COnstrained Data Extrapolation (CODE): A new approach for high definition vascular imaging from low resolution data Journal Articles uri icon

  • Overview
  • Research
  • Identity
  • Additional Document Info
  • View All


  • PURPOSE: To introduce a new approach to reconstruct high definition vascular images using COnstrained Data Extrapolation (CODE) and evaluate its capability in estimating vessel area and stenosis. MATERIALS AND METHODS: CODE is based on the constraint that the full width half maximum of a vessel can be accurately estimated and, since it represents the best estimate for the width of the object, higher k-space data can be generated from this information. To demonstrate the potential of extracting high definition vessel edges using low resolution data, both simulated and human data were analyzed to better visualize the vessels and to quantify both area and stenosis measurements. The results from CODE using one-fourth of the fully sampled k-space data were compared with a compressed sensing (CS) reconstruction approach using the same total amount of data but spread out between the center of k-space and the outer portions of the original k-space to accelerate data acquisition by a factor of four. RESULTS: For a sufficiently high signal-to-noise ratio (SNR) such as 16 (8), we found that objects as small as 3 voxels in the 25% under-sampled data (6 voxels when zero-filled) could be used for CODE and CS and provide an estimate of area with an error <5% (10%). For estimating up to a 70% stenosis with an SNR of 4, CODE was found to be more robust to noise than CS having a smaller variance albeit a larger bias. Reconstruction times were >200 (30) times faster for CODE compared to CS in the simulated (human) data. CONCLUSION: CODE was capable of producing sharp sub-voxel edges and accurately estimating stenosis to within 5% for clinically relevant studies of vessels with a width of at least 3pixels in the low resolution images.


  • Song, Yang
  • Hamtaei, Ehsan
  • Sethi, Sean K
  • Yang, Guang
  • Xie, Haibin
  • Haacke, Mark

publication date

  • December 2017