Hi,
Making custom priors for FIRST is doable, but requires manual segmentations.
In the case of applying FIRST to the 2 groups to construct models, if
there was a bias in FIRST then the models would reflect that bias. It is
worth keeping in mind that the models (priors) used in FIRST are rather
different than tissue probability maps. The surface models explicitly
represent the variation in the training set using a point distribution
model and search across the modes of variation (eigenvectors) whilst
predicting the expected mean and covariance of the intensity samples
conditional on shape. Also the training set was comprised of 6
heterogenous sets with different scan protocols (and potentially different
scanners):
Group | Size | Age Range | Resolution (mm) |Patient Group
1 | 37 | 16 to 72 | 1.0 x 1.5 x 1.0 | NC & SZ
2 42 | Adults | 1.0 x 1.0 x 1.0 | NC & AD
3 17 | 65 to 83 | 0.9375 x 1.5 x 0.9375 | NC & AD
4 87 | 23 to 66 | 0.9375 x 3.0 x 0.9375 | NC & SZ
5 14 | 9 to 11 | 1.0 x 1.0 x 1.0 | NC & PC
6 139 | 4.2 to 16.9 | 0.9375 x 1.5 x 0.9375 | NC & ADHD & SZ
This is not to say that no bias could exist, but I would I hope that it
would alleviate some of it. Precautions should always be taken for any
analysis, have you looked at the data to see if the segmentations look
reasonable (first_roi_slicesdir) can be helpful for this for larger
datasets)?
Cheers,
Brian
> Dear FIRST experts,
>
> I am faced with the task of comparing the sub-cortical anatomy of a group
> of
> controls to a group with known differences in brain shape, so I (or a
> reviewer)
> might wonder if the standard priors generated by the CMA brains would
> represent a bias in my comparison, since my controls will likely look more
> like
> the CMA data than my abnormal group. Is there some facility for
> generating
> customized priors in FIRST, perhaps by running my two groups through
> FIRST,
> then using the output to create an analysis-specific set of priors? From
> the
> FIRST documentation, I see the note below, but I'm not sure if it's
> relevant, or
> if so, how I might make use of the bvars output.
>
> "output_name_first.bvars: Do not delete this file. It contains the mode
> parameters and the model used. This file, along with the appropriate model
> files, can be used to reconstruct the other outputs. The mode parameters
> are
> what FIRST optimizes. This output can be used be used later as to perform
> vertex analysis or as a shape prior to segment other shapes."
>
> Any thoughts appreciated,
>
> Shane Kippenhan
>
>
|