Computational Modeling and Tractography for DBS Targeting

6 Computational Modeling and Tractography for DBS Targeting

Michael D. Staudt, Sarah Ridge, Jennifer A. Sweet

Abstract

Deep brain stimulation (DBS) has an established and efficacious role in the treatment of movement disorders, with emerging indications for neuropsychiatric disorders and epilepsy. However, the underlying effects of electrical stimulation on cellular mechanisms and widespread neural networks remain unclear. Furthermore, there are little data regarding the correlation between clinical outcomes and modulation of dysfunctional neural circuitry via the direct and indirect stimulation of axonal pathways. The advent of computational modeling for DBS has been a powerful tool for better understanding of DBS and neural circuitry, including the development of surgical targets for electrode placement and refinement of stimulation parameters. Advanced imaging techniques, including tractography and high-field MRI, allow for the specific visualization of white matter tracts and individual nuclei, potentially leading to the development of patient or symptom-specific treatment models. This chapter reviews the value of computational modeling and advanced imaging technologies for DBS therapies and discusses existing and future DBS applications.

Keywords: computational modeling, deep brain stimulation, neuroimaging, targeting tractography

6.1 Introduction

Deep brain stimulation (DBS) is the delivery of electrical impulses to deep structures in the brain via surgically implanted electrodes. While DBS has proven to be an effective therapy for movement disorders, such as Parkinson’s disease (PD),1,2,3 essential tremor (ET),^4,5 and dystonia,^6,7 as well as for emerging neuropsychiatric indications including obsessive compulsive disorder,^8,9 the exact mechanism of action of DBS still remains unknown. Since many of the current intracranial DBS targets were historically lesioned producing therapeutic effects similar to DBS, stimulation was originally thought to act via the inhibition of these gray matter targets.¹⁰ However, data suggest that DBS may also result in the excitation of surrounding white matter (WM) axons,^11,12,13 possibly contributing to the downstream effects of stimulation, seen at sites distant from the electrode target.^{14,15,16,17,18,19,20,21} Prevailing theories support the notion that high-frequency DBS ultimately disrupts aberrant neuronal patterns producing global network modulation.^11,12,22

This proposed mechanism of action, involving the stimulation of WM tracts and the propagation of widespread neuronal effects, may account for both the improvement in symptoms seen with DBS as well as many of the unwanted side effects. As such, selective activation of these complex and interconnected neural networks may further improve patient outcomes. DBS targeting can be further enhanced by using computational modeling techniques and novel imaging strategies. Such tools can improve our understanding of DBS and brain networks, and can also help in increasing accuracy, and identifying better targets for diseases that are currently treated with DBS. These tools can also aid in the discovery of new surgical targets for neurological, psychiatric, and possibly even cognitive disorders. The goal of this chapter is to review the value of computational modeling and advanced imaging technologies for DBS therapies and discuss existing and future DBS applications.

6.2 Computational Modeling Techniques

Computational modeling techniques have largely contributed to our current understanding of the mechanism of DBS in the treatment of movement disorders, allowing for more effective surgical targeting. In 1999, Grill used electrical models to demonstrate the impact of the properties of neural tissue surrounding DBS electrodes on the electric fields generated by these electrodes.23 The models demonstrated the inhomogeneity and anisotropy of the neural tissue stimulated by DBS, reinforcing the importance of electrode lead location for generating the desired effects. On the basis of these principles, McIntyre and Grill used modeling techniques to show that axons are the most excitable neural elements with stimulation.^24,25,26 They also demonstrated that knowledge of the electrical conductivity of the tissue around the DBS lead as well as familiarity of the electrode shape and position within the brain can help to predict the electric field generated by DBS and the subsequent neural response.^25,26 Thus, understanding the influence that surrounding neural tissue has on the effects of DBS could potentially lead to improved predictions of the clinical responses to stimulation.

6.2.1 Volume of Tissue Activated

Developments in computational modeling techniques have ever since utilized this principle, allowing investigators to better understand the relationship between lead placement and the delivery of stimulation to adjacent anatomical structures, and how this relationship impacts clinical outcomes.15,27,28,29 One method by which this has been done involves the use of models to allow the visualization of the volume of tissue activated (VTA) by DBS. VTAs are created using finite element modeling (FEM) that combines an anatomical model, based on imaging data from subjects who have undergone DBS implantation, with an electrical model using actual or theoretical DBS stimulation parameters to determine the voltage spread from stimulation.^11,15,27 The results depend on the composition of the tissue surrounding the electrode, such as gray matter versus WM, as this affects electrode capacitance and impedance as well as the type of stimulation and the parameters used.¹⁵ The axonal activation pattern can then be predicted from the presumed electrical field created from each active DBS electrode contact, and the threshold for generating an action potential by adjacent axons.^11,15,27 Ideally, a truly “connectomic” approach will allow for identification of the differential effects of electrical stimulation on different parts of the neuron (i.e., soma, axons, dendrites) as part of a greater neural network.30 Computational models have been developed that emulate the effects of electrical stimulation on networks of multicompartmental neurons,^31,32 although these paradigms have not yet been translated to DBS research.

