A validation approach for computational models of TMSinduced brain currents using motor evoked potentials

Abstract

The adoption of transcranial magnetic stimulation (TMS) has steadily increased in research as a tool capable to safely and non-invasively stimulate both the central and peripheral nervous systems. Initial clinical applications were limited to diagnostic use of TMS and readout signals such as electromyograms (EMG). Subsequently, repetitive TMS (rTMS) was appreciated for its therapeutics benefits as well. However, even after a decade of use of rTMS as an alternative treatment of major depression disorder in psychiatry, the mechanism of action is still not well understood. Computer models predicting the induced electric field distribution in the brain have been suggested before in the hope to resolve at least some of the uncertainty and resulting variable treatment response associated with the clinical use of TMS. We constructed a finite element model (FEM) of the head using individual volumetric tissue meshes obtained from an MRI scan and a detailed model of a TMS coil that together can predict the current induced in the head of a patient at any given location with any given coil position and orientation. We further designed several potential metrics of how a TMS induced current induced neuronal activation in the motor cortex, and added this to the model. We validated this model with motor evoked potentials (MEPs), EMG responses of the hand muscles after TMS on the motor cortex, in an experiment on 9 healthy subjects. We adopted a tailored MEP mapping protocol for model validation, which unlike traditional grid mappings, varies the TMS machine output intensity between stimulation locations. We further varied coil orientation on each point stimulated to allow exploration of the angular dependency of the model MEPs. Taken together, this approach covers a wide domain and scope of the modeled and measured responses, which are optimally suited for model validation. For each subject the motor hotspot was carefully identified using individual cortical anatomy and BOLD fMRI measurements. Modeled activation in the motor cortex did not show a good correlation to the observed magnitude of the observed MEPs, for none of the neuronal activation metrics adopted. For an activation metric that was asymmetric, taking into account induced current direction with respect to the motor cortex sulcal wall, was marginally better than other metrics. Generally all activation metrics based on induced currents performed better than a control metric agnostic of induced electric field magnitude. Our results suggest that one should take into account components of the injected currents and their relationship to the morphology of the underlying motor cortex, but the coarse metrics we used to model the relationship between induced current and neuronal activation probably did not do justice to the complex neuronal circuitry of the cortical sheet. Furthermore, it seemed MEP magnitudes in our experiment are too variable over subsequent stimulations, which could be mitigated by more repetitions per stimulation location and orientation. Further efforts to construct validated models predicting TMS effects in individual patients brains should incorporate microcircuits interactions in the cortical sheet, in addition to induced electrical field models, and take into account inherent trial to trial variability of MEPs.