Abstract
Perceptual decision making has been suggested to engage a large network of sensory and frontoparietal areas in the human brain. However, relatively little is known about the role of learning in shaping processing in these regions at different stages of decision making from sensory analysis to perceptual judgments. Here, we combine psychophysical and simultaneous EEG-fMRI measurements to investigate the spatiotemporal dynamics of learning to discriminate visual patterns. Observers were instructed to discriminate between radial and concentric Glass pattern stimuli that were either embedded in different noise levels (coarse discrimination) or varied in the spiral angle between radial and concentric patterns (fine discrimination). Our behavioral results showed that training enhanced the observers' sensitivity in the coarse task, while changed the internal decision criterion (i.e. categorical boundary) in the fine task. Information theory-based analyses of EEG single-trials revealed two temporal components that contained discriminative information between radial and concentric patterns: an early component (120 ms post-stimulus) associated with the analysis of visual stimuli and a later component (240 ms) related to the global pattern discrimination. Further, using multivariate pattern classification analysis we tested whether we could predict learning-dependent changes in the observers' choices from fMRI signals related to these EEG components. We observed learning-dependent changes in prefrontal circuits at the later EEG component for both tasks. In contrast, learning-dependent modulation in higher occipitotemporal areas (LO, KO/LOS) differed between tasks: for the coarse discrimination learning-dependent changes were associated with the first EEG component, while for the fine discrimination with the later component. These findings demonstrate that learning shapes the dynamics of neural processing in visual areas in a task-dependent manner. In particular, learning shapes sensitivity in early detection and integration processes for coarse discrimination tasks, while later decision criteria processes for fine categorical judgments.