There is no way to set a hard and fast threshold. I would recommend
running a set of them (20ish), checking them manually to assure that
they are correct, then compute the mean and stddev of the mincost. For a
new data set, you can compute a z-score of the mincost and then pick a
threshold
Hi Doug,
As an update, --init-coreg produced essentially the same result as
--init-fsl; however, that was on EPIs we've discovered were LR flipped
during reconstruction, so I've switched to a different test subject for now.
Now, for --init-fsl and --init-coreg, respectively, we get:
MinCost:
That min cost (.86) is too high for sure. My guess is that it is the fsl
init. In general, bbregister is pretty insensitive to the initialization
method (as long as the init does not fail), so I would not worry about
some being done with coreg and some with flirt
On 3/9/17 3:40 PM,
Hi Doug,
This is the final line with "cost" in it:
MinCost: 0.862331 7440.183178 7411.264107 -1.819400
We're switching from FLIRT BBR (two-pass) to bbregister when there's a
reconstructed subject available, so I'm using `--init-fsl` to try to
minimize the differences in the pipeline in the two
that should have worked. what was the bbregister final cost function? It
could have been that fsl did not provide a good initial registration. If
you have v6, you can leave off --init-fsl and it will use mri_coreg
(which is more robust).
On 03/09/2017 03:24 PM, Christopher Markiewicz wrote:
>
