By snooping in the log files in scripts/ and considering the instruction to train and then label with commands such as
mris_ca_train -n 2 -t ./iowa_gross_fs_colors.txt rh sphere.reg iowa_gross_hand_parc ${SUBJECT_LIST} ./rh.iowa_gross_cort_parc.gcs
which prepares the way for
mris_ca_label ${SUBJECT} rh sphere.reg /opt/freesurfer/average/rh.iowa_gross_cort_parc.gcs
it becomes evident that the -autorecon3 step applies the MGH standard parcellation atlas from
/opt/freesurfer/average/[lr]h.curvature.buckner40.filled.desikan_killiany.gcs
and thus, the "standard" atlas is known as "curvature.buckner40.filled.desikan_killiany".
When we evaluate our own atlas, we compare hand-traced labels with the labels produced by mris_ca_label.
Now my question: Is there a way we could access the training data for "curvature.buckner40.filled.desikan_killiany", so we can compare how mris_ca_label approximates the MGH atlas's gold standard with how it approximates the Iowa atlas's gold standard?
We are curious how many subjects went into the standard atlas at mris_ca_train, too. Is there a paper (a pdf?) where this is reported?
Greg Harris University of Iowa Psychiatry Brain Imaging Lab