A practical introduction into Dynamic Causal Modeling

Thursday, June 1st, 2023 – 12:00 pm - 1:00 pm Online
Thursday, June 8th, 2023 – 12:00 pm - 1:00 pm Online
Thursday, June 15th, 2023 – 12:00 pm - 1:00 pm Contact helen.pushkarskaya@yale.edu

Dr. Peter Zeidman, Senior Research Fellow, Chair of the Methods Group at the Wellcome Centre for Human Neuroimaging, University College of London

What?

All of our cognitive functions depend on regions of the brain interacting with one another – their “effective connectivity”. Neurological and psychiatric disorders can be understood in terms of disruption to this effective connectivity. This 3-part series focusses on how to characterize people’s neuronal responses and effective connectivity using fMRI, and how to test for changes in these measures due to a diagnosis, a treatment, or behavioural / clinical measures.

These methods are widely applied to investigate psychiatric and neurological disorders. Primarily, the aim is to characterize the mechanisms of disease and medical interventions (such as drugs or therapy) on the brain. In the longer term, these same techniques may find application as diagnostic tools or readout measures for clinical trials. Note that while we will focus on functional MRI, the same principles apply to the analysis of electrophysiological recordings (EEG/MEG) or intracranial recordings (ECoG/LFPs).

• June 1, 12 - 1pm Kangjoo Lee, PhD, Yale psychiatry, OnlineIn the first session, we will give an overview of functional localization – identifying where in the brain neural activity occurs. We will focus on the General Linear Model (GLM), in conjunction with a simple auditory task fMRI experiment.
• June 8, 12 - 1pm Peter Zeidman, PhD, University College of London, OnlineIn the second session, we will introduce dynamic causal modelling (DCM), a framework for inferring the dynamic interactions among brain regions.
• June 15, 12 - 1pm Peter Zeidman, PhD, University College of London The third session will be a practical workshop, where we will illustrate applying the GLM and DCM to investigate individual differences in effective connectivity. All analyses will be illustrated using the open source Statistical Parametric Mapping (SPM) software package, which is a toolbox for MATLAB. Nevertheless, many of the principles we will cover are common across analysis packages.

Note: Participation in the third session is by invitations only (contact helen.pushkarskaya@yale.edu). Recordings will be posted online. All tutorial participants will offer additional follow up practical tutorials during the following weeks. To sign up for these tutorials or to invite instructors to your research group, please sign up here.

Why?

Dynamic Causal Modelling (DCM) for fMRI has three key strengths. First, DCM properly distinguishes between neural and vascular contributions to fMRI data. This is important, because fMRI is a downstream consequence of underlying neuronal activity, mediated by blood oxygen. Second, DCM enables the probability for different hypotheses about neural connectivity to be quantified - for example, whether or not a drug has an impact on particular connections. This is made possible using “Bayesian” statistical methods. Third, DCM enables directed connectivity between brain regions to be inferred from the downstream fMRI data, giving a richer description of brain networks than undirected functional connectivity.

DCM is the most popular tool for effective connectivity analysis using neuroimaging data. However, it is widely regarded as being very complicated. This is because it incorporates both dynamical systems theory and Bayesian statistical methods, and many technical papers assume a high degree of mathematical awareness. With recent tutorial papers, as well as this lecture series, we hope to demystify the technology and alleviate these concerns, showing that high quality analyses can be performed with a basic grounding in the key concepts.

How?

Prerequisites:

• Basic programming skills in MATLAB (license), fundamentals of functional MRI.
  Required Tools:
• MATLAB (license)
• Statistical Parametric Mapping (SPM), open source.
  Note that SPM is highly backward compatible – the current release of SPM12 works with MATLAB R2007a (7.4), which is over 15 years old. If MATLAB is unavailable, then a compiled version of SPM is available that does not require MATLAB – please see the SPM website for details.

Datasets:

• DCM for fMRI is relevant for any resting state or task-based fMRI dataset
• Open example datasets are available here

Reference publications:

• A tutorial introduction to first level (within-subject) DCM for fMRI: Zeidman, P., Jafarian, A., Corbin, N., Seghier, M.L., Razi, A., Price, C.J. and Friston, K.J., 2019. A guide to group effective connectivity analysis, part 1: First level analysis with DCM for fMRI. Neuroimage, 200, pp.174-190. https://doi.org/10.1016/j.neuroimage.2019.06.031 and https://doi.org/10.1016/j.neuroimage.2019.06.032
• Part two of the tutorial, covering group analysis (PEB): Zeidman, P., Jafarian, A., Seghier, M.L., Litvak, V., Cagnan, H., Price, C.J. and Friston, K.J., 2019. A guide to group effective connectivity analysis, part 2: Second level analysis with PEB. Neuroimage, 200, pp.12-25, https://doi.org/10.1016/j.neuroimage.2019.06.032
• An example of a recent paper that used DCM to investigate hallucinations in Parkinson’s disease - Thomas, G.E., Zeidman, P., Sultana, T., Zarkali, A., Razi, A. and Weil, R.S., 2023. Changes in both top-down and bottom-up effective connectivity drive visual hallucinations in Parkinson’s disease. Brain Communications, 5(1), https://doi.org/10.1093/braincomms/fcac329

Registration Information

Please subscribe to the MAPs workshop via google group: yale-maps@googlegroups.com

For part III, please download https://www.fil.ion.ucl.ac.uk/~pzeidman/workshop.zip. Note that in this example, Peter selected different coordinates for one of the brain regions (V1) than appear in the SPM manual. Therefore, the results you will get are numerically different than those presented in the SPM manual, but the overall conclusions are the same.