Differential Dynamic Causal Nets: Model Construction, Identification and Group Comparisons
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Genomics Radio. Generated commentary on the latest computational biology and genomics papers.
Ines: We're looking at the paper: "Differential Dynamic Causal Nets".
Marcus: Here's the gist of the paper. Differential Dynamic Causal Nets: Model Construction, Identification and Group Comparisons presents a novel approach to construct differential causal networks directly from electroencephalogram (EEG) data by modeling interacting neuronal populations as conditionally coupled dynamic causal models The proposed network is based on conditionally coupled neuronal circuits which describe the average behaviour of interacting neuron populations that contribute to observed EEG data Model Construction and Methodology The paper introduces a novel approach to construct differential causal networks directly from electroencephalogram (EEG) data based on conditionally coupled neuronal circuits Each node in the network represents a parameterised local neural system while directed edges stand for node-wise connections with transmission parameters The network is hierarchically structured in the sense that node and edge parameters are varying in subjects but follow a mixed-effects model The underlying neural mass activities are described by stochastic Jansen-Rit models Parameter Inference and Optimization A novel evolutionary optimisation algorithm for parameter inference in the proposed method is developed using a loss function derived from Chen-Fliess expansions of stochastic differential equations The conditional estimation of the proposed model is implemented by minimising a profiled loss function based on what is called Chen-Fliess expansions of stochastic differential equations This method allows for the inference of the large number of node parameters feasible even when nodes are heterogeneous The estimation process involves three steps: first solving the Jansen-Rit neural mass differential equations, then constructing the forward pairwise loss functions, and finally minimising these loss functions by some evolutionary optimisation algorithms Identification of Differential Causal Nets Differential causal nets are identified between cases and controls through post-multivariate analysis of variance based on a package of permutational multivariate analysis of variance The significance of the case-control factor implies that case-control memberships make a contribution to the group differences while the significance of the channel-factor shows that the group difference should be accounted for by multiple channels The analysis reveals various differential dynamic causal nets associated with different parameters, such as A(ij)k / B(ilj)k and v(ij)k0nq Key Findings and Implications The analysis of the differential dynamic causal nets reveals that abnormal connectivity patterns are associated with brain diseases like epilepsy Specifically, for the parameter A(ij)k / B(ilj)k, there were 6 channel pairs to be claimed to differ cases from controls significantly The most pronounced differences between the case and control groups were in channel pairs with the output channels in the frontal regions Furthermore, an altered excitation/inhibition (E/I) balance, often involving decreased inhibition, has been identified as a robust biomarker in various neurodegenerative diseases such as Alzheimer’s disease (AD) and frontotemporal dementia (FTD) The proposed NccDCM is a novel and useful tool for detecting disease-predisposed effective connectivity in case-control study using EEG data Limitations and Future Directions The present model has limitations, including the algorithms for estimating the NccDCM and the omission of subjects’ demographic information such as age and gender A marginal screening approach is used to estimate channel-wise connection parameters to avoid overfitting but this method is at the cost of estimation efficiency Further work is needed to take age and gender as additional predictors in each ccDCM The work can be extended to the source-level, in particular, those sources in the deep brain Simulation Validation Simulation studies were conducted to evaluate the accuracy of the proposed parameter estimation using both synthetic and real EEG data The simulation results showed that our proposed estimation is promising in terms of absolute relative error (ARE) for local nodes The performance was demonstrated in the case where nodes are dynamically coupled each other Bifurcation behavior was also demonstrated, showing that the ccDCM does have bifurcation behavior when A/B, C, K, and v0 take values in some ranges The simulation study on three-nodes NccDCMs showed a good performance in the case where nodes are dynamically coupled each other Data Analysis Techniques The analysis employs various statistical methods, including post-analysis of variance (ANOVA) to test the significance of case-control and channels as two factors The analysis also involves permutation-based Wilcoxon signed-rank tests to compare NccDCM parameters between preictal and ictal periods Lagged correlation analysis was also performed, revealing 31 channel pairs with p-values less than 0.05 in testing the difference of lagged correlations between the control and case groups The results show that the NccDCM analysis has advantages over conventional correlation analysis In summary, the proposed NccDCM is a novel and useful tool for detecting disease-predisposed effective connectivity in case-control study using EEG data We have shown that the NccDCM can be used to characterise epileptic brain network at the whole brain scale These differential dynamic causal nets partially covered default mode network (DMN), highlighting the contribution of dysfunctions in the DMN to epileptic activities and cognitive behaviors in patients The findings suggest that changes in region-transmission parameters can be used for predicting onset of seizures Appendix Details The appendix provides detailed information on the optimisation algorithm JADE, which is an adaptive differential evolution with optional external archive It also includes details on lagged correlation-based functional connectivity and EEG electrode allocation on the scalp The appendix Sd provides post-ANOVA tables summarizing the results of the NccDCM fitting to each subject The simulation study on three-nodes NccDCMs is detailed in the appendix Sf The paper is supported by an EPSRC grant EP/X038297/1 The EEG data on epilepsy is publicly available in OK et al.
