Simulation-based Inference on Virtual Brain Models of Disorders

Machine Learning Science and Technology, Journal Year: 2024, Volume and Issue: 5(3), P. 035019 - 035019

Published: July 11, 2024

Abstract Connectome-based models, also known as virtual brain models (VBMs), have been well established in network neuroscience to investigate pathophysiological causes underlying a large range of diseases. The integration an individual’s imaging data VBMs has improved patient-specific predictivity, although Bayesian estimation spatially distributed parameters remains challenging even with state-of-the-art Monte Carlo sampling. imply latent nonlinear state space driven by noise and input, necessitating advanced probabilistic machine learning techniques for widely applicable estimation. Here we present simulation-based inference on (SBI-VBMs), demonstrate that training deep neural networks both spatio-temporal functional features allows accurate generative disorders. systematic use stimulation provides effective remedy the non-identifiability issue estimating degradation limited smaller subset connections. By prioritizing model structure over data, show hierarchical SBI-VBMs renders more effective, precise biologically plausible. This approach could broadly advance precision medicine enabling fast reliable prediction

Language: Английский

---

Role of hierarchical heterogeneity in shaping seizure onset dynamics: Insights from structurally-based whole-brain dynamical network models

Communications in Nonlinear Science and Numerical Simulation, Journal Year: 2023, Volume and Issue: 130, P. 107721 - 107721

Published: Nov. 22, 2023

Language: Английский

---

Dynamical models reveal anatomically reliable attractor landscapes embedded in resting state brain networks

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Jan. 16, 2024

Analyses of functional connectivity (FC) in resting-state brain networks (RSNs) have generated many insights into cognition. However, the mechanistic underpinnings FC and RSNs are still not well-understood. It remains debated whether resting state activity is best characterized as noise-driven fluctuations around a single stable state, or instead, nonlinear dynamical system with nontrivial attractors embedded RSNs. Here, we provide evidence for latter, by constructing whole-brain systems models from individual fMRI (rfMRI) recordings, using Mesoscale Individualized NeuroDynamic (MINDy) platform. The MINDy consist hundreds neural masses representing parcels, connected fully trainable, individualized weights. We found that our manifested diverse taxonomy attractor landscapes including multiple equilibria limit cycles. when projected anatomical space, these mapped onto limited set canonical RSNs, default mode network (DMN) frontoparietal control (FPN), which were reliable at level. Further, creating convex combinations models, bifurcations induced recapitulated full spectrum dynamics via fitting. These findings suggest traverses dynamics, generates several distinct but anatomically overlapping landscapes. Treating rfMRI unimodal stationary process (i.e., conventional FC) may miss critical properties structure within brain. Instead, be better captured through modeling analytic approaches. results new generative mechanisms intrinsic spatiotemporal organization networks.

Language: Английский

---

Revealing excitation-inhibition imbalance in Alzheimer’s disease using multiscale neural model inversion of resting-state functional MRI

Communications Medicine, Journal Year: 2025, Volume and Issue: 5(1)

Published: Jan. 15, 2025

Alzheimer's disease (AD) is a serious neurodegenerative disorder without clear understanding of pathophysiology. Recent experimental data have suggested neuronal excitation-inhibition (E-I) imbalance as an essential element AD pathology, but E-I has not been systematically mapped out for either local or large-scale circuits in AD, precluding precise targeting treatment. In this work, we apply Multiscale Neural Model Inversion (MNMI) framework to the resting-state functional MRI from Disease Neuroimaging Initiative (ADNI) identify brain regions with disrupted balance large network during progression. We observe that both intra-regional and inter-regional progressively cognitively normal individuals, mild cognitive impairment (MCI) AD. Also, find inhibitory connections are more significantly impaired than excitatory ones strengths most reduced MCI leading gradual decoupling neural populations. Moreover, reveal core comprised mainly limbic cingulate regions. These exhibit consistent alterations across thus may represent important biomarkers therapeutic targets. Lastly, multiple found be correlated test score. Our study constitutes attempt delineate progression, which facilitate development new treatment paradigms restore physiological The cells within brain, neurons, communicate using activity. Excitation-inhibition measure contribution communication. memory, thinking reasoning disrupted. people applied computational model imaging could potentially used treatments developed improve balance, possibly improving symptoms Li et al. multiscale modeling approach scale based on MRI. concentrates regions, long-range subjects impairment,

Language: Английский

---

Efficient Inference on a Network of Spiking Neurons using Deep Learning

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: Jan. 26, 2024

Abstract The process of making inference on networks spiking neurons is crucial to decipher the underlying mechanisms neural computation. Mean-field theory simplifies interactions between produce macroscopic network behavior, facilitating study information processing and computation within brain. In this study, we perform a mean-field model gain insight into likely parameter values, uniqueness degeneracies, also explore how well statistical relationship parameters maintained by traversing across scales. We benchmark against state-of-the-art optimization Bayesian estimation algorithms identify their strengths weaknesses in our analysis. show that when confronted with dynamical noise or case missing data presence bistability, generating probability distributions using deep density estimators outperforms other algorithms, such as adaptive Monte Carlo sampling. However, class generative models may result an overestimation uncertainty correlation parameters. Nevertheless, issue can be improved incorporating time-delay embedding. Moreover, training Neural ODEs enables system dynamics from microscopic states. summary, work demonstrates enhanced accuracy efficiency learning harnessed solve inverse problems

