The goal of this project is to develop an interactive, patient-specific modeling pipeline to simulate the effects of deep brain stimulation in near real-time and provide meaningful feedback as to how these devices influence neural activity in the brain.
Deep brain stimulation (DBS), which involves insertion of an electrode to deliver stimulation to a localized brain region, is an established therapy for movement disorders and is being applied to a growing number of disorders. Computational modeling has been successfully used to predict the clinical effects of DBS; however, there is a need for novel modeling techniques to keep pace with the growing complexity of DBS devices. These models also need to generate predictions quickly and accurately. The goal of this project is to develop an image processing pipeline to incorporate structural magnetic resonance imaging (MRI) and diffusion weighted imaging (DWI) into an interactive, patient specific model to simulate the effects of DBS. A virtual DBS lead can be placed inside of the patient model, along with active contacts and stimulation settings, where changes in lead position or orientation generate a new finite element mesh and solution of the bioelectric field problem in near real-time, a timespan of approximately 10 seconds. This system also enables the simulation of multiple leads in close proximity to allow for current steering by varying anodes and cathodes on different leads. The techniques presented in this paper reduce the burden of generating and using computational models while providing meaningful feedback about the effects of electrode position, electrode design, and stimulation configurations to researchers or clinicians who may not be modeling experts.
Deep brain stimulation (DBS) is an established therapy for movement disorders such as essential tremor1 and Parkinson's disease2. This therapy is also being investigated as a potential treatment for a growing number of disorders including traumatic brain injury3, Tourette's syndrome4, and depression5. DBS systems require surgical implantation of an electrode lead to deliver stimulation in a localized brain region to modulate ongoing neural activity6. The location of the electrodes and the stimulation parameters both have an effect on modulating neuronal circuits that provide therapeutic benefit. Small variations in location can affect the therapeutic window, possibly increasing the likelihood of adverse side-effects before therapeutic benefit is achieved7,8,9. In practice, it is often difficult to predict the effects stimulation will have on neural activity; consequently, this window of therapeutic benefit is identified on a patient-by-patient basis as the stimulation device is programmed by the clinician8,9. This process is becoming more complex as new generations of DBS devices become available. For example, novel lead designs are being introduced with more contacts10,11,12, and in some cases multiple leads are being implanted in close proximity to one another13. Hence, there is a need to be able to explore and predict the effects of DBS across a large and growing parameter space.
Computational modeling and analysis can be used to predict the physiological and clinical effects of DBS on a patient-specific basis. These models use finite element modeling (FEM) to build three dimensional representations of the brain tissue and the biophysical characteristics of the implanted electrode. FEM bioelectric field models have been successfully used to predict the effects of DBS14, but until now these have been time-consuming and computationally costly to generate. There is a need for novel modeling techniques to keep pace with the growing complexity of DBS devices. These patient-specific models should provide near real-time visual feedback on the effects of DBS as the lead location or stimulation parameters are changed. The user would obtain feedback about a lead location and stimulation setting in a few seconds, enabling continued refinement of lead placement over the course of several minutes. Patient-specificity is achieved by incorporating the patient's anatomy, their brain's shape and size, when building the FEM and applying the biophysical properties of their brain, such as anisotropic tissue conductivity. Anisotropic conductivity describes how current will spread through different brain regions and can be non-invasively measured for the entire brain similar to a typical magnetic resonance image (MRI).
DBS modeling approaches that do not use patient-specific information can provide fast, but less accurate predictions of stimulation effects, due to generalized geometries and conductivity values for brain tissue. In this approach, a single FEM is used for all patients and the predicted neural activity can be computed in advance. Patient-specific models cannot be generalized and pre-computed since a new FEM is built for each individual. These models require more effort to build but can be more accurate. Several factors limit the speed at which these models can be built and used: 1) Modifying parameters at the beginning of the model building pipeline, such as electrode position, requires manual effort to update all of the subsequent steps; and 2) The steps in the modeling pipeline are not easily integrated with one another, requiring the passing of data between multiple software packages. Often, we want to evaluate many different situations such as electrode position, stimulation parameters, or electrode designs. To provide meaningful feedback about the effect these changes have on the therapeutic effect the patient will receive, these results should be both accurate and generated quickly.
Our goal is to present new techniques for building patient-specific models that take advantage of the speed obtained in generalized models and automating many of the pipeline steps to create an interactive modeling environment that provides near real-time visual feedback about the effects of DBS. An interactive simulation allows a user to test predictions and obtain results quickly without focusing on details of model construction. This is beneficial when there is a large parameter space to explore and how these parameters influence the simulation are uncertain. We will describe the steps in the processing pipeline to generate interactive, patient-specific FEM models from magnetic resonance imaging (MRI) acquisitions. Utilizing the tools and techniques outlined in this paper will reduce the time cost for creating FEM bioelectric field models and provide a way to make these models accessible to researchers and clinicians who are not modeling experts.
