Jawata Afnan (McGill University)

MEG estimation of resting state oscillations and functional connectivity: validation with an intracranial EEG atlas

Magnetoencephalography (MEG) is a widely used non-invasive tool to estimate brain activity with high temporal resolution. However, due to the ill-posed nature of the MEG source imaging (MSI) problem, the ability of MSI to identify accurately underlying brain sources along the cortical surface is uncertain and requires validation. We aimed to validate the ability of MSI to estimate the background resting state power spectra and connectivity profiles of 45 healthy participants by comparing MSI findings to in-situ measurements obtained from the intracranial EEG (iEEG) atlas (https://mni-open-ieegatlas.research.mcgill.ca/), at the population level.
We applied wavelet-based Maximum Entropy on the Mean (wMEM) as a MSI technique dedicated to localization oscillations. Next, we converted MEG source maps into intracranial space by applying a forward model to the MEG-reconstructed source maps, and estimated virtual iEEG (ViEEG) potentials on each iEEG channel location; we finally quantitatively compared ViEEG with actual iEEG signals from the same region in terms of spectral characteristics and connectivity profiles in the canonical frequency bands. For connectivity analysis, we calculated four widely used metrics- Amplitude Envelope Correlation (AEC), orthogonalized AEC (OAEC), Phase Locking Value (PLV) and corrected imaginary part of PLV (ciPLV) and compared the cross-modal spatial correlation between MEG and iEEG connectomes.
Spectral analysis: MEG spectra were more accurately estimated in the lateral regions compared to the deeper medial regions. Regions with higher amplitude in the ViEEG than in the iEEG were more accurately recovered. In the deep regions, MEG-estimated amplitudes were largely underestimated, and the spectra were poorly recovered. Moreover, MEG largely overestimated oscillatory peaks in the alpha band, especially in the anterior and deep regions. This is possibly due to higher phase synchronization of alpha oscillations over extended regions, exceeding the spatial sensitivity of iEEG but detected by MEG. Connectivity analysis: Preliminary results showed that the cross-modal correlation between MEG and iEEG connectomes was band-specific and the highest correlations were found for the beta band when connectivity measures considered zero-lag connectivity such as AEC and PLV. In contrast, the connectivity metrics that remove the zero-lag connectivity (such as OAEC and ciPLV) significantly decreased the cross-modal correlation. Compared to iEEG, MEG showed higher zero-lag connectivity, a well-known disadvantage of non-invasive source imaging resulting from source-leakage.
This study identifies brain regions, frequency bands, and connectivity metrics for which MEG source analysis is likely to be reliable, a promising step towards resolving the uncertainty in recovering intracerebral activity from non-invasive MEG studies.

Obai Bin Ka’B Ali (Concordia University)

Dialogue mechanisms between astrocytic and neuronal networks: a whole-brain biophysical modelling approach

Astrocytes, a major type of glia, possess assorted structural and functional properties making them inseparable from their neighbouring neurons. However, most published computational models of whole-brain activity, if not all, remain focused on neurons while ignoring astrocytes. We herewith introduce a biophysical model built upon neural network mass and compartmental modelling techniques, where large-scale astrocytic and neuronal networks couple their activity through glutamatergic and GABAergic transmission systems. We formulate a network scheme where neural dynamics are constrained by a two-layered structural network interconnecting either astrocytic or neuronal populations, and we ask how astrocytic networks contribute to whole-brain activity and emerging functional connectivity patterns. By developing a biologically plausible simulation approach based on bifurcation and multilayer network theories, we demonstrate that astrocytic and neuronal networks engage in a dialogue over fast and slow fluctuations or over phase-based and amplitude-based network connectivity. Our study is a step forward for more thoroughly investigating the role of glia alongside neurons in health or disease conditions.

Zhengchen Cai (McGill University)

Estimation of fMRI Responses Related to Epileptic Discharges Using Bayesian Hierarchical Modeling

Simultaneous electroencephalography–functional MRI (EEG-fMRI) is a unique and noninvasive method for epilepsy presurgical evaluation. When selecting voxels by null-hypothesis tests, the conventional analysis may overestimate fMRI response amplitudes related to interictal epileptic discharges (IEDs), especially when IEDs are rare.

