September 2021
Volume 21, Issue 9
Open Access
Vision Sciences Society Annual Meeting Abstract  |   September 2021
GLMsingle: a turnkey solution for accurate single-trial fMRI response estimates
Author Affiliations & Notes
  • Jacob S. Prince
    Carnegie Mellon University
  • John A. Pyles
    University of Washington
  • Michael J. Tarr
    Carnegie Mellon University
  • Kendrick N. Kay
    University of Minnesota
  • Footnotes
    Acknowledgements  This research was supported by NIH P41 EB027061, NSF IIS-1822683, and NSF IIS-1822929.
Journal of Vision September 2021, Vol.21, 2831. doi:https://doi.org/10.1167/jov.21.9.2831
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Jacob S. Prince, John A. Pyles, Michael J. Tarr, Kendrick N. Kay; GLMsingle: a turnkey solution for accurate single-trial fMRI response estimates. Journal of Vision 2021;21(9):2831. doi: https://doi.org/10.1167/jov.21.9.2831.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Recent massive fMRI data collection efforts – e.g., the Natural Scenes Dataset (NSD; Allen et al., VSS 2020) and BOLD5000 (Chang et al., 2019) – have measured high-resolution brain responses to tens of thousands of naturalistic visual stimuli. These efforts enable novel, data-hungry analyses that can more finely characterize visual representations and test theories of visual function. However, one challenge for such experiments is that fMRI measurements typically suffer from low signal-to-noise ratio and a high degree of signal overlap across trials. We introduce GLMsingle, a scalable, user-friendly tool for the accurate estimation of single-trial fMRI responses. Requiring only BOLD time-series data and a design matrix as inputs, GLMsingle integrates three techniques designed to improve the accuracy of trial-wise GLM beta estimates. First, for each voxel, a custom hemodynamic response function (HRF) is identified from a library of candidate functions. Second, the selected HRFs are used in a cross-validated GLM in order to derive a set of noise regressors from voxels unrelated to the experimental paradigm (“GLMdenoise”; Kay et al., 2013). Third, to mitigate the effects of signal overlap, beta estimates are regularized on a voxel-by-voxel basis using ridge regression ("Fracridge"; Kay & Rokem, 2020). Validation on both NSD and BOLD5000 reveals that GLMsingle substantially improves the signal-to-noise ratio of beta estimates across visually-responsive cortex in all participants. Furthermore, we find that GLMsingle meaningfully impacts three higher-level aspects of the data relevant for neuroscientific analyses: it improves the decorrelation of signal estimates between trials that are nearby in time, it enhances the representational similarity between participants within and across datasets, and it boosts one-versus-many decodability of visual stimuli that overlap between NSD and BOLD5000. Together, these results indicate that GLMsingle can help improve the quality of existing or future neuroimaging datasets that sample brain activity across many experimental conditions.

×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×