AFQ_RenderFibers works for individual fiber groups, but not for whole brain

94 views
Skip to first unread message

Steven Meisler

unread,
May 6, 2020, 12:39:00 AM5/6/20
to Automated Fiber Quantification (AFQ)
Hello AFQ team,

I am running AFQ on MATLAB 2019a on a Linux HPC system with 16gb of GPU memory. I have a wholebrain_fg and fg_classified of data preprocessed in Vistasoft. For the individual fiber groups in fg_classified, I am able to render the fibers fine. However, when I try to render fibers on the whole brain, the code never seems to finish. This is true even when I limit the number of fibers to something unpractically low, such as 10. I have been able to render wholebrains before by running through the AFQ example, so I doubt that this is an AFQ or MATLAB problem.

If it's relevant, some of the fiber groups (the first 9 of the typical 20 AFQ tracts) are empty after classification. I did not collect this data, but I cannot think of why this would be the case. Please let me know if you need any more information.

Thank you,
Steven

Jason Yeatman

unread,
May 6, 2020, 11:29:02 AM5/6/20
to automated-fiber-q...@googlegroups.com
Hi Steven,

If you are missing 9 fiber groups there is (a) something wrong with the data or more likely (b) something wrong in how you preprocessed the data. Going through the archive you will see many people who had this issue. Almost every time it has to due with flipped gradient directions.

--
You received this message because you are subscribed to the Google Groups "Automated Fiber Quantification (AFQ)" group.
To unsubscribe from this group and stop receiving emails from it, send an email to automated-fiber-quantif...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/automated-fiber-quantification-afq/33a32acb-08c5-4e00-8d7f-9d145ad253ac%40googlegroups.com.

Steven Meisler

unread,
May 12, 2020, 3:12:40 PM5/12/20
to Automated Fiber Quantification (AFQ)
Hi Jason,

Thank you for your reply. I realized I did not list all of the troubleshooting steps I had tried in the original post. I made sure the phase-encoding direction was correct and set the dwParams.rotateBvecsWithCanXform = 1, since the data come a Siemens scanner. The RGB maps all looked correct, and what could be tracked looked fine (I have attached a picture of the left UF, ILF, IFOF, AF, and SLF). The T1 files were also manually ACPC aligned in Vistasoft, and I confirmed that the problems persist on the raw T1 file. I will try to drop an ROI in the callosum and see if tracking looks okay from there, but if what I said changes any of the additional troubleshooting steps I should take, please let me know.

Best,
Steven


On Wednesday, May 6, 2020 at 11:29:02 AM UTC-4, Jason wrote:
Hi Steven,

If you are missing 9 fiber groups there is (a) something wrong with the data or more likely (b) something wrong in how you preprocessed the data. Going through the archive you will see many people who had this issue. Almost every time it has to due with flipped gradient directions.

On Tue, May 5, 2020 at 9:39 PM Steven Meisler <smei...@g.harvard.edu> wrote:
Hello AFQ team,

I am running AFQ on MATLAB 2019a on a Linux HPC system with 16gb of GPU memory. I have a wholebrain_fg and fg_classified of data preprocessed in Vistasoft. For the individual fiber groups in fg_classified, I am able to render the fibers fine. However, when I try to render fibers on the whole brain, the code never seems to finish. This is true even when I limit the number of fibers to something unpractically low, such as 10. I have been able to render wholebrains before by running through the AFQ example, so I doubt that this is an AFQ or MATLAB problem.

If it's relevant, some of the fiber groups (the first 9 of the typical 20 AFQ tracts) are empty after classification. I did not collect this data, but I cannot think of why this would be the case. Please let me know if you need any more information.

Thank you,
Steven

--
You received this message because you are subscribed to the Google Groups "Automated Fiber Quantification (AFQ)" group.
To unsubscribe from this group and stop receiving emails from it, send an email to automated-fiber-quantification-afq+unsub...@googlegroups.com.
tracts.png

Jason Yeatman

unread,
May 12, 2020, 4:16:49 PM5/12/20
to automated-fiber-q...@googlegroups.com
Those fiber tracts all look good. Which 9 fiber groups are you missing?

