(F-234) Computational modeling of spinal cord stimulation for restoring somatosensation after amputation

Introduction::

Spinal cord stimulation (SCS) can evoke sensations in a missing limb and can improve functionality of prostheses [1]. Our lab is investigating the effectiveness of lateral spinal cord stimulation (LSCS), which aims to evoke responses in the dorsal rootlets to elicit focal sensations. However, determining stimulation parameters for a participant requires long iterative testing.

The aim of this project was to build a 3D patient-specific computational model of the spinal cord (SC) to simulate the effects of LSCS and determine optimal stimulation parameter combinations. The model will enable us to investigate how LSCS evokes activity in the dorsal rootlets, while also reducing time required for manual tuning of stimulation parameters.  

The computational model consists of a finite element method (FEM) model populated with equivalent-circuit models of neurons. The FEM model calculates the electrical field generated throughout the tissue by SCS. Tissues from medical images (MRI and CT) are segmented, extruded and combined with modeled electrodes. Then the overall FEM is meshed, and a solution is computed. The solution is used with the neuron model to calculate axonal response to stimulation.

This paper will describe the process of developing the patient-specific SC solid model that is the basis for the FEM model. Studies have demonstrated that patient-specific models that accurately represent individual anatomy yield results that better match clinical measurements compared with generalized SC models [2,3]. Therefore, this work represents a critical step toward creating a model that can provide insights into the mechanisms and effects of LSCS.

Materials and Methods::

The participant was a 61-65-year-old female who was implanted with 3 stimulation leads (8 electrode contacts each). A T2 weighted MRI scan with sagittal orientation, obtained prior to electrode implantation, was used to segment the SC. A CT scan of the lumbar region, obtained within a week post-implantation, was used to estimate the placement of the electrodes.

First, we co-registered the CT scan with the MRI scan using SPM-12. Then we manually segmented the cerebrospinal fluid (CSF), white matter, epidural fat, bone, and the ventral and dorsal rootlets in FSL (Fig 1b). The MRI scan consisted of 639 slices with .4375 mm thickness. We targeted the segmented region to start and stop one vertebra above and below the electrodes. Due to the low resolution of the MRI images the gray matter (GM) and dura mater could not be differentiated from neighboring tissues.

In order to model the GM, we used a T2 weighted MRI scan of the lumbar region from a 33-year-old cadaver that had been previously segmented. Cadaver GM segmentations were co-registered to the participant’s white matter. SC levels were compared and estimated based on the location where white matter was first visible. Other studies didn’t include dura mater in their models since the conductivity difference was negligible [4]. Following this, we chose to also exclude the tissue from our model.

Finally, we extracted and smoothed the segmentation in Sim4life to create a solid model. Later, modeled electrodes will be added to the positions marked from the CT scan.   

Results, Conclusions, and Discussions::

The manual segmentation included 390 slices, spanning 170.625 mm along the rostro caudal axis. This ensured sufficient space above and below the electrodes to avoid edge effects in the simulation. Once the segmentation was extracted to a solid model and smoothed in Sim4Life (Fig 2a) a visual inspection was performed. We were primarily interested in the locations and entry points of the dorsal rootlets, which are the target of LSCS. Dorsal rootlets can be seen entering at each lumbar level and will be the structure of interest during simulation. (Fig 2b).

Once the FEM solution is computed, a hybrid model can be created and used to run simulations of LSCS. One driving hypothesis in this study is that LSCS, applied in the dorsal epidural space, recruits dorsal rootlets. Employing the use of a computational model can help to determine if dorsal rootlets are recruited with clinically relevant stimulation parameters. Therefore, to test this hypothesis, the solid model should be as accurate to the participant’s anatomy as possible. Manual segmentation allows close consideration in each slice of the spinal cord, especially when targeting the dorsal rootlets and the surrounding CSF, white matter and ventral rootlets. Activation of neurons in the GM isn’t the primary target of epidural SCS and the contrast between gray and white matter in the T2-weighted MRI was low, we chose to use images from a cadaveric specimen to estimate the location of GM. Also, difficulties adding the dura mater and literature supporting the exclusion of the tissue resulted in our decision to not include it in the model. However, the use of higher resolution medical images in the future may make solid model construction more efficient.       

A patient-specific solid model was successfully created, and the electrode placements were marked. This model forms the basis of an FEM, which can be combined with a neuron model to simulate the effects of LSCS. This hybrid model will ultimately be used to determine the recruitment of neural structures in response to LSCS and help map the activation of neural structures to sensation evoked by stimulation.

Acknowledgements (Optional): :

References (Optional): : [1] Nanivadekar A. C. et al (2022). Spinal cord stimulation restores sensation, improves function, and reduces phantom pain after transtibial amputation. medRxiv
[2] Lempka S. F. et al (2020). Patient-specific analysis of neural activation during spinal cord stimulation for pain. Neuromodulation: Technology at the Neural Interface, 23(5), 572– 581.
[3] Solanes C. et al (2021). 3D patient-specific spinal cord computational model for SCS management: potential clinical applications. J. Neural Eng. 18 036017
[4] Capogrosso M. et al (2013). A computational model for epidural electrical stimulation of spinal sensorimotor circuits. Journal of Neuroscience, 33(49), 19326– 19340.