In functional neurosurgery, surgical interventions are directed at retuning brain structures or networks that are dysfunctional in chronic therapy-resistant neurological disorders (e.g. Parkinson's disease). Deep brain structures in the thalamus or basal ganglia are reached stereotactically, with procedures allowing to precisely target a group of neurons (nucleus) or a fiber tract. Visualization of fine histological structures such as individual nuclei in the thalamus or subthalamic fibre tracts are still beyond the capabilities of current MR imaging techniques. Therefore, one has still to rely on indirect targeting, usually by transferring detailed histological maps obtained from a three-dimensional anatomical atlas onto the patient's individual brain images. Anatomical and stereotactic precision is particularly important in case of radiosurgery or other neuroradiological approach like focused ultrasound (FUS), where no electrophysiological verification of the target is possible. In order to retain the precision achieved in present stereotactic operations, we intend to enhance the current, simple landmark-based targeting by creating a statistical atlas of the thalamus and its nuclei and develop methods to map it onto the MR image of the patient. Statistical shape models [COOTES95] have seen widespread use in medical image analysis. The first step of creating a statistical model is the establishment of dense correspondence across the members of the training set. In our case the individual examples are provided in the form of histological, thus highly anisotropic, slice stacks, specifying a 3D model for each thalamic nucleus. Atlas based neurosurgical planning assumes that the topological relationship between different nuclei is constant. We therefore propose to use this information to establish correspondences: The surface of each nucleus can be divided into different patches, where each patch is the surface, which the nucleus has in common with some neighbouring anatomical structure. Using patch parameterization techniques [QUICKEN00] we can then derive correspondences between the same surface patches of a nucleus. In this paper we outline our approach and investigate to which extent such patches can be reconstructed in slice stacks of different stereotactic orientation.