Clinical applications of the VTA can be valuable for determining which structures are responsible for producing the clinical effects of DBS. In 2004, McIntyre et al, combined tissue conductivity information with an FEM to show the shape and volume of stimulation with standard DBS parameters in the subthalamic nucleus (STN) for PD, demonstrating that even subtle deviations in electrode positions would result in the stimulation of different structures, thus resulting in variable clinical effects.¹² Miocinovic et al similarly used this concept to devise a computational model integrating STN anatomical data from Parkinsonian macaques to an FEM of the electric field from DBS, while also incorporating the biophysical properties of neurons in the STN, the globus pallidus interna (GPi), and the internal capsule.¹⁵ In this way, they predicted the axonal activation patterns of STN DBS and showed that although stimulation of this target resulted in both STN and GPi fiber activation, it was the volume of STN tissue activated that resulted in the specific therapeutic effects.¹⁵

Butson et al also used the VTA to study the effects of STN DBS in a PD patient by fusing the preoperative and postoperative imaging data to determine the exact location of the electrode.²⁷ They then created VTA models according to the different DBS programming parameters used and correlated this with clinical outcomes. The authors found that corticospinal tract symptoms correlated well with involvement of the corticospinal tract in the VTA, and improvement in bradykinesia and rigidity corresponded to VTAs that involved the zona incerta.²⁷ In 2009, Maks et al retrospectively studied 10 PD patients with STN DBS and determined the VTA from imaging data and reports of the active contacts used for each subject.²⁸ They showed that when the active contacts were near the dorsal border of the STN, the resultant VTAs produced optimal therapeutic benefits. In a similar manner, Mikos et al found that PD patients with STN DBS whose VTA included nonmotor areas of the STN had worsening verbal fluency compared to PD patients with VTA involving only motor STN.²⁹ Thus, the VTA can be retrospectively assessed in patients previously implanted with DBS electrodes to help ascertain what structures are being stimulated by the active contact to produce the observed effects, consequently helping to determine the optimal DBS target ( Fig. 6.1).³³

Moreover, there is great interest in creating patient-specific models that can prospectively define anatomical pathways of interest and calculate the response to DBS stimulation. Pathway activation models integrate imaging data with both tractography and the biophysics of electrical stimulation modeling to theoretically estimate pathway activation a priori.³⁵ These models calculate the axonal response to DBS in relation to electrode configuration, the characteristics of the applied stimuli, the tissue conduction properties, axonal geometry, and axonal membrane biophysics.³⁵ In comparison, conventional VTA modeling relies primarily on activation volume to generate seeds for tractography and is primarily performed retrospectively. As such, pathway activation models are much more time- and resource-intensive to develop.^35,36 They are potentially powerful tools of analysis, but lack in their current inability to quantify the effects of DBS stimulation on a network level.

6.2.2 Whole-Brain Network Models

Other computational modeling techniques have aided in our understanding of the influence of DBS on widespread, complex neural networks. In 2004, Rubin and Terman created a computational network model to determine how DBS of the STN for the treatment of the motor symptoms in PD results in the disruption of downstream pathological thalamic rhythms.37 By simulating the molecular environment surrounding neurons in the STN, the GPi, and the thalamus under conditions of a healthy state, a parkinsonian state, and a parkinsonian state with STN DBS, the authors demonstrated that STN DBS restores abnormal oscillations within the basal ganglia, ultimately normalizing thalamic relay processes.³⁷ In 2010, Hahn and McIntyre evaluated the VTA of STN DBS by incorporating presumed cortical and striatal inputs to the basal ganglia into their computational model, thus allowing for investigation into the influence of networks as opposed to isolated cell-to-cell interactions.³⁸ These authors found that the STN VTA plays a key role in correcting pathologic GPi bursting in PD, thus influencing cortico-striatal-thalamic networks. They concluded that there may be a critical VTA required to produce the observed outcomes of DBS.

Humphries and Gurney devised a computational model of the basal ganglia to show that DBS of the STN results in a mixture of excitatory and inhibitory responses from the basal ganglia output nuclei.³⁹ They postulated that this diversification of responses from the basal ganglia ultimately produces far-reaching network effects accounting for the clinical effects of DBS. Thus, these studies demonstrate complexity of the widespread signaling pathways in PD, the downstream effects of DBS on these networks, and the value of computational modeling in discovering such processes.

Fig. 6.1 Visualization of active electrode contact (a) in subthalamic nucleus (STN) deep brain stimulation model and volume of tissue activated (b; dark pink) from active electrode contact in relation to surrounding nuclei. Thalamus shown in yellow, STN in green, red nucleus in red. (Adapted from Sweet et al.34)

Fig. 6.2 Visualization of the dentatorubrothalamic tract traversing from dentate nucleus (a), to contralateral red nucleus (b); yellow boxes represent regions of interest.