Ines: (2021) The codes to reproduce simulation results are in https://github.com/zhangjsib/NccDCM The EEG electrode pairs and their brain regions are detailed in Table 3 The post-ANOVA tables are summarized in Appendix Sd The supplementary material is available at https://github.com/zhangjsib/NccDCM The paper is published on arXiv as 2601.21478v1
q-bio.NC: 29 Jan 2026 The authors have no competing interests The research of K.Y.
Marcus: and J.Z is supported by the Engineering and Physical Sciences Research Council (EPSRC) grant (EP/X038297/1) The paper is organised as follows, with details of the proposed methodology and optimisation algorithm provided in Section 2 The applications of the proposed methods to the IED dataset and synthetic data are presented in Section 3 Discussion and conclusions are made in Section 4 The discussion and conclusions are made in Section 4 The authors have no competing interests G.G is an employee of Innovision IP Ltd which provides commercial reports on individuals who may have had a head injury The EEG data on epilepsy is publicly available in OK et al.
Ines: (2021) The codes to reproduce simulation results are in https://github.com/zhangjsib/NccDCM The EEG electrode pairs and their brain regions are detailed in Table 3 The post-ANOVA tables are summarized in Appendix Sd The supplementary material is available at https://github.com/zhangjsib/NccDCM The paper is published on arXiv as 2601.21478v1
q-bio.NC: 29 Jan 2026 The authors have no competing interests The research of K.Y.
Title and authors: Ines: So we're talking about this paper titled Differential Dynamic Causal Nets Model Construction Identification and Group Comparisons. What does that even mean for us right now?
Marcus: It sounds like they’re taking EEG data and building these differential causal networks directly, which is a big step because modeling brain interactions at that level is usually too complex.
Ines: Exactly. Instead of just looking at isolated brain regions, this approach uses conditionally coupled neuronal circuits to describe the average behavior of interacting populations contributing to what we see in the EEG data.
Yuki: I wonder if that means they can actually map out how different brain systems influence each other across the whole scalp, not just within one area.
Marcus: Right. And they do this by parameterizing local neural systems as nodes and using directed edges for those connections with transmission parameters, which are what vary between subjects in a mixed-effects model.
Ines: It’s about moving away from static connectivity maps to dynamic ones that change based on the group you're comparing, whether it's cases versus controls.
Yuki: I think that makes sense for understanding how disease might alter those interactions over time or across different conditions.
The paper's summary: Marcus: Okay, so they’re using a novel evolutionary optimization algorithm for parameter inference here, based on the Chen-Fliess expansions of stochastic differential equations to minimize a profiled loss function. That’s a lot of heavy math to get the parameters right.
Ines: It sounds like they are tackling the challenge where some parameters in these neural mass differential equations are only partially identifiable, which is a real headache in this field.
