Print

Print
Hi Xinyang,

this is expected, as each streamline will need to check if it any of these ROIs has been crossed (more ROIs = more tasks), 
and also keep this information somewhere (more ROIS = higher memory requirements).

If you have a GPU, the last CUDA version of probtrackx2 has been optimised for network mode, and the computation is accelerated more than 100X:
https://users.fmrib.ox.ac.uk/~moisesf/Probtrackx_GPU/

Moises.
  

On 24 January 2018 at 04:02, Xinyang Liu <[log in to unmask]> wrote:
Dear FSL experts,

We have some confusions when tracing the whole-brain AAL90 and high-resolution 1024 ROI network in "probtrackx2 --network" mode.
Our understanding is, the algorithm is to send out thousands of streamlines from each voxel in seed ROIs and do probabilistic tracking. Each voxel, no matter which ROI it belongs to, would be used as seed point for once, and its connection to other ROI voxels would be recorded then. The total voxel number of AAL90 and 1024 ROIs are the same, with only difference in regional segmentation. However, the running time of 1024 network is much longer than the 90 network.

Our question is :
Why the running time of whole-brain 1024 network is much longer than whole-brain 90 ROI network although their total voxel numbers are the same?

Maybe our understanding is not correct enough. Any guidance would be very appreciated. Thank you very much.


Best regards,
Xinyang