Dear eddy experts,
We've managed to get eddy_cuda to run with old gradient echo field map scans for susceptibility correction (both FSL versions 5.0.11 & 6.0.1) but curiously 9 of 33 subjects in our data set produce segmentation errors or finish with bad output volumes (individual volumes pulled in different directions, not properly aligned etc). In all 'bad' cases, the mss error increases throughout the dwi volume-to-volume iterations and presumably leads to the segmentation error where eddy can't find a suitable reference volume (the subjects that 'finish' still have very high mss values). After reading about the possible causes of segmentation errors on here we tried running the tool without slice-to-volume correction but that didn't lead to any improvement, probably because the problem already occurs during the volume-to-volume correction stage.
We also tried to find out whether the DWI runs and/or gradient echo field maps submitted to eddy_cuda show differences for the successful vs 'bad' subjects but found only small differences in terms of the overall range of values - the 'bad' subjects' input DWI runs have smaller ranges of values in the DWI volumes (not the b0s). The gradient echo field maps don't seem to be different in any meaningful way. Notably, all 'bad' subjects finish without any problems when running eddy_cuda without susceptibility correction (no topup or field input).
Does anyone have any ideas as to why this might be happening? Does eddy apply gradient echo field maps differently from topup inputs? We suspect that somehow problems in the application of the field maps lead to a very high initial error estimates that the tool is unable to recover from.
Note that we applied 2D Gaussian smoothing with a sigma of 4 (-s flag in fugue) to the field maps produced using fsl_prepare_fieldmap (SIEMENS data) and scaled them to Hertz prior to use with eddy_cuda. We found this smoothing value to lead to the best possible unwarping (in terms of matching the T1 volumes) and field map smoothing was applied for all subjects so this shouldn't be the cause of the problem (we get the same errors when trying to run eddy_cuda with non-smoothed field maps).
Thank you very much for any suggestions you might have!
Matthias
########################################################################
To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
|