Marcus: They solve that by constructing forward pairwise loss functions by summing up squared prediction errors over increasing time instances, and then using that to minimize everything with their evolutionary optimization algorithm.
Yuki: That sounds like a lot of computational work just to get the model parameters stable enough to even start comparing groups. What kind of biological detail are they aiming for with this modeling?
Ines: They are aiming for a heterogeneous composite network model, or NccDCM, where each node is a local Jansen-Rit neural mass model describing synaptic interactions in excitatory and inhibitory populations.
Marcus: And the key thing is that they identify differential causal nets between cases and controls through post-multivariate analysis of variance using a package of permutational multivariate analysis of variance.
Yuki: That means they aren't just looking for one single connection difference; they are testing if group membership contributes to the differences across multiple channels simultaneously.
The paper's improvements: Ines: One major improvement they highlight is that this approach relaxes the previous limitations by proposing a heterogeneous composite network model, NccDCM, for groups of EEG scans instead of focusing on just one subject at a time.
Marcus: That heterogeneity is key because it allows each node to be a local Jansen-Rit neural mass model describing interactions in both excitatory and inhibitory and pyramidal neuron populations.
Ines: And they use a mixed-effects model for network parameters, which accounts for the fact that node and edge parameters vary across subjects but follow this structure.
Yuki: So, it moves beyond just looking at a single brain scan to building a framework that can characterize disease-predisposed effective connectivity across multiple cortical regions at once.
Marcus: They also developed this novel optimization algorithm using the Chen-Fliess expansion of stochastic differential equations for parameter inference, which is better than relying on pre-existing literature values for estimating those parameters directly from subject EEG data.
Ines: And they found that by focusing on the excitation inhibition ratio, A over B, they can identify core model parameters even when individual component parameters like A or B are not well-identifiable alone.
Conclusion: Marcus: So to wrap up this paper on Differential Dynamic Causal Nets Model Construction Identification and Group Comparisons, the main implication is that this NccDCM tool provides a novel way to detect disease-predisposed effective connectivity in case-control studies using EEG data.
Ines: They showed that these differential dynamic causal nets can characterize epileptic brain networks at the whole brain scale, and they even found that these nets partially covered the default mode network, highlighting dysfunction in the default mode network for patients with epilepsy.
Yuki: That suggests that changes in region-transmission parameters could potentially be used to predict seizure onset based on how those networks are wired.
Marcus: And they noted that abnormal connectivity patterns are associated with brain diseases like epilepsy, specifically finding six channel pairs to be significantly different between cases and controls for the excitation inhibition ratio A over B, with the most pronounced differences being in frontal regions.
Ines: They also identified an altered excitation inhibition balance, often involving decreased inhibition, as a robust biomarker in various neurodegenerative diseases like Alzheimer’s disease and frontotemporal dementia.
Yuki: It sounds like this approach is useful for moving beyond just correlation and looking at how networks actually drive the observed EEG differences between different patient groups.
Marcus: Exactly. Before we move on to the next paper, let's take a quick look at how they actually tested their model accuracy in simulation studies.
Ines: Right, let's see what the simulation validation showed about the performance of this NccDCM tool.
Kang You, Gary Green, Jian Zhang
School of Engineering, Mathematics and Physics, University of Kent · York Neuroimaging Centre · Innovision IP Ltd
q-bio.NC, stat.AP, stat.ML
Submitted: 2026-01-29
Updated: 2026-11-01
Comments: 40 pages, 13 figures
Code: https://github.com/zhangjsib/NccDCM
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 83/100
The gist: Differential Dynamic Causal Nets: Model Construction, Identification and Group Comparisons presents a novel approach to construct differential causal networks directly from electroencephalogram (EEG)…
Key concepts
- Differential Dynamic Causal Models (DCMs)
- These are mathematical models used to describe how different groups of neurons interact dynamically over time. In this study, they model the average behavior of interacting neuron populations contributing to the observed EEG data, allowing researchers to infer causal connections between neural systems.
