Thank you Steve for further clarifications.
Melodic help explains how to apply melodic’s mixture modelling to a statistic image without actually re-running ICA. HCP500_Parcellation_Timeseries_Netmats data provide ‘melodic_IC_sum.nii.gz’ files for each dimensionality of group-ICA parcellation. I wonder, is it possible to use a ‘melodic_IC_sum.nii.gz’ file as input to melodic to carry out mixture modelling with a specified value for mixture-model-based thresholding? For example, by running command lines: echo “1” > grot.txt melodic -i melodic_IC_sum.nii.gz --ICs= melodic_IC_sum.nii.gz --mix=grot.txt -o ‘outDir’ --Oall --report -v --mmthresh=0.5 will this generate properly thresholded images for each component corresponding to the 0.5 posterior probability activation? If yes, how then these images (a number of ‘thresh_zstat*.nii.gz’ files) can be merged into a single ‘*.dscalar.nii’ file analogous to the format of the provided unthresholded Zstats ‘melodic_IC.dscalar.nii’ files that can be viewed in wb_view? Cheers, Bagrat From: Stephen Smith [mailto:st...@fmrib.ox.ac.uk] Sent: Saturday, December 05, 2015 4:21 AM To: Bagrat Amirikian; hcp-users Subject: Re: [HCP-Users] group-ICA spatial maps Hi On 3 Dec 2015, at 20:27, Bagrat Amirikian <amiri...@umn.edu> wrote: Thank you very much Steve for your prompt response! 1. I looked into Beckman IEEE TMI 2004. Is there any way to find out the threshold ‘Z-stats’ for each IC corresponding to the 0.5 posterior probability activation in the framework of Gaussian mixture model? In other words, how to use the provided Z-stats distributions, how to find a reasonable Z-stats threshold for an individual IC? Can such a threshold selection procedure be automated, or one needs to go over all components manually and after visual inspection make a selection? If you run melodic with all output options turned on, you can find various things relating to the mixture-model fitting and related thresholding (for the web report images). However for these HCP group-ICA maps, no thresholding was applied (eg when using these as "parcellations" for generating the node timeseries. 2. I was running two instances of wb_view in Windows 7, the second one loaded with WB_1.0_Tutorial. After closing them and then opening only one instance of wb_view loaded with group-ICA, only a single number appeared. Should this be reported as a bug in wb_view for Windows? Sorry I'm not a wb_view expert so would need someone else to comment on that. Cheers, Steve. 3. Thanks for the explanation. Cheers, Bagrat From: Stephen Smith [ <mailto:st...@fmrib.ox.ac.uk> mailto:st...@fmrib.ox.ac.uk] Sent: Thursday, December 03, 2015 2:43 AM To: Bagrat Amirikian Cc: <mailto:hcp-users@humanconnectome.org> hcp-users@humanconnectome.org Subject: Re: [HCP-Users] group-ICA spatial maps Hi On 3 Dec 2015, at 04:24, Bagrat Amirikian < <mailto:amiri...@umn.edu> amiri...@umn.edu> wrote: Hi, I have downloaded and unpacked HCP500_Parcellation_Timeseries_Netmats data. My question is about group ICA spatial maps, specifically, about melodic_IC.dscalar.nii and melodic_IC_ftb.dlabel.nii files in the groupICA* subdirectories. There is a very short description of these files in the accompanying PDF file: melodic_IC.dscalar.nii ICA spatial maps (unthresholded Zstats); one "timepoint" per map. Grayordinates melodic_IC_ftb.dlabel.nii Summary "find the biggest" labels image for all ICA spatial maps. Grayordinates I have loaded these files generated for 25-dimensional group-ICA to wb_view with corresponding Q1-Q6_R440_midthikness surface files for their inspection. 1. I can appreciate Zstats maps provided for each IC by melodic_IC.dscalar.nii. How were exactly these Zstats computed? Are these regular Z-scores obtained by subtracting the vertex/voxel-wise mean and dividing by the standard deviation, or something else? Not quite - these are "z-stats" as output by MELODIC ICA tool in FSL (see Beckmann IEEE TMI 2004). The simple answer is that the ICA maps are fed into a mixture model to normalise the central/null part of the distribution across voxels. 2. When I click on the map and Information Window pops up, it shows two SCALAR values for the selected vertex. For example: VERTEX CortexLeft: 28165 ANATOMICAL XYZ: -6.28827, 59.2204, -0.919904 CIFTI SCALARS melodic_IC.dscalar.nii: 0.935668 -0.20834 What do these two numbers 0.935668 and -0.20834 actually mean? I though there should be just one number corresponding to the Zstat of the vertex. I don't know - when I try this I only get a single number. Do you have multiple similarly-named images loaded maybe? 3. It appears that melodic_IC_ftb.dlabel.nii provides a hard non-overlapping parcellation and labels each parcel (which could consist of several spatially separated regions) by a color. The number of parcels corresponds to the ICA dimension, 25 in this case. How is this parcellation obtained? How is the set of spatially continuous/diffusive maps of individual ICs give rise to this parcellation with sharp non-overlapping borders and without any gaps between the neighboring parcels? "ftb" - "find the biggest" - at each grayordinate the index number of the ICA component map having the highest value there is used. Cheers, Steve. Thank you very much for your help. Bagrat Bagrat Amirikian, Ph.D. Department of Neuroscience University of Minnesota Medical School Brain Sciences Center Minneapolis Veterans Affairs Health Care System _______________________________________________ HCP-Users mailing list <mailto:HCP-Users@humanconnectome.org> HCP-Users@humanconnectome.org <http://lists.humanconnectome.org/mailman/listinfo/hcp-users> http://lists.humanconnectome.org/mailman/listinfo/hcp-users --------------------------------------------------------------------------- Stephen M. Smith, Professor of Biomedical Engineering Head of Analysis, Oxford University FMRIB Centre FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK +44 (0) 1865 222726 (fax 222717) <mailto:st...@fmrib.ox.ac.uk> st...@fmrib.ox.ac.uk <http://www.fmrib.ox.ac.uk/~steve> http://www.fmrib.ox.ac.uk/~steve --------------------------------------------------------------------------- <http://smithinks.net/> Stop the cultural destruction of Tibet --------------------------------------------------------------------------- Stephen M. Smith, Professor of Biomedical Engineering Head of Analysis, Oxford University FMRIB Centre FMRIB, JR Hospital, Headington, Oxford OX3 9DU, UK +44 (0) 1865 222726 (fax 222717) st...@fmrib.ox.ac.uk http://www.fmrib.ox.ac.uk/~steve --------------------------------------------------------------------------- <http://smithinks.net> Stop the cultural destruction of Tibet _______________________________________________ HCP-Users mailing list HCP-Users@humanconnectome.org http://lists.humanconnectome.org/mailman/listinfo/hcp-users
