This paper evaluates a nonlinear registration method for warping a 3D histological atlas of the basal ganglia into patient data for deep brain stimulation (DBS) planning. The power of the method is the possibility to combine iconic registration with geometric constraints under a unified diffeomorphic framework. This combination aims to ensure robust and accurate atlas-to-patient warping and anatomy-preserving deformations of stimulation target nuclei. A comparison of the method with a state-of-the-art diffeomorphic registration algorithm reveals how each approach deforms low-contrasted image regions where DBS target nuclei often lie. The technique is applied to T1-weighted magnetic resonance images from a cohort of Parkinsonian subjects, including subjects with standard-size and large ventricles. Results illustrate the effects of iconic or geometric registration alone, as well as how both constraints can be integrated in order to contribute for registration precision enhancement.