- Conditionally Coupled Neuronal Circuits
- This approach describes how different local neural systems are connected and influence each other. The network is structured hierarchically, meaning parameters for nodes (local systems) and edges (connections) vary across subjects but follow a mixed-effects model, capturing individual differences.
- Chen-Fliess Expansions
- This is a mathematical technique used within the loss function to estimate the many parameters of the complex neural model. It allows for accurate conditional estimation even when nodes are heterogeneous, enabling parameter inference from EEG data through an evolutionary optimization process.
Terminology
Summary
Differential Dynamic Causal Nets: Model Construction, Identification and Group Comparisons presents a novel approach to construct differential causal networks directly from electroencephalogram (EEG) data by modeling interacting neuronal populations as conditionally coupled dynamic causal models The proposed network is based on conditionally coupled neuronal circuits which describe the average behaviour of interacting neuron populations that contribute to observed EEG data
Model Construction and Methodology
The paper introduces a novel approach to construct differential causal networks directly from electroencephalogram (EEG) data based on conditionally coupled neuronal circuits Each node in the network represents a parameterised local neural system while directed edges stand for node-wise connections with transmission parameters The network is hierarchically structured in the sense that node and edge parameters are varying in subjects but follow a mixed-effects model The underlying neural mass activities are described by stochastic Jansen-Rit models
Parameter Inference and Optimization
A novel evolutionary optimisation algorithm for parameter inference in the proposed method is developed using a loss function derived from Chen-Fliess expansions of stochastic differential equations The conditional estimation of the proposed model is implemented by minimising a profiled loss function based on what is called Chen-Fliess expansions of stochastic differential equations This method allows for the inference of the large number of node parameters feasible even when nodes are heterogeneous The estimation process involves three steps: first solving the Jansen-Rit neural mass differential equations, then constructing the forward pairwise loss functions, and finally minimising these loss functions by some evolutionary optimisation algorithms
Identification of Differential Causal Nets
Differential causal nets are identified between cases and controls through post-multivariate analysis of variance based on a package of permutational multivariate analysis of variance The significance of the case-control factor implies that case-control memberships make a contribution to the group differences while the significance of the channel-factor shows that the group difference should be accounted for by multiple channels The analysis reveals various differential dynamic causal nets associated with different parameters, such as A(ij)k / B(ilj)k and v(ij)k0nq
Key Findings and Implications
The analysis of the differential dynamic causal nets reveals that abnormal connectivity patterns are associated with brain diseases like epilepsy Specifically, for the parameter A(ij)k / B(ilj)k, there were 6 channel pairs to be claimed to differ cases from controls significantly The most pronounced differences between the case and control groups were in channel pairs with the output channels in the frontal regions Furthermore, an altered excitation/inhibition (E/I) balance, often involving decreased inhibition, has been identified as a robust biomarker in various neurodegenerative diseases such as Alzheimer’s disease (AD) and frontotemporal dementia (FTD) The proposed NccDCM is a novel and useful tool for detecting disease-predisposed effective connectivity in case-control study using EEG data
Limitations and Future Directions
The present model has limitations, including the algorithms for estimating the NccDCM and the omission of subjects’ demographic information such as age and gender A marginal screening approach is used to estimate channel-wise connection parameters to avoid overfitting but this method is at the cost of estimation efficiency Further work is needed to take age and gender as additional predictors in each ccDCM The work can be extended to the source-level, in particular, those sources in the deep brain
Simulation Validation
