About method 2 - this is the easiest method but can be very inaccurate. "Inaccurate" in the sense that a structure in tlrc atlas will correspond to many different areas in different brains. The tlrc atlas is also "funny" in some places. For example, the angular gyrus is very small. The precentral gyrus has a branch that extends too much anterirorly, and so forth. I recently mapped the tlrc atlas gyri to a surface to get a feel for what it looks like, and the results were not encougraging.
This applyis to the gryi as defined in the tlrc atlas. Using the Broadmann's areas from atlas is going to be even more inaccurate, because they are defined only for the gray matter, so they are relatively thin stripes, and could correspond to very different areas in different brains.
What is needed, it seems to me, is a probabilistic atlas in AFNI tlrc space. One needs to draw various structures in many (say ~50) brains in tlrc space and find the probability for each voxel, of being in a given structure. It is relatively easy to draw structures on inflated surfaces, and of course can be done directly on volumes too with enough experties. Based on the probabilist altals, you can select all voxels that have say 90% chance of being in a given structure.