Dynamic Causal Modelling for EEG/MEG: principles J. Daunizeau Motivation, Brain and Behaviour group, ICM, Paris, France Wellcome Trust Centre for Neuroimaging, London, UK Overview 1 DCM: introduction 2 Dynamical systems theory 3 Neural states dynamics 4 Bayesian inference 5 Conclusion Overview 1 DCM: introduction 2 Dynamical systems theory 3 Neural states dynamics 4 Bayesian inference 5 Conclusion Introduction structural, functional and effective connectivity structural connectivity functional connectivity effective connectivity O. Sporns 2007, Scholarpedia • structural connectivity = presence of axonal connections • functional connectivity = statistical dependencies between regional time series • effective connectivity = causal (directed) influences between neuronal populations ! connections are recruited in a context-dependent fashion Introduction from functional segregation to functional integration localizing brain activity: functional segregation effective connectivity analysis: functional integration A B A B ? u1 u1 u2 u1 u2 u1 X u2 « Where, in the brain, did my experimental manipulation have an effect? » « How did my experimental manipulation propagate through the network? » Introduction DCM: evolution and observation mappings Hemodynamic observation model: temporal convolution Electromagnetic observation model: spatial convolution neural states dynamics x f ( x, u , ) fMRI EEG/MEG • simple neuronal model • realistic observation model • realistic neuronal model • simple observation model inputs Introduction DCM: a parametric statistical approach • DCM: model structure y g x, x f x, u, 24 2 likelihood p y , , m 4 3 1 u • DCM: Bayesian inference parameter estimate: ˆ E y, m priors on parameters model evidence: p y m p y , , m p m p m d d Introduction DCM for EEG-MEG: auditory mismatch negativity sequence of auditory stimuli … … S S S D S S S S D S S-D: reorganisation of the connectivity structure standard condition (S) rIFG rIFG deviant condition (D) lA1 lSTG rA1 lSTG rSTG lA1 rA1 rSTG t ~ 200 ms Daunizeau, Kiebel et al., Neuroimage, 2009 Introduction DCM for fMRI: audio-visual associative learning auditory cue visual outcome or P(outcome|cue) or Put response 0 200 400 600 800 2000 time (ms) PMd PPA FFA PPA cue-dependent surprise Put FFA PMd cue-independent surprise Den Ouden, Daunizeau et al., J. Neurosci., 2010 Overview 1 DCM: introduction 2 Dynamical systems theory 3 Neural states dynamics 4 Bayesian inference 5 Conclusion Dynamical systems theory motivation u x y 1 1 21 1 2 32 13 3 13u 1 2 2 3 3 2 time t t 3 3u t 0 t u t t x t 0 ? t Dynamical systems theory exponentials Dynamical systems theory initial values and fixed points Dynamical systems theory time constants Dynamical systems theory matrix exponential Dynamical systems theory eigendecomposition of the Jacobian Dynamical systems theory dynamical modes Dynamical systems theory spirals Dynamical systems theory spirals Dynamical systems theory spiral state-space Dynamical systems theory embedding Dynamical systems theory kernels and convolution Dynamical systems theory summary • Motivation: modelling reciprocal influences • Link between the integral (convolution) and differential (ODE) forms • System stability and dynamical modes can be derived from the system’s Jacobian: o D>0: fixed points o D>1: spirals o D>1: limit cycles (e.g., action potentials) o D>2: metastability (e.g., winnerless competition) limit cycle (Vand Der Pol) strange attractor (Lorenz) Overview 1 DCM: introduction 2 Dynamical systems theory 3 Neural states dynamics 4 Bayesian inference 5 Conclusion Neural ensembles dynamics DCM for M/EEG: systems of neural populations macro-scale meso-scale Golgi micro-scale Nissl EI external granular layer external pyramidal layer EP II mean-field firing rate internal granular layer internal pyramidal layer synaptic dynamics Neural ensembles dynamics DCM for M/EEG: from micro- to meso-scale x j t : post-synaptic potential of j th neuron within its ensemble 1 N H x t j' H x t p x t dx N 1 j ' j S mean-field firing rate mean firing rate (Hz) ensemble density p(x) S(x) H(x) S(x) membrane depolarization (mV) mean membrane depolarization (mV) Neural ensembles dynamics DCM for M/EEG: synaptic dynamics post-synaptic potential membrane depolarization (mV) EPSP 1 2 2 2 2 i / e S ( ) 2 i / e 2 i / e 1 IPSP Ki 1 K