Dear Michele,
the Gaussian Process inside eddy sometimes struggle to deal with data that has been denoised. The reason is that the denoising introduces an almost total correlation between the estimates of two of the hyper-parameters, so the optimisation essentially wanders around in a completely flat valley.
I am a little surprised that it crashes because it runs out of memory. Are you sure that this is what happens, and that it isn't just that the job gets killed because it has used up too much CPU time? Anyway, from your perspective it doesn't really matter since either way it won't run to completion.
My "solution" would be to not use denoising. For "good" data denoising has very little impact on the end results, and we rarely/never use it. I guess that one can make a case for denoising if one has "old-style" data with a limited number of volumes/directions. I don't have a sure-fire solution if you want to persist with denoising, but I would suggest trying:
1. --hpcf=MML (use a different cost-function to estimate hyper-parameters)
2. --nvoxhp=10000 (use a much larger sample of voxels to estimate hyper-parameters from)
You can also combine 1 and 2. There is a third option, which is to explicitly set the hyper-parameters yourself. But that is a bit of a nuclear option, and I would need to know more about your data to be able to advice you on that.
Best regards Jesper
On 08/08/2022, 09:32, "FSL - FMRIB's Software Library on behalf of Michele Valotti" <[log in to unmask] on behalf of [log in to unmask]> wrote:
Dear FSL experts,
I am trying to process a denoised DTI image with eddy_openmp, but eddy crashes when trying to estimate the hyperparameters of the prediction maker model (the process is killed after using all the available RAM, more than 12GB)
The original DTI is denoised using MRTrix's dwidenoise and the result of dwidenoise is thresholded at 0 with fslmaths so that the image does not have any negative values
Has anyone encountered a similar problem? Or does anyone have any suggestions on how to debug this further?
Thank you,
Michele Valotti
########################################################################
To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=FSL&A=1
This message was issued to members of www.jiscmail.ac.uk/FSL, a mailing list hosted by www.jiscmail.ac.uk, terms & conditions are available at https://www.jiscmail.ac.uk/policyandsecurity/
########################################################################
To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=FSL&A=1
This message was issued to members of www.jiscmail.ac.uk/FSL, a mailing list hosted by www.jiscmail.ac.uk, terms & conditions are available at https://www.jiscmail.ac.uk/policyandsecurity/
