Limits...
Construct validation of a DCM for resting state fMRI.

Razi A, Kahan J, Rees G, Friston KJ - Neuroimage (2014)

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.

View Article: PubMed Central - PubMed

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: a.razi@ucl.ac.uk.

Show MeSH

Related in: MedlinePlus

This figure reports the Bayesian parameter averages of the effective connection strengths using the same format as in Fig. 3. Because we have pooled over 32 simulated subjects, the confidence intervals are much smaller (and also note the characteristic shrinkage one obtains with Bayesian estimators). The right column (resp. left column) shows the results for spectral DCM (resp. stochastic DCM) revealing the similarity between the Bayesian parameter averages from long runs (upper panel) and shorter runs (lower panel), of 1024 and 256 scans, respectively.
© Copyright Policy - CC BY
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4295921&req=5

f0025: This figure reports the Bayesian parameter averages of the effective connection strengths using the same format as in Fig. 3. Because we have pooled over 32 simulated subjects, the confidence intervals are much smaller (and also note the characteristic shrinkage one obtains with Bayesian estimators). The right column (resp. left column) shows the results for spectral DCM (resp. stochastic DCM) revealing the similarity between the Bayesian parameter averages from long runs (upper panel) and shorter runs (lower panel), of 1024 and 256 scans, respectively.

Mentions: In Monte-Carlo simulations, we generated the time series of various lengths ranging from 128 time points to 1024 time points with a step size of 128. We simulated 32 realisations (as a proxy for number of participants) for each run length and calculated the root mean squared error. The mean RMS error for each run length (averaging over 32 simulations for each run length) is presented in Fig. 4. For spDCM, RMS error decreases as the number of time points increases. It is satisfying to note that the RMS drops below the typical effective connectivity value of 0.1 Hz, for times series with 384 time points, corresponding to around 13 min of scanning (with a repetition time of 2 s). Whilst the sDCM RMS also decreases with longer time series, the errors for sDCM estimates are greater than those for spDCM, and fail to reach the 0.1 Hz threshold. Whilst the accuracies of the sDCM estimates are impressive, demonstrating the scheme insensitivity to the run length, the estimation accuracy (for this particular graph) was inferior to spDCM. The Bayesian parameter averages (ignoring posterior correlations) of the effective connectivity are shown in Fig. 5 as a function of session length. As has been previously noted (Friston et al., 2014), these Bayesian parametric averages do not change greatly when comparing, for example, run lengths of 256 and 1024. As we are pooling over many simulations, the confidence intervals around these Bayesian parameter averages are fairly small. This suggests that when a sufficient number of participants are used, shorter run lengths may provide reasonably accurate estimates.


Construct validation of a DCM for resting state fMRI.

Razi A, Kahan J, Rees G, Friston KJ - Neuroimage (2014)

This figure reports the Bayesian parameter averages of the effective connection strengths using the same format as in Fig. 3. Because we have pooled over 32 simulated subjects, the confidence intervals are much smaller (and also note the characteristic shrinkage one obtains with Bayesian estimators). The right column (resp. left column) shows the results for spectral DCM (resp. stochastic DCM) revealing the similarity between the Bayesian parameter averages from long runs (upper panel) and shorter runs (lower panel), of 1024 and 256 scans, respectively.
© Copyright Policy - CC BY
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4295921&req=5

f0025: This figure reports the Bayesian parameter averages of the effective connection strengths using the same format as in Fig. 3. Because we have pooled over 32 simulated subjects, the confidence intervals are much smaller (and also note the characteristic shrinkage one obtains with Bayesian estimators). The right column (resp. left column) shows the results for spectral DCM (resp. stochastic DCM) revealing the similarity between the Bayesian parameter averages from long runs (upper panel) and shorter runs (lower panel), of 1024 and 256 scans, respectively.
Mentions: In Monte-Carlo simulations, we generated the time series of various lengths ranging from 128 time points to 1024 time points with a step size of 128. We simulated 32 realisations (as a proxy for number of participants) for each run length and calculated the root mean squared error. The mean RMS error for each run length (averaging over 32 simulations for each run length) is presented in Fig. 4. For spDCM, RMS error decreases as the number of time points increases. It is satisfying to note that the RMS drops below the typical effective connectivity value of 0.1 Hz, for times series with 384 time points, corresponding to around 13 min of scanning (with a repetition time of 2 s). Whilst the sDCM RMS also decreases with longer time series, the errors for sDCM estimates are greater than those for spDCM, and fail to reach the 0.1 Hz threshold. Whilst the accuracies of the sDCM estimates are impressive, demonstrating the scheme insensitivity to the run length, the estimation accuracy (for this particular graph) was inferior to spDCM. The Bayesian parameter averages (ignoring posterior correlations) of the effective connectivity are shown in Fig. 5 as a function of session length. As has been previously noted (Friston et al., 2014), these Bayesian parametric averages do not change greatly when comparing, for example, run lengths of 256 and 1024. As we are pooling over many simulations, the confidence intervals around these Bayesian parameter averages are fairly small. This suggests that when a sufficient number of participants are used, shorter run lengths may provide reasonably accurate estimates.

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.

View Article: PubMed Central - PubMed

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: a.razi@ucl.ac.uk.

Show MeSH
Related in: MedlinePlus