Search tips
Search criteria 


Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
Neuron. Author manuscript; available in PMC 2009 August 31.
Published in final edited form as:
PMCID: PMC2735208

Different dynamics of performance and brain activation in the time course of perceptual learning


Perceptual learning (PL) is regarded as a manifestation of plasticity in the visual system. Yet, its underlying neural mechanism is unclear. Here we show different dynamics of performance and brain activation in V1 over the time course of PL development. Within initial few weeks of training of a visual task, both activation in the sub-region of the human V1 for the trained location and performance increased. However, after that the activation increase in V1 disappeared, while the performance enhancement was maintained. These results suggest an important dynamics of neural activity in the time course of PL. The first phase of training increases strength or number of synaptic connections in local network and enhances performance, and the second phase maintains the high performance with a smaller number of synapses that survived overall synaptic downscaling while decreasing activation. That is, there are at least two phases in the development of PL.


A central goal of neuroscience is to establish a link between behavior and the underlying neural mechanism. Perceptual learning (PL) is defined as performance or sensitivity increase in a sensory feature as a result of repetitive training or exposure to the feature and is regarded as manifestation of sensory plasticity (Fahle and Poggio, 2002; Gilbert et al., 2001; Karni and Sagi, 1991; Schoups et al., 1995; Watanabe et al., 2001). Thus, a large number of studies have been devoted to examining PL in the hope of clarifying the link between performance improvement of a perceptual task and the underlying plasticity (Ghose et al., 2002; Mukai et al., 2007; Schiltz et al., 1999; Schoups et al., 2001). So far most neuroscientific approaches to PL have focused on clarifying in which area(s) and how response properties of the area(s) change as a result of the formation of PL. However, it is not clear how sensory plasticity occurs and changes on a long-term basis.

The purpose of the present study was to examine how activation changes in the visual cortex over a long time course of PL for the first time. In the present study, we used the texture discrimination task (Censor et al., 2006; Karni and Sagi, 1991, 1993; Ofen et al., 2007; Schwartz et al., 2002; Walker et al., 2005), a standard visual PL task that is known to involve V1 (Schwartz et al., 2002; Walker et al., 2005). By measuring brain activation in a long time course of PL, we found that dynamics of performance and brain activation in V1 are different in the time course of PL. Within initial few weeks, activation in the sub-region of V1 that retinotopically corresponds to the trained location increased and performance enhancement was also observed. However, after the initial period, the activation enhancement disappeared while the performance increase was maintained. These results suggest that there are two stages in visual plasticity. Initially, the strength or number of synaptic connections increases and results in enhancement of both performance and BOLD signals. After performance saturation occurs, the high performance is maintained by a smaller number of synapses that survived overall synaptic downscaling (Censor et al., 2006; Tononi and Cirelli, 2003). The visual system reorganizes a local neural network to acquire and consolidate learning paying extra cost, and after the completion of these processes the network is optimized and high performance is given without further reorganization and consolidation with paying extra cost.


We used a texture discrimination task, a standard task in studies of visual perceptual learning (Censor et al., 2006; Karni and Sagi, 1991; Karni et al., 1994; Mednick et al., 2002; Ofen et al., 2007; Schwartz et al., 2002; Stickgold et al., 2000; Walker et al., 2005). As illustrated in Figure 1, the study involved six behavioral training sessions (black bars) and four fMRI sessions (gray bars) in Experiment 1. In each training trial, we presented a textured display along with a target array that was consistently displayed in the same visual field quadrant (i.e. the upper left visual field). Subjects (n=6) performed two types of tasks. One was to identify a letter presented at the central fixation point, whether the letter ‘T’ or the letter ‘L’. This task was used primarily to ensure subjects’ eye fixation. The other task was to report the orientation of a target array presented in a peripheral position for short duration. Performance in this second task was subject to learning. The subjects were asked to respond first to the letter task and then to the orientation task. There, a mask appeared at an interval after target array presentation. We refer to the interval as stimulus-to-mask-onset asynchrony, or SOA. We employed various SOAs in the training sessions in order to obtain a psychometric function for determining an 80% threshold SOA. Shortening of the threshold after repetitive performance was considered an indication that PL had taken place (Fig. 2A).

