GLMsingle

GLMsingle estimates single-trial beta responses from fMRI time-series to improve the reliability and interpretability of neural response measurements for systems and cognitive neuroscience analyses.

Key Features:

• Input requirements: Requires fMRI time-series data and a design matrix as primary inputs.
• Voxelwise HRF selection: Identifies an optimal Hemodynamic Response Function (HRF) for each voxel from a library of candidate HRFs to refine beta estimates.
• Noise regressors via cross-validation: Derives noise regressors from voxels unrelated to the experiment using cross-validation to reduce confounding noise.
• Ridge regression regularization: Applies voxel-wise ridge regression regularization to stabilize beta estimates for closely spaced trials.

Scientific Applications:

• Visual fMRI datasets: Improves reliability of beta estimates across visually-responsive cortex in datasets such as the Natural Scenes Dataset and BOLD5000.
• Auditory fMRI datasets: Enhances beta estimate reliability in auditory datasets such as StudyForrest.
• Higher-level neuroscience analyses: Boosts representational similarity within and across subjects and increases one-versus-many decoding performance for stimulus representations.

Methodology:

Given fMRI time-series and a design matrix, GLMsingle identifies voxelwise optimal HRFs from a candidate library, derives noise regressors from non-task voxels via cross-validation, and applies voxelwise ridge regression regularization to produce single-trial beta estimates.