Project Summary: The goal of this proposal is to advance our ability to accurately infer the properties of neu- ral-level responses from the more coarse-grained information obtained with non-invasive imaging in humans. To achieve this goal, the project will capitalize on feature-selective cortical responses. For example, many neu- rons in visual cortex exhibit a tuning function such as a response profile in which firing rate is greatest for one orientation of a line, and falls off for orientations progressively less similar to that orientation. Promising new methods for analyzing functional Magnetic Resonance Imaging (fMRI) data reveal analogous feature-tuning in the blood oxygenation level-dependent (BOLD) signal. Because these voxel-level tuning functions (VTFs) are superficially analogous to the neural tuning functions (NTFs) observed with electrophysiology, it is tempting to interpret VTFs as mirroring the characteristics of the underlying NTFs that contribute to them. However, this interpretation is not justified because there are multiple alternative accounts by which changes in the NTFs could produce a given change in the VTF. To distinguish between these accounts, we need a means of map- ping VTFs back to NTFs. That is, for fMRI to provide insights into neural-level mechanisms, the inverse prob- lem of mapping voxel-level fMRI signals back to neural-level responses must be solved. The proposed work will tackle this inverse problem by considering a plausible set of neural-level changes that may give rise to an observed change in voxel-level fMRI responses, and determining which model of neural-level change is most likely using either model recovery or hierarchical Bayesian estimation, followed by model selection. The goal will be accomplished via three specific aims: (1) Determine the conditions that allow us to distinguish between alternative models of neural-level modulation for a simple modulation of orientation-selective VTFs (stimulus contrast); (2) Identify the neural-level mechanisms underlying modulations of orientation-selective VTFs induced by other manipulations of perceptual or cognitive state; and (3) Identify the neural-level mecha- nisms underlying modulation of two further classes of VTF. The approach for all three aims entails: i) collecting optimal fMRI data, ii) applying alternative models of neural-level modulation to the fMRI data to account for voxel-level modulations, iii) performing model selection based upon model recovery or hierarchical Bayesian estimation, iv) comparing the outcome of model selection with ‘ground truth’ from electrophysiology. The pro- ject will thereby develop an experimental and model selection procedure for revealing the neural-level mecha- nisms that underlie modulations in feature-selective voxel responses observed with fMRI. Moreover, it will ena- ble the comparison of data from animal studies investigating fine-grained neural mechanisms with data from non-invasive imaging in humans, for a range of perceptual and cognitive phenomena.

Project Narrative The proposed research is relevant to the BRAIN initiative because it seeks to bridge across different scales of neurobiological mechanism. Functional Magnetic Resonance Imaging (fMRI) is a non-invasive experimental technique which measures brain activation with a spatial resolution of around 2mm, such that the signal recorded with fMRI averages over the responses of many individual neurons in the brain. By developing a procedure to understand the way in which neural responses are mapped onto the coarse- grained signals recorded with fMRI, this research will advance our ability to accurately and precisely infer the properties of neural-level responses from data obtained via non-invasive fMRI in humans.