Dynamic Causal Modeling
Dynamic Causal Modeling for fMRI Brain Networks · Also known as: DCM, Dynamic Causal Model
Dynamic Causal Modeling (DCM) is a Bayesian framework for specifying and inverting generative models of brain connectivity from neuroimaging data. Introduced by Karl Friston and colleagues in 2003, DCM treats brain regions as dynamical systems and estimates effective connectivity by fitting observed fMRI time series to a biophysically plausible model of neuronal interactions.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
DCM is most appropriate when testing mechanistic hypotheses about directed connectivity between specific brain regions, when experimental designs include multiple conditions that might modulate connectivity, and when prior neuroscientific knowledge can constrain the model structure. Avoid DCM when the number of regions exceeds the information content of the data (typically keep models to 3–6 regions) or when connectivity is poorly characterized in the literature.
Strengths & limitations
- Explicitly incorporates biophysical constraints and generates testable predictions about neuronal mechanisms
- Provides probabilistic model comparison, allowing formal assessment of competing hypotheses about brain organization
- Estimates effective connectivity (directional, causal-like relationships) rather than mere correlation
- Naturally accommodates contextual modulation of connectivity across experimental conditions
- Computationally expensive; fitting a single model can require hours of computation
- Requires careful a priori specification of regional nodes and connection structure; misspecification can bias results
- Sensitive to preprocessing steps and regional definitions; small changes can alter posterior estimates
- Typically restricted to small networks (fewer than 8 regions) due to computational and inferential constraints
Frequently asked
What is the difference between DCM and functional connectivity?
Functional connectivity reflects correlations between brain regions and is undirected. DCM estimates effective connectivity—directed, causal-like interactions—by fitting a generative model. DCM is more computationally costly but yields mechanistic insights; functional connectivity is simpler and better suited to large-scale network characterization.
How do I choose which brain regions to include in my DCM model?
Begin with prior literature and theoretical predictions. Include regions activated in your experimental contrast and known to interact anatomically. Start with 3–4 regions; more regions exponentially increase model space. Use localizer scans or independent datasets to define regions objectively, avoiding circularity.
Can DCM tell me if connectivity is truly causal?
DCM estimates effective connectivity, which reflects directed statistical associations in the fitted generative model. While it can distinguish directionality, it does not prove true causality without experimental interventions (e.g., transcranial stimulation). Always interpret DCM estimates as mechanistic hypotheses supported by the data.
How many trials or time points do I need for reliable DCM?
DCM requires sufficient data to constrain the posterior distribution. Rule of thumb: at least 50–100 scans per condition; more for complex models. If time series are very short, results become unstable. Longer experiments (5–10 minutes per condition) yield more robust estimates.
Sources
- Friston, K. J., Harrison, L., & Penny, W. (2003). Dynamic causal modelling. NeuroImage, 19(4), 1273–1302. DOI: 10.1016/S1053-8119(03)00202-7 ↗
- Stephan, K. E., & Mathys, C. (2015). Computational approaches to neuroscience. Current Opinion in Neurobiology, 25, 85–92. link ↗
How to cite this page
ScholarGate. (2026, June 3). Dynamic Causal Modeling for fMRI Brain Networks. ScholarGate. https://scholargate.app/en/neuroimaging/dynamic-causal-modeling
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Graph Brain Network AnalysisNeuroimaging↔ compare
- Structural Equation ModelingResearch Statistics↔ compare