Abstract
People can learn arbitrary visual feature conjunctions when the features to be conjoined are present at the same time. This learning is associated with increasing activity in visual cortex (Frank et al., 2013, Human Brain Mapping). Can people also integrate temporally non-coincident features into a coherent conjunction that spans space and time? If so, what brain areas are involved in spatio-temporal conjunction learning? We addressed these questions by training participants to search for a single dot moving down, then up, amongst distractor dots moving up, then down presented in an eccentric radial array. Twenty participants performed the visual search task during an MRI scan, then completed ten days of practice on the task, after which they performed the task in the scanner again. With training, target detectability increased, as measured by d'. At the same time reaction time decreased dramatically. The difference in BOLD signal between target present vs. target absent trials increased in higher-order visual motion areas MST and MT, but not in retinotopic cortex. This activation difference in MST and MT was significantly correlated with learning rate: participants who learned quickly had a larger signal difference after training than those who learned slowly. In retinotopic visual cortex, target locations showed more BOLD activity than distractor locations after learning, but this increase in BOLD signal-to-noise was not correlated with learning rate.
Meeting abstract presented at VSS 2014