search

mckib2 / pygrappa

Python implementations of GRAPPA-like algorithms.
https://pygrappa.readthedocs.io/en/latest/
MIT License
64 stars 13 forks source link

bad result in vcgrappa when R=3 #71

Closed kingaza closed 4 years ago

kingaza commented 4 years ago

I am quite confused... the code in basic_vcgrappa.py is modified as 

    kspace[1::3, ...] = 0
    kspace[2::3, ...] = 0

vcgrappa_x3

While the results are quite good when R equals to 2 or 4 vcgrappa_x2 vcgrappa_x4

mckib2 commented 4 years ago

I can recreate this issue. Is this due to the center line of kspace being omitted? I think in general you want to include the center of kspace.

Try:

# R=3
kspace[0::3, ...] = 0
kspace[2::3, ...] = 0

Using the above I get the following: image

kingaza commented 4 years ago

yes, I got the same result as you. Thank you for the hint. But why vcgrappa is so sensitive to the center line? I don't remember the paper mentioned about it...

mckib2 commented 4 years ago

I can't say why it's so sensitive, that's an interesting result. Do you know of any other reference VC-GRAPPA implementations that we could test against? Might also be worth attempting the experiment on real data, simulated data is not always representative of how algorithms perform on real datasets.

kingaza commented 4 years ago

I will try more data later. thanks again. There could be some other solution, i.e. generating calib data away from center. But it seems another topic than virtual coil.

mckib2 commented 4 years ago

Related to #72

kingaza commented 4 years ago

I tested it with real data, and it worked. But the improvement is not always significant as mentioned in the paper. This issue is closed as I have no idea how to evaluate it further. Anyway, it can be traced by the new one #72