Figure 1
Experimental design. We conducted four fMRI sessions. Experiment 1 (n=6) involved six training sessions conducted until post 2 (each session is represented as a black bar). The average time intervals (± standard errors) between the initial training ...
Figure 2
Results. (A) The averaged threshold SOA (± standard errors) across all subjects in Experiment 1. (B) Mean performance (± standard errors) in the trained (circles) and untrained (squares) locations in Experiment 1. (C, D) Mean location-specific ...

Four separate measurements of each subject’s brain activation with 3 Tesla fMRI and assessment of task performance during the fMRI acquisition (for details, see Experimental procedures) were made before the start of training (pre-training), 10–25 hours (next day) after initial training (post-training 1), 10–14 days after initial training (post-training 2), and 4 weeks after initial training (post-training 3) (see Fig. 1). Each fMRI session involved two conditions for the location of stimulus presentation (for detail see Experimental procedures). The texture discrimination task is known to have location specificity (Karni and Sagi, 1991). To estimate a location specific training effect in V1, we presented the target arrays not only in the trained location (upper left visual field), but also in an untrained location (lower right visual field) only in the fMRI sessions. While the two location conditions were presented from trial to trial in a random order in an event-related fMRI paradigm, SOA was constant at 100 msec, as determined from our preliminary data.

Figure 2A shows the threshold SOAs observed in the behavioral training sessions. The threshold SOA reached asymptotes in 5–6 days, corresponding to the original literature (Karni and Sagi, 1991). Figure 2B indicates that while performance improvement is observed in the trained condition (ANOVA with repeated measurement, F(3,15)=17.28, p<0.001; post-hoc t-tests, post 2 vs pre-training, p<0.003; post 3 vs pre-training, p<0.03), no significant improvement was observed in the untrained condition. Figure 2C shows location-specific performance or fMRI activation that are defined as f(1, j)/f(1, 0) −f(0, j)/f(0, 0), where f(i,j) represents fMRI performance or BOLD signal in a location i (0=untrained location, 1=trained location) and in post j (post 0=pre-training phase, post 1, 2, 3=post-training phases), respectively (see Supplementary Figure 1 for BOLD signal changes).

Phase effects in both performance and fMRI activation were significant (ANOVA, p<0.01, p<0.05, respectively) in Experiment 1. The location-specific performance in posts 2 and 3 was significantly higher than in the pre-training stage (post-hoc t-tests, p<0.01, p<0.05, respectively).

To our surprise, however, location-specific fMRI activation in V1 (Fig. 2C), which was boosted at posts 1 and 2 (post-hoc t-test; p<0.01 and p<0.05 for pre- vs post 1-training and pre- and post 2- training, respectively), decreased to the baseline level (defined as the location-specific activation in V1 in the pre-training phase) at post 3 (p<0.70 for pre- and post 3- training). A significant quadratic trend (F(1,5)=8.9, p<0.05) was obtained. Thus, there are two distinctive patterns of dynamic relations between performance enhancement and neural activation changes in V1 in different stages of the time course of PL. For the initial few weeks after the onset of training, performance improvement and activation increase in V1 occurred. After that period, the V1 activation enhancement vanishes while the improved performance is maintained.

Note that a consistent SOA (100 msec) was used throughout fMRI phases to evaluate and compare brain activation and performance in the sessions. A relatively short or difficult SOA had to be used to avoid a ceiling effect in later phases such as post 2 and 3. The 100 msec SOA might have been so short that learning effects, if any, were not shown until thresholds became below 100 msec. Thus, although the correct response ratio was low in the trained location in the post 1 fMRI session (Fig. 2B), this does not necessarily indicate that learning did not occur from the pre-training to post 1 phases. Fig. 2A indicates that the 80% threshold SOA in the Day 2 training phase was significantly lower than in the Day 1 training phase but still higher than the 100 msec SOA. This indicates significant learning indeed occurred from the Day1 to Day2 training phases, as originally shown (Karni and Sagi, 1991). There was a high correlation (r=0.82) between the amount of SOA threshold improvement from the Day 1 to Day 2 training phases (Day1 –Day 2 thresholds) and the amount of changes in location-specific fMRI activation from the pre-training to post1 phases (see Supplementary Fig. 2 for correlations between threshold changes and MRI signal changes).

