Dissociable effects of transcranial direct current stimulation (tDCS) on early and later stages of visual motion perceptual learning

perceptual


Introduction
Visual perceptual learning (VPL) refers to the phenomenon that perceptual experience induces long-term enhancement of human visual perceptual abilities (Dosher and Lu, 2017;Lu et al., 2016). Over the past decades, an increasing number of studies have translated VPL techniques from the laboratory to real-world applications (Lu et al., 2016). VPL has been found to enhance or rehabilitate patients with various types of vision loss, such as amblyopia (Astle et al., 2011;Huang et al., 2008), visual field defects (Casco et al., 2018), myopia (Casco et al., 2014;Tan and Fong, 2008;Yan et al., 2015), and presbyopia (Sterkin et al., 2018). However, adequate performance enhancement usually requires hundreds of thousands of trials over a long period of training (Herpich et al., 2019), which limits practical applications of VPL.
Fortunately, transcranial direct current stimulation (tDCS) has been found to boost VPL when coupled with behavioral training (Karlaftis et al., 2021;Pirulli et al., 2013). However, previous studies have investigated the beneficial effect of tDCS at early limited training sessions, and little is known about the tDCS effects at the later stage of the whole training series.
As a kind of noninvasive transcranial electrical stimulation, tDCS is particularly attractive due to its low cost and portability (Reinhart et al., 2016). It transiently modulates cortical excitability by altering the membrane potential of neurons Stagg and Nitsche, 2011). In a seminal study, anodal tDCS over the human middle temporal complex (hMT+) or primary motor cortex (M1) significantly increased the early learning phase, whereas cathodal stimulation had no significant effect . Additionally, application of anodal tDCS over the primary visual cortex (V1) improved visual orientation-discrimination learning and increased cortical excitability (Sczesny-Kaiser et al., 2016). Recently, Karlaftis et al. (2021) applied anodal tDCS or sham tDCS over the right occipito-temporal cortex during training on a signal-in-noise task and found that anodal tDCS boosted learning by decreasing GABA+ and altering local processing in the visual cortex. For patients with unilateral visual field loss, occipital cortical tDCS combined with visual field rehabilitation enhanced visual functional outcomes compared with visual rehabilitation alone (Plow et al., 2011;Plow et al., 2012). In the above studies, stimulation was administered while participants underwent training. Additionally, some studies delivered tDCS to participants immediately after training (early consolidation) was completed and still found that tDCS facilitated VPL (Yang et al., 2022).
However, some tDCS studies on visual learning revealed diverging effects. Some studies could not show tDCS effects on VPL when participants were trained on visual perception tasks (Fertonani et al., 2011;Herpich et al., 2019;Larcombe et al., 2018). Additionally, overnight consolidation of visual learning was even blocked by anodal tDCS applied while participants were trained on the contrast detection task (Peters et al., 2013). It is difficult to explain the causes of these inconsistent results since many factors, such as learning tasks, time sequence and stimulation parameters, were different among these previous studies. It is necessary to accumulate more empirical research evidence in this area. Thus, the first objective of this study was to further investigate whether tDCS is able to boost VPL.
Previous studies have provided a basis for understanding how tDCS benefits VPL. However, these studies focused on tDCS effects within predetermined and limited sessions, i.e., the early stage of training sessions. One unanswered question is whether tDCS can improve the learning effects at later time points of the entire training period. The later stage in this study is defined by the measurements when performance has reached a plateau. The training sessions in the early studies were limited (generally fewer than 5 training sessions), and it is unclear whether the participants' performance had reached a plateau. Therefore, there is a possibility that tDCS only increases the learning rate within a limited time, which may have resulted in VPL being facilitated by tDCS in previous studies. After extending the training time to reach a plateau, the observed tDCS facilitation may disappear. Answering this question is of practical significance. Specifically, we suggest that the later learning effects may be more meaningful and important than the increase in learning speed in a limited time. This is because learning at a slow speed also achieves the expected training effects as long as the amount of training (or training time) is increased to a sufficient level. It is possible to extend the training time in practical application situations. For example, extra days to weeks of training may be easily accepted by clinical patients with vision loss. In contrast, the later learning effects directly determine the rehabilitation level of the patients' visual functions.
It has been found that VPL is associated with changes in mechanisms in multiple stages. For example, in a certain phase of training, neural correlates of improved performance are found in the visual areas, whereas in a different phase, neural changes are observed in cognitive areas or in the connectivity between the visual and cognitive areas (Shibata et al., 2014;Watanabe and Sasaki, 2015). Thus, the brain region that plays a major role may dynamically change during training. It is reasonable to speculate that the tDCS effect is not constant from the early stage to the later stage. Thus, the second aim of this study was to explore the tDCS effect at the later stage of training.
In summary, the current study aimed to explore the facilitated effects of anodal tDCS on visual motion perceptual learning at distinct temporal stages. To investigate the tDCS effect at the later stage of training, we proposed two hypotheses. First, participants were administered tDCS during the entire training sessions while they were trained until a plateau was reached. If anodal tDCS improved participants' performance on tests or their plateau levels of the learning curve, the notion that tDCS benefits the later learning effects would be supported (Hypothesis 1). Second, participants were trained until they reached a plateau. After that, they continued to train while tDCS was administered. If tDCS can further increase participants' performance or plateau levels after continuous training, the notion would also be supported (Hypothesis 2).

