Freesurfer May 2008

freesurfer@nmr.mgh.harvard.edu

63 participants
95 discussions

out of memory issue with selxavg3-sess
by Schadwinkel, Stefan 05 May '08

05 May '08

Hi, I've got a general question about selxavg3-sess. In short, when I try to run the analysis, MATLAB exits with an out of memory issue. My previous studies used a TR of 8 seconds, about 120 Acquisitions per run, 5 runs, and everything worked just fine. I now ran two different pilot studies with a TR of 1 second, 4 runs with 840 Acquisitions each. On this data, sexlavg3-sess causes MATLAB to exit with an out of memory message. My main question is, is this just too much data to process or is there a way to configure MATLAB/change the processing scripts to free more memory or use more swap space to allow processing or could it be an issue with the analysis setup creating way too much data? Here are some more details: I use FS-Fast from the freesurfer 4.0.1. rh9 32bit package. When I run the analysis on my machine (x86_64 linux with 1 GB RAM), I get following error, followed by a Java VM fatal exception: ----------- . . OLS Residual Pass run 1 t= 0.0 Loaded /home/WELTRAUM/STUDY2_TR1_TEST/GMOC/bold/008/gm_run_er.par as a par2 file. ??? Error using ==> unknown Out of memory. Type HELP MEMORY for your options. Error in ==> fast_selxavg3 at 413 rsserun = sum(rrun.^2); >> Unable to open file for Java VM logging Opening log file: /home/stefan/java.log.10568 pure virtual method called terminate called without an active exception . -------------------- I assumed, 1 GB might just not enough memory, so I tried the same analysis with just one run and turned off as much as possible of the other options (temporal whitening, motion regressors, etc.). Yet, looking at the analysis.conf resulting from that, I am not sure that turning it off in the mkanalysis-sess -gui had an effect with all options. With that configuration and just one run, the memory problem persisted. Next, I tried it on another linux box with 2GB RAM (same freesurfer version), with similar errors during whitening: ---------------------- . . Whitening Computing whitening matrices ??? Error using ==> zeros Out of memory. Type HELP MEMORY for your options. Error in ==> fast_selxavg3 at 501 W = zeros(ntptot,ntptot,flac0.acfbins); >> ------------------------------------------ ERROR: fast_selxavg3() failed\n ------------------------------------- Next, still assuming not enough memory to be the problem, I tried it on a Mac Pro with Mac OS X 10.5 and 4 GB RAM using the current corresponding freesurfer version (freesurfer-Darwin-leopard-i686-stable-pub-v4.0.3-full.dmg). Still, similar memory problems persisted: ----------------- . . Whitening Computing whitening matrices MATLAB(21818,0xb0801000) malloc: *** mmap(size=913952768) failed (error code=12) *** error: can't allocate region *** set a breakpoint in malloc_error_break to debug ??? Out of memory. Type HELP MEMORY for your options. Error in ==> fast_selxavg3 at 505 W = zeros(ntptot,ntptot,flac0.acfbins); >> ------------------------------------------ ERROR: fast_selxavg3() failed\n ------------------------------------ Now, although it could be just too much data to compute, I think I'm not the only person who wishes to analyse data of that size, as it is not really that much. My session consists of 14 Minutes per run with TR=1 and about 8 Blocks per run with an event-related paradigm, resulting in about 16 roughly 32s long blocks to fit with the GLM (per run). Maybe, I simply made some errors in setting up the analysis. Here's one paradigm file of the first run and the analysis.conf: ------------------ 16.000000 0 68.235200 1 99.998800 2 114.000000 0 169.248400 1 201.487000 2 217.000000 0 264.894600 1 300.822200 2 318.000000 0 331.545500 1 356.612700 1 375.654700 2 387.401400 1 410.274400 2 413.000000 0 453.019700 1 508.000000 0 515.996200 2 552.006400 1 603.000000 0 603.923700 2 652.629700 1 699.231800 2 702.000000 0 709.726400 1 753.749600 1 781.971400 2 804.000000 0 ----------------------- ----------- analysis.conf ----------------- cat analysis.cfg -delay 0.000000 -TER 1.000000 -polyfit 2 -gammafit 2.250000 1.250000 -gammaexp 2.000000 -timewindow 24.000000 -prestim 4.000000 -autostimdur ------------------------------------------------ ------------ analysis.info ---------------------- cat analysis.info analysis gmoc_analysis_1 TR 1.000000 fsd bold funcstem fmcstc runlistfile study_runs tpexclude parname gm_run_er.par designtype event-related nconditions 2 Condition 1 Condition01 Condition 2 Condition02 -------------------------------------------------- Thanks in advance, your comments on this issue and the size of the fMRI session is greatly appreciated. Best regards, Stefan Schadwinkel

