Transformer aided Dynamic Causal Model for Scalable Estimation of Effective Connectivity DOI Creative Commons
Sayan Nag, Kâmil Uludağ

Imaging Neuroscience, Journal Year: 2024, Volume and Issue: 2, P. 1 - 22

Published: Jan. 1, 2024

Abstract Dynamic Causal Models (DCMs) in functional Magnetic Resonance Imaging (fMRI) decipher causal interactions, known as Effective Connectivity, among neuronal populations. However, their utility is often constrained by computational limitations, restricting analysis to a small subset of interacting brain areas, typically fewer than 10, thus lacking scalability. While the regression DCM (rDCM) has emerged faster alternative traditional DCMs, it not without its including linearization terms, reliance on fixed Hemodynamic Response Function (HRF), and an inability accommodate modulatory influences. In response these challenges, we propose novel hybrid approach named Transformer encoder decoder (TREND), which combines with state-of-the-art physiological (P-DCM) decoder. This innovative method addresses scalability issue while preserving nonlinearities inherent equations. Through extensive simulations, validate TREND’s efficacy demonstrating ability accurately predict effective connectivity values dramatically reduced time relative original P-DCM even networks comprising up to, for instance, 100 regions. Furthermore, showcase TREND empirical fMRI dataset superior accuracy and/or speed compared other variants. summary, amalgamating Transformer, introduce pioneering determining regions, extending applicability seamlessly large-scale networks.

Language: Английский

Predicting individual traits from models of brain dynamics accurately and reliably using the Fisher kernel DOI Creative Commons
Christine Ahrends, Mark W. Woolrich, Diego Vidaurre

et al.

eLife, Journal Year: 2025, Volume and Issue: 13

Published: Jan. 31, 2025

Predicting an individual’s cognitive traits or clinical condition using brain signals is a central goal in modern neuroscience. This commonly done either structural aspects, such as connectivity cortical thickness, aggregated measures of activity that average over time. But these approaches are missing aspect function: the unique ways which unfolds One reason why dynamic patterns not usually considered they have to be described by complex, high-dimensional models; and it unclear how best use models for prediction. We here propose approach describes functional amplitude Hidden Markov model (HMM) combines with Fisher kernel, can used predict individual traits. The kernel constructed from HMM mathematically principled manner, thereby preserving structure underlying model. show here, fMRI data, HMM-Fisher accurate reliable. compare other prediction methods, both time-varying time-averaged connectivity-based models. Our leverages information about has broad applications neuroscience personalised medicine.

Language: Английский

Citations

1
A data-driven network decomposition of the temporal, spatial, and spectral dynamics underpinning visual-verbal working memory processes DOI Creative Commons
Chiara Rossi, Diego Vidaurre, Lars Costers

et al.

Communications Biology, Journal Year: 2023, Volume and Issue: 6(1)

Published: Oct. 23, 2023

The brain dynamics underlying working memory (WM) unroll via transient frequency-specific large-scale networks. This multidimensionality (time, space, and frequency) challenges traditional analyses. Through an unsupervised technique, the time delay embedded-hidden Markov model (TDE-HMM), we pursue a functional network analysis of magnetoencephalographic data from 38 healthy subjects acquired during n-back task. Here show that this inferred task-specific networks with unique temporal (activation), spectral (phase-coupling connections), spatial (power density distribution) profiles. A theta frontoparietal exerts attentional control encodes stimulus, alpha temporo-occipital rehearses verbal information, broad-band P300-like profile leads retrieval process motor response. Therefore, work provides unified integrated description multidimensional can be interpreted within neuropsychological multi-component WM, improving overall neurophysiological comprehension WM functioning.

Language: Английский

Citations

16
Comparison between EEG and MEG of static and dynamic resting‐state networks DOI Creative Commons
SungJun Cho, Mats W.J. van Es, Mark W. Woolrich

et al.

