Freesurfer

freesurfer@nmr.mgh.harvard.edu

5 participants
30076 discussions

What is the latest 'supported' CUDA version?
by June Kang 17 Nov '16

17 Nov '16

Dear freesurfer developers, Anybody know which CUDAversion is the latest one supported by freesurfer? I just found the thread below. it's written in 2014, but experiences same error.Successful detectcuda command, but core dump during very first step of recon-all. https://www.mail-archive.com/freesurfer@nmr.mgh.harvard.edu/msg36769.html Now I am trying to remove CUDA 8 and installing CUDA 5.5. Next will be the CUDA5.0If anybody knows what is the 'supported' or 'successful' late cuda version, please let me know.Bests,

2 2

load_analyze_hdr.m: incomplete hdr struct with modified hdr.hist.originator field
by Kathryn I. Alpert 17 Nov '16

17 Nov '16

Dear FS developers, Firstly, let me thank you for FreeSurfer: I use FS almost daily, and it is fantastic. Now, onto the bug report: When reading an analyze header with a modified hdr.hist.originator field, load_analyze_hdr.m sometimes “runs out” of bytes and leaves the last field (hdr.hist.smin) empty. When I pass the hdr struct to write_analyze_hdr.m, an invalid, 344 byte header is written (due to this empty hdr.hist.smin field). I will explain how to replicate the issue after first explaining what I think is happening and how I think it can be fixed. I notice that on line 119 of load_analyze_hdr.m, the comment reads “Is this correct? Should be unsigned char”; so, perhaps someone had this issue previously, or hand an inkling that the code was problematic. Because of varying character encodings, fscanf(fp,’%c’,10) may read more than 10 bytes, which means all later fields will not be set properly. The fix is simple, and there are a few options: 1) When opening the hdr file, force ‘US-ASCII’ encoding so each character is precisely 1 byte: (lines 59 and 63) fp = fopen(hdrfile,’r’,’b’,’US-ASCII’); (line 75) fp = fopen(hdrfile,’r’,’l’,’US-ASCII’); *This is the preferred solution as it preserves the “readability” of the field. It also would apply to all the other instances of fscanf.* 2) use fread with uchar (line 119) hdr.hist.originator = fread(fp,10,’uchar’); 3) use fread with char*1 (line 119) hdr.hist.originator = fread(fp,10,’char*1’); I noticed this bug in the following context. I received a nifti file with an incorrect resolution in the header. When I corrected the resolution using matlab (load_nifti.m -> correct nii.pixdim -> save_nifti.m), converted the nifti image to analyze (mri_convert -ot analyze in.nii out.img), loaded the resulting analyze header in matlab with hdr = load_analyze_hdr(‘out.hdr’), and tried to run write_analyze_hdr(hdr,’new.hdr’), the new.hdr file was an invalid, 344 byte header file. I would be happy to send along sample images if needed. I am hoping that you will add this fix to later versions of FreeSurfer, and perhaps consider updating 5.3.0, as well. Thanks and best wishes, Kate *Kathryn I. Alpert* Software Developer, Neuroimaging and Applied Computational Anatomy Lab Northwestern University Feinberg School of Medicine 312.725.2832 k-alpert(a)northwestern.edu [image: Feinberg School of Medicine] *Educate, Discover, Improve Health*

