Abstract:The functional MRI (fMRI) signal is an indirect measure of neuronal activity. In order to deconvolve the neuronal activity from the experimental fMRI data, biophysical generative models have been proposed describing the link between neuronal activity and the cerebral blood flow (the neurovascular coupling), and further the hemodynamic response and the BOLD signal equation. These generative models have been employed both for single brain area deconvolution and to infer effective connectivity in networks of multiple brain areas. In the current paper, we introduce a new fMRI model inspired by experimental observations about the physiological underpinnings of the BOLD signal and compare it with the generative models currently used in dynamic causal modeling (DCM), a widely used framework to study effective connectivity in the brain. We consider three fundamental aspects of such generative models for fMRI: (i) an adaptive two-state neuronal model that accounts for a wide repertoire of neuronal responses during and after stimulation; (ii) feedforward neurovascular coupling that links neuronal activity to blood flow; and (iii) a balloon model that can account for vascular uncoupling between the blood flow and the blood volume. Finally, we adjust the parameterization of the BOLD signal equation for different magnetic field strengths. This paper focuses on the form, motivation and phenomenology of DCMs for fMRI and the characteristics of the various models are demonstrated using simulations. These simulations emphasize a more accurate modeling of the transient BOLD responses - such as adaptive decreases to sustained inputs during stimulation and the post-stimulus undershoot. In addition, we demonstrate using experimental data that it is necessary to take into account both neuronal and vascular transients to accurately model the signal dynamics of fMRI data. By refining the models of the transient responses, we provide a more informed perspective on the underlying neuronal process and offer new ways of inferring changes in local neuronal activity and effective connectivity from fMRI.

Generalized Recurrent Neural Network accommodating Dynamic Causal Modeling for functional MRI analysis

A fast dynamic causal modeling regression method for fMRI

Physiologically informed dynamic causal modeling of fMRI data

Regression dynamic causal modeling for resting‐state fMRI

Identifying Brain Networks at Multiple Time Scales via Deep Recurrent Neural Network

A graph neural network framework for causal inference in brain networks

Convolutional Recurrent Neural Network for Dynamic Functional MRI Analysis and Brain Disease Identification

DSAM: A Deep Learning Framework for Analyzing Temporal and Spatial Dynamics in Brain Networks

Modeling Task Fmri Data Via Deep Convolutional Autoencoder

A strategy of model space search for dynamic causal modeling in task fMRI data exploratory analysis

Generalised Filtering and Stochastic DCM for Fmri

State-space model with deep learning for functional dynamics estimation in resting-state fMRI

Modularity-Constrained Dynamic Representation Learning for Interpretable Brain Disorder Analysis with Functional MRI.

Dynamic Causal Models of Time-Varying Connectivity

3d Convolutional Long-Short Term Memory Network For Spatiotemporal Modeling Of Fmri Data

Exploring Brain Hemodynamic Response Patterns Via Deep Recurrent Autoencoder

Dynamic Causal Modeling in Probabilistic Programming Languages

Deep Representations for Time-varying Brain Datasets

Neural Changes in the Cortico-Basal Ganglia Network During Online Statistical Sequence Learning: an Fmri Study

A Deep Dynamic Causal Learning Model to Study Changes in Dynamic Effective Connectivity during Brain Development

Adapting Dynamic Causal Modeling to Investigate the Cooperation Between the Ventral and Dorsal Routes of Reading Neural Network