Participants
This study recruited 38 right-handed participants (mean age: 20.56 years; age range: 19-24 years; 14 females) with normal neurology and normal or corrected-to-normal vision. The number of participants per group was based on previous studies that combined VPL and electrical stimulation (Larcombe et al., 2018;Sczesny-Kaiser et al., 2016). All participants had not previously participated in any visual perception experiments and were blinded to the aim of the study. Participants gave written informed consent before experiment initiation. The study was approved by the research ethics committee and was implemented according to the ethical principles of the Declaration of Helsinki.

Study design
The whole experimental procedure consisted of two stages. The first stage included the pretest, the 9-day consecutive training period and posttest 1 (48 h after training). After the first stage, the second stage consisted of 3 days of consecutive training and posttest 2 (48 h after training). No stimulation was delivered when the tests were conducted. There are two important points to note. First, posttest 1 and posttest 2 were carried out separately 48 h after training in each stage because previous studies frequently used time intervals of at least 48 h to clarify the potential carryover effects of tDCS (Wu et al., 2021). Tests without tDCS contributed to the determination of the pure improvement in VPL since tDCS itself could directly benefit visual motion perception (Wu et al., 2020). Second, our pilot study found that the averaged coherent performance appeared to reach a plateau after 8-9 training sessions. At this time, the amount of coherent threshold reduction in most participants (8 out of 10 participants) was less than 1 %. The averaged coherent thresholds were 24.84 ± 2.34 % in session 8 and 24.12 ± 1.91 % in session 9. Thus, participants were required to complete 9-day training sessions within the first stage to ensure that their averaged performance had reached a plateau.
Participants were randomly assigned to three experimental groups ( Fig. 1). No significant differences among the three groups were found regarding age, vision and sex, ps > 0.100. In the first group (n = 14), defined as the anodal tDCS + anodal tDCS group (AA), anodal tDCS was applied over the left hMT+ while participants completed 12 days of training (stage 1 + stage 2). In the second group (n = 12), defined as the no stimulation + anodal tDCS group (NA), participants trained over 9day consecutive sessions without application of anodal tDCS (stage 1) and then were stimulated by anodal tDCS during the 3-day training session (stage 2). In the third group (n = 12), defined as the no stimulation + sham tDCS group (NS), no stimulation was administered at stage 1, and sham tDCS was delivered at stage 2.

Apparatus
In a quiet, dark room, participants sat in front of a computer monitor to complete the entire experimental procedure. The experimental environment and apparatus were kept constant in all sessions. The visual stimuli were displayed on a gamma-corrected 60 × 34 cm monitor (spatial resolution: 1920 × 1080 pixels; refresh rate: 85 Hz) by a computer running MATLAB and PsychToolbox extensions. Participants binocularly viewed the displays subtended 6.84 • × 3.89 • at an observation distance of 75 cm, with a chinrest/forehead bar combination to keep their heads stabilized.