Eighty-two epilepsy patients who underwent EEG-fMRI scans and surgeries at the MNI were included in this study. Surgical outcomes were classified with a follow-up of at least one year. IEDs were annotated by experienced neurophysiologists on processed EEGs in which scanner artifacts were removed. Anatomical and functional MRI preprocessing were performed using fMRIPrep. We proposed a hierarchical model involving local and global hemodynamic responses to IEDs using the Schaefer2018 atlas and Freesurfer subcortical parcellation as the prior. Hence, the fMRI response in each voxel was estimated not only by its own data, but also adjusted according to the local homogeneity of responses within each region of interest (ROI). ROI-level responses were further regularized to take into account the global homogeneity using the resting state networks to which they belong. The IED-related fMRI response was represented by the relative percentage change of the blood oxygen level dependent (BOLD) signal. A voxel-wise model was compared to ours from three perspectives: 1) the “type M(agnitude) error” of the estimated effect, as assessed by the ratio of BOLD percentage changes between two models; 2) the ability to differentiate BOLD percentage changes induced by brief (e.g., spike) and long-lasting (e.g., burst) IEDs; and 3) clinical utility in presurgical evaluation, as determined by the concordance between the cluster exhibiting the highest fMRI response and the surgical cavity. All parameters of the aforementioned models (e.g., effect, nuisance and autocorrelation) were solved simultaneously using a full Bayesian approach – Hamiltonian Monte Carlo in Stan platform.

The voxel-wise model overestimated fMRI responses compared to the hierarchical model, evidenced by a practically and statistically significant difference between the estimated BOLD percentage changes. Only the hierarchical model differentiated brief and long-lasting IEDs with significantly different BOLD percentage changes. Overall, the hierarchical model outperformed the voxel-wise model on presurgical evaluation, measured by higher prediction performance. When compared with a previous study, the hierarchical model showed higher performance metric values, but the same or lower sensitivity.

Our results demonstrated the capability of the hierarchical model of providing more physiologically reasonable and more accurate estimations of fMRI response amplitudes induced by IEDs. To enhance the sensitivity of EEG-fMRI for presurgical evaluation, it may be necessary to incorporate more appropriate spatial priors and bespoke decision strategies.

Dilanjan Diyabalanage (University of Western Ontario)

Physics-Informed Probabilistic Approach to Analyzing and Modeling Neuroimaging Data

In computational neuroscience, understanding the multifaceted dynamics within neural networks remains a pressing challenge, particularly in the context of the brain’s staggering complexity, comprising approximately 100 billion neurons. While traditional models have focused on local coupling and function, there is a growing consensus that a network-centric approach is indispensable for a comprehensive understanding of brain function. Against this backdrop, this study introduces a groundbreaking approach that leverages the mathematical principles of quantum mechanics to scrutinize neuroimaging data.

Our quantum-inspired model offers a novel framework for investigating network dynamics, complementing, and extending existing network science methodologies. It provides a nuanced perspective on neural information flow, enabling a deeper understanding of brain network topology. A significant feature of this model is its ability to integrate neuro-energetics, thereby enriching our understanding of metabolic processes within these intricate networks. Particularly salient is the model’s utility in examining the temporal dynamics of resting-state networks, which are key to understanding the brain’s baseline functional connectivity. Using predefined neural networks as a template, our model dynamically tracks network behavior, aligning with the focus on neural network inference from imaging data.

Significantly, this approach uses source-localized electroencephalography (EEG) data, allowing for a broader application in both clinical and research settings. A significant expansion of this research is the incorporation of Physics-Informed Neural Networks (PINNs) to extend the capabilities of the quantum-mechanical framework. PINNs serve as a computational bridge, facilitating the integration of physical laws into the modeling process, thus enhancing the model’s predictive accuracy and interpretability.

