Task Design & Analysis: From GLM to MVPA
Anna I Blazejewska1
1A.A. Martinos Center for Biomedical Imaging, Harvard Medical School, Massachusetts General Hospital, United States

Synopsis

This lecture will discuss the motivation, methodology and limitations of different experimental designs and data analysis approaches for task-based fMRI experiments. More specifically, the GLM and MVPA approaches will be discussed and compared.

Target audience

This lecture is intended for researchers, clinicians and students in the fields of Neuroscience, Cognitive science, Neuroradiology, MR physics, and more, who are currently involved in task-based fMRI experiments or interested in learning about task-based fMRI design and analysis.

Outcomes & objectives

Understanding the motivation, methodology and limitations of different experimental designs and data analysis approaches for task-based fMRI experiments.

Introduction

Since it was first introduced in 1990s [1]–[4], the general linear model (GLM) has become the most common framework for fMRI data analysis using either block or event-related paradigms [5]–[7]. More recently, the popularity of multivoxel pattern analysis (MVPA) methodologies has been growing in the fMRI community [8]–[10], as they have been applied to decode and predict task-induced brain activity [11]–[13]. In this lecture the GLM and MVPA approaches will be discussed and compared.

Methods

The GLM framework detects task-related brain activation by fitting a linear model to BOLD fMRI time courses. The shape of the hemodynamic response function (HRF) is assumed to be known and for each voxel the model fitting provides the estimates of response amplitudes while noise is estimated from the residuals. A test statistic (such as t-statistic) can be generated and converted to p-values and a probability threshold is applied to allow statistical inference. As the time series measured in multiple voxels are analyzed independently, this approach is considered mass-univariate and model-based. On the other hand, MVPA is a classification methodology that examines all available voxels and aims to detect spatial patterns of task-related brain activation, using machine learning, pattern recognition algorithms and more recently also neural networks [14]. With no explicit modeling assumptions, it is not only multivariate, but also a data-driven approach.

Discussion

The classical model-based GLM approach is used in fMRI analysis of block-related task design experiments.It assumes a canonical HRF, while the actual hemodynamic responses are known to vary across brain regions, across time and across individuals [15]. In order to perform a linear regression without the assumptions about the HRF shape, event-related task designs can be used and finite impulse response (FIR) analysis would then be applied in order for the response shape itself be inferred from the data. As a univariate approach, GLM neglects a potential covariance of the task-related activation across neighboring voxels which could be considered a weakness of this approach. MVPA addresses this limitation by extracting spatial patterns of the activation and therefore may allow better insights in spatial functional organization of the brain. These methods however also have limitations due to the variability of the detected patterns across subjects which may require building separate classifiers for each case.

Conclusions

Both univariate and multivariate approaches to fMRI data analysis provide certain benefits and both have the limitations that should be considered while choosing the methodology of the experiment.

Acknowledgements

No acknowledgement found.

References

[1] K. J. Friston, A. P. Holmes, K. J. Worsley, J. ‐P Poline, C. D. Frith, and R. S. J. Frackowiak, “Statistical parametric maps in functional imaging: A general linear approach,” Hum. Brain Mapp., vol. 2, no. 4, pp. 189–210, 1994.

[2] P. A. Bandettini, A. Jesmanowicz, E. C. Wong, and J. S. Hyde, “Processing strategies for time‐course data sets in functional mri of the human brain,” Magn. Reson. Med., vol. 30, no. 2, pp. 161–173, 1993.

[3] S. M. Smith, “Overview of fMRI analysis,” Br. J. Radiol., vol. 77, no. SPEC. ISS. 2, 2004.

[4] K. J. Friston, O. Josephs, G. Rees, and R. Turner, “Nonlinear Event-Related Responses in fMRI,” MRM, vol. 39, pp. 41–52, 1998.

[5] J. B. Poline and M. Brett, “The general linear model and fMRI: Does love last forever?,” Neuroimage, vol. 62, no. 2, pp. 871–880, 2012.

[6] B. R. Rosen, R. L. Buckner, and A. M. Dale, “Event-related functional MRI: Past, present, and future,” Proc. Natl. Acad. Sci. USA., vol. 95, no. 3, pp. 773–780, 1998.

[7] A. M. Dale and R. L. Buckner, “Selective averaging of individual trials using fMRI,” Neuroimage, vol. 5, no. 4 PART II, pp. 329–340, 1997.

[8] J. V. Haxby, M. I. Gobbini, M. L. Furey, A. Ishai, J. L. Schouten, and P. Pietrini, “Distributed and overlapping representations of faces and objects in ventral temporal cortex,” Science, vol. 293, no. September, pp. 2425–2430, 2001.

[9] D. D. Cox and R. L. Savoy, “Functional magnetic resonance imaging (fMRI) ‘brain reading’: Detecting and classifying distributed patterns of fMRI activity in human visual cortex,” Neuroimage, vol. 19, no. 2, pp. 261–270, 2003.

[10] J. V. Haxby, “Multivariate pattern analysis of fMRI: The early beginnings,” Neuroimage, vol. 62, no. 2, pp. 852–855, 2012.

[11] F. Tong and Y. Kamitani, “Decoding the visual and subjective contents of the human brain,” Nat. Neurosci., vol. 8, no. 5, pp. 679–685, 2005.

[12] J. D. Haynes and G. Rees, “Predicting the orientation of invisible stimuli from activity in human primary visual cortex,” Nat. Neurosci., vol. 8, no. 5, pp. 686–691, 2005.

[13] N. Kriegeskorte, R. Goebel, and P. Bandettini, “Information-based functional brain mapping,” Proc. Natl. Acad. Sci. USA, vol. 103, no. 10, pp. 3863–3868, 2006.

[14] R. Prenger, M. C. K. Wu, S. V. David, and J. L. Gallant, “Nonlinear V1 responses to natural scenes revealed by neural network analysis,” Neural Networks, vol. 17, no. 5–6, pp. 663–679, 2004.

[15] D. A. Handwerker, J. M. Ollinger, and M. D’Esposito, “Variation of BOLD hemodynamic responses across subjects and brain regions and their effects on statistical analyses,” Neuroimage, vol. 21, no. 4, pp. 1639–1651, 2004.

Proc. Intl. Soc. Mag. Reson. Med. 28 (2020)