Simulation studies were conducted to evaluate the accuracy of the proposed parameter estimation using both synthetic and real EEG data The simulation results showed that our proposed estimation is promising in terms of absolute relative error (ARE) for local nodes The performance was demonstrated in the case where nodes are dynamically coupled each other Bifurcation behavior was also demonstrated, showing that the ccDCM does have bifurcation behavior when A/B, C, K, and v0 take values in some ranges The simulation study on three-nodes NccDCMs showed a good performance in the case where nodes are dynamically coupled each other
Data Analysis Techniques
The analysis employs various statistical methods, including post-analysis of variance (ANOVA) to test the significance of case-control and channels as two factors The analysis also involves permutation-based Wilcoxon signed-rank tests to compare NccDCM parameters between preictal and ictal periods Lagged correlation analysis was also performed, revealing 31 channel pairs with p-values less than 0.05 in testing the difference of lagged correlations between the control and case groups The results show that the NccDCM analysis has advantages over conventional correlation analysis
Conclusion
In summary, the proposed NccDCM is a novel and useful tool for detecting disease-predisposed effective connectivity in case-control study using EEG data We have shown that the NccDCM can be used to characterise epileptic brain network at the whole brain scale These differential dynamic causal nets partially covered default mode network (DMN), highlighting the contribution of dysfunctions in the DMN to epileptic activities and cognitive behaviors in patients The findings suggest that changes in region-transmission parameters can be used for predicting onset of seizures
Appendix Details
The appendix provides detailed information on the optimisation algorithm JADE, which is an adaptive differential evolution with optional external archive It also includes details on lagged correlation-based functional connectivity and EEG electrode allocation on the scalp The appendix Sd provides post-ANOVA tables summarizing the results of the NccDCM fitting to each subject The simulation study on three-nodes NccDCMs is detailed in the appendix Sf The paper is supported by an EPSRC grant EP/X038297/1 The EEG data on epilepsy is publicly available in OK et al. (2021) The codes to reproduce simulation results are in https://github. p. 1 , 2019; Park et al. , 2021; Forrester et al. , 2019; Park et al. , 2021; Forrester et al. , 2019; Park et al. , 2021; Forrester et al. , 2019; Park et al. , 2021; Forrester et al. , 2019; Park et al. , 2021; Forrester et al. ,
2018) among others. M/EEG are promising tools for studying brain dynamics and
1/40
arXiv:2601. ,
2018) among others. M/EEG are promising tools for studying brain dynamics and
1/40
arXiv:2601. ,
2018) among others. M/EEG are promising tools for studying brain dynamics and
1/40
arXiv:2601. ,
2018) among others. M/EEG are promising tools for studying brain dynamics and
1/40
arXiv:2601. ,
2018) among others. M/EEG are promising tools for studying brain dynamics and
1/40
arXiv:2601. M/EEG are promising tools for studying brain dynamics and
1/40
arXiv:2601. 21478v1 [q-bio. M/EEG are promising tools for studying brain dynamics and
1/40
arXiv:2601. 21478v1 [q-bio. M/EEG are promising tools for studying brain dynamics and
1/40
arXiv:2601. 21478v1 [q-bio. M/EEG are promising tools for studying brain dynamics and
1/40
arXiv:2601. 21478v1 [q-bio. M/EEG are promising tools for studying brain dynamics and
1/40
arXiv:2601. 21478v1 [q-bio. M/EEG are promising tools for studying brain dynamics and
1/40
arXiv:2601. 21478v1 [q-bio. 21478v1 [q-bio. NC] 29 Jan 2026 21478v1 [q-bio. NC] 29 Jan 2026 21478v1 [q-bio. NC] 29 Jan 2026 21478v1 [q-bio. NC] 29 Jan 2026 21478v1 [q-bio. NC] 29 Jan 2026 21478v1 [q-bio. NC] 29 Jan 2026 The applications of
the proposed methods to the IED dataset and synthetic data are presented in Section 3. Discussion and conclusions are made in Section 4. 2 York Neuroimaging Centre, The Biocentre, York Science Park, University
of York, Heslington, York YO10 5NY, UK. 3 Innovision IP Ltd, Culham Innovation Centre, D5 Culham Science Centre,
Abingdon, Oxfordshire, OX14 3DB, UK. 2 York Neuroimaging Centre, The Biocentre, York Science Park, University
of York, Heslington, York YO10 5NY, UK. 3 Innovision IP Ltd, Culham Innovation Centre, D5 Culham Science Centre,
Abingdon, Oxfordshire, OX14 3DB, UK. 2 York Neuroimaging Centre, The Biocentre, York Science Park, University
of York, Heslington, York YO10 5NY, UK. 3 Innovision IP Ltd, Culham Innovation Centre, D5 Culham Science Centre,
Abingdon, Oxfordshire, OX14 3DB, UK. 2 York Neuroimaging Centre, The Biocentre, York Science Park, University
of York, Heslington, York YO10 5NY, UK. 3 Innovision IP Ltd, Culham Innovation Centre, D5 Culham Science Centre,
Abingdon, Oxfordshire, OX14 3DB, UK. The details of the proposed
methodology and optimisation algorithm are provided in Section 2. The applications of
