Developments in medical imaging allow use of several modalities toward more effective diagnosis. Different modalities often complementary information of the explored organ. In order to take advantage of these different techniques image registration is required. In our case study, we deal with MRI and SPECT images from the same patient. The paper presents a registration approach, without any external landmarks, that proceeds in two steps. First, image segmentation is achieved by a competitive fuzzy clustering method. In the second step we seek for brain orientation, considered as an invariant feature, by principal axis computing. In order to handle pathological cases, which present missing information, we improve brain orientation detection with a possibilistic shell fitting method.