近年放射線治療技術從強度調控(IMRT)進化到弧形治療(VMAT)，不變的是在設計治療計畫前，專科醫師必須先擬定治療區域和正常器官的劑量限制，此步驟是治療計劃系統規劃治療參數的前提。若劑量規劃不合理或評估差異過大，治療計劃系統必須花較多的時間收斂目標以及最佳化。同時間3D U-net的出現、對醫學影像預測也比過去機器學習的方式更精確，而多數研究著重在放射治療劑量預測主要針對較高要求的治療計畫的劑量驗證。對中型醫院無工作站等級的GPU電腦設備的前提之下，本研究探討小型3D U-net為主的模型在直腸腫瘤骨盆放射線治療計畫預測股骨頭劑量的可行性。利用71筆臨床治療計畫擴增的資料集再拆分成單一計畫目標體積(OnePTV)以及多重計畫目標體積(MultiPTV)兩組，分別進行集成式學習各自訓練出一個三層的3D U-Net模型(SPU-Net)和一個五層的3D U-Res-Net(SPUR-Net)模型，對於兩側股骨頭的預測劑量彼此比較。

以OnePTV 組作分析，預測兩側股骨頭最大劑量和平均劑量的平均絕對百分比誤差(MAPE)來看， SPUR-Net除右側股骨頭的最大劑量以外均優於SPU-Net。然分析MultiPTV 組，兩者在此組在預測劑量上沒有顯著差異。然在臨床上劑量設計階段的預估要求下、在10%的誤差範圍內都是可接受的模型。在推論方面，兩種模型均能在一般的辦公筆電上於一分鐘內完成單筆計畫的預測，可應用在中型醫院。

Recent advances in radiotherapy have led to the widespread adoption of volumetric modulated arc therapy (VMAT), replacing traditional intensity-modulated radiotherapy (IMRT). Despite this progress, radiation oncologists are still required to manually define dose constraints for planning target volumes (PTVs) and organs at risk (OARs).

Among various machine learning approaches, the 3D U-Net architecture has demonstrated superior accuracy in dose prediction, particularly for radiotherapy plan quality assurance. In this study, two lightweight models based on the 3D U-Net framework were developed using ensemble learning to predict femoral head doses in VMAT plans for rectal cancer. The accessibility of these models was also evaluated by assessing their inference performance on standard office laptops.

A total of 71 clinical VMAT plans were classified and augmented into two groups: single-target volume (OnePTV) and multi-target volumes (MultiPTV). In the OnePTV group, the simplified pelvic U-Res-Net model (SPUR-Net) outperformed the simplified pelvic U-Net model (SPU-Net). However, in the MultiPTV group, no significant difference was observed between the two models. Both models achieved a mean absolute percentage error (MAPE) of approximately 10% or less compared to the actual planning dose, indicating acceptable predictive performance for dose constraint support. Furthermore, inference time per case was under one minute on an office laptop, demonstrating the models' feasibility for practical implementation in mid-sized medical institutions in Taiwan.

1. Babier, A. et al. OpenKBP: The open-access knowledge-based planning grand challenge and dataset. Med. Phys. 48, 5549–5561 (2021).
2. Çiçek, Ö., Abdulkadir, A., Lienkamp, S. S., Brox, T. & Ronneberger, O. 3D U-Net: Learning Dense Volumetric Segmentation from Sparse Annotation. in Medical Image Computing and Computer-Assisted Intervention – MICCAI 2016 (eds. Ourselin, S., Joskowicz, L., Sabuncu, M. R., Unal, G. & Wells, W.) 424–432 (Springer International Publishing, Cham, 2016). doi:10.1007/978-3-319-46723-8_49.
3. He, K., Zhang, X., Ren, S. & Sun, J. Deep Residual Learning for Image Recognition. in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 770–778 (2016). doi:10.1109/CVPR.2016.90.
4. Hou, Z. et al. Deep learning-powered radiotherapy dose prediction: clinical insights from 622 patients across multiple sites tumor at a single institution. Radiat. Oncol. 20, 80 (2025).
5. Huizenga, H. & and Storchi, P. R. M. The Use of Computed Tomography Numbers in Dose Calculations for Radiation Therapy. Acta Radiol. Oncol. 24, 509–519 (1985).
6. Jones, D. ICRU Report 50—Prescribing, Recording and Reporting Photon Beam Therapy. Med. Phys. 21, 833–834 (1994).
7. Kandalan, R. N. et al. Dose Prediction with Deep Learning for Prostate Cancer Radiation Therapy: Model Adaptation to Different Treatment Planning Practices. Preprint at https://doi.org/10.48550/arXiv.2006.16481 (2020).
8. Kearney, V., Chan, J. W., Haaf, S., Descovich, M. & Solberg, T. D. DoseNet: a volumetric dose prediction algorithm using 3D fully-convolutional neural networks. Phys. Med. Biol. 63, 235022 (2018).
9. Liu, S., Zhang, J., Li, T., Yan, H. & Liu, J. Technical Note: A cascade 3D U-Net for dose prediction in radiotherapy. Med. Phys. 48, 5574–5582 (2021).
10. Momin, S. et al. Knowledge‐based radiation treatment planning: A data‐driven method survey. J. Appl. Clin. Med. Phys. 22, 16–44 (2021).
11. Ronneberger, O., Fischer, P. & Brox, T. U-Net: Convolutional Networks for Biomedical Image Segmentation. in Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015 (eds. Navab, N., Hornegger, J., Wells, W. M. & Frangi, A. F.) vol. 9351 234–241 (Springer International Publishing, Cham, 2015).
12. Song, Y. et al. Dose prediction using a deep neural network for accelerated planning of rectal cancer radiotherapy. Radiother. Oncol. 149, 111–116 (2020).
13. Soomro, M. H., Alves, V. G. L., Nourzadeh, H. & Siebers, J. V. DeepDoseNet: A Deep Learning model for 3D Dose Prediction in Radiation Therapy. Preprint at https://doi.org/10.48550/arXiv.2111.00077 (2021).
14. Wang, H. et al. Enhanced dose prediction for head and neck cancer artificial intelligence‐driven radiotherapy based on transfer learning with limited training data. J. Appl. Clin. Med. Phys. 26, e70012 (2025).
15. Wang, M., Pan, Y., Zhang, X. & Yang, R. Exploring the impact of network depth on 3D U-Net-based dose prediction for cervical cancer radiotherapy. Front. Oncol. 14, 1433225 (2024).
16. Web-Content. DICOM History. DICOM https://www.dicomstandard.org/history.
17. Xu, W., Fu, Y.-L. & Zhu, D. ResNet and its application to medical image processing: Research progress and challenges. Comput. Methods Programs Biomed. 240, 107660 (2023).
18. Zhou, J. et al. A method of using deep learning to predict three‐dimensional dose distributions for intensity‐modulated radiotherapy of rectal cancer. J. Appl. Clin. Med. Phys. 21, 26–37 (2020).
19. Zimmermann, L., Faustmann, E., Ramsl, C., Georg, D. & Heilemann, G. Technical Note: Dose prediction for radiation therapy using feature‐based losses and One Cycle Learning. Med. Phys. 48, 5562–5566 (2021).