人工智能培训

搜索

人工智能教程:用于免引用EPI幻影校正的K-SPACE深度学习(k-Space Deep Learning for Reference-free EPI Gho

[复制链接]
zwb521 发表于 2018-6-4 08:36:07 | 显示全部楼层 |阅读模式
zwb521 2018-6-4 08:36:07 1020 0 显示全部楼层
人工智能教程:用于免引用EPI幻影校正的K-SPACE深度学习(k-Space Deep Learning for Reference-free EPI Ghost Correction)奈奎斯特鬼影伪影在EPI图像中起源于偶数和奇数回波之间的相位失配。然而,使用参考扫描的传统校正方法经常产生错误的结果,尤其是在高场MRI中,由于非线性和时变的局部磁场变化。已经表明,鬼影校正问题可以转化为k-spacedata插值问题,该问题可以使用基于歼灭滤波器的低秩Hankel结构矩阵完备方法(ALOHA)来解决。另一个最近的发现显示深度卷积神经网络与数据驱动的汉克尔矩阵分解密切相关。通过对这些发现进行有力结合,我们提出了一种k-空间重叠学习方法,可以在不进行参考扫描的情况下立即纠正k-空间相位失配。使用7T体内数据的重建结果显示,与现有方法相比,用于EPI鬼影校正的所提出的无参考k空间深度学习方法显着地提高了图像质量,并且计算时间快几个数量级。
Nyquist ghost artifacts in EPI images are originated from phase mismatchbetween the even and odd echoes.However, conventional correction methods usingreference scans often produce erroneous results especially in high-field MRIdue to the non-linear and time-varying local magnetic field changes.It hasbeen shown that the problem of ghost correction can be transformed into k-spacedata interpolation problem that can be solved using the annihilatingfilter-based low-rank Hankel structured matrix completion approach (ALOHA).Another recent discovery has shown that the deep convolutional neural networkisclosely related to the data-driven Hankel matrix decomposition.Bysynergistically combining these findings, here we propose a k-space deeplearning approach that immediately corrects the k- space phase mismatch withouta reference scan.Reconstruction results using 7T in vivo data showed that theproposed reference-free k-space deep learning approach for EPI ghost correctionsignificantly improves the image quality compared to the existing methods andthe computing time is several orders of magnitude faster.人工智能教程:用于免引用EPI幻影校正的K-SPACE深度学习(k-Space Deep Learning for Reference-free EPI Ghost Correction) WIIw2PWrwWWBA42C.jpg
URL地址:https://arxiv.org/abs/1806.00153     ----pdf下载地址:http://arxiv.org/pdf/1806.00153    ----人工智能教程:用于免引用EPI幻影校正的K-SPACE深度学习(k-Space Deep Learning for Reference-free EPI Ghost Correction)
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

zwb521当前离线
新手上路

查看:1020 | 回复:0

快速回复 返回顶部 返回列表