Dynamic Causal Modelling (DCM) for fMRI Klaas Enno Stephan Laboratory for Social & Neural Systems Research (SNS) University of Zurich Wellcome Trust Centre for Neuroimaging University College London SPM Course, FIL 13 May 2011 Structural, functional & effective connectivity • anatomical/structural connectivity = presence of axonal connections Sporns 2007, Scholarpedia • functional connectivity = statistical dependencies between regional time series • effective connectivity = directed influences between neurons or neuronal populations Some models of effective connectivity for fMRI data • Structural Equation Modelling (SEM) McIntosh et al. 1991, 1994; Büchel & Friston 1997; Bullmore et al. 2000 • regression models (e.g. psycho-physiological interactions, PPIs) Friston et al. 1997 • Volterra kernels Friston & Büchel 2000 • Time series models (e.g. MAR/VAR, Granger causality) Harrison et al. 2003, Goebel et al. 2003 • Dynamic Causal Modelling (DCM) bilinear: Friston et al. 2003; nonlinear: Stephan et al. 2008 Dynamic causal modelling (DCM) • DCM framework was introduced in 2003 for fMRI by Karl Friston, Lee Harrison and Will Penny (NeuroImage 19:1273-1302) • part of the SPM software package • currently more than 160 published papers on DCM Dynamic Causal Modeling (DCM) Hemodynamic forward model: neural activityBOLD Electromagnetic forward model: neural activityEEG MEG LFP Neural state equation: dx F ( x , u, ) dt fMRI simple neuronal model complicated forward model EEG/MEG complicated neuronal model simple forward model inputs Example: a linear model of interacting visual regions x3 x1 FG left LG left FG right LG right x4 LG = lingual gyrus FG = fusiform gyrus x2 Visual input in the - left (LVF) - right (RVF) visual field. RVF LVF u2 u1 x1 a11 x1 a12 x2 a13 x3 c12u2 x2 a21 x1 a22 x2 a24 x4 c21u1 x3 a31 x1 a33 x3 a34 x4 x4 a42 x2 a43 x3 a44 x4 Example: a linear model of interacting visual regions x3 x1 FG left LG left { A, C} LG right x4 LG = lingual gyrus FG = fusiform gyrus x2 Visual input in the - left (LVF) - right (RVF) visual field. RVF LVF u2 u1 state changes x Ax Cu FG right effective connectivity system state input parameters external inputs x1 a11 a12 a13 0 x1 0 c12 x c x a a u 0 a 0 1 24 2 21 2 21 22 x3 a31 0 a33 a34 x3 0 0 u2 0 a a a x x 0 0 42 43 44 4 4 Extension: bilinear model x3 FG left FG right x4 m x ( A u j B( j ) ) x Cu j 1 x1 LG left LG right x2 RVF CONTEXT LVF u2 u3 u1 0 b12(3) x1 a11 a12 a13 0 x a a 0 a 0 0 24 2 21 22 u3 0 0 x3 a31 0 a33 a34 x4 0 a42 a43 a44 0 0 0 0 0 0 b34(3) 0 0 0 x1 0 c12 x c 0 2 21 x3 0 0 x4 0 0 0 u1 0 u2 0 u3 0 y y BOLD y activity x2(t) neuronal states hemodynamic model x integration modulatory input u2(t) t Neural state equation endogenous connectivity t λ activity x3(t) activity x1(t) driving input u1(t) y modulation of connectivity direct inputs x ( A u j B( j ) ) x Cu x x x u j x A B( j) x C u Bilinear DCM driving input modulation Two-dimensional Taylor series (around x0=0, u0=0): dx f f f f ( x, u) f ( x0 ,0) x u ux ... dt x u xu 2 Bilinear state equation: m dx A ui B ( i ) x Cu dt i 1 f A x u 0 2 f B xu f C u x 0 DCM parameters = rate constants Integration of a first-order linear differential equation gives an exponential function: dx ax dt x(t ) x0 exp(at ) Coupling parameter a is inversely proportional to the half life of z(t): x( ) 0.5 x0 The coupling parameter a thus describes the speed of the exponential change in x(t) 0.5x0 x0 exp( a ) a ln 2 / ln 2 / a Example: context-dependent decay stimuli u1 context u2 + - x1 + u1 u1 u2 u2 Z1 x Z2 1 x2 + x2 - x Ax u2 B (2) x Cu1 - Penny et al. 2004, NeuroImage x1 x a 21 2 2 a12 b11 x u 2 0 0 c1 0 u1 x u 2 0 0 b 22 2 The problem of hemodynamic convolution Goebel et al. 2003, Magn. Res. Med. Hemodynamic forward models are important for connectivity analyses of fMRI data Granger causality DCM David et al. 2008, PLoS Biol. u The hemodynamic model in DCM stimulus functions t m dx A u j B ( j ) x Cu dt j 1 neural state equation 0.4 0.2 vasodilatory signal 0 s x s γ( f 1) f 0 2 4 6 8 10 12 s s N RBM N, = 1 CBM , = 1 N RBM , = 2 1 flow induction (rCBF) 0.5 f s hemodynamic state equations N CBM N, = 2 0 f Balloon model changes in volume τv f v1 /α v ( q, v ) 14 RBM N, = 0.5 CBM , = 0.5 v 0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14 0.2 changes in dHb τq f E ( f,E0 ) qE0 v1/α q/v q 0 -0.2 -0.4 -0.6 S q V0 k1 1 q k2 1 k3 1 v S0 v k1 4.30 E0TE k2 r0 E0TE k3 1 BOLD signal change equation Stephan et al. 2007, NeuroImage How interdependent are neural and hemodynamic parameter estimates? 