The V1 activation reduction in Experiment 1 occurred after training was terminated. One hypothesis for the reduction is that the enhancement in V1 activation in posts 1 and 2 is related to training and the reduction in post 3 is due to the termination of training. If this is the case, additional training until post 3 will produce enhanced V1 index at post 3. The counter-hypothesis is that the reduction in post 3 is not due to the termination of training. To test which hypothesis is correct, we conducted a control experiment, Experiment 2, in which a new group of subjects (n=5) received continued training between posts 2 and 3; all other conditions were identical to those of the Experiment 1. The subject participated in 14 behavioral training sessions over a span of four weeks (black and white bars in Fig. 1, see Supplementary Fig. 3 for the threshold SOA in the behavioral training sessions and fMRI sessions in Experiment 2). Fig. 2D indicates generally the same tendency as those in Experiment 1, allowing us to thus conclude that the V1 activation reduction is not due to termination of training. Note that the mean V1 and performance indices in post 1 in Experiment 2 were higher than in Experiment 1. This may be attributed to initially better performance with the subjects overall in Experiment 2 (See Supplementary Fig. 3A). The 80% thresholds in the initial training phase (Day 1) with control subjects were lower than with the subjects in Experiment 1. The initially higher individual performance in Experiment 2 than in Experiment 1 may have led the onsets of both performance saturation (post 1) and the V1 index drop (post 2) in Experiment 2 to occur more early than in Experiment 1. That is, the onsets of learning saturation and the V1 index drop may be influenced by individual differences to some degree.

We analyzed the reaction times in the trained and untrained locations in the fMRI experiments. Figure 3A shows results on reaction time to the texture orientation task in Experiment 1. The results of 2-way ANOVA (phase and location) with repeated measurement showed a significant phase effect (p<0.01) but did not show a significant effect in location. Post-hoc t-tests showed that there were significant differences between the pre-training and post 2 for both the trained location (p< 0.01) and the untrained location (p< 0.01), between the pre-training and post 3 for both the trained location (p<0.001) and the untrained location (p< 0.002). The same tendency was seen with the reaction times in Experiment 2, shown in Supplementary Fig. 4A.

Figure 3
Reaction time to the orientation task, activated region size, and correct response ratio for the fixational letter task in Experiment 1. (A) The reaction time to the orientation task was defined as the time interval from the onset of the target stimulus ...

Did activated size in V1 in Experiment 1 change over time? Figure 3B shows the activated region size in V1. The results of 2-way ANOVA with repeated measurement (phase and location) showed no significant difference in either factor. The activated size in V1 in Experiment 2 is shown in Supplementary Fig. 4B, indicating that there were no significant phase and location effects, either.

Did the subjects improve the fixation task? We analyzed the performance for the fixation letter task over time. Figure 3C shows the correct response ratio for the fixational letter task in Experiment 1. The results of 1-way ANOVA with repeated measurement showed no significant effects in the phase factor. The ratios were consistently high throughout the training, indicating that the subjects fixated very well during experiments. The correct response ratio for the fixational letter task in Experiment 2 (Supplementary Fig. 4C) showed the same tendency. Furthermore, we conducted Experiment 3 in which reaction times as well as accuracy to the letter task were measured with 4 subjects and “fMRI sessions” were conducted in a mock scanner so that only performance was obtained, with the otherwise same procedure as in Experiment 1. The results of one-way ANOVA with repeated measurement applied separately to the RTs and accuracy data in Experiment 3 did not show any significant effects of either RTs or accuracy (Supplementary Fig. 5). These results suggest that performance for the central task remained constant and that the performance benefit for the peripheral task was not due to a differential allocation of attentional resources throughout the different sessions.

Do other brain areas including the visual areas such as V2 and V3 and higher cognitive areas such as the intraparietal sulcus, superior parietal gyrus, and middle frontal gyrus show similar activation changes during the time course of learning like V1 where a novel dynamic activation during the time course of perceptual learning was found? We combined the data from Experiments 1 and 2, since there was no significant difference between the two experiments as a result of ANOVA. First, we applied ANOVA with repeated measurement (phase) to indices for each of these areas. No significant differences were found in any of the areas except for V1 (Fig. 4). The V1 indices at post 1 and post 2 were both significantly higher than the baseline (t-tests, p<0.001, p<0.005, respectively). The amount of difference between the V1 indices in posts 2 and 3 was significantly larger than that between posts 1 and 3 (t(10)=2.281, p<0.05). Second, we applied a trend analysis to each of the areas. V1 showed a significant quadratic trend (F(1,10)=12.06, p<0.001). In contract, none of the other areas showed any of linear, quadratic, or cubic trends. Third, correlation coefficients between V1 and each of V2, V3, the intraparietal sulcus, superior parietal gyrus and the middle frontal gyrus across the subjects were rather small, 0.33, 0.12, 0.07, −0.15 and 0.09, respectively. At the same time, the results of two-way ANOVA with repeated measurement (phase and area) to each combination of V1 and one of the five other areas indicated that except for area as a factor for V1 vs the middle-frontal gyrus (F(1, 10)=7.49, p<0.021) with no significant interaction between the phase and area factors, no other significant effect was found in any combination between V1 and one of the five areas. These results seem to be due to much larger standard errors across the subjects in the five areas than V1 (Fig. 4). Thus, we conclude to state that no clear result was obtained to discuss a tendency in the analyzed areas other than V1.