the proposed methods to the IED dataset and synthetic data are presented in Section 3. The details of the proposed
methodology and optimisation algorithm are provided in Section 2. The applications of
the proposed methods to the IED dataset and synthetic data are presented in Section 3. The details of the proposed
methodology and optimisation algorithm are provided in Section 2. The applications of
the proposed methods to the IED dataset and synthetic data are presented in Section 3.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Pruning redundant reasoning tokens
The existing imaging studies often focused on a single subject study, either with models for a small number of cortical columns or with models in a large scale but involving only excitatory/inhibitory neuron populations or with models in a large scale but for homogeneous brain regions, where all nodes in the network models are identical In this paper, to relax the above limitations, we have proposed a heterogeneous composite network model, NccDCM, for groups of EEG scans respectively In the NccDCM, each node represents a local Jansen-Rit neural mass model for describing synaptic interactions in excitatory/inhibitory and pyramidal neuron populations To account for individual subject-variability, we have introduced a mixed-effects model for network parameters
How the training signal is built
The estimation of NccDCMs is implemented in three steps: first solve the Jansen-Rit neural mass differential equations, then construct the forward pairwise loss functions, and finally minimising these loss functions by some evolutionary optimisation algorithms The loss function is constructed by use of the Chen-Fliess expansion The forward pairwise loss function is constructed by the summation of the squared prediction errors over the set of increasing time-instances Tq
Where the gains come from
The differential dynamic causal nets are quantified by post-analysis of variance on estimated parameters in cases and controls The most pronounced differences between the case and control groups were in channel pairs with the output channels in the frontal regions for several parameters The excitation/inhibition ratio A(ij)k / B(ilj)k showed significant differences between cases and controls An altered excitation/inhibition (E/I) balance, often involving decreased inhibition, has been identified as a robust biomarker in various neurological diseases The proposed NccDCM is a novel and useful tool for detecting disease-predisposed effective connectivity in case-control study using EEG data The paper is published on arXiv as 2601.21478v1 [q-bio.NC] 29 Jan 2026 The authors have no competing interests The research of K.Y. and J.Z is supported by the Engineering and Physical Sciences Research Council (EPSRC) grant (EP/X038297/1) The paper is organised as follows, with details of the proposed methodology and optimisation algorithm provided in Section 2 The applications of the proposed methods to the IED dataset and synthetic data are presented in Section 3 Discussion and conclusions are made in Section 4 The authors have no competing interests G.G is an employee of Innovision IP Ltd which provides commercial reports on individuals who may have had a head injury The EEG data on epilepsy is publicly available in OK et al. (2021) The codes to reproduce simulation results are in https://github.com/zhangjsib/NccDCM The EEG electrode pairs and their brain regions are detailed in Table 3 The post-ANOVA tables are summarized in Appendix Sd The supplementary material is available at https://github.com/zhangjsib/NccDCM The paper is published on arXiv as 2601.21478v1 [q-bio.NC] 29 Jan 2026 The authors have no competing interests The research of K.Y. and J.Z is supported by the Engineering and Physical Sciences Research Council (EPSRC) grant (EP/X038297/1) The paper is organised as follows, with details of the proposed methodology and optimisation algorithm provided in Section 2 The applications of the proposed methods to the IED dataset and synthetic data are presented in Section 3 Discussion and conclusions are made in Section 4 The authors have no competing interests G.G is an employee of Innovision IP Ltd which provides commercial reports on individuals who may have had a head injury The EEG data on epilepsy is publicly available in OK et al. (2021) The codes to reproduce simulation results are in https://github.com/zhangjsib/NccDCM The EEG electrode pairs and their brain regions are detailed in Table 3 The post-ANOVA tables are summarized in Appendix Sd The supplementary material is available at https://github.com/zhangjsib/NccDCM The paper is published on arXiv as 2601.21478v1 [q-bio.NC] 29 Jan 2026 The authors have no competing interests The research of K.Y. and J.
