Hi Russ,
It looks like flirt cannot extract boundary points from the freesurfer segmentation.
The first things I would check are: (i) whether there are any NaNs in the freesurfer segmentation, as NaNs anywhere in the image really muck up FSL, and (ii) that the values in the background are zero and the WM is one or higher (with a consistent value). If it doesn't seem that either of these things are the problem then upload the data to us (see our Support page for the link if you haven't used it before) and we'll have a look.
Oh, and if you see this error don't bother letting it continue running as it is certainly going to fail.
All the best,
Mark
On 5 Feb 2013, at 18:25, Russ Poldrack <[log in to unmask]>
wrote:
> hi - I'm running into a problem using BBR registration (via epi_reg in 5.0.2) to align an EPI image (2.4 mm iso resolution) to a T1-weighted anatomy (1mm iso resolution). The wmseg file is actually obtained from freesurfer and looks good (I think), as does the wmedge file. The FLIRT pre-alignment registration (saved to blah_init.mat) also looks reasonable. However, when it gets to the BBR registration section, I see lots of messages like this:
>
> Running BBR
> ERROR::set_bbr_seg: could not find any boundary points!
> ERROR::set_bbr_seg: could not find any boundary points!
> ERROR::set_bbr_seg: could not find any boundary points!
> ERROR::set_bbr_seg: could not find any boundary points!
> WARNING:: Found NaN in BBR cost (2)
> ERROR::set_bbr_seg: could not find any boundary points!
> WARNING:: Found NaN in BBR cost (2)
> ERROR::set_bbr_seg: could not find any boundary points!
> WARNING:: Found NaN in BBR cost (2)
> ERROR::set_bbr_seg: could not find any boundary points!
> WARNING:: Found NaN in BBR cost (2)
> ERROR::set_bbr_seg: could not find any boundary points!
> WARNING:: Found NaN in BBR cost (2)
>
> if I let it keep running it will do so for several hours, finally finishing, but the resulting output image is badly misregistered (worse than the initial pre-alignment). Any suggestions for how to address this problem would be most appreciated.
> best,
> russ
|