Toward defining deep brain stimulation targets in MNI space: A subcortical atlas based on multimodal MRI, histology and structural connectivity

Highlights

• Composite subcortical atlas based on a multimodal, high definition MNI template series, histology and tractography.

• High definition atlas of DBS targets matching MNI 152 NLIN 2009b space.

• Tractography based parcellation of the two primary DBS target regions STN and GPi into functional zones.

• Multimodal subcortical segmentation algorithm applied to MNI template.

Abstract

Three-dimensional atlases of subcortical brain structures are valuable tools to reference anatomy in neuroscience and neurology. For instance, they can be used to study the position and shape of the three most common deep brain stimulation (DBS) targets, the subthalamic nucleus (STN), internal part of the pallidum (GPi) and ventral intermediate nucleus of the thalamus (VIM) in spatial relationship to DBS electrodes. Here, we present a composite atlas based on manual segmentations of a multimodal high resolution brain template, histology and structural connectivity. In a first step, four key structures were defined on the template itself using a combination of multispectral image analysis and manual segmentation. Second, these structures were used as anchor points to coregister a detailed histological atlas into standard space. Results show that this approach significantly improved coregistration accuracy over previously published methods. Finally, a sub-segmentation of STN and GPi into functional zones was achieved based on structural connectivity. The result is a composite atlas that defines key nuclei on the template itself, fills the gaps between them using histology and further subdivides them using structural connectivity. We show that the atlas can be used to segment DBS targets in single subjects, yielding more accurate results compared to priorly published atlases. The atlas will be made publicly available and constitutes a resource to study DBS electrode localizations in combination with modern neuroimaging methods.

Introduction

Three-dimensional subcortical atlases are valuable tools to reference anatomy in the brain. In the field of deep brain stimulation (DBS), atlases of certain target structures may be used to study the relationship of electrode placement to its target structure in subcortical space (Merkl et al., 2015, Horn and Kühn, 2015, Neumann and Jha, 2015, Neumann and Staub, 2015, Eisenstein et al., 2014, Barow, 2014, Welter et al., 2014, Butson, 2007). Most of the available atlases have been defined either by histology or magnetic resonance imaging (MRI). For instance, the atlases by Yelnik et al. (2007) and Chakravarty et al. (2006) were defined using histological stacks of a single brain. Mai et al. (2015) defined a histological atlas based on three brains and Morel et al. used maps derived from six brains (Morel, 2013). A similar approach applied by Xiao and colleagues was to construct an MNI template for Parkinson's Disease to which a histological atlas was coregistered (Xiao et al., 2014). Unlike these histological atlases, there are others that were defined using only MRI data. These include the basal ganglia human area template (BGHAT; based on manual segmentations of a single MRI; Prodoehl et al., 2008) the subcortical shape atlas from the Computational Functional Anatomy lab at National University of Singapore (CFA atlas; Qiu et al., 2010; based on manual delineation of 41 subjects), the ATAG atlas (Keuken et al. (2014); based on multimodal high-field MRI that additionally includes three probabilistic age-varying maps of the subthalamic nucleus (STN); Keuken et al., 2013). Others have endeavored to include white matter connections into their studies; work by Lenglet et al. presented a comprehensive atlas of basal ganglia structures and their white-matter interconnections using structural and diffusion-weighted MRI (Lenglet et al., 2012). Finally, the Multimodal Imaging-Based Detailed Anatomical (MIDA) model (Iacono et al., 2015), provides a more comprehensive model of the whole head including blood vessels, muscles and bones, in addition to subcortical structures of the brain.

However, despite the high precision of anatomical labels in histological atlases and the probabilistic nature of atlases derived from larger cohorts of subjects, a shortcoming of the aforementioned atlases is that their spatial definition of structures – such as the STN – do not exactly match the position, size and shape of the same structures defined by the MNI template commonly used in neuroimaging. This is a crucial point and is illustrated in Fig. 1 (see S1 for details). Specifically, by using the combination of a template (used for nonlinear warps) and an atlas (used to define structures) that don’t exactly correspond to each other, misregistrations are prone to happen. This problem is further aggravated in the field of DBS imaging where i) anatomical structures are small in size and ii) inaccuracies measuring only a few millimeters can significantly alter clinical outcome.

To overcome the spatial discrepancy between templates and atlases, we present a different approach to define subcortical target regions in standard space. Here, the goal was to manually define an atlas that maximally agrees with the spatial position, shape and size of the structures defined by the template. We devised an algorithm that simultaneously used the T1- and T2-weighted as well as proton density (PD) and T2 relaxometry (T2rlx) modalities of the ICBM 2009 nonlinear templates. The algorithm created multimodal intensity fingerprints for each structure based on a few manually defined sample points and then calculated probability maps that exhibit tissue probability for each voxel of each structure on the template. These maps were then used as basis for the subsequent manual segmentation of the target regions.

The resulting atlas exhibits high spatial agreement to the MNI template but lacks anatomical detail and only includes structures visible on MRI. To overcome this limitation, in a second step, we used the template-based atlas as an additional anchor point to nonlinearly coregister a histology based intensity matched dataset into standard space. Such histology-to-template registration has previously been achieved (without the additional anchor point) by intensity matching histologically defined labels to corresponding structures on the colin27 MNI template (Chakravarty et al., 2008, Chakravarty et al., 2006). This procedure results in pseudo-MRIs defined by histology that resemble the MRI template and can then be nonlinearly coregistered to it. To increase precision of the warps especially at and around DBS regions of interest, we extended the approach introduced by Chakravarty and colleagues to include an additional registration that solely included the DBS structures on the pseudo-MRI as defined by histology and warped these to the corresponding manually segmented structures on the template (see. Fig. 2). The result was a patchwork atlas defining regions visible on MRI using the information of the template itself but relying on histology for regions not discernible on MRI.

Finally we used structural connectivity to further refine anatomical parcellations. This concept has previously been applied to define functional zones within anatomical structures (e.g. Behrens et al., 2003; Choi et al., 2012; Zhang et al., 2008). To take advantage of this additional source of information, STN and GPi nuclei were further subdivided into functional zones based on their structural connectivity. The STN was subdivided based on connectivity to cortical zones. A state-of-the art diffusion spectrum imaging dataset acquired in 30 healthy subjects was used to derive a tripartite segmentation of the STN into sensorimotor, associative and limbic functional zones. The same analysis was replicated based on diffusion dataset of 90 patients suffering from Parkinson's Disease. The GPi was subdivided based on its efferent fibers to seven functional regions of the thalamus (Fig. 3, Fig. 4, Fig. 5).

Our final atlas constitutes a composite dataset that uses histology, structural MRI and connectivity where appropriate. The atlas both maximally agrees with the 2009b MNI template but still exhibits great anatomical detail. To demonstrate the use and application of our atlas, we investigated the accuracy of a nonlinear coregistration of STN and GPi to a group of healthy subjects using our atlas and two published atlases.