Hi Freesurfer list-
I was trying to compare hand-drawn amygdalae and amygdala volumes generated by mri_ca_label/aseg (similar to what Morey et al., 2009 did) and wanted some advice. My amygdalae are drawn by hand in pathological plane, with a quite labor intensive tracing routine. And I wondered the best way to get my aseg volumes into that (path plane) space (b/c I didn't want to move the hand-drawn masks)?
Initially, I just aligned my raw T1w to pathological plane T1s and ran recon-all on these images, but the bivariate correlations between automated aseg volumes and hand-drawn amygdala dropped a fair amount (r=.5 in raw space, to r=.3 in path plane). I presume the re-alignment to path plane is adding some additional smoothing (which lowers mri_ca_label accuracy?).
I didn't think it would be a fair comparison to calculate dice-coefficients, intersection/union, etc. with aseg pathological plane amygdala (compared to my hand-drawn amygdala). Is the best practice to segment in original space, calculate the transform on the T1 from orig to path-plane, and then apply that transformation matrix to the aseg volume generated from the raw T1? Are there other suggestions? Any advice is greatly appreciated! Thanks much!
Best, jamie.