Hello, We are running FS v3.0.5 on some elderly brains, and the talairach transform (as assessed via tkregister2 with fstal flag) is inaccurate in an appreciable number of the scans we have examined so far. Frequently, the inaccuracies involve primarily scaling or translation, although in one instance the rotation was way-off (such that the cardinal axes were basically swapped). However, even for this worst case, the white and pial surfaces, as well as the aseg, look reasonable. (I haven't run autorecon3 to create the surface parcellations yet).
To what extent is talairach.xfm important for the autorecon2 & 3 stages? If surfaces and aseg look ok, can an inaccurate talairach.xfm just be ignored if we are not going to be averaging functional data? Will the surface parcellations perhaps be affected somehow?
Looking at ReconAllDevTable it appears that talairach.lta and talairach.m3z are the primary inputs to stages 2 and 3. So, I'm wondering how the .xfm relates to the .lta and .m3z files...?
Lastly, the particularly bad .xfm under v3.0.5 looks fine when created using v4.0.1. If we need or want accurate .xfm's would it be appropriate to run autorecon1 under v4 but then switch to v3 for autorecon2 & 3 so as to maintain compatibility with other subjects processed under v3?
thanks, Mike H.