The overall goal of this project is to develop an interactive patient-specific modeling pipeline and stimulate the effects of deep brain stimulation on neural fiber pathways in near real time. Deep brain stimulation is a therapeutic intervention that's been FDA approved and used for over a decade for movement disorders such as Parkinson's disease, and it's been investigated for a wide range of other neurological and psychiatric conditions. Computational modeling of deep brain stimulation has been a tool that's been quite useful, and the use of it is grown in order to better understand and predict the effects of stimulation in individual subjects.
One of the persistent problems with computational modeling is the amount of time that it's taken to construct and run computational model experiments. The purpose of this protocol is to explain ways that the computational modeling can be done in near real time and in a much more interactive manner, and this is achieved through a few innovations. The major one is the ability to do interactive meshing of whole heads of individual subjects.
The protocol in this video will describe how to build a patient-specific finite element model from acquired MRI volumes and then simulate the electric field induced by an implanted deep brain stimulation electrode. The main steps in the creation of these models are to, one, build a finite element model representing the patient's brain and the implanted electrode, two, add biophysical properties of the brain and stimulation parameters of the deep brain stimulation device to the finite element mesh, and three, solve for the voltage generated by the electrode in the model. The electrode modeled in this project is a Medtronic 3387 DBS Lead.
However, any electrode geometry can be used with this approach. The 3387 is a cylindrical lead with four contacts separated by non-conductive material. Surface representations for each of these components are created for the meshing process.
Closed surfaces are created for the four contacts, the five shaft segments, the encapsulation layer, and two surrounding boxes which are used to control mesh density around the electrode. It is important to create a high density surface mesh of the contacts to preserve the cylindrical geometry and accurately solve the bioelectrical field problem. The voltage distribution through the brain induced by the electrode is computed by solving a Poisson equation which through the application of finite element modeling simplifies to a linear system of equation, Ax equals B, where A is the stiffness matrix that represents the conductivity and geometry of the mesh, x is the voltage solution at each node on the mesh, and B is modified based upon boundary conditions and current sources.
The encapsulation layer, shaft segments, and the four contacts must be assigned conductivity values as well based upon their material properties shown in the table on the screen. Fixed boundary conditions are used to calculate the bioelectric field. Point surfaces are placed in the center of each of the four DBS contacts, and monopolar or bipolar stimulation can be applied through any combination.
The outer brain surface is set to zero volts to simulate a distant return electrode. These conditions represent the known voltage on the finite element model. We know the voltage at each of the four contacts and at the return electrode, which is the implantable pulse generator.
The linear system of equations is solved using a conjugate gradient method to calculate the voltage solution in the model for a set of boundary conditions. SCIRun is an advanced problem solving environment for modeling, simulation, and visualization of several classes of scientific problems. The visual programming syntax allows users to connect software models together to process and visualize scientific data.
For this protocol, SCIRun is used to create a finite element mesh of the brain and deep brain stimulation lead and solve the bioelectric field problem in an interactive environment. One key aspect of SCIRun is the ability to interactively visualize data at any point in the pipeline, which we will demonstrate. The first step in building the finite element model is to determine the structural anatomy of the brain.
To perform this step, we chose to use FreeSurfer, an open source software suite for processing and analyzing MRI volumes for human brains. To begin, patient T1 MRI is passed to FreeSurfer as a DICOM stack with the following command. Next, the following command is used to start the automated processing and segmentation.
FreeSurfer creates a new image volume of the brain tissue with the skull removed. Intensity thresholding and isosurfacing are performed on this new volume to obtain a surface representation of the brain which is then used to define the outer boundary of the finite element model. The FreeSurfer pipeline also segments white matter, gray matter, subcortical nuclei, and ventricles.
Each brain region is assigned a unique integer label. The next pre-processing step is to estimate diffusion tensors from the diffusion weighted imagining data and co-register this volume with the TI MRI. DWI is a measurement of water diffusivity through brain tissue in multiple directions.
For this step, we chose to use 3DSlicer, a free open source software platform for medical image processing and 3D visualization of imaging data. First, convert the DICOM series into a single DWI volume using the DWI converter module in Slicer. In order to eliminate background and noise from the tensor reconstruction, create a mask with the diffusion weighted volume masking module.
This will allow tensors to only be calculated within the brain tissue. The next step is to use the DWI to DTI Estimation module to generate the diffusion tensors. Set the tensor estimation algorithm to least squares and check Shift Negative Eigenvalues.
This will remove tensors with negative eigenvalues related to noise or acquisition error. The 3D surface geometries for the electrode model can be built using any 3D modeling software, which will not be demonstrated in this video. Here we created a closed surface mesh for each of the electrode contacts, shaft segments, encapsulation layer, and two surrounding boxes that are used to control mesh density.
The brain surface created in part one and the electrode geometry are combined to create a finite element model of the brain with an implanted electrode using the meshing tool TetGen. Pipe the brain surface and the electrode geometry into the InterfaceWithTetGen module. Apply volume constraints to the tetrahedral element generation in each region of the model, as shown in the table on the screen.
