Computer Engineering and Applications ›› 2021, Vol. 57 ›› Issue (1): 227-233.DOI: 10.3778/j.issn.1002-8331.1911-0287

Previous Articles     Next Articles

Joint Segmentation of Full Convolutional Deep Migration Network

LI Yan, LUI Jun   

  1. 1.Hubei Key Laboratory of Intelligent Information Processing and Real Time Industrial Systems(Wuhan University of Science and Technology), Wuhan 430065, China
    2.School of Computer Science and Technology, Wuhan University of Science and Technology, Wuhan 430065, China
  • Online:2021-01-01 Published:2020-12-31



  1. 1.智能信息处理与实时工业系统湖北省重点实验室(武汉科技大学),武汉 430065
    2.武汉科技大学 计算机科学与技术学院,武汉 430065


In recent years, with the application of ultrasonic cavitation in medical treatment, ultrasonic cavitation therapy has returned to people’s field of vision, and its extensive advantages in dealing with vascular occlusion have caused extensive discussion and research. In order to achieve accurate real-time treatment, a method of ultrasonic vessel segmentation based on Fully Convolutional Networks(FCN) is proposed. The Full Convolutional deep Aggregation Migration Network(AMFCN) uses a symmetric network connection to the full convolutional network, deeply extracts image features in depth aggregation mode, optimizes the data enhancement mode, adds a migration learning model, and selects a suitable loss function. The method effectively utilizes existing data for data expansion and alleviates the influence of too little medical image data. The experimental results show that the research method can achieve better segmentation performance on the ultrasound blood vessel image and accurately segment the blood vessel region.

Key words: ultrasonic therapy, Full Convolutional Network(FCN), deep aggregation, small data set, migration learning


近年来,随着超声空化在医疗上的应用,超声空化治疗又重新回到了人们的视野,因其在处理血管阻塞方面的天然的优势,引起了广泛的讨论和研究。为实现准确的实时治疗,提出了一种基于全卷积网络(Fully Convolutional Networks,FCN)的超声血管分割方法。全卷积深度迁移分割网络(Full Convolutional deep Aggregation Migration Network,AMFCN)通过对全卷积网络使用对称网络连接,深度聚合模式以深度提取图像特征,并优化数据增强方式,添加迁移学习模型等方法,有效地利用已有数据进行数据拓展,缓解医学图像数据过少的影响。实验结果表明,该研究方法在超声血管图像上取得了较好的分割性能,能准确地分割出血管区域。

关键词: 超声治疗, 全卷积网络, 深度聚合, 小数据集, 迁移学习