That is a interesting question. I see that you are working with volume source space. Under certain circumstances these two methods are almost the same. Let me expand.
On the first approach, "From subject anatomy", there is a MRI volume already in the subject space (either from the MRI segmentation, or the anatomical parcellation was imported from the MNI space), then for each vertex in the volume source space, a label is assigned according to label of the anatomical parcellation in the subject space.
For the second approach, "Load atlas", as you can see there are several options to import. In your image, you are importing it as dilated, MNI space. Thus, for each vertex in the volume source space, a label is assigned according to its MNI coordinates, and the label of the anatomical parcellation in that location. With the dilated option, each ROI in the anatomical parcellation is inflated before, see the full explanation in here:
https://neuroimage.usc.edu/brainstorm/Tutorials/TutVolSource#Volume_atlases
So, if in the second option, you choose no-overlap instead of dilated, the results are almost the same as with the approach 1.
Moreover, check these other forum posts: