基于深度学习的医学影像配准综述

图像配准是医学影像处理与智能分析领域中的重要环节和关键技术.传统的图像配准算法由于复杂性较高、计算代价较大等问题,无法实现配准的实时性要求.随着深度学习方法的发展,基于学习的图像配准方法也取得显著效果.文中系统总结基于深度学习的医学图像配准方法.具体地,将方法归为3类:监督学习,无监督学习和对偶监督/弱监督学习.在此基础上,分析和讨论各自优缺点.进一步,着重讨论近年来提出的正则化方法,特别是基于微分同胚表示的正则和基于多尺度的正则.最后,根据当前医学图像配准方法的发展趋势,展望基于深度学习的医学图像配准方法.

Deep Learning Based Medical Image Registration: A Review

Image registration is a key technology in the field of medical image processing and intelligent analysis. The real-time registration cannot be accomplished due to the high complexity and computational cost of traditional registration methods. With the development of deep learning, learning based image registration methods achieve remarkable results. In this paper, the medical image registration methods based on deep learning are systematically summarized and divided into three categories, including supervised learning, unsupervised learning and dual supervised learning. On this basis, the advantages and disadvantages for each category are discussed. Furthermore, the regularization methods proposed in recent years are emphatically discussed, especially based on diffeomorphism and multi-scale regularization. Finally, the medical image registration methods based on deep learning are prospected according to the development trend of the current medical image registration methods.