Language: Английский

---

Leveraging Brain Modularity Prior for Interpretable Representation Learning of fMRI

IEEE Transactions on Biomedical Engineering, Journal Year: 2024, Volume and Issue: 71(8), P. 2391 - 2401

Published: Feb. 27, 2024

Resting-state functional magnetic resonance imaging (rs-fMRI) can reflect spontaneous neural activities in the brain and is widely used for disorder analysis. Previous studies focus on extracting fMRI representations using machine/deep learning methods, but these features typically lack biological interpretability. The human exhibits a remarkable modular structure networks, with each module comprised of functionally interconnected regions-of-interest (ROIs). However, existing learning-based methods cannot adequately utilize such modularity prior. In this paper, we propose modularity-constrained dynamic representation framework interpretable analysis, consisting graph construction, via novel network (MGNN), prediction biomarker detection. designed MGNN constrained by three core neurocognitive modules ( i.e. , salience network, central executive default mode network), encouraging ROIs within same to share similar representations. To further enhance discriminative ability learned features, encourage preserve topology input graphs reconstruction constraint. Experimental results 534 subjects rs-fMRI scans from two datasets validate effectiveness proposed method. identified connectivities be regarded as potential biomarkers aid clinical diagnosis.

Language: Английский

---

Parameter estimation in a whole-brain network model of epilepsy: Comparison of parallel global optimization solvers

PLoS Computational Biology, Journal Year: 2024, Volume and Issue: 20(7), P. e1011642 - e1011642

Published: July 11, 2024

The Virtual Epileptic Patient (VEP) refers to a computer-based representation of patient with epilepsy that combines personalized anatomical data dynamical models abnormal brain activities. It is capable generating spatio-temporal seizure patterns resemble those recorded invasive methods such as stereoelectro EEG data, allowing for the evaluation clinical hypotheses before planning surgery. This study highlights effectiveness calibrating VEP using global optimization approach. approach utilizes SaCeSS, cooperative metaheuristic algorithm parallel computation, yield high-quality solutions without requiring excessive computational time. Through extensive benchmarking on synthetic our proposal successfully solved set different configurations models, demonstrating better scalability and superior performance against other solvers. These results were further enhanced Bayesian framework hyperparameter tuning, significant gains in terms both accuracy cost. Additionally, we added scalable uncertainty quantification phase after model calibration, used it assess variability estimated parameters across problems. Overall, this has potential improve estimation pathological areas drug-resistant epilepsy, thereby inform decision-making process.

Language: Английский

---

Macroscale intrinsic dynamics are associated with microcircuit function in focal and generalized epilepsies

Communications Biology, Journal Year: 2024, Volume and Issue: 7(1)

Published: Feb. 1, 2024

Abstract Epilepsies are a group of neurological disorders characterized by abnormal spontaneous brain activity, involving multiscale changes in functional organizations. However, it is not clear to what extent the epilepsy-related perturbations activity affect macroscale intrinsic dynamics and microcircuit organizations, that supports their pathological relevance. We collect sample patients with temporal lobe epilepsy (TLE) genetic generalized tonic-clonic seizure (GTCS), as well healthy controls. extract massive features fMRI BOLD time-series characterize dynamics, simulate neuronal used large-scale biological model. Here we show whether dysfunction differed epilepsies, how these linked. Differences gradient prominent primary network default mode TLE GTCS. Biophysical simulations indicate reduced recurrent connection within somatomotor microcircuits both subtypes, even more further demonstrate strong spatial correlations between differences epilepsies. These results emphasize impact on high-order networks, suggesting systematic abnormality hierarchical organization.

Language: Английский

---

Brain Temporal-Spectral Functional Variability Reveals Neural Improvements of DBS Treatment for Disorders of Consciousness

IEEE Transactions on Neural Systems and Rehabilitation Engineering, Journal Year: 2024, Volume and Issue: 32, P. 923 - 933

Published: Jan. 1, 2024

Deep brain stimulation (DBS) is establishing itself as a promising treatment for disorders of consciousness (DOC). Measuring changes crucial in the optimization DBS therapy DOC patients. However, conventional measures use subjective metrics that limit investigations treatment-induced neural improvements. The focus this study to analyze regulatory effects and explain mechanism at functional level Specifically, paper proposed dynamic temporal-spectral analysis method quantify DBS-induced variations Functional near-infrared spectroscopy (fNIRS) promised evaluate levels was used monitor fNIRS-based experimental procedure with auditory stimuli developed, activities during from thirteen patients before after were recorded. Then, networks formulated sliding-window correlation phase lag index. Afterwards, respect temporal global regional networks, variability efficiency, local clustering coefficient extracted. Further, converted into spectral representations by graph Fourier transform, energy diversity assess variability. results showed under exhibited increased significantly associated Moreover, right regions had stronger enhancements than left regions. Therefore, well signifies patients, may serve biomarkers evaluations

Language: Английский

---

A Data-Driven Framework for Whole-Brain Network Modeling with Simultaneous EEG-SEEG Data

IFIP advances in information and communication technology, Journal Year: 2024, Volume and Issue: unknown, P. 329 - 342

Published: Jan. 1, 2024

Language: Английский

---