Abstract
It has been proposed that the cortical representations of visual stimuli are altered by experience. We tested this idea by recording fMRI data from 18 participants performing a Pavlovian fear conditioning paradigm. The paradigm consisted of three sessions. In the initial habituation session, two Gabor patches (45 and 135 degrees) were repeatedly presented in random order for a total of 120 trials. In the acquisition session, which followed the habituation session, the 45-degree Gabor patch (CS+) was paired with a loud scream (US) whereas the 135-degree Gabor patch (CS-) was never paired with the US. In the final extinction session, the CS+ and CS- were again repeatedly presented in random order (120 trials; no US). Applying the MVPA decoding method to fMRI data from the primary visual cortex in a sliding trial window fashion, we observed the following. First, the accuracy of decoding between the two Gabor patches was at chance level (50%) early in habituation, and became progressively higher as the participant experienced more repetitions of the two stimuli, reaching ~60% at the end of habituation. Second, for extinction, the decoding accuracy started at ~60% and became progressively lower, reaching chance level at the end of extinction. Third, MVPA classifiers trained on data from the end of habituation could not decode data from the beginning of extinction and vice versa. These results are consistent with the idea that the neural representations of visual stimuli undergo fundamental changes through learning: in habituation, repeated presentations without associative learning sharpened the sensory and perceptual distinctiveness of the two Gabor patches, whereas in extinction, the initially high representational distinctness of CS+ and CS-, reflecting the motivational quality of the CS+ and CS- (threat vs safety) acquired through associative learning, became diminished with extinction learning through potentially an active de-sharpening process.