Incorrect Result in Registration

I am doing an affine registration using mutual information and I am getting incorrect results when trying to register the MNI atlas 1mm with the vanderbuild database.
The images of the vanderbuilt data base differ from the others because their image have just 25 slices, the head is not scanned completely and their orientation is different.

I also tried others similarity measures, but the result continuous incorrect. So, how can I solve this problem? Is there any way of creating another similarity measure?


I often found that if one of the images cover a larger region than the other then registration may fail (does not converge to the right location) and cropping them to cover approximately the same region helps.

Also, for me, registration methods in ITK always required careful parameter tuning, while Elastix (registration package built on ITK) worked out of the box, using default parameters, on a wide range of images. There are other ITK-based registration packages, such as ANTs, which might work better for certain tasks.