 |
 |
This doesn't seem to have seen a reply previously, but I am running into the same issue and I suspect there may be a bug in fsl_regfilt dealing specifically with number of components. I have a number of subjects in a study with 900 timepoints that fail with segmentation faults on the fsl_regfilt step of ICA_AROMA. The failing subjects generally have many artifact components. During troubleshooting I specifically re-tried those subjects using subsets of their artifact component lists and found that fsl_regfilt fails anytime the number is around 270 or greater (but there's not a specific consistent cutoff across subjects). I have 384GB of memory on my system, and the process doesn't come anywhere close to using too much memory so I don't believe it is a memory limitation issue, but rather a bug somewhere due to number of components being removed. Any thoughts on this? Using FSL version 5.0.10 -Mike -----Original Message----- From: FSL - FMRIB's Software Library [mailto:[log in to unmask]] On Behalf Of Eric Claus Sent: Friday, November 18, 2016 11:27 AM To: [log in to unmask] Subject: [FSL] fsl_regfilt memory issues? Dear FSL experts, I am trying to clean some functional runs using ICA-AROMA, and while this has worked for several runs of data, I am running into problems with a couple runs where the number of components to be regressed out is relatively large (i.e. between 270 and 325 components out of ~520 extracted components). Although ICA-AROMA does not appear to give any error, there is no cleaned data set present when the python script finishes. I have tried running the fsl_regfilt step with the appropriate inputs (fsl_regfilt -i ../filtered_func_data.nii.gz -d melodic.ica/melodic_mix -f `cat classified_motion_ICs.txt` -m melodic.ica/mask.nii.gz -o cleaned_data), but it crashes with "Segmentation fault (core dumped)" on a Linux machine and "Abort trap: 6" on a mac, which both seem to be related to RAM limitations. Is there any way to estimate the amount of RAM needed to get fsl_regfilt to run? My filtered_func_data is 1220 volumes (TR = 460 msec) with 3.024390x3.024390x3 mm^3 voxels. Thanks, Eric ********************************************************** Electronic Mail is not secure, may not be read every day, and should not be used for urgent or sensitive issues