数据与计算发展前沿 ›› 2026, Vol. 8 ›› Issue (3): 1-14.

doi: 10.11871/jfdc.issn.2096-742X.2026.03.001

• 专刊：第二十一届全国科学计算与信息化会议征文 • 上一篇    下一篇

基于迁移学习与Attention U-Net的同步辐射非人灵长类脑血管图像分割算法研究

叶静1,2,3(),王春鹏1,2,3(),李沁桐1,陈卓1,4,李宗泽1,张家如1,张祥志1,2,3,胡宇光1,*(),邰仁忠1,2,3,5,*()   

  1. 1 中国科学院上海高等研究院， 上海 201210
    2 中国科学院上海应用物理研究所， 上海 201800
    3 中国科学院大学， 北京 101408
    4 南京信息工程大学， 江苏 南京 210044
    5 上海科技大学， 上海 201210
  • 收稿日期:2025-11-05 出版日期:2026-06-20 发布日期:2026-06-18
  • 通讯作者: 胡宇光,邰仁忠
  • 作者简介:叶静，中国科学院上海高等研究院，工程师，中国科学院应用物理研究所，博士研究生，主要研究方向为同步辐射大数据技术、机器学习在同步辐射中的应用。
    本文中负责论文撰写、算法设计及代码开发。
    YE Jing is an engineer at the Shanghai Advanced Research Institute, Chinese Academy of Sciences and a Ph.D. student at the Shanghai Institute of Applied Physics, Chinese Academy of Sciences. Her research interests include synchrotron radiation big data technology and machine learning application in the field of synchrotron radiation.
    In this paper, she is mainly responsible for manuscript writing, algorithm design, and code development.
    E-mail: yej@sari.ac.cn|王春鹏，中国科学院上海高等研究院，研究员，博士，上海同步辐射光源中心，主任助理、大数据与实验辅助中心主任。主要研究方向为同步辐射大数据技术及计算平台。
    本文中负责算法设计及测试。
    WANG Chunpeng, Ph.D., is a researcher at the Shanghai Advanced Research Institute, Chinese Academy of Sciences. He is also the assistant director of Shanghai Synchrotron Radiation Facility and the director of Department of Big Data and Experiment Assist System Center. His research focuses on synchrotron radiation big data technology and high-performance computing center.
    In this paper, he is mainly responsible for algorithm design and test.
    E-mail: wangcp@sari.ac.cn|胡宇光，中国科学院上海高等研究院，研究员，博士，主要研究方向为同步辐射成像技术。
    本文中负责成像系统及实验设计、数据采集及数据手工标注。
    HU Yuguang, Ph.D., is a researcher at the Shanghai Advanced Research Institute, Chinese Academy of Sciences. His research focuses on synchrotron radiation imaging technology.
    In this paper, he is mainly responsible for imaging system and experimental design, data acquisition, and manual data annotation.
    E-mail: huyg@sari.ac.cn|邰仁忠，中国科学院上海高等研究院，博士，副院长、研究员、博导，上海同步辐射光源常务副主任。主要研究方向为同步辐射X射线方法学。
    本文中负责成像系统及实验设计。
    TAI Renzhong, Ph.D., is the vice president, a researcher, and a doctoral supervisor at the Shanghai Advanced Research Institute, Chinese Academy of Sciences. He also serves as the executive deputy director of the Shanghai Synchrotron Radiation Facility. His research focuses on synchrotron radiation X-ray methodologies.
    In this paper, he is mainly responsible for imaging system and experimental design.
    E-mail: tairz@sari.ac.cn
  • 基金资助:
    中国科学院上海高等研究院创新基金“同步辐射脑成像数据中血管与神经关联定位方法研究”(2024CP003)

Segmentation of Non-Human Primate Cerebrovascular Images from Synchrotron Radiation Micro-Tomography Using Transfer Learning and Attention U-Net

YE Jing1,2,3(),WANG Chunpeng1,2,3(),LI Qintong1,CHEN Zhuo1,4,LI Zongze1,ZHANG Jiaru1,ZHANG Xiangzhi1,2,3,HU Yuguang1,*(),TAI Renzhong1,2,3,5,*()   

  1. 1 Shanghai Advanced Research Institute, Chinese Academy of Sciences, Shanghai 201210, China
    2 Shanghai Institute of Applied Physics, Chinese Academy of Sciences, Shanghai 201800, China
    3 University of Chinese Academy of Sciences, Beijing 101408, China
    4 Nanjing University of Information Science and Technology, Nanjing, Jiangsu 210044, China
    5 ShanghaiTech University, Shanghai 201210, China
  • Received:2025-11-05 Online:2026-06-20 Published:2026-06-18
  • Contact: HU Yuguang,TAI Renzhong

RichHTML

7

PDF

37

摘要：

【目的】针对同步辐射显微断层成像（SR-μCT）非人灵长类脑血管影像高质量标注样本稀缺及伪影干扰难题，本研究提出一种基于迁移学习与层级复合加权策略的自动化分割方法。 【方法】采用全局百分位归一化抑制伪影并保持信号一致性；构建“预训练-微调”框架，创新引入层级化复合权重策略，协同强化对微小血管与边界特征的捕获能力。 【结果】实验表明，该方法全面超越nnU-Net 3D，Dice系数达0.8686，召回率高达96.93%，拓扑连通性指标clDice达0.8848，显著解决了微血管漏检与断裂问题。 【结论】该算法有效克服了小样本与高分辨成像的脑血管同步辐射影像的分割挑战，为更大尺度的非人灵长类及人类亚微米级同步辐射全脑成像数据分析和神经血管网络量化分析奠定了技术基础和数据基础。

关键词: 脑血管分割, 同步辐射显微断层成像, 迁移学习, 层级化复合加权, 拓扑连通性

Abstract:

[Objective] To address the challenges of scarce high-quality annotations and severe artifact interference in Synchrotron Radiation Micro-Tomography (SR-μCT) imaging of non-human primate cerebrovasculature, this study proposes an automated segmentation method based on transfer learning and a hierarchical combined weighting strategy. [Methods] A global percentile normalization strategy is employed to suppress artifacts while maintaining signal consistency. A “pre-training and fine-tuning” framework is established, innovatively incorporating a hierarchical combined weighting strategy to synergistically reinforce the capture of micro-vessels and boundary features. [Results] Experimental results demonstrate that the proposed method comprehensively outperforms nnU-Net 3D, achieving a Dice coefficient of 0.8686, a recall of 96.93%, and a topological connectivity metric (clDice) of 0.8848. These results signify a substantial resolution to the issues of micro-vessel miss-detection and disconnection. [Conclusions] This algorithm effectively overcomes the segmentation challenges associated with small-sample and high-resolution SR-μCT cerebrovascular imaging, laying a solid technical and data foundation for future large-scale sub-micron whole-brain imaging analysis and neurovascular network quantification in non-human primates and humans.

Key words: cerebrovascular segmentation, synchrotron radiation micro-tomography, transfer learning, hierarchical combined weighting, topological connectivity