Figure 4
The mean activation indices (± standard errors) for fMRI responses in V1 (A), V2 and V3 (B), and IPS (the intraparietal sulcus), SPG (the superior parietal gyrus) and MFG (the middle frontal gyrus) (C), from the results from Experiment 1 combined ...

Can the fMRI activation drop in V1 from posts 2 and 3 be attributed to modified attention? There were no significant differences in either correct response ratios (Fig. 2B) or reaction times for orientation tasks (Fig. 3A) between posts 2 and 3 (therefore, no evidence for task-load change). Furthermore, as aforementioned, there was no significant fMRI activation change in the time course of learning in other analyzed areas than V1. Thus, there is no clear evidence that the activation drop in V1 from posts 2 to 3 was due to modified attention.


The present study shows different patterns of BOLD signals and performance changes in the long time course of perceptual learning. In the initial stage, BOLD signals and performance increase (although not necessarily at the same time, as discussed below). However, in the second stage that occurs after performance saturation, BOLD signals decrease to the level obtained before training.

What underlying mechanism is suggested by the results of the first stage? Note that recently it has been pointed out that the degree of BOLD activation is indicative of the degree of synaptic activity that may reflect the number of synapses and/or strength of synaptic connections (Logothetis et al., 2001; Viswanathan and Freeman, 2007). If this is the case and PL is the result of synaptic changes, our results would be in accord with the hypothesis that the initial stage of training produces increased synaptic connections or their strength and these synapses both help behavior and increase fMRI signal. Note that with the 100msec SOA neither performance improvement in the trained location (Fig. 2B) nor location specific performance improvement in post 1 (in the scanner) was observed (Fig. 2C) while trained location specific fMRI activation was enhanced in post 1 and a largest amount of performance improvement (SOA threshold decrease) was observed in Day 2 in the behavioral training (Fig. 2A). One possible explanation is that the synaptic increase/strengthening occurs gradually within initial weeks of training and that while the increase/strengthening was sufficient for SOA threshold to decrease by Day 1 and for trained specific fMRI activation to increase in post 1, it was not sufficient for performance with the short fixed 100msec to increase by post 2.

What underlying mechanism is suggested by the results of the second stage? The finding of the present study may be as well explained by synaptic downscaling (Censor et al., 2006; Tononi and Cirelli, 2003) that occurs after performance saturation. The reduction of BOLD activation to the level that was before the training without accompanying reduction of once increased performance in the second stage of our results may be explained by the synaptic downscaling. That is, after performance became saturated, the number and/or strength of overall synapses is reduced (downscaled)(Censor et al., 2006; Tononi and Cirelli, 2003). However, only the synapses that are most critical for the task survive the downscaling. If the degree of BOLD activation is indicative of the degree of synaptic activity (Logothetis et al., 2001; Viswanathan and Freeman, 2007), the BOLD activation reduction in our results is in accord with synaptic downscaling.

It should be noted that the activated region size did not expand in the trained V1 as learning proceeded. The absence of expansion in the activated region suggests that the learning and reorganization were localized. Interestingly, these results are contrasted to the result of a study of the motor skill learning (Karni et al., 1995) which indicated that initial fast reduction in the size of the fMRI activated region was then followed by expansion. These contrasting results may be related to difference between the modalities and tasks in the two studies. In the present study, a visual task was used to test plasticity in the visual cortex including V1 which has a highly retinotopic structure. Learning of the texture discrimination task used here is highly specific to the location and this suggests that the learning involves highly localized network (Karni and Sagi, 1991). Thus, learning and synaptic changes may occur within the localized network.