1 0

CUDA problem in ubuntu with M4000 card.
by June Kang 17 Nov '16

17 Nov '16

Dear Freesurfer + CUDA users, I am trying to run cuda-enabled recon-all procedure.My system is Ubuntu 14.04, NVIDIA M4000, with freesurfer stable beta 20161028 + CUDA 8.0 driver. I get the error message like below. $ recon-all -all -s DEP030 -use-gpu Testing for CUDA device: nvcc: NVIDIA (R) Cuda compiler driver Copyright (c) 2005-2012 NVIDIA Corporation Built on Fri_Sep_21_17:28:58_PDT_2012 Cuda compilation tools, release 5.0, V0.2.1221Driver : 8.0 Runtime : 8.0Acquiring CUDA device Using default device CUDA device: Quadro M4000 Segmentation fault (core dumped) Linux DTI 3.13.0-101-generic #148-Ubuntu SMP Thu Oct 20 22:08:32 UTC 2016 x86_64 x86_64 x86_64 GNU/Linuxrecon-all -s DEP030 exited with ERRORS at 2016. 11. 18. (금) 07:54:00 KSTFor more details, see the log file To report a problem, see http://surfer.nmr.mgh.harvard.edu/fswiki/BugReporting the detectcuda command seems can get M4000 properly.Detecting CUDA... There are 0 devices supporting CUDA:Device 0: "Quadro M4000" CUDA Driver Version: 8.0 CUDA Runtime Version: 8.0 CUDA Capability Major revision number: 5 CUDA Capability Minor revision number: 2 Total amount of global memory: 4219863040 bytes Number of multiprocessors: 13 Number of cores: 572 Total amount of constant memory: 65536 bytes Total amount of shared memory per block: 49152 bytes Total number of registers available per block: 65536 Warp size: 32 Maximum number of threads per block: 1024 Maximum sizes of each dimension of a block: 1024 x 1024 x 64 Maximum sizes of each dimension of a grid: 2147483647 x 65535 x 65535 Maximum memory pitch: 2147483647 bytes Texture alignment: 512 bytes Clock rate: 0.77 GHz Concurrent copy and execution: Yes Run time limit on kernels: Yes Integrated: No Support host page-locked memory mapping: Yes Compute mode: Default (multiple host threads can use this device simultaneously)

1 0

Problem with wm/pial surfaces
by Sneha Pandya 17 Nov '16

17 Nov '16

Hi team, I am having problem with almost all the subjects in which wm/pial surfaces are not estimated correctly. These subjects are suffering from Corticobasal degeneration or progressive supranuclear palsy. I tried running sample of these subjects with multiple T1s as input (as we had them) and with additional parameters: -FLAIR, -FLAIRpial, -norm2-b 20, -norm2-n 5, -3T, and -bigventricles. These are either mprage or spgr sequences with 1x1x1 resolution. Flair input did not help instead made pial surfaces worst and convoluted (blue) as opposed to *.woFLAIR.pial surfaces (magenta), and even with addition of other flags it did not further improve estimation of wm/pial surfaces as seen in attached screenshots. The only option being adding control points. But these are about 200+ subjects, and I am not sure if to trust the manual edits on such a big scale. Can anyone please suggest any other tweaks I could try? [cid:01f72aae-1e98-4f12-bcf1-d2e3639dc992] [cid:d1acff28-76d5-4cbf-aa2b-8fb55a07abfe] Thanks, Sneha

2 2

creating label that is the overlap between two other labels
by Harry R. Smolker 17 Nov '16

17 Nov '16

Hello, I have two labels (generated from qdec analyses) that have some overlapping vertices. I am interested in creating a label that is just the overlap of the two aforementioned labels. Is there a way to do this using Free Surfer tools? Thanks, Harry Smolker

2 1

Problem using mri_cor2label
by Sophie Wohltjen 17 Nov '16

17 Nov '16

Hi, I am trying to create cortical surface labels from several brain masks. All are .mgz files, separated by hemisphere, and aligned to fsaverage. Here is the command I entered to create the labels: mri_cor2label --i <filename>.mgz --l <labelname>.label --id 1 When I run this command, I successfully create the label, but I am unable to open the label in tksurfer. I've tried opening both from the GUI menu and from command line, and each time I get the following error: -- ******************** 0 nonzero vertices found ******************** label stat field identically zero - setting to 1 13245 unassigned vertices in label - building spatial LUT... assigning vertex numbers to label... MatrixMultiply: m2 is null! -- And then tksurfer closes. I'm not sure why this is happening -- any help would be greatly appreciated! Thanks, Sophie

