Construct validation of a DCM for resting state fMRI.
Bottom Line: Dynamic causal modelling (DCM) is a framework that allows for the identification of the causal (directed) connections among neuronal systems--known as effective connectivity.We also simulated group differences and compared the ability of spectral and stochastic DCMs to identify these differences.We show that spectral DCM was not only more accurate but also more sensitive to group differences.
Affiliation: The Wellcome Trust Centre for Neuroimaging, University College London, 12 Queen Square, London WC1N 3BG, UK; Department of Electronic Engineering, NED University of Engineering and Technology, Karachi, Pakistan. Electronic address: email@example.com.Show MeSH
Related in: MedlinePlus
Mentions: The posterior estimates of the effective connectivity are shown in Fig. 3. Posterior expectations are presented as grey bars with pink bars indicating the 90% Bayesian confidence intervals. We have also superimposed the true connectivity as black bars for comparison. The upper left panel shows the posterior expectations for the spDCM inversion (shown as spectral in title), whilst the lower left panel shows the results for the stochastic scheme using generalised filtering. Clearly, the spectral DCM's estimates are very accurate, with most of the extrinsic connection strengths falling within 90% confidence intervals. The intrinsic connections (i.e. the self-connections) are modelled as a (log) scale parameter and have a prior mean of zero. These connections are still estimated with good accuracy showing around a 20% underestimation of the self-connectivity. The stochastic scheme also performed well, with estimates tending towards the true values but not as accurately as the deterministic (spectral) scheme. This reiterates the point that stochastic DCM can find it difficult to recover effective connectivity from data generated from graphs with reciprocal connectivity. We have also presented these results in a scatter plot to illustrate the relative accuracies of the spectral and stochastic estimates (the posterior estimates of spDCM are closer to the true parameters than those generated by the model based on sDCM). It can also be seen that the stochastic model underestimates the parameters, a behaviour which has previously been reported (Li et al., 2011) and is generally characteristic of approximate Bayesian inference schemes that contend with conditional dependencies.
Affiliation: The Wellcome Trust Centre for Neuroimaging, University College London, 12 Queen Square, London WC1N 3BG, UK; Department of Electronic Engineering, NED University of Engineering and Technology, Karachi, Pakistan. Electronic address: firstname.lastname@example.org.