Group-level current source reconstruction of the MNI iEEG atlas
Poster No:
1657
Submission Type:
Abstract Submission
Authors:
Ying Wang1, Min Li2, Deirel Paz-Linares1, Ariosky Areces-Gonzalez1, Jorge Bosch-Bayard3, Maria Bringas-Vega1, Pedro Valdes-Sosa1
Institutions:
1University of Electronic Science and Technology of China, Chengdu, Sichuan, 2Hangzhou Dianzi University, Hangzhou, Zhejiang, 3McGill University, Chengdu, Quebec
First Author:
Co-Author(s):
Introduction:
Multichannel imaging tools, such as fMRI and EEG/MEG, capture brain functional activity. However, these measures rely on indirect physical responses or face limitations in spatial/temporal resolution. iEEG offers direct observation but exhibits heterogeneous coverage due to invasiveness. This paper addresses the incomplete observation problem by proposing a method for whole-brain activity construction. Our approach reconstructs activity in uncovered regions and corrects amplitude differences across patients, enabling the generation of a comprehensive activation map across the entire brain. This method is compatible with other imaging modalities.
Methods:
We employ Electrophysiological Source Imaging (ESI) for whole-brain activity reconstruction. Due to electrode number and position limitations, individual brain activity is often reconstructed in nearby patches. Instead of Monte-Carlo simulations for confidence intervals, this paper opts for the screening method (Fan and Lv 2008) for high-dimensional regression. Individual target regions are obtained by truncating the cumulative correlation between a smooth simulation and its inverse solution, in which the correlation is sorted by leadfield gain.
Selected patches per subject are small; most cannot estimate the entire brain individually. A group-level approach is necessary to create a whole-brain activation map due to individual differences and system deviations. Scaling each subject's data to the same level is crucial. In scalp EEG analysis, the Global Scale Factor (GSF) (Hernández et al., 1994) normalizes amplitudes, facilitating population metrics analysis (e.g., spatial patterns, norms (Li et al. 2022)). GSF, however, falls short for sEEG and ECoG due to randomly distributed electrodes. Our method estimates the scale factor for sEEG and ECoG by assuming energy distribution fluctuates from a similar spatial pattern across subjects. The aligned space for this scale estimation is the group-averaged cortex or volume space, exemplified by the MNI cortical space.
The leadfield, obtained from Denuro FEM via the Brainstorm (Medani et al. 2023), was converted from bipolar to unipolar montage. Activation in selected regions for each subject was estimated using the eLORETA (Pascual-Marqui 2007) estimator.
In this group-level space, we address the scale issue between subjects by modeling the source log spectrumlog10(s ̂_jj (i)) for each subject using IOLMM (Incomplete Observation Linear Mixed Effect Model) outlined in Eq (1) in the figure.
The X(i) is a binary selection matrix for each subject, choosing confident sources based on the above criteria for reconstruction. The population mean log10(s ̅_jj ) is modeled with the fixed effect. The observation scale problem is addressed with a univariate random effect, where z(i) is 1 and b(i) is the scalar scale factor for each subject in the log scale. The additive independent error ε(i) is modeled with Gaussian noise N(0,σ^2 ). The problem is solved by maximizing the likelihood with the EM (Expectation Maximization) algorithm (McLachlan and Krishnan 2008).
The scale factor for the original signal is obtained as c(i)=√(10^b(i) ) , along with the group pattern s ̅_jj. The figure with a bipolar colormap illustrates the entire process.
Selected patches per subject are small; most cannot estimate the entire brain individually. A group-level approach is necessary to create a whole-brain activation map due to individual differences and system deviations. Scaling each subject's data to the same level is crucial. In scalp EEG analysis, the Global Scale Factor (GSF) (Hernández et al., 1994) normalizes amplitudes, facilitating population metrics analysis (e.g., spatial patterns, norms (Li et al. 2022)). GSF, however, falls short for sEEG and ECoG due to randomly distributed electrodes. Our method estimates the scale factor for sEEG and ECoG by assuming energy distribution fluctuates from a similar spatial pattern across subjects. The aligned space for this scale estimation is the group-averaged cortex or volume space, exemplified by the MNI cortical space.
The leadfield, obtained from Denuro FEM via the Brainstorm (Medani et al. 2023), was converted from bipolar to unipolar montage. Activation in selected regions for each subject was estimated using the eLORETA (Pascual-Marqui 2007) estimator.
