NKI-AI/direct

JOSS manuscript comments

sairamgeethanath opened this issue · 2 comments

sairamgeethanath commented

Is your feature request related to a problem? Please describe.
Clarifications and request for edits

  1. Accelerated MRI Reconstruction section - Please clarify the benefits of going from iterative (gradient approaches), analytical approaches (such as compressed sensing) to DL implementations - such as the savings in recon. time - iterative vs inference

  2. Retrospectively undersampled vs. prospectively undersampled - It is not clear to me at this time whether DIRECT has been used to reconstruct undersampled data from an MR scanner or mostly reconstructed from retrospectively undersampled k-space data.

  3. Citations - It will be best to include the seminal works of Candes, Romberg, Tao and Lustig et al to fully acknowledge their contributions to compressed sensing and subsequent application to MRI.

  4. Knee example image - The authors may consider including the image shown on the documentation page after clarifying if the CS implementation was custom or from another open source repository.

georgeyiasemis commented

Thanks @sairamgeethanath for your comments. I will answer all of them as soon as possible!

georgeyiasemis commented

Hi @sairamgeethanath. Regarding your comments:

  1. Accelerated MRI Reconstruction section - Please clarify the benefits of going from iterative (gradient approaches), analytical approaches (such as compressed sensing) to DL implementations - such as the savings in recon. time - iterative vs inference

We have added references that might clarify this.

  1. Retrospectively undersampled vs. prospectively undersampled - It is not clear to me at this time whether DIRECT has been used to reconstruct undersampled data from an MR scanner or mostly reconstructed from retrospectively undersampled k-space data.

DIRECT (currently) can be used with retrospectively sub-sampled data or fully-sampled data that can be (retrospectively) sub-sampled using our sampling schemes. DIRECT is only intended for research.

  1. Citations - It will be best to include the seminal works of Candes, Romberg, Tao and Lustig et al to fully acknowledge their contributions to compressed sensing and subsequent application to MRI.

Thanks for the suggestion. We have included the relevant references.

  1. Knee example image - The authors may consider including the image shown on the documentation page after clarifying if the CS implementation was custom or from another open source repository.

Clarified.