2 3

recon-all error using GPU in OS X Sierra
by June Kang 17 Nov '16

17 Nov '16

Hello FreeSurfer Developers, I'm attempting to run recon-all using -use-gpu option in Mac Pro 5,1 with OS X 10.12.1.I am using GTX980 rom-flashed edition, and it supports CUDA properly with other tools.Current cuda version is 8.0, and cudadetect command works perfect. Device 0: "GeForce GTX 980" CUDA Driver Version: 8.0 CUDA Runtime Version: 5.0 CUDA Capability Major revision number: 5 CUDA Capability Minor revision number: 2 Total amount of global memory: 4294770688 bytes Number of multiprocessors: 16 Number of cores: 448 Total amount of constant memory: 65536 bytes Total amount of shared memory per block: 49152 bytes Total number of registers available per block: 65536 Warp size: 32 Maximum number of threads per block: 1024 Maximum sizes of each dimension of a block: 1024 x 1024 x 64 Maximum sizes of each dimension of a grid: 2147483647 x 65535 x 65535 Maximum memory pitch: 2147483647 bytes Texture alignment: 512 bytes Clock rate: 1.22 GHz Concurrent copy and execution: Yes Run time limit on kernels: Yes Integrated: No Support host page-locked memory mapping: Yes Compute mode: Default (multiple host threads can use this device simultaneously) However, when I run recon-all -all -s DEP030 -qcache -use-gpu, I face repeatedly the error reading transform from transforms/talairach.m3zERROR: cannot find or read transforms/talairach.m3zNo such file or directory I found it might be caused since talairach.m3z couldn’t created. I also found an error above.Cuda error: MRIconvolveKernelX failed! in file 'mriconvolve_cuda.cu' in line 536 : unspecified launch failure.freeing gibbs priors...done.average std[0] = 5.0 I also attach full log file too. Thanks.

2 1

Re: [Freesurfer] tracula reinit
by Lars M. Rimol 17 Nov '16

17 Nov '16

Hi Anastasia! That makes sense, thank you! So the question is how do I evaluate the noisiness of a path? 1) Most of my isosurfaces look a lot noisier than your example on the wiki. There's obviously no hard and fast rule here but is there any way to gauge the noisiness of a reconstructed path? And can that information be used for quality control? 2) Another source of information on outliers is the log files that are produced when running -trac-all -stat (based on the shape of the tract). It doesn't seem to flag tracts that are (almost) completely missing (which are spotted by visual inspection in Freeview) - but they may fall under the minimum threshold and therefore not be evaluated? Would you recommend automatically excluding the subjects that are flagged as outliers here? >Hi Lars - A path with fewer control points will generally be smoother. So >if the tract is pretty much a straight line, you might get a noisier >result if you increase the number of control points. But if the tract is >more convoluted than a straight line, then you need more control points to >define it accurately. So it's hard to predict the result of increasing the >control points in general, it depends on the specific case. >BTW, if you change the number of control points, you don't need to use the >reinit variable - that's only for rerunning with the same control points >as before. > >Best, >a.y sincerely yours, Lars M. Rimol, PhD Senior researcher, Norwegian Advisory Unit for functional MRI Department of Radiology, St. Olav's University hospital, 7006 Trondheim, Norway

1 0

[Announcement] Freesurfer v6.0 beta release and testing
by Z K 16 Nov '16

16 Nov '16

Hello All, The Freesurfer development team seeks beta testers for the beta version of Freesurfer v6.0. This is a beta release to the Freesurfer community to help identify bug fixes and other unknown issues before the stable release of Freesurfer v6.0. PLEASE REMEMBER THAT THIS IS NOT AN "OFFICIAL" RELEASE AND CHANGES MAY BE MADE THAT WOULD REQUIRE THAT YOU REANALYZE YOUR DATA. Download and installation instructions can be found here: https://surfer.nmr.mgh.harvard.edu/fswiki/DownloadAndInstall6.0-beta We ask beta testers to submit any problems or issues they may encounter to the Freesurfer support list. And follow our protocols for submitting a bug report. http://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer https://surfer.nmr.mgh.harvard.edu/fswiki/BugReporting It would also be helpful to know if the issue you encounter exists in v5.3 as well. We anticipate an approximately 2 week beta testing period. Thanks for your help, -Freesurfer Development Team

1 0

Manual vs Automatic Cortical Thickness
by Donna Cosgrove 16 Nov '16

16 Nov '16

Hi all, I was hoping to get some advice about FS edits, namely for cortical thickness analysis. We have just under 300 subjects and have run recon-all, and extracted the outliers. I am currently visually inspecting each image slice by slice to look for any artefacts/abnormalities; however, I am wondering if it is better NOT to manually edit the images unless there is something VERY obvious and easily corrected (e.g. pial surface/skull strip included)? I am concerned about bringing subjectivity into this process by carrying out manual editing of particularly the surfaces in potentially ambiguous pial/white matter boundaries. Possibly the more minor segmentation errors made by Freesurfer are more objective through the entire dataset than my manual corrections could be? Any thoughts would be appreciated on this. Apologies if this is duplicating a previously answered query! Thank you Donna

1 0

Jump to page:

2026

2025

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

2006

2005

2004

Freesurfer