FastMRI tutorial question

I’m working on the single coil challenge and became confused by the tutorial, mainly about slices vs. coil. Please help me to understand the tutorial.

From Input 5 in the notebook, we take a SLICE(20) from the volume of slices. As far as I’m concerned the tutorial only works on that one slice the entire notebook. Yet the next function is ‘show_slices()’. Should this not be show_coils? Later on it proceeds to use the ‘root_sum_of_squares()’ function on that slice, and it looks like the function sums up that particular slice (so it sums up the coils into one complete image).

Is the root_sum_of_squares() function only relevant to the multicoil data? My intuition tells me that slices are like the Z height plane dimension and the coils give the XY plane length-width dimension in MRI’s but I am no expert.

Finally, how is the ground-truth reached for our datasets? Is it just our ifft on the fully sampled data? When I do this the mean squared error between this and the ground truth per pixel seems to be around 1e-24 which seems reasonable

Thanks in advance,

Hi Kevin

Thanks for bringing this to our attention. The function should indeed be called show_coils and not show_slices since it displays different coils of the same slice. Anything related to coils is only relevant for the multi-coil data. In particular, the Root Sum of Squares (RSS) operation only applies to multi-coil data. Also, you are correct about the slices being along the Z-dimension.

Reg ground-truth: For single-coil data, you simply have to apply IFFT followed by computing the absolute value of the fully-sampled data to get the ground truth. For multi-coil, you also have to apply RSS after computing the absolute value to combine all the coils. For either dataset, you should apply a center crop (to get the central 320 pixel x 320 pixel region) in the end.