mmmm

then, probably that computing node cannot allocate all the ROIs (I do not know if you are using NifTI volume or surfaces).

I would check if this is the problem running the tool with less ROIs.

Then, the obvious solution would be to use a computer with more memory or reduce the number of ROIs.

HCP data is very high-resolution data, so, it is expected to require a lot of memory.

The option seeds_to_target gives you a matrix with the connectivity from each individual seed point to any ROI.

You could divide the matrix into different parts, use this option and sum the values of the seeds that belong to the same ROI.

But, this is going to be very time-consuming, and very tedious.

Maybe someone has a better idea.

Moises

On 8 January 2018 at 22:52, Xinyang Liu <[log in to unmask]> wrote:

Thanks.
I reran the command without --omatrix1 and --opd. But the running was killed again. Here is the information which is same as before.
Running in network mode
load seeds
terminate called after throwing an instance of 'std::bad_alloc'
   what(): std::bad_alloc
Aborted

Best regards,
Xinyang

At 2018-01-09 11:36:28, "Moises Hernandez" <[log in to unmask]> wrote:

--omatrix1 is the problem.

This option generates a matrix with connectivity from each seed point to any other seed point.
I think you are not looking for this, but a matrix with the connectivity from each ROI to any other ROI.
With --network option is enough to get this ROI x ROI matrix (file fdt_network_matrix)

Also, if you do not need a map with the path distribution, you can drop option --opd.

Moises.

On 8 January 2018 at 22:26, Xinyang Liu <[log in to unmask]> wrote:
Hi, Thanks a lot for your feedback.
We use volumes to run the 1024 ROI network. The data is the diffusion MRI from the Human Connectome Project.
The command is:
probtrackx2 --network -s /.../Diffusion.bedpostX/merged -m /.../nodif_brain_mask.nii.gz -x /.../seedmasks_1024_100206.txt -o 100206_network1024 --dir=/.../output_network_1024 --forcedir --opd --omatrix1
Could you please check where is the problem? Thank you very much.

Best regards,
Xinyang

At 2018-01-09 00:54:50, "Moises Hernandez" <[log in to unmask]> wrote:

Hi,
A matrix of 1024 x 1024 does not require much memory:
1024 X 1024 x 4 (bytes) = 4MB

This is the size of the matrix created with --network option.
However, the tool needs to allocate the ROIs. Are you using volumes or surfaces?
Can you show us how are you running probtrackx2?

Moises

On 8 January 2018 at 03:58, Xinyang Liu <[log in to unmask]> wrote:
Dear FSL experts,

We want to build a 1024*1024 brain network using probtrackx2. However, the running was killed under the network mode probably due to limited memory. Therefore, we decide to split the computing to smaller sections. Now we have some questions:

1. We change to trace from one or m(<N) seed ROIs to N target ROIs, we want to get the number of streamlines separately reaching each target ROI. However, the output file "waytotal" only contains the streamlines from the seed ROI. Which command could we use to acquire separate streamline numbers corresponding to each target ROI? We want to get a connectivity matrix like the one in network mode but in m*N size.
2. I am not very clear about the default settings of seeds_to _targets classification and network mode. Is their default computings equal to waypoint tracking or stop tracking or both?
3. Is the killed running of 1024*1024 network due to limited computer memory? The information given was:
terminate called after throwing an instance of 'std::bad_alloc'
   what(): std::bad_alloc
Aborted
Our single CPU core has a 16G RAM, which cannot support the 1024*1024 network computing. Could you please suggest an available way for such running?

Any guidance would be very appreciated. Thanks!

Best regards,
Xinyang Liu