In this group-level space, we address the scale issue between subjects by modeling the source log spectrumlog10(s ̂_jj (i)) for each subject using IOLMM (Incomplete Observation Linear Mixed Effect Model) outlined in Eq (1) in the figure.
The X(i) is a binary selection matrix for each subject, choosing confident sources based on the above criteria for reconstruction. The population mean log10(s ̅_jj ) is modeled with the fixed effect. The observation scale problem is addressed with a univariate random effect, where z(i) is 1 and b(i) is the scalar scale factor for each subject in the log scale. The additive independent error ε(i) is modeled with Gaussian noise N(0,σ^2 ). The problem is solved by maximizing the likelihood with the EM (Expectation Maximization) algorithm (McLachlan and Krishnan 2008).
The scale factor for the original signal is obtained as c(i)=√(10^b(i) ) , along with the group pattern s ̅_jj. The figure with a bipolar colormap illustrates the entire process.
·The procedures and model for the reconstruction.
Results:
Utilizing the MNI atlas dataset (Frauscher et al. 2018), our figure displayed rescaled data results. The left map reveals a global energy pattern congruent with EEG source localization of alpha rhythm in the eye-closed condition, with heightened occipital lobe activation. On the right, a measure of signal nonlinearity, max bicoherence, demonstrating the generalized nonlinearity of iEEG courses. This nonlinearity is particularly prominent in the parietal lobe.
·The left figure is the energy (0-80Hz) distribution estimated from 105 patients. The right figure is the maximum bicoherence (0-80Hz panel) distribution estimate from 105 subjects.
Conclusions:
We offer a robust tool for constructing group-level intracranial activation from iEEG data across various subjects. This approach facilitates the integration and comparison of direct biophysical measurements of neural activity with other brain imaging modalities.
Modeling and Analysis Methods:
Activation (eg. BOLD task-fMRI)
EEG/MEG Modeling and Analysis 1
Methods Development
Multivariate Approaches 2
Novel Imaging Acquisition Methods:
EEG
Keywords:
Data analysis
Data Organization
ELECTROCORTICOGRAPHY
Electroencephaolography (EEG)
ELECTROPHYSIOLOGY
Source Localization
Spatial Normalization
Statistical Methods
1|2Indicates the priority used for review
Provide references using author date format
Fan, Jianqing, and Jinchi Lv. 2008. “Sure Independence Screening for Ultrahigh Dimensional Feature Space.” Journal of the Royal Statistical Society: Series B (Statistical Methodology) 70 (5): 849–911. https://doi.org/10.1111/j.1467-9868.2008.00674.x.
Frauscher, Birgit, Nicolas von Ellenrieder, Rina Zelmann, Irena Doležalová, Lorella Minotti, André Olivier, Jeffery Hall, et al. 2018. “Atlas of the Normal Intracranial Electroencephalogram: Neurophysiological Awake Activity in Different Cortical Areas.” Brain 141 (4): 1130–44. https://doi.org/10.1093/brain/awy035.
Li, Min, Ying Wang, Carlos Lopez-Naranjo, Ronaldo Cesar Garcia Reyes, Aini Ismafairus Abd Hamid, Alan C. Evans, Alexander N. Savostyanov, et al. 2022. “Harmonized-Multinational qEEG Norms (HarMNqEEG).” NeuroImage, April, 119190. https://doi.org/10.1016/j.neuroimage.2022.119190.
McLachlan, Geoffrey J., and Thriyambakam Krishnan. 2008. The EM Algorithm and Extensions. 2nd edition. Hoboken: Wiley-Interscience.
Medani, Takfarinas, Juan Garcia-Prieto, Francois Tadel, Marios Antonakakis, Tim Erdbrügger, Malte Höltershinken, Wayne Mead, et al. 2023. “Brainstorm-DUNEuro: An Integrated and User-Friendly Finite Element Method for Modeling Electromagnetic Brain Activity.” NeuroImage 267 (February): 119851. https://doi.org/10.1016/j.neuroimage.2022.119851.
Pascual-Marqui, Roberto D. 2007. “Discrete, 3D Distributed, Linear Imaging Methods of Electric Neuronal Activity. Part 1: Exact, Zero Error Localization,” October. http://arxiv.org/abs/0710.3341.