Coherent motion direction identification testing and training
At the beginning of each trial, a 300 ms blank screen was presented accompanied by a brief tone (Fig. 2). After that, 400 white moving dots (0.18 deg in diameter with a speed of 10 deg/s) were displayed against a gray background (mean luminance: 26 cd/m 2 ) within a 200 ms interval that consisted of 17 frames in a temporal sequence (with a frame duration of 11.76 ms). The first frame contained randomly positioned moving dots within a round window (8 deg in diameter). Once dots moved outside of the window, new dots would replace the other dots quickly at different, random locations within the window; thus, the density of moving dots was kept constant (7.96 dots/deg 2 ). The coherent dots moved at random in one of four directions: 45 • , 135 • , 225 • , and 315 • ; the other dots moved randomly in any direction. Participants were required to make a four-alternative forced-choice (4AFC) judgment of the direction of coherently moving dots displayed among randomly moving distractor dots.
The coherent threshold (i.e., percentage of coherently moving dots) was controlled by an adaptive three-down/one-up staircase method such that answers were correct 79.4 % of the time. If participants responded correctly every three consecutive times, the coherence threshold decreased by 10 %. Correspondingly, if the participant incorrectly responded once, the coherence threshold increased by 10 %. We recorded the reversals once the direction of the staircase changed, i. e., from a decreasing to increasing threshold or vice versa. We removed the first four (if the total number of reversals was even) or five (if odd) reversals and averaged the remaining reversals to assess the coherent threshold for identifying the direction of coherently moving dots.
During the test periods, a brief tone occurred after each response regardless of its accuracy. The test consisted of 100 trials and took approximately 4 min to complete. Participants adequately practiced before the formal test to become familiar with the tasks and did not stop until the reversals remained relatively stable. The starting threshold was set near the expected coherent threshold (i.e., 30 %) based on the pilot tests. During the training periods, a brief tone appeared only following each right response. Auditory feedback was provided for correct responses during training, not during pretest or posttest assessments of the threshold. Each training session had 6 blocks of 80 trials lasting approximately 17.6 ± 0.39 min, which was shorter than the stimulation time (20 min), to ensure that the training and stimulation were conducted simultaneously. Participants relaxed between two blocks and decided their own start time for the next block.

Fig. 1.
Experimental design and procedure. The experiment was a between-subjects design with three groups: AA, NA and NS. The yellow arrows represent the administration of anodal tDCS during daily training. The gray arrows denote sham tDCS. The black rectangles represent tests of motion direction identification that were separately carried out before and 48 h after training. The white rectangles denote training sessions.

Fig. 2.
Description of the coherent motion direction identification task. Each trial began with 300 ms of fixation, followed by a 200 ms stimulus period. Dots moved coherently but randomly in one of four directions: 45 • , 135 • , 225 • , and 315 • . After that, participants responded in an untimed user response window. The next trial started immediately after a blank screen was presented for 900 ms.

Stimulation protocols
Coherent motion tasks are frequently employed to investigate visual perception and its neural mechanisms since this type of perception has a relatively distinct brain region (e.g., the hMT+). Multiple techniques, such as electrophysiology (Britten et al., 1992), lesion assessment (Newsome and Pare, 1988), functional magnetic resonance imaging (fMRI; Chen et al., 2016;Chen et al., 2017) and electrical stimulation , have confirmed the involvement of hMT+ in motion visual perception. High-definition tDCS (HD-tDCS) was administered by a 4 × 1 stimulator (Soterix Medical, NY, USA) to more focally generate hMT+ since this approach has been confirmed to have the advantage of focusing more current on the target brain regions through small electrodes compared with the conventional approach of using two large sponge electrodes (Dmochowski et al., 2011). Zito et al. (2015) applied HD-tDCS over the right hMT+ , which was used to guide the electrode montage in this study. We stimulated only the left hemisphere that was symmetric to Zito's region of interest. As shown in Fig. 3A, the central electrode was placed at PO7, which was surrounded by four return electrodes (P3, O Z , TP7, and PO9, according to the 10-10 standard EEG system). For anodal tDCS, a 1.5 mA direct current was delivered from the anode (PO7) to the remaining four electrodes (20 min; fade in/out: 30 s). In sham tDCS, the direct current was ramped up over 30 s at the beginning and ramped down over 30 s at the end of the 20 min period. The conductivity was increased by injecting conductive gel into the electrode casings (diameter 1 cm). The impedance values of each electrode were all kept at less than 5 kΩ during the entire session. Fig. 3B presents the simulated current flow of anodal tDCS with HD-Explore software (Soterix Medical Inc., New York).
During the training sessions, participants were asked to report tDCSinduced sensations by answering the following question: What is your sensation in the stimulated region, including any sensations such as burning, itching, tingling, pain and so on? Sensation intensity was measured on a 10-point scale as follows: 0 = none, 10 = strong and intolerable. Additionally, we asked participants whether they thought they had received sham or real stimulation at the end of the whole experimental procedure.
Since the stimulator was operated by an experimenter, he or she was unblinded to whether the participant was receiving anodal or sham stimulation. However, this experimenter was blinded to the purpose and the experimental design of the current study.

