Thus, we conducted the experiment [37] to test whether there is neural activation specific to consolidation during sleep after intensive training of TDT. After spending two adaptation nights to acclimate subjects to the experimental environment to alleviate the first night effect, which refers to the subjective difficulties of falling asleep in a novel environment [112–115], subjects (n = 7) slept inside the MRI scanner on the third night (the pre-training sleep session). On the evening of the fourth day, subjects were trained on the TDT for more than 1 h. During the training period, only one quadrant within 5° eccentricity of the visual field (counterbalanced across subjects) was trained. The subjects slept again on the fourth night, after the visual training (post-training sleep).

During both the pre- and post-training sleep periods, neural activation was measured for 90 min by fMRI; concurrently, PSG was acquired using a system compatible with the high magnetic field of the MRI scanner [116]. Sleep stages were scored according to the classical sleep staging criteria [39] and we confirmed that all subjects slept in the pre- and post-training sleep sessions more than 80 % of the time.

We hypothesized that the trained region of V1 would be the area where neural activation is modulated during sleep—if in fact sleep consolidates perceptual learning—for two reasons. First, the TDT is known to have a robust location-specific training effect, indicating that the neural changes associated with learning must have a robust and fine location representation, that is, in V1 [100]. Second, neural activation in the trained part of V1 is enhanced after sleep [108, 109], as shown previously [111]. We measured BOLD signal in V1 during pre-training and post-training sleep sessions and calculated relative BOLD changes during sleep compared to BOLD during the awake period after sleep, in both the trained and untrained regions in V1. The trained and untrained regions of V1 were determined based on subjects’ retinotopic representation, which was localized in advance in a separate fMRI session, using a standard retinotopic mapping technique [117–119].

During the post-training sleep period, the relative activity change in the trained V1 region was significantly higher than in the untrained region, which serves as a control region. Meanwhile, there was no significant difference in activation between the untrained and trained regions of V1 during in the pre-training sleep period. In addition, there was a tendency for increased SWS during the post-training sleep period as compared to the pre-training sleep period, although the tendency was not statistically significant. In the re-test session of TDT, which was conducted immediately after post-training sleep, subjects’ performance was significantly higher than during the initial training period. A significant correlation was found between performance and brain activity in the trained region of V1 (r = 0.810, p < 0.03). Since no performance improvement was observed in the control condition, where no sleep took place in the same time interval as in the main experiment, performance improvement can be attributed to the presence of sleep rather than to the mere passage of time between initial training and the re-test sessions. These results indicate that during sleep the learning-consolidation processing on TDT occurs in a retinotopically highly localized circuit specific to the trained location. The results strongly indicate that performance improvement after sleep did not result from reduced fatigue in general, since reduced fatigue would not produce modified cortical activation in the highly localized cortical region.