1 A 0.8 5 0.6 10 B 0.4 15 C 0.2 20 0 25 -0.2 h ε 30 -0.4 35 -0.6 -0.8 40 5 10 15 20 25 30 35 40 -1 Stephan et al. 2007, NeuroImage DCM is a Bayesian approach new data prior knowledge p( y | ) p ( ) p( | y ) p( y | ) p( ) posterior likelihood ∙ prior Bayes theorem allows one to formally incorporate prior knowledge into computing statistical probabilities. In DCM: empirical, principled & shrinkage priors. The “posterior” probability of the parameters given the data is an optimal combination of prior knowledge and new data, weighted by their relative precision. stimulus function u Overview: parameter estimation • • • • Combining the neural and hemodynamic states gives the complete forward model. An observation model includes measurement error e and confounds X (e.g. drift). Bayesian inversion: parameter estimation by means of variational EM under Laplace approximation Result: Gaussian a posteriori parameter distributions, characterised by mean ηθ|y and covariance Cθ|y. neural state equation x ( A u j B j ) x Cu activity - dependent vasodilatory signal s z s γ( f 1) s s f parameters flow - induction (rCBF) hidden states z {x, s, f , v, q} state equation f s h { , , , , } f n { A, B1...B m , C} { h , n } z F ( x , u, ) changes in volume τv f v1 /α v changes in dHb τq f E ( f, ) q v1/α q/v q v ηθ|y y (x ) y h(u , ) X e modelled BOLD response observation model Inference about DCM parameters: Bayesian single-subject analysis • Gaussian assumptions about the posterior distributions of the parameters • posterior probability that a certain parameter (or contrast of parameters cT ηθ|y) is above a chosen threshold γ: cT y p N cT C y c • By default, γ is chosen as zero ("does the effect exist?"). Bayesian single subject inference LD|LVF 0.13 0.19 FG left LD p(cT>0|y) = 98.7% 0.34 0.14 FG right 0.44 0.14 0.29 0.14 LG left 0.01 0.17 RVF stim. Stephan et al. 2005, Ann. N.Y. Acad. Sci. LD LG right -0.08 0.16 LD|RVF LVF stim. Contrast: Modulation LG right LG links by LD|LVF vs. modulation LG left LG right by LD|RVF Inference about DCM parameters: Bayesian parameter averaging (FFX group analysis) Likelihood distributions from different subjects are independent Under Gaussian assumptions this is easy to compute: one can use the posterior from one subject as the prior for the next group posterior covariance p | y1 y N p y1 y N p individual posterior covariances N p p yi i 1 N p y1 p yi i 2 N p y1 , y2 p yi p y1 1 | y1 ,..., y N C | y ,..., y 1 N i 3 y N 1 p y N “Today’s posterior is tomorrow’s prior” group posterior mean N C|1yi i 1 N 1 C | yi | yi C | y1 ,..., y N i 1 individual posterior covariances and means Inference about DCM parameters: RFX group analysis (frequentist) • In analogy to “random effects” analyses in SPM, 2nd level analyses can be applied to DCM parameters: Separate fitting of identical models for each subject Selection of (bilinear) parameters of interest one-sample t-test: parameter > 0 ? paired t-test: parameter 1 > parameter 2 ? rmANOVA: e.g. in case of multiple sessions per subject definition of model space inference on model structure or inference on model parameters? inference on individual models or model space partition? optimal model structure assumed to be identical across subjects? yes FFX BMS comparison of model families using FFX or RFX BMS inference on parameters of an optimal model or parameters of all models? optimal model structure assumed to be identical across subjects? yes no FFX BMS RFX BMS no RFX BMS Stephan et al. 2010, NeuroImage FFX analysis of parameter estimates (e.g. BPA) RFX analysis of parameter estimates (e.g. t-test, ANOVA) BMA What type of design is good for DCM? Any design that is good for a GLM of fMRI