To unsubscribe from this group and stop receiving emails from it, send an email to automated-fiber-quantif...@googlegroups.com.

--
You received this message because you are subscribed to the Google Groups "Automated Fiber Quantification (AFQ)" group.
To unsubscribe from this group and stop receiving emails from it, send an email to automated-fiber-quantif...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/automated-fiber-quantification-afq/5ed1958d-914b-4618-b814-c2c3a1f0b92e%40googlegroups.com.

Steven Meisler

unread,
May 12, 2020, 4:20:35 PM5/12/20
to Automated Fiber Quantification (AFQ)
I am missing callosum forceps major as well as bilateral thalamic radiation, cortical spinal tracts, cingulum cingulate, and cingulum hippocampus.

Steven


On Tuesday, May 12, 2020 at 4:16:49 PM UTC-4, Jason wrote:
Those fiber tracts all look good. Which 9 fiber groups are you missing?

On Tue, May 12, 2020 at 12:12 PM Steven Meisler <smei...@g.harvard.edu> wrote:
Hi Jason,

Thank you for your reply. I realized I did not list all of the troubleshooting steps I had tried in the original post. I made sure the phase-encoding direction was correct and set the dwParams.rotateBvecsWithCanXform = 1, since the data come a Siemens scanner. The RGB maps all looked correct, and what could be tracked looked fine (I have attached a picture of the left UF, ILF, IFOF, AF, and SLF). The T1 files were also manually ACPC aligned in Vistasoft, and I confirmed that the problems persist on the raw T1 file. I will try to drop an ROI in the callosum and see if tracking looks okay from there, but if what I said changes any of the additional troubleshooting steps I should take, please let me know.

Best,
Steven


On Wednesday, May 6, 2020 at 11:29:02 AM UTC-4, Jason wrote:
Hi Steven,

If you are missing 9 fiber groups there is (a) something wrong with the data or more likely (b) something wrong in how you preprocessed the data. Going through the archive you will see many people who had this issue. Almost every time it has to due with flipped gradient directions.

On Tue, May 5, 2020 at 9:39 PM Steven Meisler <smei...@g.harvard.edu> wrote:
Hello AFQ team,

I am running AFQ on MATLAB 2019a on a Linux HPC system with 16gb of GPU memory. I have a wholebrain_fg and fg_classified of data preprocessed in Vistasoft. For the individual fiber groups in fg_classified, I am able to render the fibers fine. However, when I try to render fibers on the whole brain, the code never seems to finish. This is true even when I limit the number of fibers to something unpractically low, such as 10. I have been able to render wholebrains before by running through the AFQ example, so I doubt that this is an AFQ or MATLAB problem.

If it's relevant, some of the fiber groups (the first 9 of the typical 20 AFQ tracts) are empty after classification. I did not collect this data, but I cannot think of why this would be the case. Please let me know if you need any more information.

Thank you,
Steven

--
You received this message because you are subscribed to the Google Groups "Automated Fiber Quantification (AFQ)" group.
To unsubscribe from this group and stop receiving emails from it, send an email to automated-fiber-quantification-afq+unsub...@googlegroups.com.

--
You received this message because you are subscribed to the Google Groups "Automated Fiber Quantification (AFQ)" group.
To unsubscribe from this group and stop receiving emails from it, send an email to automated-fiber-quantification-afq+unsub...@googlegroups.com.

Jason Yeatman

unread,
May 12, 2020, 4:34:42 PM5/12/20
to automated-fiber-q...@googlegroups.com
Weird. I've never heard of that happening unless there is an issue with the data. Maybe someone else has input

On Tue, May 12, 2020 at 1:20 PM Steven Meisler <smei...@g.harvard.edu> wrote:
I am missing callosum forceps major as well as bilateral thalamic radiation, cortical spinal tracts, cingulum cingulate, and cingulum hippocampus.

Steven

On Tuesday, May 12, 2020 at 4:16:49 PM UTC-4, Jason wrote:
Those fiber tracts all look good. Which 9 fiber groups are you missing?

