aguirreg

Review of: Distinguishing conjoint and independent neural tuning for stimulus features with FMRI adaptation.

Link to original F1000 review: http://www.f1000biology.com/article/id/1161286/evaluation

Tech Advance

The question of whether the brain encodes stimulus dimensions independently or conjointly is a central issue in cognitive neuroscience. The authors present a carefully constructed and innovative functional magnetic resonance imaging (fMRI)-adaptation procedure for studying this issue, and they validate their procedure using new fMRI data.

In the perceptual domain, numerous studies have been devoted to establishing whether particular kinds of objects (e.g. faces, houses, words) are perceived holistically or in terms of their parts, and explaining why this is the case {1}. Independent versus conjoint coding is also a crucial issue with regard to learning: computational models of learning posit that some tasks will elicit formation of conjunctive representations (e.g. “exclusive or” problems where subjects have to respond positively when one dimension or the other has a high value, but negatively when both dimensions have a high value) but other tasks will not {2}. The method presented here will make it possible to see, non-invasively, where conjunctive representations are formed, and it will make it possible to track how these conjunctive representations develop over the course of learning. The logic of the authors' procedure is relatively simple: if stimulus dimensions are represented independently, the effect of changing both dimensions on neural activity should be proportional to the change in dimension X plus the change in dimension Y (“city-block distance”). If neurons represent points in the X, Y space conjointly (i.e. their receptive fields are radially symmetric in the X, Y space) the effect of changing both dimensions should be proportional to the Euclidean distance between the old point and the new point. Other studies have taken a similar approach to studying neural coding of low-level visual features {3}. The key virtue of the approach taken here is that the authors present a very specific recipe (with regard to both the behavioral paradigm and the analysis) for efficiently computing whether coding is independent versus conjoint. Also, the authors are very sophisticated about potential confounds relating to nonlinearities in the BOLD fMRI response, and they propose post-hoc tests to address these confounds. Taken together, these factors make this a very practical and powerful technique for studying neural coding with fMRI.

References

1. Farah et al. Psychol Rev 1998, 105:482-98 PMID:9697428.
2. O'Reilly and Rudy, Psychol Rev 2001, 108:311-45 PMID:11381832.
3. Engel SA, Neuron 2005, 45:613-23 PMID:15721246.

Competing interests

None declared

Evaluated 19 Jun 2009