Cardiac Segmentation Workflow

This tool uses a fully convolutional neural network to automatically segment the left ventricle, the left ventricle myocardium and the right ventricle of cine-MRI cardiac images from three possible views (short axis, four-chamber long axis and two-chamber long axis). It is based on the Github repository DLTK models: UKBB cardiac segmentation cine-MRI.


The tool detects the end-systolic and end-diastolic frames in the image and uses the information to compute a set of clinical values:

  • Left ventricular end-diastolic volume (LVEDV)
  • Left ventricular end-systolic volume (LVESV)
  • Left ventricular myocardial mass (LVM)
  • Right ventricular end-diastolic volume (RVEDV)
  • Right ventricular end-systolic volume (RVESV)

The full workflow includes: - DICOM files to NIfTI conversion and reorientation to standard. - Cardiac segmentation using fully convolutional neural network.

Required inputs

  • One or more cine-MRI cardiac images.
  • The inputs must be labeled as either:
    • cardiac short_axis 4d
    • cardiac long_axis 2ch 4d
    • cardiac long_axis 4ch 4d

Outputs

For each of the input images the following files are outputted:

  • Original image in 4D
  • Segmentation of the original image
  • End-diastolic frame of the original image
  • Segmentation of the end-diastolic frame
  • End-systolic frame of the original image
  • Segmentation of the end-systolic frame
  • Clinical measures in a csv file
  • Report for quality control purposes

Advanced settings

  • DCM2NII:
    • Preferred DICOM to NIfTI conversion tool (drop-down selection):
      • DCM2niix (Default)
      • Mrtrix
      • DCM2nii
      • diffunpack
      • MRIConvert

References

  • [1] W. Bai, et al. Human-level CMR image analysis with deep fully convolutional networks. arXiv:1710.09289. arxiv: 1710.09289

  • [2] S. Petersen, et al. Reference ranges for cardiac structure and function using cardiovascular magnetic resonance (CMR) in Caucasians from the UK Biobank population cohort. Journal of Cardiovascular Magnetic Resonance, 19:18, 2017. DOI: 10.1186/s12968-017-0327-9

 


Create free account now!

Sign Up