Connectivity analysis with structural equation modelling: an example of the effects of voxel selection

Structural equation modelling (SEM) of neuroimaging data is commonly applied to a network of distributed brain regions. We applied SEM to an fMRI dataset to identify condition-specific effects in a simple experiment composed of visual stimulation and baseline conditions. The visual network was composed of three well-defined anatomical regions (V1, V2, and V5) and three path connections (V1 --> V2, V1 --> V5, and V2 --> V5). This network was used to test four hypotheses: (1) whether the condition-specific effects for all three connections vary according to the data selected for modelling; (2) whether the "summary" measures that are often used are indeed appropriate; (3) whether measures taken from the voxel timecourse can reliably predict the condition-specific effects for each one of the three path connections, and (4) whether all voxels within an anatomical region yield equivalent SEM outcomes. There was some variability in the significance of the condition-specific effects across randomly selected voxels within regions. However, the SEM outcome from the "summary" measures was comparable to the most frequent pattern of condition-specific effects. Magnitude, delay, spread, and goodness-of-fit measures taken from a gamma fit to the voxel time courses predicted reliably the significance of the SEM condition-specific effects for each connection. This result enabled us to identify spatially coherent regions at the boundaries of V2 that displayed different condition-specific effects from those seen in the majority of the voxels. Although the generality of these results awaits further investigation, this example highlights a number of important issues for SEM. We have provided further evidence that the SEM outcome does vary somewhat according to the voxels selected and that, although the use of summary measures can give a generalised view of the connectivity pattern, they could fail to capture functional differences within specialised areas.

[1]  A. E. Maxwell,et al.  Factor Analysis as a Statistical Method. , 1964 .

[2]  S. Kapur,et al.  Mapping Neural Interactivity onto Regional Activity: An Analysis of Semantic Processing and Response Mode Interactions , 1998, NeuroImage.

[3]  Karl J. Friston,et al.  Effective Connectivity and Intersubject Variability: Using a Multisubject Network to Test Differences and Commonalities , 2002, NeuroImage.

[4]  P. Jezzard,et al.  A method for determining venous contribution to BOLD contrast sensory activation. , 2002, Magnetic resonance imaging.

[5]  S Makeig,et al.  Analysis of fMRI data by blind separation into independent spatial components , 1998, Human brain mapping.

[6]  Jieun Kim,et al.  Effects of Verbal Working Memory Load on Corticocortical Connectivity Modeled by Path Analysis of Functional Magnetic Resonance Imaging Data , 2002, NeuroImage.

[7]  A. Basilevsky,et al.  Factor Analysis as a Statistical Method. , 1964 .

[8]  Robert Sekuler,et al.  Corticolimbic Interactions Associated with Performance on a Short-Term Memory Task Are Modified by Age , 2000, The Journal of Neuroscience.

[9]  K J Friston,et al.  The predictive value of changes in effective connectivity for human learning. , 1999, Science.

[10]  C. Büchel,et al.  Modulation of connectivity in visual pathways by attention: cortical interactions evaluated with structural equation modelling and fMRI. , 1997, Cerebral cortex.

[11]  William R. Dillon,et al.  Offending Estimates in Covariance Structure Analysis: Comments on the Causes of and Solutions to Heywood Cases , 1987 .

[12]  Karl J. Friston,et al.  Learning-related neuronal responses in prefrontal cortex studied with functional neuroimaging. , 1999, Cerebral cortex.

[13]  K. Amunts,et al.  Brodmann's Areas 17 and 18 Brought into Stereotaxic Space—Where and How Variable? , 2000, NeuroImage.

[14]  E. Haacke,et al.  Identification of vascular structures as a major source of signal contrast in high resolution 2D and 3D functional activation imaging of the motor cortex at l.5T preliminary results , 1993, Magnetic resonance in medicine.

[15]  Eric R. Ziegel,et al.  Generalized Linear Models , 2002, Technometrics.

[16]  E. Bullmore,et al.  How Good Is Good Enough in Path Analysis of fMRI Data? , 2000, NeuroImage.

[17]  Ingrid S. Johnsrude,et al.  Can Meaningful Effective Connectivities Be Obtained between Auditory Cortical Regions? , 2001, NeuroImage.

[18]  Chantal Delon-Martin,et al.  fMRI Retinotopic Mapping—Step by Step , 2002, NeuroImage.

[19]  Leslie G. Ungerleider,et al.  Modulation of sensory suppression: implications for receptive field sizes in the human visual cortex. , 2001, Journal of neurophysiology.

[20]  F. Gonzalez-Lima,et al.  Structural equation modeling and its application to network analysis in functional brain imaging , 1994 .

[21]  G. Orban,et al.  Attention to Speed of Motion, Speed Discrimination, and Task Difficulty: An fMRI Study , 2000, NeuroImage.

[22]  R. Cabeza,et al.  Analysis of neural interactions explains the activation of occipital cortex by an auditory stimulus. , 1998, Journal of neurophysiology.

[23]  Scott L. Zeger,et al.  Non‐linear Fourier Time Series Analysis for Human Brain Mapping by Functional Magnetic Resonance Imaging , 1997 .

[24]  Scott T. Grafton,et al.  Network analysis of motor system connectivity in Parkinson's disease: Modulation of thalamocortical interactions after pallidotomy , 1994 .

[25]  A R McIntosh,et al.  Functional interactions between the medial temporal lobes and posterior neocortex related to episodic memory retrieval. , 1998, Cerebral cortex.

[26]  J V Haxby,et al.  Network analysis of PET-mapped visual pathways in Alzheimer type dementia. , 1995, Neuroreport.

[27]  E. Marg.  A VISION OF THE BRAIN , 1994 .

[28]  Karl J. Friston,et al.  Event-related fMRI , 1997 .