This protocol describes how to build a patient-specific finite element model from acquired MRI volumes, and then simulate the electric field induced by a DBS electrode. The main steps in the creation of these models are to: 1) build a finite element model (FEM) representing the patient's brain and the implanted electrode, 2) add biophysical properties of the brain and stimulation parameters of the DBS device to the FEM, and 3) solve for the voltage generated by the electrode in the model. Two imaging modalities are needed to build a patient specific model for DBS simulation. A T1 MRI is used to construct segmentations of the brain surface, ventricles, and specific nuclei. Diffusion weighted imaging (DWI), a measure of water diffusivity, is used to estimate diffusion tensors throughout the brain tissue15. The diffusion tensors are converted into conductivity tensors which quantify the inhomogeneous, anisotropic biophysical properties of the tissue on a voxel by voxel basis16. The voltage distribution throughout the brain induced by the electrode is computed by solving the Poisson equation, which through the application of FEM simplifies to a linear system of equations Ax = b where A is a stiffness matrix that represents the conductivity and geometry of the mesh, x is the voltage solution at each node in the mesh, and b is modified based upon boundary conditions and current sources.
This protocol has demonstrated techniques to reduce the time burden and computational cost of building patient specific models for DBS to a degree that allows near real-time feedback of simulation results. Rapid feedback enables the exploration of a large parameter space to better understand how parameter changes affect the model results. These parameters include lead location, choice of active contacts, and the amplitude, pulse width, and frequency of the stimulation waveform at those contacts. The key features of the proposed tool are: 1) A simple user interface to adjust the model parameters with near real-time visualization of how these parameters influence the simulation, and 2) Automation of model creation from a small set of inputs: the brain surface, brain tissue conductivity tensors, and a surface representation of the electrode geometry. This automation expedites the creation of patient-specific models for many individuals who have different brain geometries and tissue conductivities as well as evaluating the effects of inserting different electrode designs into existing models. The image pre-processing steps described in this protocol were not fully automated and can take up to a day of processing time. However, once completed the data generated from these steps are considered static, meaning this data is not modified during the simulations. The automation of model creation arises from the ability of the system to apply this data to the patient-specific FEM without manual effort. The SCIRun network to perform all of the model generation, simulation, visualization steps only needs to be built once. Therefore, only the image pre-processing steps need to be performed again to generate a patient-specific model for an incoming patient.
The performance gain in generating results from the modeling pipeline is due to the integration of mesh generation, bioelectric field calculations, and visualization of the solution into a single software environment. Existing modeling techniques such as adaptive mesh generation were used to create higher mesh density around the electrode and lower density farther from the electrode which reduces the time to build and solve the FEM. The software, SCIRun, also enables automation of the mesh generation and bioelectric field calculations. User-defined movement of the electrode using the interactive widgets triggers the construction of a new mesh with the updated electrode position. This includes modifying the boundary conditions and conductivity values for the new electrode position.
The electrode geometry is treated as a free-moving object inside of the brain volume before its position is integrated into the FEM. A key implication of this approach for mesh construction is that multiple electrodes can easily be inserted into the model. For example, a second copy of the electrode geometry can be placed several millimeters away and both will be included in the FEM. In recent research, two electrodes have been implanted in close proximity to treat multiple sclerosis tremor13 and have been used in non-human primate experiments to explore effective stimulation targets21. The benefit of using multiple electrodes is to provide better control of the electric field generated in the tissue over a larger area. Stimulating with active contacts on both electrodes can steer current towards the target region and away from regions that would lead to negative side effects. Finer control of stimulation over a larger area is also useful to explore different stimulation locations when the exact location of the intended target is unknown, as is the case with many of the emerging therapy applications of DBS. However, determining the parameters to achieve therapeutic stimulation is more challenging than a single electrode due to the increase in an already large parameter space.
We envision that this interactive modeling tool could provide benefit during pre-operative planning for DBS implantation. Feedback about the extent of stimulation in the brain tissue can allow surgeons to modify the electrode location in their surgical plan to provide therapeutic stimulation to the target region. While DBS therapy was the primary motivating factor for developing this tool, the techniques presented in this paper can be applied to any FEM bioelectric field model with different stimulation or recording paradigms. Stimulation therapies such as transcranial direct current stimulation for depression22 or the use of depth electrodes for the treatment of epilepsy23 share the same challenges as DBS in determining the best location of stimulation to achieve therapeutic results. Electrocorticography, a recording technique with arrays of electrodes on the surface of the brain to identify seizure onset regions, has the challenge of determining where to place the electrodes to record from target regions in the brain24. All of these applications are dependent on electrode position while dealing with the uncertainty of how current flows through the brain tissue. The techniques presented in this paper reduce the burden of generating and using computational models while providing meaningful feedback to researchers and clinicians using these devices who are not modeling experts.
The authors have nothing to disclose.
This project was supported by the National Institute of Health Grants UH3, NS095554. Technical support was provided by the Center for Integrative Biomedical Computing at the Scientific Computing and Imaging Institute and was made possible in part by software developed from the NIH P41-GM103545, Center of Integrative Biomedical Computing.
Gratitude is extended to Lexie Floor and Nathan Galli at the Scientific Computing and Imaging Institute for production and editing of the submission video, and also to Theresa Lins for assistance with manuscript preparation.
FreeSurfer | Athinoula A. Martinos Center for Biomedical Imaging | https://surfer.nmr.mgh.harvard.edu/ | |
3D Slicer | BWH, Harvard University | https://www.slicer.org/ | |
SCIRun | University of Utah Center for Integrative Biomedical Computing | http://www.sci.utah.edu/cibc-software/scirun.html |