Human Brain Mapping, Journal Year: 2024, Volume and Issue: 45(13)

Published: Sept. 1, 2024

Abstract The characterisation of resting‐state networks (RSNs) using neuroimaging techniques has significantly contributed to our understanding the organisation brain activity. Prior work demonstrated electrophysiological basis RSNs and their dynamic nature, revealing transient activations with millisecond timescales. While previous research confirmed comparability identified by electroencephalography (EEG) those magnetoencephalography (MEG) functional magnetic resonance imaging (fMRI), most studies have utilised static analysis techniques, ignoring nature Often, these use high‐density EEG systems, which limit applicability in clinical settings. Addressing gaps, medium‐density systems (61 sensors), comparing both network features obtained from a MEG system (306 sensors). We assess qualitative quantitative EEG‐derived MEG, including ability capture age‐related effects, explore reproducibility within across modalities. Our findings suggest that offer comparable descriptions, albeit offering some increased sensitivity reproducibility. Such two modalities remained consistent qualitatively but not quantitatively when data were reconstructed without subject‐specific structural MRI images.

Language: Английский

Citations

5
The GLM-spectrum: A multilevel framework for spectrum analysis with covariate and confound modelling DOI Creative Commons
Andrew J. Quinn, Lauren Atkinson, Chetan Gohil

et al.

Imaging Neuroscience, Journal Year: 2024, Volume and Issue: 2, P. 1 - 26

Published: Jan. 1, 2024

Abstract The frequency spectrum is a central method for representing the dynamics within electrophysiological data. Some widely used estimators make use of averaging across time segments to reduce noise in final spectrum. core this approach has not changed substantially since 1960s, though many advances field regression modelling and statistics have been made during time. Here, we propose new approach, General Linear Model (GLM) Spectrum, which reframes averaged spectral estimation as multiple regression. This brings several benefits, including ability do confound modelling, hierarchical significance testing via non-parametric statistics. We apply dataset EEG recordings participants who alternate between eyes-open eyes-closed resting state. GLM-Spectrum can model both conditions, quantify their differences, perform denoising through single step. application scaled up from channel whole head recording and, finally, applied age differences large group-level dataset. show that lends itself rigorous within- between-subject contrasts well interactions, model-projected spectra provides an intuitive visualisation. flexible framework robust multilevel analysis power spectra, with adaptive covariate modelling.

Language: Английский

Citations

4
Dynamic network analysis of electrophysiological task data DOI Creative Commons
Chetan Gohil, Oliver Kohl,

Rukuang Huang

et al.

Imaging Neuroscience, Journal Year: 2024, Volume and Issue: 2, P. 1 - 19

Published: Jan. 1, 2024

Abstract An important approach for studying the human brain is to use functional neuroimaging combined with a task. In electrophysiological data, this often involves time-frequency analysis, in which recorded activity transformed and epoched around task events of interest, followed by trial-averaging power. While simple can reveal fast oscillatory dynamics, regions are analysed one at time. This causes difficulties interpretation debilitating number multiple comparisons. addition, it now recognised that responds tasks through coordinated networks areas. As such, techniques take whole-brain network perspective needed. Here, we show how responses from conventional approaches be represented more parsimoniously level using two state-of-the-art methods: HMM (Hidden Markov Model) DyNeMo (Dynamic Network Modes). Both methods frequency-resolved millisecond resolution. Comparing DyNeMo, HMM, traditional response identify activations/deactivations other fail detect. offers powerful new method analysing data dynamic networks.

Language: Английский

Citations

4
Predicting individual traits from models of brain dynamics accurately and reliably using the Fisher kernel DOI Open Access
Christine Ahrends, Mark W. Woolrich,

Diego Vidaurr

et al.

Published: Jan. 15, 2025

Predicting an individual’s cognitive traits or clinical condition using brain signals is a central goal in modern neuroscience. This commonly done either structural aspects, such as connectivity cortical thickness, aggregated measures of activity that average over time. But these approaches are missing aspect function: the unique ways which unfolds One reason why dynamic patterns not usually considered they have to be described by complex, high-dimensional models; and it unclear how best use models for prediction. We here propose approach describes functional amplitude Hidden Markov model (HMM) combines with Fisher kernel, can used predict individual traits. The kernel constructed from HMM mathematically principled manner, thereby preserving structure underlying model. show here, fMRI data, HMM-Fisher accurate reliable. compare other prediction methods, both time-varying time-averaged connectivity-based models. Our leverages information about has broad applications neuroscience personalised medicine.

Language: Английский

Citations

0
Evidence for Transient, Uncoupled Power and Functional Connectivity Dynamics DOI Creative Commons
R. Huang, Chetan Gohil, Mark W. Woolrich

et al.