Adrien Dubois (Université de Montréal)

Normative modelling on large scale EEG

Neurodevelopmental disorders (NDDs) present a variety of symptoms and etiologies, which poses challenges for accurate diagnosis and personalized treatment. Although electrophysiology has demonstrated potential in revealing the neural basis of NDDs, inconsistent findings and methodological heterogeneity have impeded its practical applications. Furthermore, reproducibility and interpretability remain critical concerns in medical research. Our study aims to overcome these challenges by creating comprehensive models and integrating multiple databases into a single framework. We hypothesize that an enhanced approach to electrophysiology can identify reliable biomarkers of NDDs, enhance biological interpretability, and advance the development of actionable biomarkers and better understanding of biological mechanisms. To achieve this, we will establish standardized growth charts of electrophysiological measures using normative modeling on a large-scale dataset (N=5000) obtained from diverse cohorts. This approach will account for nonlinear age-related effects on EEG measures, gender and sex variations, and different cohorts, enabling rigorous statistical analyses of large-scale multi-site data. In addition to the psychiatric diagnosis, our databases include genetic conditions, which allows us to map the deviations and thus bridge the gap between genetic markers and neurophysiological outcomes. By leveraging advanced statistics techniques, including unsupervised and supervised learning, we intend to dissect the complex interplay between genotypic variability and phenotypic expression in NDDs. This multidimensional analysis aims to illuminate the underlying genetic contributions to observed electrophysiological patterns, paving the way for genotype-informed interventions and precision medicine in the realm of NDDs.

Maeva Gacoin (McGill University)

Effects of fluoxetine on macaque cortical functional connectivity following reward-based spatial learning

Despite the prevailing notion of stable synaptic connectivity in the adult brain, research indicates that residual plasticity can persist and be amplified through behavioral and pharmacological interventions. These interventions aim to heighten the brain’s responsiveness to environmental stimuli by manipulating the balance between excitation and inhibition, with glutamate as the excitatory agent and gamma-aminobutyric acid (GABA) serving as the inhibitory counterpart. Our study aimed to explore the network effects of re-establishing plasticity through intense reward-based spatial behavioral training, both with and without fluoxetine, a selective serotonin reuptake inhibitor known for reducing GABA levels. In a previous study (Gacoin and Ben Hamed 2022), we demonstrated that fluoxetine enhances reward selectivity when learning a spatial priority map. To investigate how serotonin availability impacts spatial priority maps and their interference with parieto-frontal and occipital function in visual decision-making, we conducted a longitudinal study with two conscious macaques. This study involved three distinct time points. First, a control baseline condition where we measured brain resting-state activity before any manipulations. Then, an assessment of resting-state brain activity after manipulating the spatial priority map through reward-based learning, effectively increasing the average reward in one visual hemifield compared to the other. Lastly, an evaluation of resting-state brain activity after reversing the reward-based spatial priority map manipulation while administering systemic fluoxetine injections, reversing the average rewards between hemifields.
We employed both ROI-to-ROI data-driven and ROI-to-whole brain hypothesis-driven functional connectivity analyses to understand the specific interactions between reward-based prioritization and fluoxetine on brain functional connectivity. Our findings show that the effects of reward-based learning on brain functional connectivity depend on whether fluoxetine is administered during the learning phase. In the hemisphere associated with the high-reward hemifield, fluoxetine enhanced functional connectivity between the dorsal lateral prefrontal cortex and striate and extra-striate cortex. In contrast, in the hemisphere associated with the low-reward hemifield, fluoxetine increased functional connectivity between the dorsal lateral prefrontal cortex and the parietal cortex. Our seed-to-whole brain analyses also revealed that reward-based learning induced selective changes in functional connectivity within both dorsal and ventral visual networks.
This global reweighing of functional brain connectivity during reward-based learning likely contributes to the behavioral effects observed with fluoxetine, as reported in our previous study. In summary, our observations enhance our understanding of how fluoxetine, in the context of learning-based plasticity, impacts the adult brain.

Laetitia Jeancolas (Concordia University)

Memory and Connectivity: Episodic and Semantic Memory and Their Interdependence in Normal Aging and Alzheimer’s Disease

The existence of a relationship between episodic and semantic memory has long been debated and is even less understood in the context of Alzheimer’s disease (AD). Here, we investigated the interaction between semantic knowledge and episodic memory processes associated with faces in normal elderly and mild-stage AD subjects, through the spectrum of electrophysiological and connectivity analyses.

A total of 19 subjects [mean age 69 ± 4 yrs] were included in this study (HM-TC project, Pitié-Salpêtrière, Paris), composed of nine patients with probable AD and ten age-matched healthy controls. The subjects underwent a T1-weighted scan and MEG acquisition during memory tasks, consisting in faces’ recognition. They were composed of blocks of semantic tasks, during which the participants had to discriminate famous from unknown faces, and blocks of episodic tasks, during which they had to indicate whether they saw the faces (which could be famous or unknown) in the previous semantic task. These tasks involved three types of memory: pure semantic retrieval, pure episodic retrieval, and a mixed operation of episodic retrieval of semantic information. Continuous MEG signals were segmented into epochs time-locked to stimulus onset.

