BCN Neuroimaging Centre University of Groningen The Netherlands PPI Friston (1997) Gitelman (2003)
BCN Neuroimaging Centre University of Groningen The Netherlands Basic fMRI refreshments
BCN Neuroimaging Centre University of Groningen The Netherlands Friston et al 1997
Ni C October 2008 Friston 1997 n Introduction u Aim: define PPI Address interpretation u Basic idea: Correlation between areas changes as context changes.
Ni C October 2008 Effective connectivety efficasy and contributions n Functional specialization n Functional integration n Functional connectivity u (correlation) n Effective connectivity u (taking into account full model)
Ni C October 2008 Effective connectivity efficacy and contributions Test on : H 0 : ik =0 i.e., test correlation between regions Note if more regions, towards effective connectivity.
Ni C October 2008 Factorial designs and Psychological interactions n Imagine 1 task (g r ), two conditions (g a ) Note g r and g a are mean corrected
Ni C October 2008 Physiological interaction n Imagine 2 areas (g r, and g a ) gaga grgr BRAIN
Ni C October 2008 Physiological interaction n Example in paper: u g r =PP u g a =V1 u Responding area: V5 n Note this is interaction and not only due to PP, PP activity is a confound
Ni C October 2008 Non linear models SKIP
Ni C October 2008 Psychophysiological interaction n x k : source region (V1) n g p : task (-1 or +1 label)
Ni C October 2008 x k : V1 g p : task (attention)
Ni C October 2008 fit V1 V5 attentionNo attention
Ni C October 2008 Once more be aware V1 V5 V1 V5 ?
Ni C October 2008 Summary n Psychophysiological interaction u Predict activity in area B by area A as a function of context u PPI effective connectivity u PPI=contribution (c.f. correlation) n Note on interpretations. u Connection A B influenced by task u Influence Task B is modified by activity in A u No guarantee that connections are direct.
BCN Neuroimaging Centre University of Groningen The Netherlands Gitelman et al 2003 (where Friston went “wrong”)
Ni C October 2008 Aim n Show importance of deconvolution n How to deconvolve properly
Ni C October 2008 Introduction n Don’t analyze interactions on raw BOLD signal. (using SEM PPI etc) n “veridical models of neuronal interactions require the neural signal or at least a well- constrained approximation to it. “
Ni C October 2008 A simulation (see examples) Time shift (0-8 s)
Ni C October 2008 Convolved with exp decay & hrf
Ni C October 2008 Deconvolve A&B Interaction Reconvolve
Ni C October 2008 Conclusion Interaction with the convolved signal Interaction at neural representation + convolution
Ni C October 2008 Noise effect
Ni C October 2008 Noise effect
Ni C October 2008 conclusion n Noise has more effect on HRF interactions n Deconvolution reduces noise
Ni C October 2008 Real data ER
Ni C October 2008 Conclusion n There is an effect for event related designs. n Not so strong as simulations.
Ni C October 2008 Real data Block
Ni C October 2008 Conclusion n Effect on BLOCK design data is not dramatic. n In short: u Calculating interactions at neural representation pays especially for ER designs. u Friston was wrong, but not that far off because of block design in his experiment.
Ni C October 2008 Theory interaction on the convoluted signal (i.e. BOLD signal)
Ni C October 2008
Ni C October 2008 How to obtain x A from y A NOTE 112 columns Basis set
Ni C October 2008 How to obtain x A from y A X has too many columns over determined matrix not one unique solution
Ni C October 2008 Solution n Biased estimation. (bayesian stat.) n I start to get lost…..
Ni C October 2008 What I do understand n High frequencies are a problem in deconvolution. u Convolution is low pass filter. high frequency information is lost high frequency estimates are unstable/unreliable. n High frequencies were also the most troubling in interactions based on BOLD signal (cf ER & BLOCK designs) n High frequencies are regularized using bayesian stat.