Deep learning-based head and neck deformable image registration using spatio-temporal analysis and self attention
Physics and Imaging in Radiation Oncology · 2026-01 · 37 · 100928
DOI: 10.1016/j.phro.2026.100928 ↗Kurz gefasst
Fachartikel in der Zeitschrift Physics and Imaging in Radiation Oncology, 2026-01. Verfasst von Donghoon Lee und 7 weiteren Autor:innen. Beteiligte Einrichtung: Department of Medical Physics, Memorial Sloan Kettering Cancer Center, New York, NY 10065, USA. Die Arbeit wurde bisher 1-mal zitiert (Crossref) und nennt 29 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: NCI NIH HHS, NIH NCI.
Abstract (Originalsprache)
Background and purpose: Significant anatomical changes during head and neck cancer (HNC) radiotherapy challenge accurate dose delivery. Deformable image registration (DIR) is essential for adaptive radiotherapy (ART), yet conventional methods are too slow for online clinical use. This study proposed a novel deep learning-based DIR algorithm for longitudinal HNC imaging. Materials & methods: We used sixty HNC patient datasets, each containing a planning CT (pCT) and six weekly cone-beam CTs (CBCTs). Fifty datasets were used for training with cross-validation, and the remaining ten were reserved for testing. The proposed DIR algorithm is a patch-based model that integrates 3D convolutional neural networks, self-attention, and a convolutional Long Short Term Memory to model temporal deformations. The model predicted bidirectional deformation vector fields and was trained with a composite loss function combining image similarity, DVF smoothness, and inverse consistency. Performance was benchmarked against the large deformation diffeomorphic metric mapping (LDDMM) algorithm using Dice similarity coefficient (DSC), Hausdorff distance, and Jacobian analysis.
Results
The proposed method achieved significantly faster inference, performing bidirectional DIR between the pCT and all six weekly CBCTs in under 3 min and averaging about 30 s per patient, while matching or exceeding LDDMM's accuracy. DSC remained above 0.8 for all key structures, and the method demonstrated improved DVF consistency with lower mean and 95th percentile Hausdorff distances. Unlike LDDMM, it required no manual parameter tuning, providing consistent results.
Conclusion
The proposed DIR algorithm enabled rapid, accurate, and consistent image registration, supporting real-time ART workflows and retrospective dose accumulation in personalized radiotherapy.
Schlagwörter
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Department of Medical Physics, Memorial Sloan Kettering Cancer Center, New York, NY 10065, USA.
- Department of Radiation Oncology, Memorial Sloan Kettering Cancer Center, New York, NY 10065, USA.
- Division of Radiation Oncology, Department of Radiology, Faculty of Medicine Siriraj Hospital, Mahidol University, Bangkok 10700, Thailand.
Förderung
- NCI NIH HHS
- NIH NCI
Zitieren
APA
Lee, D., Hu, Y. C., TreeChairusame, T., Oh, J. H., Lee, N., Aristophanous, M., Cerviño, L., & Zhang, P. (2026). Deep learning-based head and neck deformable image registration using spatio-temporal analysis and self attention. Physics and Imaging in Radiation Oncology, 37, 100928. https://doi.org/10.1016/j.phro.2026.100928
Vancouver
Lee D, Hu YC, TreeChairusame T, Oh JH, Lee N, Aristophanous M, et al. Deep learning-based head and neck deformable image registration using spatio-temporal analysis and self attention. Physics and Imaging in Radiation Oncology. 2026;37:100928. doi:10.1016/j.phro.2026.100928
BibTeX
@article{Lee2026,
author = {Lee, Donghoon and Hu, Yu-Chi and TreeChairusame, Teeradon and Oh, Jung Hun and Lee, Nancy and Aristophanous, Michalis and Cerviño, Laura and Zhang, Pengpeng},
title = {Deep learning-based head and neck deformable image registration using spatio-temporal analysis and self attention},
journal = {Physics and Imaging in Radiation Oncology},
year = {2026},
volume = {37},
pages = {100928},
doi = {10.1016/j.phro.2026.100928}
}
EUVIMED