We performed source reconstructions, using SPM12b. To compute the forward gain matrix, we employed the local spheres model and constructed a canonical mesh including the cortex and the hippocampus. To solve the inverse problem, we assumed IID and distributed sources, and conducted a group inversion.
The effective connectivity analyses were conducted using Dynamic Causal Modelling. We performed Bayesian Model Selection to compare the connectivity models (ROIs were selected using the source reconstruction and the connections were chosen based on a connectivity atlas). Then we performed ANOVAs on the connectivity modulations to assess the changes due to the type of memory and AD.

The source activations and connectivity in the mixed condition differed from pure episodic condition and showed some similarity with the semantic task, depicting the interaction between episodic and semantic memories. Moreover, AD group showed a decreased hippocampus activity and connectivity, but also an increase in occipito-parietal region during episodic memory.

Our study characterized the active brain sources and the connectivity patterns involved in semantic and episodic memories and in their interaction. We also showed how these memory neural correlates were affected in AD. Analyses on larger databases are necessary to confirm these promising findings, for a comprehensive understanding of the memory underlying mechanisms in normal aging and AD.

Ahmed Khan (McGill University)

Individualized modeling of neurotransmitter receptor influence on multi-modal measures of Alzheimer’s disease progression

Neurodegenerative disorders such as Alzheimer’s disease (AD) can be clinically heterogeneous and involve multiple interacting physiological systems, from metabolic, vascular and functional dysregulation to misfolded protein accumulation and atrophy. To resolve potential disease mechanisms and treatment targets, the molecular features underlying macroscopic physiological and symptomatic variability must be identified. As critical mediators of cellular signalling and pharmacological response, multiple neurotransmitter receptors have been associated with AD and targeted for symptomatic therapy, but we lack a comprehensive understanding of their role in disease progression and symptom severity. In this work, we combined multi-modal imaging data (tau, amyloid-β and glucose PET, and structural, functional and arterial spin labelling MRI) with the healthy spatial distributions of 15 neurotransmitter receptors from autoradiography in individualized computational brains, to characterize receptor mediation of multi-factorial brain alterations in the AD spectrum. We demonstrated that receptor architecture helps explain long-term brain re-organization in a heterogeneous aged population (including healthy controls, MCI subjects and AD patients). Furthermore, we identified 2 distinct yet overlapping axes linking receptor-mediated physiological interactions in AD with i) executive dysfunction and ii) memory, language and visuospatial impairment. This data-driven modeling approach can be used to infer latent molecular involvement and guide personalized treatment design.

Lu Wen Da (McGill University)

Tract-specific g-ratio using COMMIT: comparison with conventional g-ratio tractometry

Introduction: By combining diffusion-weighted imaging and magnetization transfer, the myelination of white matter tracts can be estimated using an analysis pipeline known as tractometry [1], where quantitative MRI maps are projected onto reconstructed white matter tracts to investigate its mirostructural properties. There are limitations associated with this technique; 60-90% of image voxels in the white matter contain multi-fiber configurations [2], leading to partial volume effects that bias the measurements of each fiber and conceal potential fiber differences. The COMMIT framework has been used to minimizing the bias stemming from crossing fibers [3-5]. We build on this work to calculate tract-specific g-ratio, a ratio between the inner and outer radius of the myelin sheath, and will compare it to conventional tractometry.
Methods: 10 healthy subjects (6 males, 29.2± 6.29 years old) were scanned and rescanned (< 3 weeks inter-scan interval) on a 3 Tesla scanner. The data was preprocessed using micapipe [6]. Tract-specific g-ratio connectivity matrices were calculated using the axonal and myelin volumes obtained from a three-stage COMMIT filtering pipeline [3-5] using the Schaefer-200 parcellation [7]. The conventional g-ratio tractometry was computed using the same streamlines and a voxel-wise g-ratio map [8].
Results and Discussion: Tract-specific data, with a wider dynamic range than tractometry, reflects the removal of partial volume effects causing smoothing in tractometry. The repeatability is high for medium to large tracts; and the smaller caliber tracts are excluded from subsequent analysis using an 80th percentile tract cutoff across all subjects. The largest 20% of tracts show good repeatability (correlation coefficient: 0.651). It is important to note that tractometry is more repeatable but sacrifices anatomical specificity.
The differences between tract-specific and tractometry g-ratios are most significant in temporal and frontal regions. We observe a relationship between tract-specific g-ratio and tract length and caliber: g-ratio is higher for longer tracts and lower for tracts with larger caliber. These results imply that the longer and smaller caliber tracts are associated with thinner myelin sheaths relative to axon caliber. This relationship is not seen in the case of tractometry where g-ratio does not change with respect to both tract length and caliber.
Conclusion: Our novel processing pipeline produced tract-specific g-ratio connectivity matrices with a wider g-ratio dynamic range compared to conventional g-ratio tractometry. By disentangling the g-ratio of tracts crossing throughout the brain, we were able to detect relationships between g-ratio and tract length and caliber that were previously concealed.

