The results from the default processing looked good over much of the brain, but, unfortunately, not in the area in which you are interested. I've tried several approaches that match that area better.
"-giant_move", "-giant_move -edge", "-giant_move -cost nmi"
I've attached an image of the 4 methods for a sagittal slice. The left image is the default output from align_epi_anat.py. The other images match the options in the same order listed above from left to right. The images show the original epi dataset overlaid on the edge-enhanced (Display menu option) aligned anatomical image for each of the methods. Try each of these solutions for yourself. Other options include using mask datasets or alternative weight datasets to compute the alignment to limit the area to base the alignment.