Poster No:
129
Submission Type:
Abstract Submission
Authors:
Parsa Oveisi1,2,3, Davide Momi1, Zheng Wang1, Sorenza Bastiaens1,4, Taha Morshedzadeh1,4, Christoph Zrenner2,3,5, John Griffiths1,4,3,5
Institutions:
1Krembil Centre for Neuroinformatics, Centre for Addiction and Mental Health (CAMH), Toronto, Canada, 2Temerty Centre for Therapeutic Brain Intervention, Centre for Addiction and Mental Health (CAMH), Toronto, Canada, 3Institute of Biomedical Engineering, University of Toronto, Toronto, Canada, 4Institute of Medical Science, University of Toronto, Toronto, Canada, 5Department of Psychiatry, University of Toronto, Toronto, Canada
First Author:
Parsa Oveisi
Krembil Centre for Neuroinformatics, Centre for Addiction and Mental Health (CAMH)|Temerty Centre for Therapeutic Brain Intervention, Centre for Addiction and Mental Health (CAMH)|Institute of Biomedical Engineering, University of Toronto
Toronto, Canada|Toronto, Canada|Toronto, Canada
Co-Author(s):
Davide Momi
Krembil Centre for Neuroinformatics, Centre for Addiction and Mental Health (CAMH)
Toronto, Canada
Zheng Wang
Krembil Centre for Neuroinformatics, Centre for Addiction and Mental Health (CAMH)
Toronto, Canada
Sorenza Bastiaens
Krembil Centre for Neuroinformatics, Centre for Addiction and Mental Health (CAMH)|Institute of Medical Science, University of Toronto
Toronto, Canada|Toronto, Canada
Taha Morshedzadeh
Krembil Centre for Neuroinformatics, Centre for Addiction and Mental Health (CAMH)|Institute of Medical Science, University of Toronto
Toronto, Canada|Toronto, Canada
Christoph Zrenner
Temerty Centre for Therapeutic Brain Intervention, Centre for Addiction and Mental Health (CAMH)|Institute of Biomedical Engineering, University of Toronto|Department of Psychiatry, University of Toronto
Toronto, Canada|Toronto, Canada|Toronto, Canada
John Griffiths
Krembil Centre for Neuroinformatics, Centre for Addiction and Mental Health (CAMH)|Institute of Medical Science, University of Toronto|Institute of Biomedical Engineering, University of Toronto|Department of Psychiatry, University of Toronto
Toronto, Canada|Toronto, Canada|Toronto, Canada|Toronto, Canada
Introduction:
The growing potential of computational brain models in clinical and research applications is increasingly evident. A crucial advantage for these models is their ability to generalize from one scenario to predict other untested conditions. For instance, modeling a patient's data to simulate their response to various treatments can help identify the most effective treatment protocol.
Transcranial magnetic stimulation (TMS) therapy is emerging as a promising treatment for neuropsychiatric disorders like depression1. A key factor affecting brain engagement in TMS is stimulation intensity, which engages individual brains differently2. Our group has developed a framework for whole-brain connectome-based neural mass modeling in PyTorch ('whobpyt'), previously used to investigate TMS propagation patterns across brain networks3. The present study leverages whobpyt to investigate TMS-evoked brain dynamics at different intensities, and the model's predictive and generalization capabilities. This research will enhance our understanding of TMS-brain interactions and facilitate in-silico testing of TMS treatments for better clinical outcomes.
Methods:
We collected single-pulse TMS-EEG data (64 channels, stimulating at the dorsolateral prefrontal cortex) from 21 healthy participants. Pulses were randomized at five different intensities in each session based on participants' unique resting motor threshold (%80 to %120).
Empirical analyses: We quantify the TMS-induced brain activity, as well as its response profiles to varying stimulation intensities in terms of: a) localized peak latencies and amplitude changes near the site of stimulation (via Pearson correlation and linear regression), and b) broader spatiotemporal activity changes at different stimulation intensities through permutation cluster testing4.
Modeling analyses: We train our model at all 5 intensity conditions to assess if any results in better goodness-of-fit (assessed via cosine similarity (CS) with empirical data). We then assess each of the 5 fitted models' overall prediction accuracy, as well as their ability to replicate the intensity scaling patterns quantified in the empirical analyses. Lastly, we examine relationships between various biologically-interpretable model parameters and intensity of TMS stimulation (via Pearson correlation) to infer potential underlying causes leading to the brain's intensity scaling patterns.
Results:
The model showed overall high fitting capabilities (mean CS=0.88, SD=0.04), across different intensities, with performance increasing with intensity (r=0.39, p<0.001). Predictive accuracy, though generally high (mean CS=0.77, SD=0.12), was highest when the difference between fitted and simulated intensities was minimal. For the model's intensity scaling profiles, while the same empirical peak latencies were replicated, similar amplitude scaling was observed primarily for early peaks at around 75ms (r=-0.27, p<0.05). In spatiotemporal patterns, the models trained on higher intensities partially replicated certain empirical clusters, particularly those observed in frontocentral regions around 200ms. However, these simulated clusters did not reach statistical significance.
Finally, we also found significant correlations between TMS stimulation intensity and the following model parameters: pyramidal-to-excitatory interneuron gains (r=+0.22, p<0.05), pyramidal-to-inhibitory interneuron gains (r=-0.25, p<0.01), and the inhibitory interneuron time constant (r=-0.21, p<0.05). Taken together, results suggest a role of cortical excitation-inhibition balance in TMS-related intensity response profiles.
Conclusions:
Here, we demonstrated the capability of our model to reliably capture and replicate the dynamics of TMS across various intensities, albeit with potential for further refinement. Furthermore, we illustrated the utility of analyzing model parameters to deduce the underlying biological mechanisms governing TMS engagement.
Brain Stimulation:
TMS 1
Modeling and Analysis Methods:
EEG/MEG Modeling and Analysis 2
Task-Independent and Resting-State Analysis
Keywords:
ADULTS
Computational Neuroscience
Data analysis
Electroencephaolography (EEG)
Machine Learning
Modeling
Transcranial Magnetic Stimulation (TMS)
1|2Indicates the priority used for review
Provide references using author date format
1. Saini, R. et al. (2018) ‘Transcranial magnetic stimulation: A review of its evolution and current applications’, Industrial Psychiatry Journal, 27(2), p. 172. doi:10.4103/ipj.ipj_88_18.
2. Pellegrini, M., Zoghi, M. and Jaberzadeh, S. (2018) ‘The effect of transcranial magnetic stimulation test intensity on the amplitude, variability and reliability of motor evoked potentials’, Brain Research, 1700, pp. 190–198. doi:10.1016/j.brainres.2018.09.002.
3. Momi, D., Wang, Z. and Griffiths, J.D. (2023) ‘TMS-evoked responses are driven by recurrent large-scale network dynamics’, eLife, 12. doi:10.7554/elife.83232.
4. Maris, E. and Oostenveld, R. (2007) ‘Nonparametric statistical testing of EEG- and Meg-Data’, Journal of Neuroscience Methods, 164(1), pp. 177–190. doi:10.1016/j.jneumeth.2007.03.024.