Indeed melodic is a single threaded application that takes quite some time to run the PCA portion of the ICA, even on the CIFTI data. Hopefully one day it will be multi-threaded. I think you would have to use fewer subjects or a matlab-based MIGP (this is what I use).
You can actually run the whole ICA in matlab using the FAST ICA toolbox, but this loses you some of the nice things that melodic does for you (e.g. sorting components based on strength and flipping them all to be positive along with mixture modeling for z-stats). Peace, Matt. From: Jenny Gilbert <[email protected]<mailto:[email protected]>> Date: Tuesday, March 28, 2017 at 11:08 AM To: Matt Glasser <[email protected]<mailto:[email protected]>>, "[email protected]<mailto:[email protected]>" <[email protected]<mailto:[email protected]>> Subject: Re: [HCP-Users] ICA failure with volumetric ICA-FIX denoised files Hi Matt, I appreciate your help with working through ICA using HCP data. Following your recommendation, I converted CIFTI files to NIFTI and am trying to run through ICA, but I am still experiencing problems. Running ICA for 62 participants (rest1_RL session) on a high power computing server with 72 GB mem and a time limit of 5 days, the ICA times out, and cannot pass the first stage of removing the mean image and normalizing by the voxel-wise variance. This is the script I'm using: melodic -i /projects/niblab/scripts/twin/input_ica.txt --migp --sep_vn -o /projects/niblab/scripts/twin/ica_out -v --nobet --bgthreshold=10 --tr=1 -d 50 --report --mmthresh=0.5 --Oall >From your experience, should we expect ICA to take 5+ days to run, or is there >a way to speed this up? Thanks for your help! Jenny On Tue, Mar 7, 2017 at 3:48 PM, Glasser, Matthew <[email protected]<mailto:[email protected]>> wrote: Hi Jenny, Indeed you can use wb_command -cifti-convert -to-nifti to convert the files to NIFTI, run melodic, do dual regression, etc, and then when you have your output maps, you convert them back to CIFTI with wb_command -cifti-convert -from-nifti and view them in Workbench. The only thing to keep in mind is that spatial relationships will not be 3D volumetric in these converted CIFTI files, so things like smoothing would not be appropriate to do on them. ICA or dual regression doesn’t use spatial neighborhood information (though you will need to select your own ICA dimensionality). Peace, Matt. From: Jenny Gilbert <[email protected]<mailto:[email protected]>> Date: Tuesday, March 7, 2017 at 2:08 PM To: Matt Glasser <[email protected]<mailto:[email protected]>> Subject: Re: [HCP-Users] ICA failure with volumetric ICA-FIX denoised files Hi Matt, Thanks for your help! How do I complete ICA the CIFTI MSMAII files? Is the first step to convert the files to NIFTI format in workbench? Also, I planned to complete dual regression with the ICA generated components, so how does dual regression work with CIFTI format data? Can all of this be done by converting CITFI to NIFTI, and then running the usual commands? I'm new to workbench, and would love all the help that you experts can offer! Thank you so much! Jenny On Tue, Mar 7, 2017 at 2:05 PM, Glasser, Matthew <[email protected]<mailto:[email protected]>> wrote: I think there is a bug in melodic, which I also encountered. I would report this to the FSL list, but it is likely the issue will be fixed in the next FSL release. Also, if you are running melodic in the volume and combining across subjects, this is not recommended because much of the brain is not well aligned in the volume. It is instead recommended that you use the CIFTI data that has been aligned with MSMAll. ICA gives very different results depending on whether the data are well aligned or not. Peace, Matt. From: <[email protected]<mailto:[email protected]>> on behalf of Kajsa Igelström <[email protected]<mailto:[email protected]>> Date: Tuesday, March 7, 2017 at 12:40 PM To: "[email protected]<mailto:[email protected]>" <[email protected]<mailto:[email protected]>> Subject: Re: [HCP-Users] ICA failure with volumetric ICA-FIX denoised files Hi Jenny and others! I'm also unable to run these files through Melodic. The most common problem is that it doesn't converge, and under some circumstances it seems like Melodic just stops in the middle of normalizing individual files, at points that are not logical to me, or reproducible. Best, Kajsa On Tue, Mar 7, 2017 at 12:24 PM, Jenny Gilbert <[email protected]<mailto:[email protected]>> wrote: Hello HCP! I am trying to complete ICA using the volumetric ICA-FIX de noised data (rfMRI_REST1_RL_hp2000_clean.nii.gz files), and melodic keeps failing. It looks like melodic gets hung up on some participants cannot progress while removing the mean image and normalizing by the voxel-wise variance. Has anyone come across this same problem? Any advice on how to fix? Thank you! Jenny _______________________________________________ HCP-Users mailing list [email protected]<mailto:[email protected]> http://lists.humanconnectome.org/mailman/listinfo/hcp-users _______________________________________________ HCP-Users mailing list [email protected]<mailto:[email protected]> http://lists.humanconnectome.org/mailman/listinfo/hcp-users ________________________________ The materials in this message are private and may contain Protected Healthcare Information or other information of a sensitive nature. If you are not the intended recipient, be advised that any unauthorized use, disclosure, copying or the taking of any action in reliance on the contents of this information is strictly prohibited. If you have received this email in error, please immediately notify the sender via telephone or return mail. _______________________________________________ HCP-Users mailing list [email protected]<mailto:[email protected]> http://lists.humanconnectome.org/mailman/listinfo/hcp-users -- Jenny Gilbert [email protected]<mailto:[email protected]> | 443.846.3687<tel:(443)%20846-3687> ________________________________ The materials in this message are private and may contain Protected Healthcare Information or other information of a sensitive nature. If you are not the intended recipient, be advised that any unauthorized use, disclosure, copying or the taking of any action in reliance on the contents of this information is strictly prohibited. If you have received this email in error, please immediately notify the sender via telephone or return mail. -- Jenny Gilbert [email protected]<mailto:[email protected]> | 443.846.3687 ________________________________ The materials in this message are private and may contain Protected Healthcare Information or other information of a sensitive nature. If you are not the intended recipient, be advised that any unauthorized use, disclosure, copying or the taking of any action in reliance on the contents of this information is strictly prohibited. If you have received this email in error, please immediately notify the sender via telephone or return mail. _______________________________________________ HCP-Users mailing list [email protected] http://lists.humanconnectome.org/mailman/listinfo/hcp-users
