Purchase this article with an account.
Rosemary Cowell, David Huber, Garrison Cottrell; Virtual Multi-Unit Electrophysiology: Inferring neural response profiles from fMRI data. Journal of Vision 2010;10(7):931. doi: https://doi.org/10.1167/10.7.931.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
We present a method for determining the underlying neural code from population response profiles measured using fMRI. This technique uses orientation tuning functions for single voxels in human V1 (Serences et al., 2009; Kay and Gallant, 2008), which superficially resemble the electrophysiological tuning functions of V1 neurons. However, a voxel tuning function (VTF) is a summed population response, and does not specify the underlying neural responses. For example, the same bell-shaped VTF may arise from a population of neurons that are (1) tuned to a range of preferred stimulus orientations, with each preferred orientation present in varying proportions, or (2) uniformly distributed across preferred orientations, but neurons with a particular preferred orientation are tuned more sharply. The reported technique gains traction on this “inverse problem” by modeling the underlying neural responses across a range of tested orientations, and can be used to model task-induced changes in VTFs. We assume a set of underlying neural tuning curves, centered on orientations spaced evenly between 0° and 180°, and sharing a common standard deviation (SD). For a given SD, we use least-squares linear regression to solve for the 'coefficients' of the neural tuning curves (i.e., the relative weighting of each neural tuning curve present in the voxel) underlying the BOLD responses of an experimentally observed voxel. We find coefficients at a range of SD values, then determine the best-fitting SD to give the best estimate of the SD and coefficients. In future work, we will scan human subjects performing visual attention tasks, then use the present method to generate and test models of how the population response in visual cortex changes (e.g., Scolari and Serences, 2009). The technique effectively extracts “virtual” simultaneous multi-unit recordings from fMRI data – albeit with the usual fMRI limitations – and may help to address fundamental questions of neural plasticity.
This PDF is available to Subscribers Only