Limits...
Dynamic causal modelling of effective connectivity from fMRI: are results reproducible and sensitive to Parkinson's disease and its treatment?

Rowe JB, Hughes LE, Barker RA, Owen AM - Neuroimage (2010)

Bottom Line: In "off" patients, action selection was associated with enhanced connectivity from prefrontal to lateral premotor cortex.Together, these results suggest that DCM model selection is robust and sensitive enough to study clinical populations and their pharmacological treatment.However, caution is required when comparing groups or drug effects in terms of the connectivity parameter estimates, if there are significant posterior covariances among parameters.

View Article: PubMed Central - PubMed

Affiliation: University of Cambridge Department of Clinical Neurosciences, CB2 2QQ, UK. james.rowe@mrc-cbu.cam.ac.uk

Show MeSH

Related in: MedlinePlus

Random effects and Fixed effects Bayesian Model Selection approaches to group model selection for Healthy older controls, PD patients “on” and PD patients “off,” for the leading models E2, E1 and C2. Fixed effects analyses results are presented as log-evidences and Posterior Model Probability (PMP). Random effects analyses are presented as Expected Posterior Probability (EPP) or Exceedance Probabilities (ExPr). It can be seen that by both approaches, model E2 is preferred in Healthy Controls and PD “on” patients, but that model E1 is preferred in PD “off” patients. The difference between fixed and random effects models is seen for PD “on” patients, for whom model E1 is second most likely by the random effects method, but very unlikely by the fixed effects method.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC3021391&req=5

fig4: Random effects and Fixed effects Bayesian Model Selection approaches to group model selection for Healthy older controls, PD patients “on” and PD patients “off,” for the leading models E2, E1 and C2. Fixed effects analyses results are presented as log-evidences and Posterior Model Probability (PMP). Random effects analyses are presented as Expected Posterior Probability (EPP) or Exceedance Probabilities (ExPr). It can be seen that by both approaches, model E2 is preferred in Healthy Controls and PD “on” patients, but that model E1 is preferred in PD “off” patients. The difference between fixed and random effects models is seen for PD “on” patients, for whom model E1 is second most likely by the random effects method, but very unlikely by the fixed effects method.

Mentions: Among healthy controls, the random effects method of Bayesian model selection again identified model E2 as most likely. A comparison of the fixed effects and random effects approaches is shown in Figs. 3 and 4. The results first confirm the clustering of models in to families of structural similarity. For the three more likely models (E2, E1, C2) the random effects and fixed effects approaches produce similar rankings, but the differences between models are easily appreciated with the random effects approach. Considering the three most likely models in Fig. 4, for any given healthy subject, one would be ∼ 57% likely to identify model E2 as the generator of the data, and for the group as a whole, one is 90% confident that E2 is the most likely model that generated subjects' data.


Dynamic causal modelling of effective connectivity from fMRI: are results reproducible and sensitive to Parkinson's disease and its treatment?

Rowe JB, Hughes LE, Barker RA, Owen AM - Neuroimage (2010)

Random effects and Fixed effects Bayesian Model Selection approaches to group model selection for Healthy older controls, PD patients “on” and PD patients “off,” for the leading models E2, E1 and C2. Fixed effects analyses results are presented as log-evidences and Posterior Model Probability (PMP). Random effects analyses are presented as Expected Posterior Probability (EPP) or Exceedance Probabilities (ExPr). It can be seen that by both approaches, model E2 is preferred in Healthy Controls and PD “on” patients, but that model E1 is preferred in PD “off” patients. The difference between fixed and random effects models is seen for PD “on” patients, for whom model E1 is second most likely by the random effects method, but very unlikely by the fixed effects method.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC3021391&req=5

fig4: Random effects and Fixed effects Bayesian Model Selection approaches to group model selection for Healthy older controls, PD patients “on” and PD patients “off,” for the leading models E2, E1 and C2. Fixed effects analyses results are presented as log-evidences and Posterior Model Probability (PMP). Random effects analyses are presented as Expected Posterior Probability (EPP) or Exceedance Probabilities (ExPr). It can be seen that by both approaches, model E2 is preferred in Healthy Controls and PD “on” patients, but that model E1 is preferred in PD “off” patients. The difference between fixed and random effects models is seen for PD “on” patients, for whom model E1 is second most likely by the random effects method, but very unlikely by the fixed effects method.
Mentions: Among healthy controls, the random effects method of Bayesian model selection again identified model E2 as most likely. A comparison of the fixed effects and random effects approaches is shown in Figs. 3 and 4. The results first confirm the clustering of models in to families of structural similarity. For the three more likely models (E2, E1, C2) the random effects and fixed effects approaches produce similar rankings, but the differences between models are easily appreciated with the random effects approach. Considering the three most likely models in Fig. 4, for any given healthy subject, one would be ∼ 57% likely to identify model E2 as the generator of the data, and for the group as a whole, one is 90% confident that E2 is the most likely model that generated subjects' data.

Bottom Line: In "off" patients, action selection was associated with enhanced connectivity from prefrontal to lateral premotor cortex.Together, these results suggest that DCM model selection is robust and sensitive enough to study clinical populations and their pharmacological treatment.However, caution is required when comparing groups or drug effects in terms of the connectivity parameter estimates, if there are significant posterior covariances among parameters.

View Article: PubMed Central - PubMed

Affiliation: University of Cambridge Department of Clinical Neurosciences, CB2 2QQ, UK. james.rowe@mrc-cbu.cam.ac.uk

Show MeSH
Related in: MedlinePlus