6.2.3 Beyond Conventional Stimulation

Computational models have also been developed to determine the effects of different stimulation paradigms beyond conventional parameters such as frequency, current/voltage, and pulse width. One novel paradigm, coordinated reset, is a desynchronization technique that specifically targets pathological parkinsonian neuronal synchrony.40 Extensively described by Tass et al, the function of coordinated reset is via the delivery of brief, high-frequency pulse trains through different DBS electrode contacts, resulting in unlearning or “resetting” of pathological neuronal synchrony and synaptic connectivity.^40,41,42 First described in computational model-based stimulations, coordinated reset has been translated to early human studies in PD patients, demonstrating promising acute and cumulative improvements in motor function.^43,44

6.3 Advanced Imaging Techniques

6.3.1 Diffusion-Weighted Imaging and Tractography

The advent of sophisticated neuroimaging technologies has also improved our knowledge of complex neural circuitry and the role of DBS in modulating such pathways. As previously discussed, the clinical effects of DBS are at least in part due to the activation of adjacent WM fiber tracts from stimulation of the electrode. Visualization of these WM pathways is now increasingly possible using MR-based diffusion-weighted imaging (DWI) and tractography techniques.33 DWI demonstrates the diffusion of water molecules within the brain. Since water diffuses more readily along the direction of a cellular barrier, such as an axon, rather than across it, it can be assumed that the path of water will follow axonal pathways. Thus, DWI approximates the course of WM fibers within the brain.^45,46 From the diffusion of water, a tensor model can be applied to reveal information pertaining to the directionality of the water diffusion, allowing for diffusion tensor imaging (DTI)⁴⁶. However, while DTI may be sufficient to view large and known WM pathways, it is less helpful for visualizing smaller, more complex, and/or unknown fiber tracts. Thus, tractography techniques utilize data-driven algorithms, via various computer software platforms, to precisely identify specific WM tracts traversing between set regions of interest (ROIs) based on the raw DWI data incorporated with structural T1 MRI data sets ( Fig. 6.2).^46,47

The ability to model these axonal fibers with tractography lends itself to the application of DBS. Tractography can be used to improve understanding about anatomical networks affected by DBS, and to aid in preoperative surgical targeting as well as in postoperative assessments of outcomes. Pouratian et al used probabilistic tractography-based thalamic segmentation to evaluate the connectivity of the region of ventralis intermedius nucleus (Vim) implanted with DBS electrodes for tremor control.⁴⁸ Though the authors expected the region of Vim with the active electrode contact to have greater connectivity to the motor cortex, instead they found that there was greater connectivity to premotor cortex, thus providing insight into the connectivity networks involved in tremor control. In 2014, Rozanski et al assessed the connectivity pattern of the GPi in patients implanted with DBS electrodes for dystonia to determine why the ventral GPi is a more effective target than the dorsal region.⁴⁹ They used fiber tractography to visualize the specific WM fibers comprising the ventral and dorsal regions of the GPi, as well as their efferent projections. They found notable somatotopy within the GPi, such that the ventral region had greater connectivity to primary sensory and posterior motor cortices, whereas the dorsal region showed more connectivity to motor and premotor cortices. Therefore, these studies demonstrate the utility of tractography in furthering the understanding of connectivity and the networks affected by DBS for movement disorders.

In addition, knowledge of the location of certain WM tracts can allow for such tracts to be selectively targeted or avoided via DBS to produce the desired effects of stimulation. In 2014, Coenen et al studied 11 patients with tremor of various etiologies, who were implanted with DBS electrodes, using micro-electrode recording and awake testing to localize the optimal electrode position within the Vim.⁴⁷ The authors then performed tractography from the preoperative imaging to visualize the dentatorubrothalamic tract (DRT) and computational modeling to view the electric field of the active contacts. They found that the most effective targets were within or adjacent to the DRT and the electric fields involved the DRT, thus concluding that tractography can be used to aid in surgical targeting using DBS. Similarly, Sweet et al assessed tremor outcomes in 14 tremor-predominant PD patients implanted with STN DBS electrodes.³⁴ They determined the active contacts from programming sessions and used computational modeling to find the VTA, which was then combined with tractography to visualize the DRT ( Fig. 6.3), and found that greater tremor control correlated with a closer proximity of the active electrode contact to the WM fibers of the DRT.³⁴ Furthermore, it is also possible to combine visualization of different tracts to maximize clinical effectiveness and minimize or avoid side effects. For example, Hana et al described the preoperative determination of the DRT in relation to the corticospinal tract for DBS planning.⁵⁰ Combination tractography has also been described in target selection for high-frequency-focused ultrasound ablation, with pre- and postoperative determination of the corticospinal tract, medial lemniscus, and DRT.⁵¹