Patterns 2020

Virtual Monoenergetic CT

Virtual Monoenergetic CT Imaging via Deep Learning

Wenxiang Cong, Yan Xi, Paul Fitzgerald, Bruno De Man, Ge Wang

Current or former lab member. Select a name to see all of their projects.

Method overview of Virtual Monoenergetic CT
Architecture of the modified residual network, from the figure file in the repository.

Abstract

Conventional single-spectrum CT shows morphology but not the elemental composition of tissue. Dual-energy CT provides energy-selective and material-selective images, at the cost of a more complex system and a higher radiation dose. This work uses a modified residual network, trained on clinical dual-energy data, to map single-spectrum CT images to virtual monoenergetic images at chosen energy levels.

The relative error of the virtual monoenergetic images is below 2%. The model also supports decomposition into three tissue classes with accuracy comparable to dual-energy CT.

What’s in the repository

  • The residual network model, with training and validation scripts
  • Data archives and the labels for two energy levels

Citation

@article{cong2020virtual,
  title   = {Virtual Monoenergetic CT Imaging via Deep Learning},
  author  = {Cong, Wenxiang and Xi, Yan and Fitzgerald, Paul and Man, Bruno De and Wang, Ge},
  journal = {Patterns},
  volume  = {1},
  number  = {8},
  pages   = {100128},
  year    = {2020},
  doi     = {10.1016/j.patter.2020.100128}
}
Back to projects