Thalamo-centric causal connectivity mapping in human brain with intracranial electrical stimulation
Presented During:
Monday, June 24, 2024: 5:45 PM - 7:00 PM
COEX
Room:
Grand Ballroom 104-105
Poster No:
33
Submission Type:
Abstract Submission
Authors:
Dian Lyu1, Josef Parvizi1
Institutions:
1Stanford University, Palo Alto, CA
First Author:
Co-Author:
Introduction:
The brain's spatiotemporal architecture, marked by functional connectivity motifs, is key to brain health and consciousness. Emerging theories highlight the thalamus' neuromodulatory role in shaping cortical connectivity motifs (Shine et al., 2023). However, testing these theories is challenging due to the small size, deep location and functional complexity of subcortical areas, where non-invasive neuroimaging techniques face limitations. In Stanford, we pioneered multi-site stimulation and recording in the thalamus using deep intracranial electrodes for mapping thalamic-centric causal connectivity. Thalamic stimulations are shown to evoke distinct EEG profiles than cortical stimulations from a recent mice study (Claar et al., 2023). Meanwhile, we have limited knowledge from direct human thalamic measurements. Therefore, our goal is to extract meaningful neural features from stimulation evoked potentials (SEP), then infer whole-brain causal connectivity.
Methods:
In this study, we recruited 27 participants with focal epilepsy with implanted electrodes for clinical purpose. Employing our standard single pulse electrical stimulation protocol, we investigated causal connectivity by stimulating a bipolar electrode pair while recording from all others.
We utilized UMAP algorithm to encode neural signals (McInnes et al., 2020), with the input of time-variant power and inter-trial phase coherence spectrograms of SEPs. We derived activation labelling by employing semi-supervised learning, after partially labelling activations manually based on preset criteria. Then, we employed group-level supervised UMAP to map the activated spectrograms to the anatomical labels: THAL-ipsi, THAL-contr, COR-ipsi, and COR-contr (stimulating from THAL/COR in the ipsi/contralateral hemisphere). Category-specific spectral features were determined with cluster-based permutation significance testing.
To decode individual evoked signals using these features, we used the spectral information from each significant cluster as a template and apply sliding-window cross-correlation to track dynamic changes in feature proximity for all trials.
We utilized UMAP algorithm to encode neural signals (McInnes et al., 2020), with the input of time-variant power and inter-trial phase coherence spectrograms of SEPs. We derived activation labelling by employing semi-supervised learning, after partially labelling activations manually based on preset criteria. Then, we employed group-level supervised UMAP to map the activated spectrograms to the anatomical labels: THAL-ipsi, THAL-contr, COR-ipsi, and COR-contr (stimulating from THAL/COR in the ipsi/contralateral hemisphere). Category-specific spectral features were determined with cluster-based permutation significance testing.
To decode individual evoked signals using these features, we used the spectral information from each significant cluster as a template and apply sliding-window cross-correlation to track dynamic changes in feature proximity for all trials.
·Fig. 1. Illustration of neural feature encoding and decoding processes.
Results:
Neural features 1 and 2, corresponding to gamma (~40ms) and high theta (75-165ms), respectively, were distinguished in ipsilateral vs. contralateral recordings of cortical stimulations (COR-ipsi vs. COR-contr). These clusters align with documented N1 and N2 components in brain stimulation literature (Keller et al., 2014). Notably, a third cluster (Feature3) in thalamus stimulations, peaking late (>165ms) and lasting ~250ms in the theta band, is distinct from the N2 component.
Examining whole-brain causal connectivity matrices based on feature presentation, we observe: (1) modularity within adjacent anatomical areas in the Feature-1 matrix, suggesting direct connectivity (Fig. 3b); (2) widespread Feature-2 representations across regions and hemispheres, maintaining first-order modularity, indicating indirect connectivity building on the initial connectivity; (3) Feature-3 matrix showing whole-brain connectivity from thalamus without first-order modularity, suggesting persisting thalamocortical feedback. Comparing thalamic subdivisions, anterior thalamus (antTH) exhibits stronger Feature 1 and 3 representations with frontal areas, indicating recurrent connectivity, while posterior thalamus (pstTH) has more connections with parietal and occipital areas than antTH.
Examining whole-brain causal connectivity matrices based on feature presentation, we observe: (1) modularity within adjacent anatomical areas in the Feature-1 matrix, suggesting direct connectivity (Fig. 3b); (2) widespread Feature-2 representations across regions and hemispheres, maintaining first-order modularity, indicating indirect connectivity building on the initial connectivity; (3) Feature-3 matrix showing whole-brain connectivity from thalamus without first-order modularity, suggesting persisting thalamocortical feedback. Comparing thalamic subdivisions, anterior thalamus (antTH) exhibits stronger Feature 1 and 3 representations with frontal areas, indicating recurrent connectivity, while posterior thalamus (pstTH) has more connections with parietal and occipital areas than antTH.
·Fig. 2. Whole-brain causal connectivity captured by feature representation (r).
Conclusions:
We encoded whole-brain stimulation-evoked potentials into 3 neural features, representing direct connectivity, indirect connectivity via cortex, and indirect connectivity via thalamocortical feedback. Decoding revealed that the thalamus receives direct connectivity from the whole brain, while its direct cortical projection is limited to the same hemisphere. In contrast, the indirect thalamocortical late feedback spans the entire brain, acting as a propagator of theta oscillations persisting in cortical signals for approximately 200 ms post-thalamic excitation.
Brain Stimulation:
Direct Electrical/Optogenetic Stimulation 1
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural) 2
Multivariate Approaches
Neuroanatomy, Physiology, Metabolism and Neurotransmission:
Anatomy and Functional Systems
Subcortical Structures
Keywords:
Consciousness
Data analysis
Machine Learning
Other - thalamocortical connectivity
1|2Indicates the priority used for review
Provide references using author date format
Keller, C. J., Honey, C. J., Mégevand, P., Entz, L., Ulbert, I., & Mehta, A. D. (2014). Mapping human brain networks with cortico-cortical evoked potentials. Philosophical Transactions of the Royal Society B: Biological Sciences, 369(1653), 20130528. https://doi.org/10.1098/rstb.2013.0528
McInnes, L., Healy, J., & Melville, J. (2020). UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction (arXiv:1802.03426). arXiv. https://doi.org/10.48550/arXiv.1802.03426
Shine, J. M., Lewis, L. D., Garrett, D. D., & Hwang, K. (2023). The impact of the human thalamus on brain-wide information processing. Nature Reviews Neuroscience, 1–15. https://doi.org/10.1038/s41583-023-00701-0