This technique creates an adaptive size mesh with a higher mesh density around the electrode tissue interface where the electric field gradient is high. The result is a tetrahedral finite element mesh of the electrode model embedded in brain tissue. SCIRun enables the movement of objects with geometric transform modules and interactive widgets in the viewer window.
To change the angle of the electrode geometry, use the CreateGeometricTransform module, combined with the TransformMesh module. Set the X rotation axis to one, and set the degree of rotation for theta to control the angle. Connect a GenerateSinglePointProbe module directly to the ViewScene to create a sphere widget and also to another TransformMesh module to control the electrode tip location.
In the view scene, Shift Click and drag the widget to move the location of the electrode tip. These transforms are applied to the electrode geometry before it is passed to TetGen, so the updated electrode location and orientation is built into the finite element model. The next step is to assign conductivity values to the finite element model.
Load the DTI data that was processed in 3DSlicer and set the conductivity values for both the electrode model and the brain tissue onto the tetrahedral mesh created in the previous section. After conductivity values are applied to the mesh, pipe the field into BuildFEMatrix. This is the first step in creating the system of equations for the bioelectrical field calculation.
For voltage controlled stimulation, we will now set the known voltage boundary conditions on the finite element mesh. Set the brain surface used as the outer boundary of our mesh to zero volts to simulate the distant return electrode. InsertVoltageSource is then used to apply this boundary condition on the mesh.
For monopolar stimulation, assign a voltage value to a point in the center of the bottom contact. Again, use InsertVoltageSource to apply this boundary condition on the mesh. Combine the output of BuildFEMatrix and InsertVoltageSource into SolveLinearSystem.
Use the conjugate gradient solving method and execute the module to calculate the voltage solution. Place the solution to the system of equations back onto the nodes of the finite element mesh using the SetFieldData module. The voltage solution can be viewed directly on the nodes of the mesh but does not provide a meaningful representation of its distribution through the brain tissue.
A better visualization is to create isovoltage surfaces similar to contour lines on a map, which also provides information about the orientation and strength of the electric field. Pass the finite element mesh voltage solution into ExtractSimpleIsosurface and set the quantity to 10. This will create 10 evenly spaced isovoltage surfaces around the contact.
With this technique, we can visualize the effect of adding anisotropic conductivity tensors to the model and how this parameter influences the voltage distribution throughout the brain tissue. Integration of the finite element construction, bioelectrical field calculations, and visualization of the results into one environment enables near real time feedback of how parameter changes such as electrode position and stimulation amplitude influence how the bioelectric field interacts with the surrounding tissue. After the electrode position is modified and the bioelectric field is recalculated, the solution is mapped onto target fiber bundles.
The activating function is calculated to predict fiber bundle activation for a given stimulation amplitude. We collect fiber bundle activation predictions across all contacts and multiple stimulation amplitudes and quickly determine which stimulation configuration would be most effective in activating the target fibers for a given electrode position. We then compare these activation profiles for multiple electrode positions during the planning stages of DBS surgery to define the electrode trajectory that will effectively stimulate the target fiber pathway.
Here we see that position two predicts a more robust activation of the target fiber bundle have lower stimulation amplitudes across all contacts compared to position one. This approach can be used to model much more complex scenarios. A novel feature of this system is the ability to quickly switch the electrode geometry in the model and the ability to simulate multiple leads in close proximity to one another.
Here we are investigating lead locations to stimulate central thalamic fiber pathways involved in arousal regulation as a potential target for patients with traumatic brain injury while avoiding nearby areas such as the sensory nuclei. For a static electrode position, we simulated the effects of changing to a directional electrode model to gain more focused control of fiber activation locations and avoid stimulation of other fiber bundles. Looking at a cross-sectional area around the electrode, we see that the directional electrode is able to effectively stimulate the target fibers with minimal spread of activation into fibers we wish to avoid.
This approach would also be useful for subthalamic nucleus targeting to treat symptoms of Parkinson's disease while avoiding the nearby internal capsule. Here we simulate the placement of three independent DBS leads and predict the activation of similar central thalamic fiber pathways for surgical planning in non-human primate experiments. The benefit of using multiple electrodes is to provide control of stimulation over a larger area.
Both directional electrodes and multiple leads increase the difficulty of determining electrode stimulation parameters. However, utilizing the approach outlined in this video enables faster and more guided exploration of the complex parameter space to identify therapeutic stimulation settings. The patient-specific modeling pipeline outlining the steps from patient imaging to interactive models that we showed you today will help close the gap between engineers and clinicians.
The development of new DBS devices with more than four contacts, including the insertion of multiple electrodes, adds more complexity to an already difficult problem. The ability to quickly visualize the effects of different electrode designs in stimulation configurations can provide valuable feedback to better understand how these devices operate in the brain.
