Latent causal modelling of neuroimaging data
Morten Mørup, Kristoffer Madsen and Lars Kai Hansen
In: NIPS 2009 workshop: Connectivity Inference in Neuroimaging, 10 December, Whistler, Canada.
We establish a close connection between Granger causality and convolutive bilinear models. Contrary to Granger
causality that considers causal relations between the measurement variables the convolutive model operate with causal relations to underlying latent sources. We derived a Bayesian approach to estimate the model parameters and demonstrated its success on real and artificial EEG data. The proposed SLCM approach readily generalize to fMRI data.