Can neuroimaging really tell us what the human brain is doing? The relevance of indirect measures of population activity.

Neuroimaging studies using positron emission tomography (PET) and functional magnetic resonance imaging (fMRI) give an indication towards the localization of mental representations and processes in the human brain. It is not clear to what extent such global measures of neuronal activity, pooling across large populations of neurons, can reveal how certain computations are implemented by the neurons in such population ('computational neuroimaging'). Population activity is related tightly to single-cell activity when all neurons in the population have similar response properties. We describe some evidence from single-cell recordings in monkeys that indicates that neurons with similar response properties are not scattered randomly throughout the visual cortex. Notwithstanding this clustering, populations of nearby neurons are still rather heterogeneous, requiring some prudence in deriving single-cell response properties from population activity. The following review of recent neuroimaging studies of the visual system describes to what degree inferences about computations and representations can be drawn from these studies.