Frequently Asked Questions - FAQ

1. There are different lengths in the slice-encode direction of the data. They vary between 170 and 180. However, only one kind of sampling mask is provided with length 170. How should I proceed? Do I need to create new sampling masks?

(Updated 27 Feb 2020) Answer: We included new sampling masks in the github repository to account for the different dimensions in the slice-encode direction.

2. I noticed some residual signal after 85 % of the k-space slices in the slice-encode direction. What does it mean?

Answer: We used vendor supplied tools (Orchestra Toolbox; GE Healthcare) to load the raw data and save as HDF5 files. Unfortunately these tools are a black-box and we do not have access to all details of their code. We assume this residual comes from the Fourier Transform used to put the data in the x-ky-kz format. You should explicitly do the zero-filling in the slice-encode direction. Sample code is shown below. The zero-filling is already done for the test sets provided.

3. I found some k-space centering issues in some samples of the the test. Is it supposed to be like that?

Answer: No. We thank Mr. Nianmao Du for pointing this issue to our team. The fixed test set should be available for download on 02 March 2020. If you have the old test set, you need to download it again. The new test set should be used on your submission!

4. I am having issues downloading the data (GDrive disk quota error, link expired, etc.). What should I do?

Answer: Our dataset is now available for download from OneDrive, GDrive, and the Canadian Open Neuroscience Platform. If your old download links expired, just submit the download form again. If you still have difficulties accessing the data, contact roberto.medeirosdeso@ucalgary.ca

5. My method requires the sampling masks to work and I noticed the masks for the test set are not provided. What should I do?

Answer: You can compute the sampling mask from the test set provided. Let's say you loaded your test k-space in a variable called kspace. To recover the corresponding mask, you can just do: sampling_mask = ~( np.abs(kspace).sum( axis = (0, -1) ) == 0). I am assuming you are using Python and NumPy, but you can code this in any language.

Updated: 17 June 2020