Fitting computational models to fMRI

Many computational models in psychology predict how neural activation in specific brain regions should change during certain cognitive tasks. The emergence of fMRI as a research tool provides an ideal vehicle totest these predictions. Before such tests are possible, however, significant methodological problems must be solved. These problems include transforming the neural activations predicted by the model into predicted BOLD responses, identifying the voxels within each region of interest against which to test the model, and comparing the observed and predicted BOLD responses in each of these regions. In the present article, methods are described for solving each of these problems.