FusionSwinUNETR多模态深度学习模型在儿童腹部肿块分割中的应用
FusionSwinUNETR: A multimodal deep learning model for abdominal mass segmentation in pediatric CT
FusionSwinUNETR多模态模型在儿童腹部肿块分割中优于纯影像模型,DSC达0.900±0.044,边界误差HD95降至4.79±3.31毫米,多中心验证显示临床元数据与影像互补提升分割精度。
- Article
- Open access
- Published: 23 October 2026
- Seung Wan Moon1 na1,
- Dayoung Ko2,5 na1,
- Seulgi Park3,
- Young Jae Kim4,
- Joong Kee Youn2,
- Hee Beom Yang6,
- Hyun Young Kim2,5 &
- …
- Kwang Gi Kim1,7
Scientific Reports (2026) Cite this article
We’re sharing this article early to provide faster access to peer-reviewed, accepted research. It is citable and carries a permanent DOI. This version is subject to further edits and will be replaced automatically by the final Version of Record. All legal disclaimers apply.
Abstract
Research on computational methods for pediatric solid tumors is constrained by data scarcity, a focus on single tumor types, and the limits of imaging alone. We present FusionSwinUNETR, a multimodal segmentation model that conditions a SwinUNETRv2 backbone on patient metadata and clinician-written findings alongside CT images, and we evaluate it on a multi-center paired dataset of 750 pediatric patients spanning 12 mass types — a considerably broader diagnostic range than previous single-tumor studies. Against an image-only baseline evaluated on the same 60 held-out patients, the fusion model improved every metric: DSC 0.900 ± 0.044 versus 0.882 ± 0.064, precision 0.928 ± 0.051 versus 0.913 ± 0.062, recall 0.880 ± 0.081 versus 0.866 ± 0.110, and HD95 4.79 ± 3.31 versus 8.93 ± 6.22 mm, a 46% reduction in boundary error. Paired within-patient testing with correction for multiple comparisons confirmed these gains: HD95 improved significantly for all three multimodal configurations, and the fusion model was the only configuration whose DSC improvement survived correction (Hodges–Lehmann + 0.012, q = 5.6 × 10⁻4) and the only one to improve precision and recall concurrently, indicating that metadata and clinical findings contribute complementary rather than redundant information. Incorporating clinical information already available at the time of image interpretation therefore yields more accurate and more reliable segmentation across a wide range of pediatric abdominal masses.
Acknowledgements
The authors would like to express their gratitude to everyone who contributed to this paper, including the clinical and administrative staff of Seoul National University Hospital, Pusan National University Yangsan Hospital, and Korea University Hospital, who assisted in the multi-institutional data collection process.
Funding
This work was supported by the Digital Medical Products Development Based on Medical Data Synthesis and AI Technologies Program (RS-2025-02305698, Development of On-Device AI Digital Medical Products Utilizing Synthetic Technology and Synthetic Data for Atypical Medical Data) funded by the Ministry of Trade, Industry & Energy (MOTIE) of Korea, and by the National Research Foundation of Korea (NRF) grant funded by the Korea government (MSIT) (RS-2026-25493516).
Author information
Author notes
Seung Wan Moon and Dayoung Ko are contributed equally as first authors.
Authors and Affiliations
Department of Biomedical Engineering, College of IT Convergence, Gachon University, Seongnam-Si, 13120, Republic of Korea
Seung Wan Moon & Kwang Gi Kim
Division of Pediatric Surgery, Department of Surgery, Seoul National University Hospital, Seoul, Korea
Dayoung Ko, Joong Kee Youn & Hyun Young Kim
Department of Nursing, College of Nursing, Gachon University, Incheon, 21999, Republic of Korea
Seulgi Park
Department of Gachon Biomedical & Convergence Institute, Gachon University Gil Medical Center, Incheon, 21565, Republic of Korea
Young Jae Kim
Department of Surgery, Seoul National University College of Medicine, Seoul, Korea
Dayoung Ko & Hyun Young Kim
Department of Surgery, Seoul National University Bundang Hospital, Seongnam-Si, 13620, Republic of Korea
Hee Beom Yang
Department. of Health Sci. & TechnolGachon Advanced Institute for Health Sci. & Technology(GAIHIST), Gachon UniversityLee Gil Ya Cancer and Diabetes Institute, 155 Gaetbeol-Ro, Yeonsu-Gu, Incheon, Korea
Kwang Gi Kim
Authors
- Seung Wan Moon
- Dayoung Ko
- Seulgi Park
- Young Jae Kim
- Joong Kee Youn
- Hee Beom Yang
- Hyun Young Kim
- Kwang Gi Kim
Corresponding authors
Correspondence to Hyun Young Kim or Kwang Gi Kim.
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher's note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
About this article
Cite this article
Moon, S.W., Ko, D., Park, S. et al. FusionSwinUNETR: A multimodal deep learning model for abdominal mass segmentation in pediatric CT. Sci Rep (2026). https://doi.org/10.1038/s41598-026-69756-0
Received: 11 May 2026
Accepted: 30 August 2026
Published: 23 October 2026
DOI: https://doi.org/10.1038/s41598-026-69756-0
Keywords
来源:Nature 健康护理主题 · nature.com