北京生物医学工程2026，Vol.45Issue(3)：221-228,246,9.DOI:10.3969/j.issn.1002-3208.2026.03.001
基于nnU-Net的肾脏区域分割方法及其术中应用
Kidney region segmentation method based on nnU-Net and its intraoperative application
王文韬 1张海康 2黄清明 3黄亦成 4徐天宇 5胡敏捷 1沈兵6
作者信息
- 1. 上海理工大学健康科学与工程学院(上海 200093)
- 2. 上海理工大学健康科学与工程学院(上海 200093)||上海理工大学上海介入医疗器械工程技术研究中心(上海 200093)
- 3. 上海理工大学健康科学与工程学院(上海 200093)||上海健康医学院医学影像学院(上海 201318)
- 4. 上海理工大学健康科学与工程学院(上海 200093)||同济大学附属第十人民医院泌尿外科(上海 200072)
- 5. 上海理工大学健康科学与工程学院(上海 200093)||上海申康医院发展中心(上海 200041)
- 6. 上海理工大学健康科学与工程学院(上海 200093)||上海理工大学上海介入医疗器械工程技术研究中心(上海 200093)||同济大学附属第十人民医院泌尿外科(上海 200072)
- 折叠
摘要
Abstract
Objective Explore the application of deep learning methods based on nnU-Net in kidney surgery,particularly the feasibility of intraoperative image segmentation.As the complexity of kidney tumor surgery increases,precise intraoperative image segmentation becomes crucial for surgical navigation.The research goal is to improve the automatic recognition and localization accuracy of the renal region during surgery using the nnU-Net framework,thereby providing technical support for intelligent assisted surgery systems.Methods Based on the left-eye endoscopic images obtained during surgery,a 2D semantic segmentation model was constructed using the nnU-Net framework.The training data were generated by extracting individual frames from the intraoperative images and combining them with manually annotated masks.The annotated regions include renal parenchyma,tumors,and surface-adhered fat,which were identified as a unified region of Interest(ROI).For data validation,an independent test set was used,consisting of five kidney surgery cases that did not participate in the training and validation.The model's performance in intraoperative image segmentation was evaluated by comparing it with manually annotated masks.Results The model performed well on the independent test set,with an average Dice similarity coefficient(DSC)of 0.9335,intersection over union(IoU)of 0.878 4,precision of 0.951 7,recall of 0.921 1,and F1 score of 0.933 5.These results indicate that nnU-Net can effectively segment the renal region and maintain high segmentation accuracy under various conditions.In terms of inference speed,the average single-frame segmentation time of the model is approximately 0.38 seconds,demonstrating strong processing capability,making it suitable for rapid intraoperative image updates and localization tasks.Conclusions The nnU-Net model maintains high accuracy and robustness even in complex intraoperative image environments,validating its potential for real-time image segmentation tasks.nnU-Net can provide reliable technical support for intelligent assisted surgery systems,showing promising clinical application prospects.关键词
深度学习/图像分割/肾脏区域/术中图像/nnU-NetKey words
deep learning/image segmentation/kidney region/intraoperative image/nnU-Net分类
医药卫生引用本文复制引用
王文韬,张海康,黄清明,黄亦成,徐天宇,胡敏捷,沈兵..基于nnU-Net的肾脏区域分割方法及其术中应用[J].北京生物医学工程,2026,45(3):221-228,246,9.