Data analysis
A power function was used to obtain the learning curves (Dosher and Lu, 2007;Zhang et al., 2018): where C 0 is the initial threshold, t is the training session number, ρ is the learning rate, and ι is the asymptotic line.
A nonlinear least square method was used to minimize the sum of squared differences between the model predictions and measured values. The goodness of fit was estimated by where y measured and y predicted represent the measured and predicted values, respectively. mean(y measured ) is the mean of all the measured values.
We compared the learning curves in a nested-model testing framework. An F test was used to statistically compare the nested models: where df 1 = k full − k reduced , df 2 = N − k full , k full and k reduced are the numbers of parameters of the full and reduced models, respectively, and N is the number of data points. The model that was statistically equivalent to other models and had minimum parameters was defined as the best-fitting model. Given that each participant had multiple thresholds that resulted in repeated measurement data, thresholds did not meet the requirement of random independence. In addition to the commonly used analysis of variance (ANOVA), a mixed effects model was performed with SPSS software. According to our two hypotheses, we needed to analyze the difference in thresholds among the three groups at posttest 1 or posttest 2. Thus, we established the linear mixed effects model, with groups (AA, NA and NS) as predictor variables and thresholds at pretest as covariates. Additionally, we added a pretest × group interaction term as the predictor variable since participants with different pretest thresholds may have different tDCS effects (Wu et al., 2020).

The coherent threshold at pre-and posttests
To investigate the effect of tDCS on the performance measured by posttests, we conducted a two-way ANOVA on the coherent threshold with 3 tests (pre, post1 and post2) and 3 groups (AA, NA and NS) as the two main effects (Fig. 4). Only the main effect of tests was significant, F (2,70) = 416.41, p < 0.001, η 2 = 0.92, and neither the main effect of group nor the interaction effect was significant, Fs < 1. The post hoc least significant difference (LSD) test showed that the coherent threshold of the pretest was significantly higher than that of posttest 1, p < 0.001, and posttest 2, p < 0.001. Additionally, the coherent threshold of posttest 1 was obviously less than that of posttest 2, p < 0.001.
More importantly, we mainly focused on the performance among the three groups on different tests. The LSD test showed that the thresholds among the three groups were not significantly different at any of the three tests (ps > 0.1). Specifically, the thresholds of the three groups at pretest were not significantly different from each other, indicating successful random grouping. Additionally, there were no significantly different thresholds between AA and NS at posttest 1, suggesting that anodal tDCS is not able to increase performance after 9 days of training. Finally, the thresholds of the three groups at posttest 2 were not significantly different, indicating no increased performance by anodal tDCS after 12 days of training (AA vs. NS; Hypothesis 1 was not supported) or after the continued 3-day training period (NA vs. NS; Hypothesis 2 was not supported).
For posttest 1 thresholds, the analyses with the mixed effects model showed a significant positive coefficient for the pretest threshold term (estimate = 0.35, SE = 0.14, t = 2.52), indicating that as the initial threshold increased, participants produced a higher posttes1 threshold. Additionally, the main effects for group and interaction terms missed significance. Importantly, pairwise comparisons did not show any significant differences in the posttest 1 threshold among the three groups (ps > 0.1). For posttest 2 thresholds, the mixed effects model also showed a significant positive coefficient for the pretest threshold term (estimate = 0.39, SE = 0.13, t = 3.17). In addition, no other significant terms were found. Similarly, there were no significant differences in the posttest 2 threshold among the three groups (ps > 0.1). Thus, the results of the mixed effects model were consistent with those of ANOVA. In conclusion, the above results do not provide evidence for the facilitative effect of anodal tDCS on performance measured by posttests during the plateau period.

Effect of tDCS on the learning curves
We established learning curves for the AA, NA and NS groups (Fig. 5). To better describe changes in the plateau, two stages were presented on the learning curves with different dimensions: the threshold as a function of training sessions (first stage) and the threshold as a function of training blocks (second stage).

