Statistical modelling of brain activity for ERP data analysis
D. Causeur, Professor, Agrocampus Ouest and IRMAR (UMR 6625 CNRS), Rennes, France
C.-F. Sheu, Professor, National Cheng-Kung University, Tainan, Taiwan.
Research project grant by Collège de Recherche Hubert Curien
D. Causeur (Agrocampus Ouest, Applied Mathematics Department)
M.-C. Chu (NCKU, Institute of Education)
S. Lê (Agrgocampus Ouest, Applied Mathematics Department)
C.-F. Sheu (NCKU, Institute of Education)
S. Hsieh (NCKU, Institute of Cognitive Science)
Brain activity by electroencephalography (EEG)
How brain components interact with each other and how brain activity changes in time-course experiments are important topics in psychophysiology. This variability of brain activity can be modelled using technologies such as functional magnetic resonance imaging (fMRI) or electroencephalography (EEG). For example, fMRI is used to understand the spatial reorganization of brain activity after stroke. EEG offers especially interesting applications in psychophysiological experiments focusing on the variations of brain activity over time. Figure 1 shows the network of 128 sensors used in a EEG experiment and gives some examples of EEG waveforms
Fig 1.1. Ellectrode locations of a 128 channels EEG (from Cunningham et al, 2005).
Fig 1.2. illustrative EEG signals for different channels (From Dolcos et Cabeza, 2002)
Changes of EEG signals in respond to stimulus events during the experiment are of particular interest for psychophysiologists. These altered EEG signals are called Event Related Potentials (ERP, see Handy, 2004 for a review on ERP). Depending on the objective of the psychological experiment, the stress can be either visual, auditory or for example induced by a brutal change in the task executed by the subject.
ERP data analysis generally highlights recurrent patterns which can be positive or negative peaks which intensity and location on the time axis can reveal some special type of brain activity. For example, a positif peak at 300 ms (the P300 ERP component) after the onset of a stimulus is expected in case of cognitive tasks.
Analysis of the dynamics of a complex system such as the brain by technologies generating high-throughput data such as EEG has to handle the large heterogeneity of the data and well-known statistical challenges for high-dimensional data. Normal brain activity can indeed vary from a subject to another one and even for a given subject, from an experiment to another one made later. Alteration of this background activity on a short period of time can therefore be hard to detect by standard statistical methods. Moreover, EEG signals, viewed as continuous functions along time, are observed for each subject on 128 channels, which generates a large amount of data with respect to the usually moderate number of subjects. Therefore, efficient procedures identifying ERP components necessarily rely on specific statistical models. However, the methodology for ERP data usually derives from standard procedures (see Lage-Castellanos et al, 2010).
Statistical models for ERP data analysis
Comparison of high-dimensional profiles using heterogeneous data generated by high-throughput technologies can be viewed as analogous with microarray data analysis in genomic issues. Inspired from Friguet et al. (2009), Blum et al. (2010) show that unsupervised identification of independant components of heterogeneity using a factor model leads to marked improvements with respect to standard procedures. A similar approach on ERP data however necessitates to account for specific dependence structures related to the space-time brain activity dynamics.
Our aim is to propose a functional statistical model for ERP data, inspired from the functional ANOVA model using P-splines as proposed by Bugli and lambert (2006), with an additional error term modelling the heterogeneity of the ERP data. A firts comparative study (see Chen-Chu et al., 2010) based on simulations shows promising improvements. Figure 2 illustrates this power gain in the context of a task-switching experiment described in Hsieh and Wu (2010) : the identification of heterogeneity compoents clearly helps in detecting a larger number of significant intervals with respect to standard procedures.
Fig 2. 2-group comparison of ERP profiles in a task-switching experiment (see Hsieh and Wu, 2010). Significant time intervals by a standard method (the FDR controlling Benjamini-Hochberg procedure) are in green and by the present method in yellow.
References
- Blum Y., Le Mignon G., Lagarrigue S. and Causeur D. (2010) A Factor Model to Analyze Heterogeneity in Gene Expression. BMC Bioinformatics, 11 :368.
- Bugli, C. and Lambert, P. (2006). functional ANOVA with random functional effects : an application to event-related potentials modelling for electroencephalograms analysis. Statistics in Medicine. 25, 3718-3739.
- Chu, M.-C., Causeur, D., Hsieh S. and Sheu, C.-F. (2010). A Comparison of Two Multiple Testing Procedures in ERP Data Analysis. 49th Taiwanese Psychology Association Meeting. Taiwan.
- Cunningham, W.A., Espinet, S.D., DeYoung, C.G., & Zelazo, P.D.(2005). Attitudes to the right- and left: Frontal ERP asymmetries associated with stimulus valence and processing goals. NeuroImage, 28, 827-834.
- Dolcos, F. and Cabeza, R. (2002). Event-related potentials of emotional memory: Encoding pleasant,
unpleasant, and neutral pictures. Cognitive, Affective, & Behavioral Neuroscience 2 (3), 252-263. - Friguet, C., Kloareg, M. and Causeur, D. (2009). A factor model approach to multiple testing under dependence. Journal of the American Statistical Association, 104 :488, 1406-1415
- Guthrie and Buchwald (1991). Significance testing of difference potentials. Psychophysiology 28, 240U244.
- Handy, T. (2004). Event-Related Potentials. The MIT Press : Cambridge.
- Hsieh, S. and Wu, M. (2010). Age differences in switching the relevant stimulus dimensions in a speeded same-different paradigm. Acta Psychologica, 135(2), 140-149.
- Lage-Castellanos, A., MartÃnez-Montes, E., Hernandez-Cabrera, J.-A. and Galan, L. (2010). False discovery rate and permutation test : an evaluation in ERP data analysis. Statistics in Medicine. 29, 63-74.