data. GLM vs. DCM DCM tries to model the same phenomena (i.e. local BOLD responses) as a GLM, just in a different way (via connectivity and its modulation). No activation detected by a GLM → no motivation to include this region in a deterministic DCM. However, a stochastic DCM could be applied despite the absence of a local activation. Stephan 2004, J. Anat. Multifactorial design: explaining interactions with DCM Stim 1 Stim 2 Stimulus factor Task factor Stim1/ Task A Stim2/ Task A Task A Task B TA/S1 TB/S1 X1 X2 TA/S2 TB/S2 Stim 1/ Task B Stim 2/ Task B X1 X2 Let’s assume that an SPM analysis shows a main effect of stimulus in X1 and a stimulus task interaction in X2. Stim1 How do we model this using DCM? Stim2 Task A Task B GLM DCM Simulated data X1 Stimulus 1 – +++ – + X1 Stimulus 2 + +++ +++ Task A X2 Stim 1 Task A + Task B X2 Stephan et al. 2007, J. Biosci. Stim 2 Task A Stim 1 Task B Stim 2 Task B X1 Stim 1 Task A Stim 2 Task A Stim 1 Task B Stim 2 Task B X2 plus added noise (SNR=1) DCM10 in SPM8 • DCM10 was released as part of SPM8 in July 2010 (version 4010). • Introduced many new features, incl. two-state DCMs and stochastic DCMs • This led to various changes in model defaults, e.g. – inputs mean-centred – changes in coupling priors – self-connections: separately estimated for each area • For details, see: www.fil.ion.ucl.ac.uk/spm/software/spm8/SPM8_Release_Notes_r4010.pdf • Further changes in version 4290 (released April 2011) to accommodate new developments and give users more choice (e.g. whether or not to meancentre inputs). The evolution of DCM in SPM • DCM is not one specific model, but a framework for Bayesian inversion of dynamic system models • The default implementation in SPM is evolving over time – better numerical routines for inversion – change in priors to cover new variants (e.g., stochastic DCMs, endogenous DCMs etc.) To enable replication of your results, you should ideally state which SPM version you are using when publishing papers. Factorial structure of model specification in DCM10 • Three dimensions of model specification: – bilinear vs. nonlinear – single-state vs. two-state (per region) – deterministic vs. stochastic • Specification via GUI. bilinear DCM non-linear DCM modulation driving input driving input modulation Two-dimensional Taylor series (around x0=0, u0=0): dx 2 f x2 f f 2 f f ( x, u ) f ( x0 ,0) ... x u ux ... 2 dt x 2 x u xu Bilinear state equation: m dx A ui B ( i ) x Cu dt i 1 Nonlinear state equation: m n dx (i ) ( j) A ui B x j D x Cu dt i 1 j 1 Neural population activity 0.4 0.3 0.2 u2 0.1 0 0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 0.6 u1 0.4 x3 0.2 0 0.3 0.2 0.1 0 x1 x2 3 fMRI signal change (%) 2 1 0 Nonlinear dynamic causal model (DCM) 0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100 4 3 m n dx (i ) ( j) A ui B x j D x Cu dt i 1 j 1 2 1 0 -1 3 2 1 Stephan et al. 2008, NeuroImage 0 attention MAP = 1.25 0.10 0.8 0.7 PPC 0.6 0.26 0.5 0.39 1.25 stim 0.26 V1 0.13 0.46 0.50 V5 0.4 0.3 0.2 0.1 0 -2 motion Stephan et al. 2008, NeuroImage -1 0 1 2 3 4 p( DVPPC 5,V 1 0 | y ) 99.1% 5 motion & attention static motion & no attention dots V1 V5 PPC observed fitted Two-state DCM Single-state DCM Two-state DCM input u x1E x1E x1 x1I x1I x x Cu ij ij exp( Aij uBij ) x x Cu ij Aij uBij 11 1N N 1 NN Marreiros et al. 2008, NeuroImage x1 x x N EE 11 IE 11 EE N 1 0 EI 11 1EEN II 11 0 0 EE NN 0 IE NN Extrinsic (between-region) coupling 0 0 EE NN IINN Intrinsic (within-region) coupling x1E I x1 x E xN xI N -1 Stochastic DCM 0 200 400 600 800 1000 hidden states - neuronal 0.1 excitatory signal 0.05 dx f x, u , dt • accounts for stochastic neural fluctuations • can be fitted to resting state data • has unknown precision and smoothness additional hyperparameters 1200 0 -0.05 -0.1 0 200 400 600 800 1000 1200 hidden states - hemodynamic 1.3 flow volume dHb 1.2 1.1 1 0.9 0.8 0 200 400 600 800 1000 1200 predicted BOLD signal 2 observed predicted 1 0 -1 -2 -3 Friston et al. (2008, 2011) NeuroImage Daunizeau et al. (2009) Physica D 0 200 400 600 time (seconds) 800 1000 1200 Li et al. (2011) NeuroImage Thank you