Functional imaging and neural information coding.

Measuring functional magnetic resonance imaging (fMRI) responses to parametric stimulus variations in imaging experiments can elucidate how sensory information is represented in the brain. However, a potential limitation of this approach is that fMRI responses reflect only a regional average of neuronal activity. For this reason stimulus-induced changes in fMRI signal may not always reflect how sensory information is encoded by neuronal population activity. We investigate the potential problems induced by the finite spatial resolution of the fMRI signal by combining the principles of Information Theory with a computational model of neuronal activity based on known tuning properties of sensory cortex and assuming a linear spike rate to fMRI signal relationship. We found that the relationship between neuronal information and fMRI signal is highly nonlinear. It follows that the brain voxel experiencing the largest fMRI signal change is not necessarily the voxel encoding the most sensory information. Results also show that functional imaging data can be better interpreted in terms of neural information processing if the fMRI data and some knowledge about the tuning properties of the underlying neuronal populations are incorporated into a computational model. We discuss how imaging techniques themselves may provide an estimation of neuronal tuning properties.