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Differential causal nets between cases and controls can be used to identify
intrinsic and extrinsic dynamic connectivity patterns that are associated with brain diseases.
This allows for identifying specific network-level alterations in parameters like excitation/inhibition ratio, within-channel connectivity, between-channel connectivity, and synaptic transmission contrasts. -
The system can detect changes in dynamic causal nets between the preictal period and the ictal period to gain
deeper insights into the mechanisms underlying epileptic seizures.
This enables prediction of seizure onset by observing parameter increases from the preictal to the ictal state. -
A novel network architecture, NccDCM, can be used for
detecting disease-predisposed effective connectivity in case-control study using EEG data.
This model allows for awhole brain scale
characterization of epileptic networks by modeling local Jansen-Rit neural mass models across multiple cortical regions. -
The AI system can perform parameter estimation using a
novel evolutionary optimisation algorithm
based on the Chen-Fliess expansion of stochastic differential equations to estimate parameters directly from subject EEG data, rather than relying on pre-existing literature values. -
The system can identify and reparameterize core model parameters, such as the excitation/inhibition balance ratio, as these are found to be
not well-identifiable although their ratio is identifiable.
This allows for a more robust characterization of underlying neural circuit function despite limitations in estimating individual component parameters like A or B.
Abstract
Pathophysiological modelling of brain systems from microscale to macroscale remains difficult in group comparisons partly because of the infeasibility of modelling the interactions of thousands of neurons at the scales involved. Here, to address the challenge, we present a novel approach to construct differential causal networks directly from grouped electroencephalogram (EEG) data. In the proposed modular network, each node represents a locally averaged regional neural system via a canonical Jansen-Rit model while directed edges stand for node-wise input-output connections. The network is hierarchically structured in the sense that node and edge parameters are random, varying in subjects, but follow a mixed-effects model. We show that unlike the classical Jansen-Rit models, the proposed node models are locally structurally identifiable. An evolutionary optimisation algorithm for parameter inference in the proposed method is developed using a loss function derived from Chen-Fliess expansions of stochastic differential equations. The method is demonstrated by fitting the proposed modular network to grouped epileptic EEG data in both sensor- and source- levels. We identify significant changes in the parameters that characterise dynamic networks within and between brain regions. We reveal network disruptions, due to imbalance of excitatory-inhibitory interneurons and other altered epileptic brain connectivities, before and during seizure periods, and contrast them between case and control groups. The performance of the proposed method is also evaluated on synthetic data.
Related papers
- Self-Supervised Learning of Local Predictive Directions as a Neural Computation
- Synchronized Hearts, Desynchronized Brains: A Nonlinear Dynamics Perspective for Function and Failure
- Message interaction dynamics covary with brain volume across mammalian connectomes
- Planning as Dynamics Relaxation: Hippocampal Recurrent Network Realizes Optimal Goal-Directed Navigation
- The Platonic brain bridge hypothesis: brain networks as architectural priors for multimodal large language models
- BrainPFN: Amortized Brain-Behaviour Prediction via Prior-Data Fitted Networks