Content area

Abstract

Clinical target volume (CTV) delineation is essential in cervical cancer brachytherapy (BT), as accurate and automatic segmentation can improve treatment effect for patients with locally advanced disease while reducing the workload of clinicians. We propose BCTVNet (BT CTV),, a 3D hybrid neural network that integrates convolutional neural network (CNN) and transformer branches to combine strong local feature extraction with global context modeling. The 3D architecture enables the model to capture spatial relationships across slices, which is crucial for accurately identifying CTV boundaries in volumetric computed tomography (CT) data. In addition, a 3D contrast limited adaptive histogram equalization preprocessing step is applied to enhance the local contrast of soft tissues, improving anatomical structure visibility and facilitating boundary recognition. Experiments on a private BT CT dataset of 95 patients show that BCTVNet achieves superior performance compared with popular CNN-based and transformer-based segmentation models, reaching a Dice similarity coefficient (DSC) of 83.23% and a Hausdorff distance 95th percentile of 3.53 mm. Evaluation on the publicly available SegTHOR dataset further confirms its strong generalizability, achieving the highest average score among all compared methods, with a DSC of 87.09%. Multiple ablation experiments verify the effectiveness of both the hybrid architecture and the adaptive preprocessing strategy. These results demonstrate that BCTVNet provides accurate and stable CTV delineation, making it a reliable tool for clinical BT and a valuable approach for wider medical image segmentation tasks.

Details

1009240
Title
BCTVNet: a 3D Hybrid segmentation neural network for clinical target volume delineation of cervical cancer brachytherapy
Volume
6
Issue
4
First page
045056
Number of pages
20
Publication year
2025
Publication date
Dec 2025
Publisher
IOP Publishing
Place of publication
Bristol
Country of publication
United Kingdom
e-ISSN
26322153
Source type
Scholarly Journal
Language of publication
English
Document type
Journal Article
Publication history
 
 
Milestone dates
2024-10-22 (Received); 2025-10-30 (Rev-Recd); 2025-11-20 (Accepted); 2025-11-04 (Oa-Requested)
ProQuest document ID
3278275493
Document URL
https://www.proquest.com/scholarly-journals/bctvnet-3d-hybrid-segmentation-neural-network/docview/3278275493/se-2?accountid=208611
Copyright
© 2025 The Author(s). Published by IOP Publishing Ltd. This work is published under https://creativecommons.org/licenses/by/4.0/ (the "License"). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.
Last updated
2025-12-05
Database
2 databases
  • Coronavirus Research Database
  • ProQuest One Academic