytorch.
Ibragimov, B. and Xing, L. (2017). Segmentation of organs-at-risks in head and neck CT images using convolutional neural networks. Medical physics, 44(2):547–557.
Kamnitsas, K., Ledig, C., Newcombe, V. F., et al. (2017). Efficient multi-scale 3D CNN with fully connected CRF for accurate brain lesion segmentation. Medical Image Analysis, 36:61–78.
Kass, M., Witkin, A., and Terzopoulos, D. (1988). Snakes: Active contour models. International Journal of Computer Vision, 1(4):321–331.
Liang, S., Tang, F., Huang, X., et al. (2019). Deep-learning-based detection and segmentation of organs at risk in nasopharyngeal carcinoma computed tomographic images for radiotherapy planning. European Radiology, 29(4):1961–1967.
Mikhailov, I., Chauveau, B., Bourdel, N., et al. (2024). A deep learning-based interactive medical image segmentation framework with sequential memory. Computer Methods and Programs in Biomedicine, 245:108038.
Otsu, N. et al. (1979). A threshold selection method from gray-level histograms. Automatica, 11(285-296).
Podobnik, G., Ibragimov, B., Peterlin, P., et al. (2024). OARi ability: Interobserver and intermodality variability analysis in OAR contouring from head and neck CT and MR images. Medical Physics, 51(3):2175–2186.
Podobnik, G., Strojan, P., Peterlin, P., et al. (2023). HaN-Seg: The head and neck organ-at-risk CT and MR segmentation dataset. Medical Physics, 50(3):1917–1927.
Putz, F., Beirami, S., Schmidt, M. A., et al. (2025). The Segment Anything foundation model achieves favorable brain tumor auto-segmentation accuracy in MRI to support radiotherapy treatment planning. Strahlentherapie und Onkologie, 201(3):255–265.
Rahman, M. A. and Wang, Y. (2016). Optimizing intersection-over-union in deep neural networks for image segmentation. In International Symposium on visual Computing, pages 234–244. Springer.
Ronneberger, O., Fischer, P., and Brox, T. (2015). U-net: Convolutional networks for biomedical image segmentation. In International Conference on Medical image computing and computer-assisted intervention, pages 234–241. Springer.
Sharp, G., Fritscher, K. D., Pekar, V., et al. (2014). Vision 20/20: perspectives on automated image segmentation for radiotherapy. Medical Physics, 41(5):050902.
Simonyan, K. and Zisserman, A. (2014). Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556.
Srivastava, N., Hinton, G., Krizhevsky, A., et al. (2014). Dropout: a simple way to prevent neural networks from overfitting. The Journal of Machine Learning Research, 15(1):1929–1958.
Sudre, C. H., Li, W., Vercauteren, T., et al. (2017). Generalised dice overlap as a deep learning loss function for highly unbalanced segmentations. In International Workshop on Deep Learning in Medical Image Analysis, pages 240–248. Springer.
Van Rooij, W., Dahele, M., Brandao, H. R., et al. (2019). Deep learning-based delineation of head and neck organs at risk: geometric and dosimetric evaluation. International Journal of Radiation Oncology* Biology* Physics, 104(3):677–684.
Wilcoxon, F. (1945). Individual comparisons by ranking methods. Biometrics bulletin, 1(6):80–83.
Xie, S., Girshick, R., Dollár, P., et al. (2017). Aggregated residual transformations for deep neural networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1492–1500.
Radiation Physics and Engineering
Volume 7, Issue 2
Spring 2026
Pages 77-90

  • Receive Date 12 November 2025
  • Revise Date 18 February 2026
  • Accept Date 02 May 2026

Home

Submit Manuscript

Reviewers

Contact Us