On Tue, May 12, 2020 at 12:12 PM Steven Meisler <smei...@g.harvard.edu> wrote:
Hi Jason,

Thank you for your reply. I realized I did not list all of the troubleshooting steps I had tried in the original post. I made sure the phase-encoding direction was correct and set the dwParams.rotateBvecsWithCanXform = 1, since the data come a Siemens scanner. The RGB maps all looked correct, and what could be tracked looked fine (I have attached a picture of the left UF, ILF, IFOF, AF, and SLF). The T1 files were also manually ACPC aligned in Vistasoft, and I confirmed that the problems persist on the raw T1 file. I will try to drop an ROI in the callosum and see if tracking looks okay from there, but if what I said changes any of the additional troubleshooting steps I should take, please let me know.

Best,
Steven


On Wednesday, May 6, 2020 at 11:29:02 AM UTC-4, Jason wrote:
Hi Steven,

If you are missing 9 fiber groups there is (a) something wrong with the data or more likely (b) something wrong in how you preprocessed the data. Going through the archive you will see many people who had this issue. Almost every time it has to due with flipped gradient directions.

On Tue, May 5, 2020 at 9:39 PM Steven Meisler <smei...@g.harvard.edu> wrote:
Hello AFQ team,

I am running AFQ on MATLAB 2019a on a Linux HPC system with 16gb of GPU memory. I have a wholebrain_fg and fg_classified of data preprocessed in Vistasoft. For the individual fiber groups in fg_classified, I am able to render the fibers fine. However, when I try to render fibers on the whole brain, the code never seems to finish. This is true even when I limit the number of fibers to something unpractically low, such as 10. I have been able to render wholebrains before by running through the AFQ example, so I doubt that this is an AFQ or MATLAB problem.

If it's relevant, some of the fiber groups (the first 9 of the typical 20 AFQ tracts) are empty after classification. I did not collect this data, but I cannot think of why this would be the case. Please let me know if you need any more information.

Thank you,
Steven

--
You received this message because you are subscribed to the Google Groups "Automated Fiber Quantification (AFQ)" group.
To unsubscribe from this group and stop receiving emails from it, send an email to automated-fiber-quantif...@googlegroups.com.

--
You received this message because you are subscribed to the Google Groups "Automated Fiber Quantification (AFQ)" group.
To unsubscribe from this group and stop receiving emails from it, send an email to automated-fiber-quantif...@googlegroups.com.

--
You received this message because you are subscribed to the Google Groups "Automated Fiber Quantification (AFQ)" group.
To unsubscribe from this group and stop receiving emails from it, send an email to automated-fiber-quantif...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/automated-fiber-quantification-afq/6a5a728c-e75d-44bd-a3df-0136bc2acac1%40googlegroups.com.

Steven Meisler

unread,
May 13, 2020, 1:05:32 PM5/13/20
to Automated Fiber Quantification (AFQ)
Hi Jason,

I just remembered that I had been tinkering with some the AFQ code earlier which led to it ignoring some of the fiber tracts. I fixed the problem now and everything seems to work.

I don't know if this is specific to a version of AFQ, but for some reason the roi1 and roi2 names for the CGC in the default AFQ_Create parameters end with .mat, while all other roi names leave the file extension out. This causes an error when computing tract properties, since AFQ_LoadROIs will append '.mat' to the roi name automatically, so the program looks for CGC roi names that end with '.mat.mat'. It's easy enough to fix by going into the parameter struct and changing these roi1 and roi2 names (indices 5 and 6), but I figured I should bring this to your attention.

Best,
Steven


On Tuesday, May 12, 2020 at 4:34:42 PM UTC-4, Jason wrote:
Weird. I've never heard of that happening unless there is an issue with the data. Maybe someone else has input

On Tue, May 12, 2020 at 1:20 PM Steven Meisler <smei...@g.harvard.edu> wrote:
I am missing callosum forceps major as well as bilateral thalamic radiation, cortical spinal tracts, cingulum cingulate, and cingulum hippocampus.

