Hi,
Unfortunately the segmentation is still not perfect, so we
do expect some voxel to be misclassified, though it should
not be many. Are you running FIRST with multiple structures?
This can help as there is a final stage which deals specifically
with voxels that border more than one structure. So I would
recommend running FIRST with all structures, if you are not
already doing that.
As for error rate, our leave-one-out tests on our manual
segmentations gives a median overlap of around 85% with
the manual segmentations for the Caudate. This is comparable
to or better than many automated and manual inter-rater reliability
measures.
I hope this helps.
All the best,
Mark
On 9 Jul 2009, at 20:08, Anastasia A Ford wrote:
> Hi,
>
> I am running FIRST to segment out the caudate and I see that there
> are a few
> voxels throughout the mask on the caudate that are actually not part
> of the
> structure. For example, voxels from the thalamus get assigned to the
> body of
> the caudate. Also, on more inferior portions of the caudate, close
> to Nuc
> Accumbens, portions of the anterior limb of the internal capsule get
> grouped
> with the caudate. Is there an example of the overall quality of the
> segmentation and what would be an acceptable error rate?
>
> Sincerely,
>
> Anastasia
>
|