e 1 time (ms) Neural ensembles dynamics DCM for M/EEG: intrinsic connections within the cortical column 7 8 Golgi external granular layer 8 3 e2 S ( 0 ) 2 e 8 e2 7 Nissl inhibitory interneurons external pyramidal layer internal granular layer internal pyramidal layer spiny stellate cells 4 3 1 4 4 1 e2 S ( 0 ) 2 e 4 e2 1 1 intrinsic connections pyramidal cells 0 5 6 2 5 5 2 e2 S ( 1 ) 2 e 5 e2 2 3 x6 6 4 i2 S ( 7 ) 2 i 6 i2 3 2 Neural ensembles dynamics ( i ) t ( i ) r, t lateral (homogeneous) density of connexions DCM for M/EEG: from meso- to macro-scale r1 r2 local wave propagation equation (neural field): 2 3 2 2 (i ) 2 2 c r, t c (i ) r, t 2 t 2 t (i ) ii ' S (i ') i' 0th-order approximation: standing wave Neural ensembles dynamics DCM for M/EEG: extrinsic connections between brain regions extrinsic lateral L S ( 0 ) connections 7 8 8 e2 (( B L 3 I ) S ( 0 )) 2 e 8 e2 7 inhibitory interneurons spiny stellate cells extrinsic forward connections F S ( 0 ) 4 3 1 4 4 e2 (( F L 1 I ) S ( 0 ) u u ) 2 e 4 e2 1 1 pyramidal cells 2 0 5 6 2 5 5 e2 (( B L ) S ( 0 ) 2 S ( 1 )) 2 e 5 e2 2 3 x6 6 i2 4 S ( 7 ) 2 i 6 i2 3 extrinsic backward B S ( 0 ) connections Overview 1 DCM: introduction 2 Dynamical systems theory 3 Neural states dynamics 4 Bayesian inference 5 Conclusion Bayesian inference forward and inverse problems forward problem p y , m likelihood posterior distribution p y , m inverse problem Bayesian inference the electromagnetic forward problem y t L(i ) w (0i ) j (ij ) t t i j Bayesian paradigm deriving the likelihood function - Model of data with unknown parameters: y f - But data is noisy: e.g., GLM: f X y f - Assume noise/residuals is ‘small’: f 1 p exp 2 2 2 P 4 0.05 → Distribution of data, given fixed parameters: 2 1 p y exp 2 y f 2 Bayesian paradigm likelihood, priors and the model evidence Likelihood: Prior: Bayes rule: generative model m Bayesian inference model comparison Principle of parsimony : « plurality should not be assumed without necessity » Model evidence: y = f(x) p y m p y , m p m d y=f(x) x model evidence p(y|m) “Occam’s razor” : space of all data sets Bayesian inference the variational Bayesian approach ln p y m ln p , y m S q DKL q ; p y, m q free energy : functional of q mean-field: approximate marginal posterior distributions: q , q 1 2 p 1 , 2 y , m 2 p 1 or 2 y, m 1 q 1 or 2 Bayesian inference DCM: key model parameters 21 1 2 32 13 3 13u 3u u 21,32 ,13 state-state coupling 3u input-state coupling 13u input-dependent modulatory effect Bayesian inference model comparison for group studies ln p y m1 ln p y m2 differences in log- model evidences m1 m2 subjects fixed effect assume all subjects correspond to the same model random effect assume different subjects might correspond to different models Overview 1 DCM: introduction 2 Dynamical systems theory 3 Neural states dynamics 4 Bayesian inference 5 Conclusion Conclusion back to the auditory mismatch negativity sequence of auditory stimuli … … S S S D S S S S D S S-D: reorganisation of the connectivity structure standard condition (S) rIFG rIFG deviant condition (D) lA1 lSTG t ~ 200 ms rA1 lSTG rSTG lA1 rA1 rSTG Conclusion DCM for EEG/MEG: variants input second-order mean-field DCM 1st and 2d order moments depolarization 250 0 0 200 -20 -20 150 -40 -40 100 -60 -60 50 -80 -80 0 0 100 200 300 time (ms) auto-spectral density LA -100 0 100 200 300 time (ms) -100 0 100 200 300 time (ms) auto-spectral density CA1 cross-spectral density CA1-LA frequency (Hz) frequency (Hz) DCM for steady-state responses frequency (Hz) DCM for induced responses DCM for phase coupling Conclusion planning a compatible DCM study • Suitable experimental design: – any design that is suitable for a GLM – preferably multi-factorial (e.g. 2 x 2) • e.g. one factor that varies the driving (sensory) input • and one factor that varies the modulatory input • Hypothesis and model: – define specific a priori hypothesis – which models are relevant to test this hypothesis? – check existence of effect on data features of interest – there exists formal methods for optimizing the experimental design for the ensuing bayesian model comparison [Daunizeau et al., PLoS Comp. Biol., 2011] Many thanks to: Karl J. Friston (UCL, London, UK) Will D. Penny (UCL, London, UK) Klaas E. Stephan (UZH, Zurich, Switzerland) Stefan Kiebel (MPI, Leipzig, Germany)