Temporal multivariate pattern analysis (tMVPA): A single trial approach exploring the temporal dynamics of the BOLD signal

Abstract

Background: fMRI provides spatial resolution that is unmatched by non-invasive neuroimaging techniques. Its temporal dynamics however are typically neglected due to the sluggishness of the hemodynamic signal. New Methods: We present temporal multivariate pattern analysis (tMVPA), a method for investigating the temporal evolution of neural representations in fMRI data, computed on single-trial BOLD time-courses, leveraging both spatial and temporal components of the fMRI signal. We implemented an expanding sliding window approach that allows identifying the time-window of an effect. Results: We demonstrate that tMVPA can successfully detect condition-specific multivariate modulations over time, in the absence of mean BOLD amplitude differences. Using Monte-Carlo simulations and synthetic data, we quantified family-wise error rate (FWER) and statistical power. Both at the group and single-subject levels, FWER was either at or significantly below 5%. We reached the desired power with 18 subjects and 12 trials for the group level, and with 14 trials in the single-subject scenario. Comparison with existing methods: We compare the tMVPA statistical evaluation to that of a linear support vector machine (SVM). SVM outperformed tMVPA with large N and trial numbers. Conversely, tMVPA, leveraging on single trials analyses, outperformed SVM in low N and trials and in a single-subject scenario. Conclusion: Recent evidence suggesting that the BOLD signal carries finer-grained temporal information than previously thought, advocates the need for analytical tools, such as tMVPA, tailored to investigate BOLD temporal dynamics. The comparable performance between tMVPA and SVM, a powerful and reliable tool for fMRI, supports the validity of our technique.

Statistical Power or More Precise Insights into Neuro-Temporal Dynamics? Assessing the Benefits of Rapid Temporal Sampling in fMRI

“…Multivariate analyses methods have been shown to have increased sensitivity for the analyses of spatial maps (Kriegeskorte and Bandettini, 2007;Vizioli et al, 2020a) in fMRI. There is evidence that this is also the case for the temporal domain, with prior work finding earlier identification of statistically significant task differences on both real (Ramon et al, 2015) and synthetically simulated data (Vizioli et al, 2018).…”

“…Analogous to multivariate pattern analysis for spatial maps, a recently developed method considers the temporal corollary of this, termed temporal MVPA (tMVPA) (Vizioli et al, 2018). tMVPA uses single trial was not certified by peer review) is the author/funder.…”

“…Analogous to multivariate pattern analysis for spatial maps, a recently developed method considers the temporal corollary of this, termed temporal MVPA (tMVPA) (Vizioli et al, 2018). tMVPA uses single trial response time courses or single run FIR estimates to compute single trial Representation Dissimilarity Matrices (RDM) independently per condition within a given ROI.…”

“…tMVPA uses single trial response time courses or single run FIR estimates to compute single trial Representation Dissimilarity Matrices (RDM) independently per condition within a given ROI. To infer statistically significant differences across time courses, tMVPA builds on a sliding window approach (extensively tested on real and synthetic data elsewhere - Vizioli et al, 2018) that allows for the precise identification of the temporal window of an effect and whether this encompasses only a few time points or is sustained over a larger time window. Multivariate analyses methods have been shown to have increased sensitivity for the analyses of spatial maps (Kriegeskorte and Bandettini, 2007;Vizioli et al, 2020a) in fMRI.…”

Grasping with a twist: Dissociating action goals from motor actions in human frontoparietal circuits

“…Here we examined (univariate) activation levels and (multivariate) activation patterns (using representational similarity analysis, RSA (Kriegeskorte, 2008). Moreover, we also used a new methodological approach, temporal MVPA (tMVPA; Ramon et al, 2015;Vizioli et al, 2018) to examine the representation similarities across trials for the same and different points in time, separately for isolated vs. sequential actions.…”