You can construct your own collision shape for a sprite, but that wouldn't really help if you need to detect different areas.
The straightforward option is what batvink suggests, having invisible sprites positioned where you need them, and move them with the rendered sprite.
Another option really only applies if you can use rectangular areas for hit detection. What I would consider doing, is sketching out the rendered sprite on paper, graph paper if you have it - then work out the position of the mouse in relation to the rendered sprite, so it's 0.0 - 1.0, to make up for sprite scaling. Then you display the difference on the screen, and take note of the coordinates. Then I'd fill and array with these coordinates and would use that to detect where the mouse is.
The code is dark and full of errors
