Referenceless distortion correction of gradient-echo echo-planar imaging under inhomogeneous magnetic fields based on a deep convolutional neural network.

Journal: Computers in biology and medicine
PMID:

Abstract

Single-shot gradient-echo echo-planar imaging (GE-EPI) plays a significant role in applications where high temporal resolution is necessary. However, GE-EPI is susceptible to inhomogeneous magnetic fields that will cause image distortion. Most existing methods either need additional acquisitions for field mapping or cannot correct the distortion at high field. Here, we propose a new algorithm based on a deep convolutional neural network (CNN) to solve this problem without additional acquisitions. The residual learning and the cascaded structure improved the performance of the CNN on distortion correction. A simulated dataset was used for training. The simulated and experimental results demonstrate that the proposed method can correct the image distortion caused by field inhomogeneity.

Authors

  • Pu Liao
    Department of Electronic Science, Fujian Provincial Key Laboratory of Plasma and Magnetic Resonance, Xiamen University, Xiamen, 361005, China.
  • Jun Zhang
    First School of Clinical Medicine, Shandong University of Traditional Chinese Medicine, Jinan, China.
  • Kun Zeng
    College of Physical Science and Technology, Department of Electronic Science, Fujian Provincial Key Laboratory of Plasma and Magnetic Resonance, Xiamen University, Xiamen, 361005, China.
  • Yonggui Yang
    Department of Medical Imaging, The 2nd Hospital of Xiamen, Xiamen, 361021, China.
  • Shuhui Cai
    Department of Electronic Science, Fujian Provincial Key Laboratory of Plasma and Magnetic Resonance, Xiamen University, Xiamen, China.
  • Gang Guo
    Department of Medical Imaging, The 2nd Hospital of Xiamen, Xiamen, 361021, China.
  • Congbo Cai
    Department of Electronic Science, Fujian Provincial Key Laboratory of Plasma and Magnetic Resonance, Xiamen University, Xiamen, China.