PT - JOURNAL ARTICLE AU - Siobhan Ewert AU - Andreas Horn TI - Three-dimensional definition of two prominent deep brain stimulation targets based on a multimodal high-definition MNI template AID - 10.1101/062851 DP - 2016 Jan 01 TA - bioRxiv PG - 062851 4099 - http://biorxiv.org/content/early/2016/07/08/062851.short 4100 - http://biorxiv.org/content/early/2016/07/08/062851.full AB - Three-dimensional atlases of subcortical brain structures are valuable tools to reference anatomy in neuroscience and neurology. In the special case of deep brain stimulation (DBS), the two most prominent targets are the subthalamic nucleus (STN) and the internal part of the pallidum (GPi). With the help of atlases that define their position and shape within a well-defined stereotactic space, their spatial relationship to implanted deep brain stimulation (DBS) electrodes may be analyzed on a single-subject level and in group studies. One standard approach to define such atlases is to manually segment structures on MR data of large cohorts and coregister them to a standard anatomy template. Other approaches include using histological stacks that are again co-registered to the template. However, most of the atlases derived in this fashion show a substantial descripancy to the anatomy that is defined by the template itself. Here, an algorithm that automatically segments the template itself by simultaneously using its T1, T2, proton density and T2 relaxometry modalities is introduced. Based on a low number of manually placed point fiducials within each structure, this three-level algorithm was able to robustly segment subcortical anatomical structures. The first level consisted of a multimodal region-growing algorithm that intended to increase the number of sample points for robustness reasons. The second level computed Mahalanobis distances from each voxel in the brain to the multimodal intensity distribution of first level points set and thus assigned target structure similarity values to each voxel. Finally, the third level assigned each super-threshold voxel to one of the subcortical structures in a winner-takes-all fashion based on second level probability maps. The algorithm was able to segment the STN, the pallidum and the red nucleus directly on the MNI template. The algorithm was not able to automatically segment the internal and external parts of the pallidum. However, given the practical importance of the GPi, a second version of the atlas was defined in high resolution by manual segmentation. To maintain a maximally possible observer-independence, this process was largely informed by automatically generated probability maps described above. We argue that the resulting atlas is a valuable tool to define DBS target structures within standard anatomical space. Furthermore, the atlas may additionally be used as an anchor-point to co-register more detailed (e.g. histological) atlases into standard space. Both the automated and manual versions of the atlas will be made publicly available under an open license.HighlightsHigh definition atlas of DBS targets exactly matching MNI 152 NLin 2009 spaceMultimodal subcortical segmentation algorithm applied to MNI templateProfound literature overview about available subcortical atlases and their use cases and potential limitations