2 1

thickness file data structure
by Yaroslav Halchenko 03 May '08

03 May '08

Dear Freesurfers, First thanks everyone who helped me out with reading-in .mgh data from within Python (especial thanks to Krish). Now I wonder what actually I got ;-) and please pardon my ignorance When I read in thickness file in I get an array of (23406, 7, 58) dimensionality, where 58 makes sense to me since we have 58 subjects. P.S. I've got the same dimensions whenever I didn't use Krish's python binding but converted to minc and loaded using NetCDF module (in python though again, but python's numpy.ndarray dimensionality is not limited by short) What I wonder about is that why 23406 and 7? ok - the product is equal to the surface number of vertices:23406*7=163842, so it seems that I can simply collapse those two dimensions (which one goes first?) but why is that done so? according to http://surfer.nmr.mgh.harvard.edu/fswiki/FsTutorial/MghFormat and mri.h in particular I don't see any limit on width/hight to be of short (not int) type and int (4 bytes) should be good enough to store full index. Is that due to some historical perspective? or am I just completely missing smth? Thanks in advance for the hints! -- Yaroslav Halchenko Research Assistant, Psychology Department, Rutgers-Newark Student Ph.D. @ CS Dept. NJIT Office: (973) 353-5440x263 | FWD: 82823 | Fax: (973) 353-1171 101 Warren Str, Smith Hall, Rm 4-105, Newark NJ 07102 WWW: http://www.linkedin.com/in/yarik

4 4

mris_preproc: multiple volumes from one subject
by Jascha Swisher 02 May '08

02 May '08

Hello all - I"ve come across a slight issue when using mris_preproc to register multiple volumes from one subject onto a common surface. The temporary files created by mri_surf2surf have the name $tmpdir/$subj.mgh, which works just fine when each volume being transformed comes from a different subject. However when transforming multiple volumes from the same subject, each file converted overwrites the previous. The result is that the final call to mri_concat concatenates together multiple copies of the same volume, rather than the separate volumes specified as input. A simple fix is to change the name of the temp files from $tmpdir/$subj.mgh to $tmpdir/$subj.$tinc.mgh, where $tinc is incremented after each volume. I've attached a diff against version 1.37.2.2 of mris_preproc (from Freesurfer 4.0.2) that implements this. cheers... -- jds 82,83d81 < @ tinc = 0; # JDS < 209,210c207 < set tval = $tmpdir/$subj.$tinc.mgh #JDS: added tinc to avoid same-subject collisions < @ tinc = $tinc + 1 # JDS: increment tinc --- > set tval = $tmpdir/$subj.mgh -- Jascha D Swisher Research Associate - Tong Lab Department of Psychology Vanderbilt University 422 Wilson Hall 111 21st Ave S Nashville TN 37203 office: 615.322.2835 cell: 617.584.9881

2 1

about nucleus of thalamus segmentation tool
by abel cruz vadell 02 May '08

02 May '08

Hello I have read in the freeSurferWiki (http://surfer.nmr.mgh.harvard.edu/fswiki/Thalamus#head-f0c8409c96fe3e19d4b7…) about a project to segment the different nucleus of the thalamus. The page have ex-vivo and in-vivo Data Locations but I can´t find them in FreeSurfer. Does FS v 4.02 have these tool? How could I use it? Thanks Abel ______________________________________________ Enviado desde Correo Yahoo! La bandeja de entrada más inteligente.

2 1

vertex correspondence
by Mishkin Derakhshan 01 May '08

01 May '08

Hi, I know that at some point in the pipeline, the surfaces are registered to a common sphere, and thus I'm hopping there is a one-to-one vertex correspondance between subjects. My question is, let's say I look at vertex X, on subject1, how do I find out the corresponding vertex location on subjectN. thanks, mishkin

2 1

2026

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

Freesurfer May 2008