Categorization training leads to sharpening tuning of shape-specific tuning in the lateral occipital cortex and learning of category-selective representations in the prefrontal cortex

Abstract

Categorization is a fundamental human cognitive function. However, little is known about the neural mechanism underlying category learning. A recent computational model (Riesenhuber and Poggio, Nat Neurosci, 2000) proposed a two-layer framework, in which categorization is performed by category-selective circuits putatively located in the prefrontal cortex that receives input from shape-selective neurons in visual cortex tuned to representatives of the relevant object classes. To investigate the neural bases of human category learning, we trained subjects (n=15) for a mean of 5 hrs on a delayed match to category task using morphed images ("cars"), and assessed the learning-related shape- and category-selective cortical representations using an fMRI rapid-adaptation (RA) paradigm, in which pairs of cars were presented. When subjects performed an apparent motion task on the cars - for which category membership was irrelevant - we found reduced activations across the visual cortex during post-training scans relative to pre-training scans, and more importantly, a sharpening of the tuning to car shapes in the lateral occipital cortex (LOC). No explicit category-selectivity was found in LOC. However, category-selectivity was found at the inferior and middle gyri of dorsolateral prefrontal cortex while subjects (n=12) performed a categorization task. Interestingly, activity in LOC appeared to be modulated by subjects' judgment of category membership during the categorization task, putatively due to top-down influences from prefrontal cortex, paralleling recent findings from a monkey study (Freedman, Riesenhuber, Poggio, Miller, J Neuro 2003). These preliminary results appear to confirm the simple two-stage model.

Footnotes

 Support Contributed By: NIMH