Adaptation method and internal organization of representations

While we await non-invasive methods that would enable us to see cortical columns or individual neurons in humans, an indirect strategy plays an important role in studying the internal organization of cerebral representations. This is the adaptation method, also known as the priming method (Grill-Spector & Malach, 2001; Naccache & Dehaene, 2001). It is based on the neurophysiological finding that most cortical neurons, whatever the coding level at which they are involved, show a decrease in response when the same stimulus is repeated several times. In fMRI, a similar adaptation phenomenon can be observed: the brain activation signal decreases as the same stimulus is presented several times, relative to a control situation in which different stimuli are presented on each trial.

The study of the fine-grained properties of adaptation enables two useful inferences in cognitive psychology (Naccache & Dehaene, 2001). If the signal is smaller in a given brain region when stimulus A is followed by A, than when B is followed by A, it is first that the brain region in question discriminates between stimuli and B, so (presumably) that it contains partially different populations of neurons coding for A and B. This inference may be valid, even though imaging does not have sufficient resolution to see these two populations directly. In this way, the adaptation method offers a form, albeit indirect, of hyper-resolution.

Secondly, adaptation makes it possible to study the metric of similarity between stimuli for a given region, and thus the level of information coding, by examining which variations in the stimulus do or do not count as repetition. In the ventral occipitotemporal cortex, for example, we can repeat the same image at different sizes, and observe unchanged adaptation (work by Kalanit Grill-Spector, Patrick Vuilleumier, etc.). Better still, in the left occipitotemporal region, the same adaptation occurs when a word is repeated identically (house-house) or in a different case (house-HOUSE) (Dehaene et al., 2001). We can then infer that this region codes for images or words in a way that is invariant to surface properties such as the exact size or shape of the object.

The lecture presented numerous illustrations of the potential of this adaptation method, borrowed in particular from the field of digital cognition. Using this method, it is indeed possible to show that the intraparietal cortex codes for numbers with a metric of semantic proximity: as numbers get closer together, adaptation increases, indicating neural coding by overlapping sets of neurons. The use of subliminal priming also improves the method insofar as it prevents attentional strategies, non-local to the brain region under study, from globally modifying the level of the fMRI signal.
However, the adaptation method must be used with caution if strong inferences are to be drawn about neural coding. Indeed, the method assumes that there is a close relationship between the neurons' tuning curve and their adaptation curve in response to stimuli more or less distant from the identical repetition. However, a recent study (Sawamura, Orban, & Vogels, 2006) directly tested this hypothesis using electrophysiology in awake monkeys, and showed that there may be conditions under which it is not valid.