There is a possibility that not all aspects of behavioral improvement between post 1 and post 2 fMRI phases is due to learning. In the present study, a texture discrimination task with relatively long sessions of 1520 trials was conducted. Recent results with this task show that the amount of practice within a session can strongly affect performance, in the way that a larger number of trials cause less improvements possibly due to suppressive processes related to adaptation in the visual system (Censor et al., 2006; Ofen et al., 2007). Thus, one might suggest that since the SOA was gradually changed from high to low within a session in the present study, a lower staring point may produce lower thresholds as well as that the too much adaptation causes less performance in the next training session. The starting SOAs in training sessions (Days 2 to 6) that were conducted between posts 1 and 2 fMRI phases were indeed shortened and, therefore, the above mentioned possibility can not be entirely ruled out. At the same time, there was one or two days interval between one training session and the next, and it has been reported that suppressive processes may be largely eliminated during sleep after each training session (Censor et al., 2006). Variable suppression may affect the behavioral results but not the fMRI sessions in our study because we used a constant number of trials and a constant SOA in the fMRI sessions. Thus, the performance improvement between post 1 and 2 fMRI phases may be largely due to learning, if not all.

One might point out that the size of the learning effect as shown in Fig. 2 appeared to be smaller than those reported by Schwartz et al. (2002) that showed smaller behavioral effects. This difference is likely to be attributed to differences between methods used in the Schwartz study and ours. For example, while stimuli in Schwartz et al. (2002) were presented monocularly, in the present study stimuli were presented binocularly. While the SOA used in Schwartz et al (2002) was 200 msec, the SOA in fMRI sessions of the present study was 100 ms. In addition, methods to analyze data were different in various ways. For example, while the standardized brain anatomy such as Talairach space was used in Schwartz et al. (2002), individual brain anatomy and localization of retinotopy were used in the present study for calculating activation in early visual cortical areas. Thus, it is difficult to simply compare between the sizes of the learning effects between the two studies.

Fig. 2 (C, D) shows location-specific indexes that were based on the subtraction of BOLD signals in the untrained location from those in the trained location. However, this leaves the possibility that increase in the index from the pre to post 1 phases in Fig. 2 (C, D) was due to decreased activity in the untrained location rather than increased activity in the trained location. To test whether this is likely the case or not, we normalized fMRI signals in each the untrained and trained quadrant by the response in one of the other quadrants, which would serve as a neutral measure of overall fMRI response. The results (see Supplementary Fig. 6) are in accord with the hypothesis that the increase in the location-specific fMRI index from the pre to post 1 phases (Fig. 2 (C, D)) is due to increase in fMRI response in the trained location rather than decrease in the untrained location.

For summary, in the present study we measured BOLD activation and performance in a long time course of PL for the first time and have found that the shapes of the dynamics of BOLD activation and performance differ. While during initial few weeks of training, trained location specific activation in V1 and performance increased. However, after the performance increase saturated, while the performance enhancement was maintained, the activation increase disappeared. These results are in accord with the hypothesis that the increased strength or number of synaptic connections in local network during the initial period. After performance saturation, the high performance is maintained by smaller number of synapses that survived overall synaptic downscaling.

Experimental procedures


A total of 15 subjects (8 females) with normal or corrected-to-normal vision were employed. Eleven subjects (age range: 22–39 yr, 5 females) participated in both behavioral training and fMRI sessions of Experiment 1 (n=6) and Experiment 2 (n=5). Additional 4 subjects (age range: 22–28 yr, 3 females) participated Experiment 3. All subjects gave written informed consent for their participation in the experimental protocol approved by the Institutional Review Board at Massachusetts General Hospital and Boston University.

