网站首页            期刊简介             编委会             投稿指南             期刊订阅             下载中心             在线留言            联系我们             English
  2025年4月5日 星期六  
文章快速检索
中国生物医学工程学报  2020, Vol. 39 Issue (5): 524-531    DOI: 10.3969/j.issn.0258-8021.2020.05.002
  论著 本期目录 | 过刊浏览 | 高级检索 |
基于可解释卷积神经网络的胚胎血管新生时间模式研究
吕雪倩1, 赵沈佳1, 李佩伦2, 方路平1, 宁钢民2#, 潘清1#*
1(浙江工业大学信息工程学院,杭州 310023)
2(浙江大学生物医学工程系,杭州 310027)
Study of Temporal Pattern of Embryonic Angiogenesis Based on Interpretable Convolutional Neural Networks
Lv Xueqian1, Zhao Shenjia1, Li Peilun2, Fang Luping1, Ning Gangmin2#, Pan Qing1#*
1(College of Information Engineering,Zhejiang University of Technology,Hangzhou 310023,China)
2(Department of Biomedical Engineering,Zhejiang University,Hangzhou 310027,China)
全文: PDF (11628 KB)   HTML (1 KB) 
输出: BibTeX | EndNote (RIS)      
摘要 理解血管网络新生的时间模式,有助于生物体发育机制研究和肿瘤等疾病的生理病理研究。提出以可解释卷积神经网络(CNN)研究鸡胚胎卵黄膜的血管新生时间模式的方法。基于CNN建立受精3d后(3dpf)和4d后(4dpf)的鸡胚胎血管网络图像的分类模型,以梯度加权的类激活映射(Grad-CAM)技术解释发育过程中血管网络形态拓扑的变化模式,并以此分类模型分析3dpf~4dpf之间血管新生的时间特性。实验共计观察17枚受精卵,结果显示最优模型区分3dpf与4dpf的血管图像的准确率达到98.62%。通过Grad-CAM技术对不同时期血管图像的特征进行可视化,发现3dpf~4dpf的发育过程主要表现为毛细血管网的生长发育。这些鸡胚胎卵黄膜在3dpf~4dpf时间段内,前12h血管新生较为剧烈,随后趋于平稳。这些结果可为血管新生研究提供新的技术手段,并辅助血管新生机制、肿瘤发病机理和器官衰老过程等的相关研究。
服务
把本文推荐给朋友
加入我的书架
加入引用管理器
E-mail Alert
RSS
作者相关文章
吕雪倩
赵沈佳
李佩伦
方路平
宁钢民
潘清
关键词 血管新生卷积神经网络鸡胚胎卵黄膜梯度加权的类激活映射    
Abstract:Understanding temporal patterns of vascular network angiogenesis facilitates the study of biological development mechanisms and the pathophysiology of tumors. In this paper,an interpretable convolutional neural network (CNN) was proposed to study the temporal pattern of angiogenesis of the chicken vitelline vascular networks. We constructed a model for classifying the vascular images of chicken vitelline at 3 days post-fertilization (3dpf) and 4 days post-fertilization (4dpf) based on CNN,explained the classification results using gradient-weighted class activation mapping (Grad-CAM),and analyzed the angiogenic pattern from 3dpf to 4dpf based on the model. A total of 17 fertilized eggs were observed in the experiment. Results showed that the accuracy of the optimal model to classify 3dpf and 4dpf vascular images was 98.62%. Using the Grad-CAM technique,we found out that the main manifestation of the vascular network development from 3dpf to 4dpf was the growth of capillary networks. Between 3dpf and 4dpf,the process of angiogenesis was more intense during the first 12 hours and then tended to be stable. This study provided new approaches for the researches on angiogenesis,assisting the physiological studies including angiogenesis mechanisms,tumor growth and organ aging.
Key wordsangiogenesis    convolutional neural network    chicken vitelline    gradient-weighted class activation mapping
收稿日期: 2019-12-10     
PACS:  R318  
基金资助:国家自然科学基金(31870938,81871454)#中国生物医学工程学会会员(Member,ChineseSocietyofBiomedicalEngineering)
通讯作者: *E-mail:pqpq@zjut.edu.cn   
引用本文:   
吕雪倩, 赵沈佳, 李佩伦, 方路平, 宁钢民, 潘清. 基于可解释卷积神经网络的胚胎血管新生时间模式研究[J]. 中国生物医学工程学报, 2020, 39(5): 524-531.
Lv Xueqian, Zhao Shenjia, Li Peilun, Fang Luping, Ning Gangmin, Pan Qing. Study of Temporal Pattern of Embryonic Angiogenesis Based on Interpretable Convolutional Neural Networks. Chinese Journal of Biomedical Engineering, 2020, 39(5): 524-531.
链接本文:  
http://cjbme.csbme.org/CN/10.3969/j.issn.0258-8021.2020.05.002     或     http://cjbme.csbme.org/CN/Y2020/V39/I5/524
版权所有 © 2015 《中国生物医学工程学报》编辑部
本系统由北京玛格泰克科技发展有限公司设计开发