[1] S Bells et al., presented at the Proc. ISMRM, (2011).[2] B Jeurissen et al., Hum Brain Mapp 34 (11), 2747 (2013).[3] S Schiavi et al., Neuroimage 249, 118922 (2022).[4] A Daducci et al., IEEE Trans Med Imaging 34 (1), 246 (2015).[5] S Schiavi et al., Science advances 6 (31), eaba8245 (2020).[6] RR Cruces et al., Neuroimage 263, 119612 (2022).[7] A Schaefer et al., Cerebral cortex 28 (9), 3095 (2018).[8] N Stikov et al., Neuroimage 118, 397 (2015).

Arsalan Rahimabadi (Concordia University)

Extended fractional-polynomial generalizations of diffusion and Fisher–KPP equations to model misfolded proteins propagation in the brain

Many practical applications require the investigation of diffusion or reaction-diffusion processes over complex structures, such as brain networks that can be modeled as weighted undirected or directed graphs. An archetypal example of such applications is the progression of neurodegenerative diseases such as Alzheimer’s disease (AD) in the brain connectome. AD and more than 25 other identified tauopathies, including Pick disease (PiD), progressive supranuclear palsy (PSP), argyrophilic grain disease (AGD), corticobasal degeneration (CBD), and frontotemporal dementia with parkinsonism-17 (FTDP-17), are characterized by the aggregation of specific microtubule-associated proteins (MAPs) called tau proteins which have been extensively studied for their function in the assembly and stabilization of microtubules providing a structural backbone for axons and dendrites of neurons. Although this protein is natively unfolded and its propensity for aggregation is negligible, post-translational modifications, such as phosphorylation and truncation, can promote the aggregation of monomeric tau. Notwithstanding the fact that it has been confirmed that tau pathology alone is sufficient to induce neurodegeneration by the identification of tau mutants in patients with FTDP-17, extensive research is still underway to determine the pathways and mechanisms underlying tau aggregation in tauopathies and their effects on the spatiotemporal pattern of tau pathology progression in the brain. The celebrated Fisher-Kolmogorov-Petrovsky-Piskunov (Fisher-KPP) reaction-diffusion equation are becoming increasingly popular for use in graph frameworks by substituting the standard graph Laplacian operator for the continuous one to study the progression of tauopathy in AD. However, due to the porous structure of neuronal fibers, the spreading of toxic species can be governed by a nonlinear diffusion process, and if this is the case, the standard graph Laplacian cannot adequately describe the dynamics of the spreading process. To capture such more complicated dynamics, we have recently proposed a diffusion equation with a nonlinear Laplacian operator and a generalization of the Fisher-KPP reaction-diffusion equation on undirected and directed networks using extensions of fractional polynomial (FP) functions. A complete analysis has been also provided for the extended FP diffusion equations, including existence, uniqueness, and convergence of solutions, as well as stability of equilibria. Moreover, for the extended FP Fisher-KPP reaction-diffusion equations, we have derived a family of positively invariant sets allowing us to establish existence, uniqueness, and boundedness of solutions. To illustrate the potential applications of the proposed extended FP equations, we have examined nonlinear diffusion on a directed one-dimensional lattice and tauopathy progression in the mouse brain.