Steven

On Tuesday, May 12, 2020 at 4:16:49 PM UTC-4, Jason wrote:
Those fiber tracts all look good. Which 9 fiber groups are you missing?

On Tue, May 12, 2020 at 12:12 PM Steven Meisler <smei...@g.harvard.edu> wrote:
Hi Jason,

Thank you for your reply. I realized I did not list all of the troubleshooting steps I had tried in the original post. I made sure the phase-encoding direction was correct and set the dwParams.rotateBvecsWithCanXform = 1, since the data come a Siemens scanner. The RGB maps all looked correct, and what could be tracked looked fine (I have attached a picture of the left UF, ILF, IFOF, AF, and SLF). The T1 files were also manually ACPC aligned in Vistasoft, and I confirmed that the problems persist on the raw T1 file. I will try to drop an ROI in the callosum and see if tracking looks okay from there, but if what I said changes any of the additional troubleshooting steps I should take, please let me know.

Best,
Steven


On Wednesday, May 6, 2020 at 11:29:02 AM UTC-4, Jason wrote:
Hi Steven,

If you are missing 9 fiber groups there is (a) something wrong with the data or more likely (b) something wrong in how you preprocessed the data. Going through the archive you will see many people who had this issue. Almost every time it has to due with flipped gradient directions.

On Tue, May 5, 2020 at 9:39 PM Steven Meisler <smei...@g.harvard.edu> wrote:
Hello AFQ team,

I am running AFQ on MATLAB 2019a on a Linux HPC system with 16gb of GPU memory. I have a wholebrain_fg and fg_classified of data preprocessed in Vistasoft. For the individual fiber groups in fg_classified, I am able to render the fibers fine. However, when I try to render fibers on the whole brain, the code never seems to finish. This is true even when I limit the number of fibers to something unpractically low, such as 10. I have been able to render wholebrains before by running through the AFQ example, so I doubt that this is an AFQ or MATLAB problem.

If it's relevant, some of the fiber groups (the first 9 of the typical 20 AFQ tracts) are empty after classification. I did not collect this data, but I cannot think of why this would be the case. Please let me know if you need any more information.

Thank you,
Steven

--
You received this message because you are subscribed to the Google Groups "Automated Fiber Quantification (AFQ)" group.
To unsubscribe from this group and stop receiving emails from it, send an email to automated-fiber-quantification-afq+unsub...@googlegroups.com.

--
You received this message because you are subscribed to the Google Groups "Automated Fiber Quantification (AFQ)" group.
To unsubscribe from this group and stop receiving emails from it, send an email to automated-fiber-quantification-afq+unsub...@googlegroups.com.

--
You received this message because you are subscribed to the Google Groups "Automated Fiber Quantification (AFQ)" group.
To unsubscribe from this group and stop receiving emails from it, send an email to automated-fiber-quantification-afq+unsub...@googlegroups.com.

Jason Yeatman

unread,
May 13, 2020, 10:01:27 PM5/13/20
to automated-fiber-q...@googlegroups.com
Not sure how that file name ended up like that but I'll fix it as soon as I have time to test

It's funny that it hasn't caused any problems before

To unsubscribe from this group and stop receiving emails from it, send an email to automated-fiber-quantif...@googlegroups.com.

--
You received this message because you are subscribed to the Google Groups "Automated Fiber Quantification (AFQ)" group.
To unsubscribe from this group and stop receiving emails from it, send an email to automated-fiber-quantif...@googlegroups.com.

--
You received this message because you are subscribed to the Google Groups "Automated Fiber Quantification (AFQ)" group.
To unsubscribe from this group and stop receiving emails from it, send an email to automated-fiber-quantif...@googlegroups.com.

--
You received this message because you are subscribed to the Google Groups "Automated Fiber Quantification (AFQ)" group.
To unsubscribe from this group and stop receiving emails from it, send an email to automated-fiber-quantif...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/automated-fiber-quantification-afq/c6b3e469-8f89-4957-a5ef-4dafedba3801%40googlegroups.com.
Reply all
Reply to author
Forward
0 new messages