Behavioral training session

With their chin and forehead fixed, each subject viewed visual displays on a screen positioned 57 cm from their eyes. All behavioral experiments were conducted in a dimly lit room. We employed a texture discrimination task (TDT), which has been widely used in visual perceptual learning studies. In each TDT trial, a test stimulus was briefly presented (13 msec) and followed by a blank screen (presentation period varied by trial), and a mask stimulus composed of randomly oriented V-shaped patterns (100 msec) at an interval after target array presentation (stimulus-to-mask-onset asynchrony, SOA). The test stimulus consisted of a centrally located letter, either a ‘T’ or a ‘L’, and a peripherally positioned horizontal or vertical array of three diagonal bars (target arrays) on a background of horizontal bars. While keeping their eyes fixated on the center of the visual field, subjects were asked to respond twice to each trial: once to identify the letter and once to indicate horizontal or vertical orientation of the target array. The fixational letter task was intended to ensure subjects’ focus and fixation at the center of the visual field; the target array discrimination task was used as a measure of perceptual learning. Each line segment was arranged within a 19-by-19 lattice in the area of a 14°-by-14° visual angle. Lines were 0.43° by 0.07°, and spaced 0.7° apart. The position of each line segment was jittered slightly, by 0–0.05°, from trial to trial. The position of the target array also varied randomly from trial to trial, but was consistently presented within a specific quadrant, and within a 2.95° – 5.15° visual angle from the center of the display. All line segments were gray (32 cd/m2), and presented on a black (0.5 cd/m2) background. Immediate auditory feedback was given only for the fixational letter task to facilitate subjects’ fixation. No feedback was given to the orientation task in this experiment to follow the original procedure of the task (Karni and Sagi, 1991) and because leaning occurs without feedback in TDT (Karni and Sagi, 1991; Sagi and Tanne, 1994).

During training sessions, the horizontal or vertical target array was presented only in the upper left visual field quadrant. Each training session contained 1520 trials, presented in 40 trial blocks (each block of 38 trials) with a constant SOA. Each training session started with a longer SOA, for instance 250 msec, which was decremented by 20–40 msec every 2–4 blocks. As training proceeded, the SOA values in each session were shortened, thus increasing the difficulty of the task. An initial SOA was determined daily and individually based on earlier performances. The percentage of correct responses was calculated for each SOA in order to construct a psychometric function for determining the threshold SOA, at which subjects reach 80% correct responses by interpolation.

Subjects took part in training sessions once every 2 or 3 days. Experiments 1 and 3 involved six training sessions. Experiments 2 included an additional eight sessions, or total 14 sessions.

FMRI experiments

Scanning sessions were conducted on four separate occasions, at each of the pre, post 1, post 2, and post 3 training phases as shown in Fig. 1. Stimuli were generated on a Mac G4 and presented via LCD projector (Sharp Note Vision 6). The fMRI experiments involved two location conditions for presentation of the target arrays, which were displayed in a random order in either the upper left visual field (trained condition) or lower right visual field (untrained condition) using an event-related fMRI paradigm. In the event-related paradigm, the timing of presenting each condition was calculated with optseq2 software (Dale, 1999; Dale et al., 1999b) so that an inter-stimulus interval was randomized from trial to trial to maximize the statistical efficiency.

We presented 128 trials for each of the two conditions during each fMRI session; a single trial lasted 2 sec. At the beginning of each trial, a blue or green fixation cross was presented for 500 msec, followed by a blank screen for 250 msec. The color of the fixation cross served as a cue for the location of a target array to follow. A blue fixation cross indicated that the target array would appear in the upper left (trained) quadrant, a green cross, that the array would appear in the lower right quadrant. A target texture was then presented approximately for 20 msec (temporal resolution limit of the display), followed by a mask for 100 msec; the SOA was a constant 100 msec. As in the behavioral training session, subjects were asked to respond to the fixational and orientation tasks by pressing a button on a box held by the subjects. Immediate auditory feedback was given only for the fixational letter task to facilitate subjects’ fixation.

Image acquisition

Subjects were scanned in a 3T MR scanner (Allegra or Trio, Siemens); a head coil was used throughout the experiments. Functional MR images were acquired using gradient echo EPI sequences (TR = 2 sec, TE = 30 ms, Flip Angle = 90°) for measurement of BOLD contrast. Thirty-five contiguous slices (3 × 3 × 3.5 mm3) oriented parallel to the AC-PC plane were acquired to cover the entire brain.

All functional data were registered to the individual anatomically reconstructed brain (Dale et al., 1999a; Fischl et al., 1999). For the anatomical reconstruction, three T1-weighted MR images (MPRAGE) were acquired (TR = 2.531 sec, TE = 3.28 msec, flip angle = 7°, TI = 1100 msec, 256 slices, voxel size = 1.3 × 1.3 × 1.0 mm3, resliced during analysis to 1 mm3). This same anatomical reconstruction was used for the brain parcellation method to localize individual gyri and sulci (Fischl et al., 2004).