Human Brain Mapping, Journal Year: 2025, Volume and Issue: 46(4)

Published: March 1, 2025

There is growing interest in studying the temporal structure brain network activity, particular, dynamic functional connectivity (FC), which has been linked several studies with cognition, demographics and disease states. The sliding window approach one of most common approaches to compute FC. However, it cannot detect cognitively relevant transient changes at time scales fast that is, on order 100 ms, can be identified model-based methods such as HMM (Hidden Markov Model) DyNeMo (Dynamic Network Modes) using electrophysiology. These new provide time-varying estimates 'power' (i.e., variance) under assumption they share same dynamics. But there no principled basis for this assumption. Using a method allows possibility power FC networks have different dynamics (Multi-dynamic DyNeMo) resting-state magnetoencephalography (MEG) data, we show are not coupled. (visual) task MEG dataset, modulated by task, coupling their significantly during task. This work reveals novel insights into evoked responses ongoing activity previous fail capture, challenging

Language: Английский

Citations

0
The brain that controls itself DOI Creative Commons
Eli J. Müller, Brandon R. Munn, James M. Shine

et al.

Current Opinion in Behavioral Sciences, Journal Year: 2025, Volume and Issue: 63, P. 101499 - 101499

Published: March 12, 2025

Language: Английский

Citations

0
Investigating brain network dynamics in state-dependent stimulation: a concurrent Electroencephalography and Transcranial Magnetic Stimulation study using Hidden Markov Models DOI Creative Commons

Saeed Makkinayeri,

Roberto Guidotti, Alessio Basti

et al.

Brain stimulation, Journal Year: 2025, Volume and Issue: unknown

Published: March 1, 2025

Systems neuroscience studies have shown that baseline brain activity can be categorized into large-scale networks (resting-state-networks, RNSs), with influence on cognitive abilities and clinical symptoms. These insights guided millimeter-precise selection of stimulation targets based RSNs. Concurrently, Transcranial Magnetic Stimulation (TMS) revealed states, measured by EEG signal power or phase, affect outcomes. However, dynamics in these are mostly limited to single regions channels, lacking the spatial resolution needed for accurate network-level characterization. We aim at mapping high temporal precision assess whether occurrence specific network-level-states impact TMS outcome. To this end, we will identify explore how their relates corticospinal excitability. This study leverages Hidden Markov Models states from pre-stimulus source space high-density-EEG data collected during targeting left primary motor cortex twenty healthy subjects. The association between fMRI-defined RSNs was explored using Yeo atlas, trial-by-trial relation excitability examined. extracted fast-dynamic unique spatiotemporal spectral features resembling major engagement different significantly influences excitability, larger evoked potentials when dominated sensorimotor network. findings represent a step forward towards characterizing network EEG-TMS both underscore importance incorporating experiments.

Language: Английский

Citations

0
Magnetoencephalography Dimensionality Reduction Informed by Dynamic Brain States DOI Creative Commons
Annie E Cathignol, Lionel Kusch, Marianna Angiolelli

et al.

European Journal of Neuroscience, Journal Year: 2025, Volume and Issue: 61(9)

Published: May 1, 2025

ABSTRACT Complex spontaneous brain dynamics mirror the large number of interactions taking place among regions, supporting higher functions. Such complexity is manifested in interregional dependencies signals derived from different areas, as observed utilising neuroimaging techniques, like magnetoencephalography. The this data produce numerous subsets active regions at any moment they evolve. Notably, converging evidence shows that these states can be understood terms transient coordinated events spread across over multiple spatial and temporal scales. Those used a proxy ‘effectiveness’ dynamics, become stereotyped or disorganised neurological diseases. However, given high‐dimensional nature data, representing them has been challenging thus far. Dimensionality reduction techniques are typically deployed to describe complex interdependencies improve their interpretability. many dimensionality lose information about sequence configurations took place. Here, we leverage newly described algorithm, potential heat‐diffusion for affinity‐based transition embedding (PHATE), specifically designed preserve system low‐dimensional space. We analysed source‐reconstructed resting‐state magnetoencephalography 18 healthy subjects represent configuration After with PHATE, unsupervised clustering via K‐means applied identify distinct clusters. topography described, represented matrix. All results have checked against null models, providing parsimonious account large‐scale, fast, aperiodic during resting‐state. study applies PHATE algorithm (MEG) reducing while preserving large‐scale neural dynamics. Results reveal configurations, ‘states’, activity, identified clustering. Their transitions characterised by This method offers simplified yet rich view interactions, opening new perspectives on health disease.

Language: Английский

Citations

0