FastReg:基于微分同胚流形上加速优化的快速非刚性配准
计算机视觉与模式识别
2019-04-25 v3
摘要
我们提出了一种医学图像微分同胚非刚性配准新方法的实现。该方法基于光流,并通过标准 内积下的梯度流对图像进行形变。为计算变换,我们依赖于微分同胚流形上的加速优化。得益于一种在时间而非空间上对梯度进行平均的新方法,我们获得了通常计算代价高昂的 Sobolev 梯度流的正则性性质。我们成功以比以往方法快数个数量级的速度配准了脑 MRI 与具有挑战性的腹部 CT 扫描。我们将代码公开于代码仓库:https://github.com/dgrzech/fastreg
引用
@article{arxiv.1903.01905,
title = {FastReg: Fast Non-Rigid Registration via Accelerated Optimisation on the Manifold of Diffeomorphisms},
author = {Daniel Grzech and Loïc le Folgoc and Mattias P. Heinrich and Bishesh Khanal and Jakub Moll and Julia A. Schnabel and Ben Glocker and Bernhard Kainz},
journal= {arXiv preprint arXiv:1903.01905},
year = {2019}
}
备注
There is an ongoing dispute about the presentation of this paper. It will be withdrawn until the dispute is resoved