Definition of region of interest (ROI)

Four retinotopic quadrants (upper left, upper right, lower right, and lower left of the visual field) of V1, V2, and V3/VP areas were localized individually in a separate fMRI session that used a standard flickering checkerboard pattern (Engel et al., 1994; Fize et al., 2003). In addition, eccentricity was localized individually using annulus stimuli of various sizes. In the subsequent analysis, regions of 3–5 degrees of eccentricity in V1, V2, and V3/VP were used. Since the retinotopy of V4 is controversial (Wandell et al., 2005), V4 was not subject to analysis.

The middle frontal gyrus, superior parietal gyrus, and interparietal sulcus were identified individually using the brain parcellation method (Fischl et al., 2004).

fMRI data analysis

Data were analyzed with FS-FAST and FreeSurfer ( software. All functional images were motion corrected (Cox and Jesmanowicz, 1999), spatially smoothed with a Gaussian kernel of 5.0 mm (FWHM), and normalized individually across scans. In this normalization process, the mean intensity for the entire functional volume was computed for each scan. The global mean of the entire brain was rescaled so that the same mean would be set across scans. A finite impulse response model (Burock and Dale, 2000) was employed to estimate hemodynamic response (timecourse) to each condition (trained or untrained locations) in each ROI at 20 1-sec interval time points. The timecourses for each condition for each ROI were then converted into the % signal changes, by subtraction and division of the mean value of the ROI. Note that the mean value of each ROI, which is a part of the entire functional volume, was not assured to be the same across different fMRI sessions.

To compute a location-specific response index in V1, we first normalized the peak hemodynamic response (4–6 sec from stimulus onset) at each phase by dividing it by the peak response in the pre-training phase. The normalized value of the upper left V1, which corresponded to the untrained location, for the untrained condition was then subtracted from the normalized value of the lower right V1, which corresponded to the trained location for the trained condition at each phase, respectively. Response indices for V2 and V3 were computed basically in the same way as for V1. Response indices for the middle frontal gyrus, superior parietal gyrus, and interparietal sulcus were computed as follows. The normalized value of the left hemisphere of each region in the untrained condition was subtracted from the normalized value of right hemisphere of each region during the trained condition.

Supplementary Material



This work was supported by the National Institutes of Health (R01EY015980, the National Science Foundation (BCS-0345746, BCS-0549036), the Human Frontier Foundation (RGP 18/2004), the NIH National Center for Research Resources (P41RR14075), the Mental Illness and Neuroscience Discovery (MIND) Institute, the Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, and the ERATO Shimojo Implicit Brain Function Project. The authors thank Nichole Eusemann for her editing of the manuscript.


Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.


