Noise/ghosting in DWI acquistion

1

Hi all,

Trying to debug an issue with our ex-vivo mouse brain acquisition where we get absolutely horrendous ghosting in the slice direction in some samples. Acquisitions are three separate shell acquisitions at b=1000, 2000, and 3000 respectively and 64 directions each.

I’m thinking it has something to do with our gradient scheme. Can you help me interpret the dirstat output?

unringed.mif.gz (b=1005.0187509520202) [ 64 directions ]

  Bipolar electrostatic repulsion model:
    nearest-neighbour angles: mean = 17.7591, range [ 16.7006 - 19.0496 ]
    energy: total = 3681.3, mean = 115.041, range [ 114.698 - 115.232 ]

  Unipolar electrostatic repulsion model:
    nearest-neighbour angles: mean = 17.8297, range [ 16.7006 - 20.1555 ]
    energy: total = 2217.25, mean = 69.289, range [ 57.3957 - 77.9783 ]

  Spherical Harmonic fit:
    condition numbers for lmax = 2 -> 8: [ 1.00267 1.01921 1.05309 1.14711 ]

  Asymmetry of sampling:
    norm of mean direction vector = 0.500489
    WARNING: sampling is strongly asymmetric - this may affect resiliance to eddy-current distortions

unringed.mif.gz (b=2004.0226353884004) [ 64 directions ]

  Bipolar electrostatic repulsion model:
    nearest-neighbour angles: mean = 17.7591, range [ 16.7006 - 19.0496 ]
    energy: total = 3681.3, mean = 115.041, range [ 114.698 - 115.232 ]

  Unipolar electrostatic repulsion model:
    nearest-neighbour angles: mean = 17.8297, range [ 16.7006 - 20.1555 ]
    energy: total = 2217.25, mean = 69.289, range [ 57.3957 - 77.9783 ]

  Spherical Harmonic fit:
    condition numbers for lmax = 2 -> 8: [ 1.00267 1.01921 1.05309 1.14711 ]

  Asymmetry of sampling:
    norm of mean direction vector = 0.500489
    WARNING: sampling is strongly asymmetric - this may affect resiliance to eddy-current distortions

unringed.mif.gz (b=3003.6484145278914) [ 64 directions ]

  Bipolar electrostatic repulsion model:
    nearest-neighbour angles: mean = 17.7591, range [ 16.7006 - 19.0496 ]
    energy: total = 3681.3, mean = 115.041, range [ 114.698 - 115.232 ]

  Unipolar electrostatic repulsion model:
    nearest-neighbour angles: mean = 17.8297, range [ 16.7006 - 20.1555 ]
    energy: total = 2217.25, mean = 69.289, range [ 57.3957 - 77.9783 ]

  Spherical Harmonic fit:
    condition numbers for lmax = 2 -> 8: [ 1.00267 1.01921 1.05309 1.14711 ]

  Asymmetry of sampling:
    norm of mean direction vector = 0.500489
    WARNING: sampling is strongly asymmetric - this may affect resiliance to eddy-current distortions

Thanks in advance, as always

2

Hi Adam,

If you’re observing ghosting in individual DWI volumes, I don’t see how the gradient table could have any responsibility for such.

Being ex vivo, is it a 3D sequence? If 2D, is there any multislice acceleration?

Providing some screenshots demonstrating the issue would likely help.

Cheers
Rob