  • Burock MA, Dale AM. Estimation and detection of event-related fMRI signals with temporally correlated noise: a statistically efficient and unbiased approach. Hum Brain Mapp. 2000;11:249–260. [PubMed]
  • Censor N, Karni A, Sagi D. A link between perceptual learning, adaptation and sleep. Vision Res. 2006;46:4071–4074. [PubMed]
  • Cox RW, Jesmanowicz A. Real-time 3D image registration for functional MRI. Magn Reson Med. 1999;42:1014–1018. [PubMed]
  • Dale AM. Optimal experimental design for event-related fMRI. Hum Brain Mapp. 1999;8:109–114. [PubMed]
  • Dale AM, Fischl B, Sereno MI. Cortical surface-based analysis. I. Segmentation and surface reconstruction. Neuroimage. 1999a;9:179–194. [PubMed]
  • Dale AM, Greve DN, Burock MA. Optimal stimulus sequences for event-realted fMRI; 5th International Conference on Functional Mapping of the Human Brain; Germany. Duesseldorf; 1999b.
  • Engel SA, Rumelhart DE, Wandell BA, Lee AT, Glover GH, Chichilnisky EJ, Shadlen MN. fMRI of human visual cortex. Nature. 1994;369:525. [PubMed]
  • Fahle M, Poggio T. Perceptual learning. Cambridge, MA: The MIT Press; 2002.
  • Fischl B, Sereno MI, Dale AM. Cortical surface-based analysis. II: Inflation, flattening, and a surface-based coordinate system. Neuroimage. 1999;9:195–207. [PubMed]
  • Fischl B, van der Kouwe A, Destrieux C, Halgren E, Segonne F, Salat DH, Busa E, Seidman LJ, Goldstein J, Kennedy D, et al. Automatically parcellating the human cerebral cortex. Cereb Cortex. 2004;14:11–22. [PubMed]
  • Fize D, Vanduffel W, Nelissen K, Denys K, Chef d'Hotel C, Faugeras O, Orban GA. The retinotopic organization of primate dorsal V4 and surrounding areas: A functional magnetic resonance imaging study in awake monkeys. J Neurosci. 2003;23:7395–7406. [PubMed]
  • Ghose GM, Yang T, Maunsell JH. Physiological correlates of perceptual learning in monkey V1 and V2. J Neurophysiol. 2002;87:1867–1888. [PubMed]
  • Gilbert CD, Sigman M, Crist RE. The neural basis of perceptual learning. Neuron. 2001;31:681–697. [PubMed]
  • Karni A, Meyer G, Jezzard P, Adams MM, Turner R, Ungerleider LG. Functional MRI evidence for adult motor cortex plasticity during motor skill learning. Nature. 1995;377:155–158. [PubMed]
  • Karni A, Sagi D. Where practice makes perfect in texture discrimination: evidence for primary visual cortex plasticity. Proc Natl Acad Sci U S A. 1991;88:4966–4970. [PubMed]
  • Karni A, Sagi D. The time course of learning a visual skill. Nature. 1993;365:250–252. [PubMed]
  • Karni A, Tanne D, Rubenstein BS, Askenasy JJ, Sagi D. Dependence on REM sleep of overnight improvement of a perceptual skill. Science. 1994;265:679–682. [PubMed]
  • Logothetis NK, Pauls J, Augath M, Trinath T, Oeltermann A. Neurophysiological investigation of the basis of the fMRI signal. Nature. 2001;412:150–157. [PubMed]
  • Mednick SC, Nakayama K, Cantero JL, Atienza M, Levin AA, Pathak N, Stickgold R. The restorative effect of naps on perceptual deterioration. Nat Neurosci. 2002;5:677–681. [PubMed]
  • Mukai I, Kim D, Fukunaga M, Japee S, Marrett S, Ungerleider LG. Activations in visual and attention-related areas predict and correlate with the degree of perceptual learning. J Neurosci. 2007;27:11401–11411. [PubMed]
  • Ofen N, Moran A, Sagi D. Effects of trial repetition in texture discrimination. Vision Res. 2007;47:1094–1102. [PubMed]
  • Sagi D, Tanne D. Perceptual learning: learning to see. Curr Opin Neurobiol. 1994;4:195–199. [PubMed]
  • Schiltz C, Bodart JM, Dubois S, Dejardin S, Michel C, Roucoux A, Crommelinck M, Orban GA. Neuronal mechanisms of perceptual learning: Changes in human brain activity with training in orientation discrimination. Neuroimage. 1999;9:46–62. [PubMed]
  • Schoups A, Vogels R, Qian N, Orban G. Practising orientation identification improves orientation coding in V1 neurons. Nature. 2001;412:549–553. [PubMed]
  • Schoups AA, Vogels R, Orban GA. Human perceptual learning in identifying the oblique orientation: retinotopy, orientation specificity and monocularity. J Physiol. 1995;483(Pt 3):797–810. [PubMed]
  • Schwartz S, Maquet P, Frith C. Neural correlates of perceptual learning: a functional MRI study of visual texture discrimination. Proc Natl Acad Sci U S A. 2002;99:17137–17142. [PubMed]
  • Stickgold R, Whidbee D, Schirmer B, Patel V, Hobson JA. Visual discrimination task improvement: A multi-step process occurring during sleep. J Cogn Neurosci. 2000;12:246–254. [PubMed]
  • Tononi G, Cirelli C. Sleep and synaptic homeostasis: a hypothesis. Brain Res Bull. 2003;62:143–150. [PubMed]
  • Viswanathan A, Freeman RD. Neurometabolic coupling in cerebral cortex reflects synaptic more than spiking activity. Nat Neurosci. 2007;10:1308–1312. [PubMed]
  • Walker MP, Stickgold R, Jolesz FA, Yoo SS. The functional anatomy of sleep-dependent visual skill learning. Cereb Cortex. 2005;15:1666–1675. [PubMed]
  • Wandell BA, Brewer AA, Dougherty RF. Visual field map clusters in human cortex. Philos Trans R Soc Lond B Biol Sci. 2005;360:693–707. [PMC free article] [PubMed]
  • Watanabe T, Nanez JE, Sasaki Y. Perceptual learning without perception. Nature. 2001;413:844–848. [PubMed]