AA and NS
To further investigate Hypothesis 1, we estimated the AA and NS learning curves (threshold as a function of 12 training sessions) by power functions with a total of six parameters. The value of each session was calculated by averaging all participants in each session. Each learning curve has three parameters: the initial threshold (C 0 ), learning rates (ρ) and asymptotic line (ι). Thus, eight models were developed by setting some parameters equal to one another (Table 1). Specifically, the model lattice consisted of 8 models, including the full 6-parameter model (M1) with independent C 0 , ρ and ι values; the reduced 5-parameter model with identical C 0 , ρ or ι (M2, M3 and M4) values; the reduced 4-parameter model with identical C 0 and ρ values (M5), C 0 and ιvalues (M6), or ρ and ι values (M7); and the reduced 3-parameter model with identical C 0 , ρ and ι (M8) values.
The third model (M3) and fourth model (M4) were not statistically worse than the full model (ps > 0.100) and better than another reduced model (ps < 0.001). Additionally, M4 (99.6%) was better fitted than M3 (99.5%) and provided the best fit. As shown in Fig. 6, M4 had two identical asymptotic lines (ι) in the AA and NS learning curves, and there was a significant difference in the initial threshold (C 0 ) and learning rates (ρ) between the AA and NS learning curves. Thus, M4 indicated that anodal tDCS did not improve the plateau level, even though it significantly decreased the coherent threshold at the beginning of the whole training series. Therefore, Hypothesis 1 was not supported.
As mentioned above, most previous research has investigated the effect of tDCS on VPL within a limited time. Therefore, we wanted to know when the tDCS effect began to disappear. Independent-samples t tests were conducted to assess the coherence thresholds of each session between the AA and NS groups. The results showed a significant difference in the threshold between the two groups during session 1, p = 0.012, and a marginally significant difference during session 2, p = 0.070. No significant differences were found during the later sessions, ps > 0.100. These results indicated that the tDCS effect on VPL  began to disappear in the third session in the current study.

NA and NS
To further explore Hypothesis 2, we conducted a 3-session (session 10, session 11 and session 12) and 2-group (NA and NS) two-way ANOVA on the training threshold when continuing the three training sessions. As shown in Fig. 7, the main effect of sessions was significant, F (2,44) = 3.62, p = 0.035, η 2 = 0.14. Specifically, the post hoc LSD test showed that the difference in the coherence threshold between session 10 and session 11 was marginally significant, p = 0.055. The threshold in session 10 was significantly less than the threshold in session 12, p = 0.027. The threshold of session 11 was not significantly different compared with that in session 12, p = 0.352. However, a significant main effect of group and interaction effect were not found, Fs < 1. These results suggest that tDCS applied during the continuous 3-day training session did not further improve the plateau level. Thus, Hypothesis 2 was not supported.

Subjective sensation intensity induced by tDCS
Each participant completed a question regarding subjective Notes: Columns 2-8 display the p values of statistical comparisons between different pairs of models. n represents the number of each model parameter. The model parameters are shown in the right column.  sensation at the midpoint of each training session. We compared the sensation intensity among the three groups with 3 sessions (session 10, session 11 and session 12) and 3 groups (AA, NA and NS) using two-way ANOVA. The main effect of group was significant, F(2,35) = 9.15, p = 0.001, η 2 = 0.34. The post hoc LSD test found that participants in the NS group felt less sensation than those in the AA group, p < 0.001, and NA group, p = 0.001. There was no obvious difference in sensation intensity between the AA and NA groups, p = 0.843. Furthermore, no significant main effect of sessions or interaction effect was found, Fs > 1.
In addition, all participants in both stimulation conditions thought they received real stimulation. Thus, there was no difference in the expectation between the sham and real stimulation. In short, the anodal tDCSinduced sensations were perceived more strongly than the shaminduced sensations. However, participants were unable to distinguish real stimulation from sham stimulation.

Discussion
This study investigated the effects of anodal tDCS over the left hMT+ on the later learning effects of coherent motion identification and failed to find significant effects. Specifically, although anodal tDCS decreased the coherent threshold during the early period of all the training sessions, it did not increase the performance and plateau level when the plateau was reached, suggesting that Hypothesis 1 was not supported. In addition, anodal tDCS did not further improve the performance and plateau level after the continuous 3-day training sessions, indicating that Hypothesis 2 was not supported. Thus, tDCS has a beneficial effect on early training effects but not later learning effects.
To our knowledge, previous studies that combined tDCS and VPL have rarely involved later training sessions. Additionally, there is no precise definition of when the later phase of perceptual learning occurs. In the current study, we regarded the plateau performance as the later phase. Thus, we focused on the later learning effects that were measured by the performance of the posttest and the plateau level of the learning curve after the participants' performance reached the plateau. This was an innovation in our study and is especially helpful for the practical application of VPL. This approach is beneficial because only the later learning effects determine the amount of gain in learning, which is closely related to the rehabilitation of patients with visual impairment. The larger the gain, the more greatly individuals improve or rehabilitate their visual function. In contrast, the change in learning speed influences the learning effects within a limited time but does not affect the later learning effects. Although increasing the learning speed is also important, extending the training time can overcome the disadvantages of slow learning. However, we did not provide evidence for the beneficial effects of tDCS on later learning effects.
Interestingly, we only found that anodal stimulation significantly decreased coherent thresholds in the first two training sessions. Additionally, the comparison of AA and NS learning curves showed a different difference in the initial threshold and learning rate. These results indicate that tDCS facilitates VPL in the early stage of the whole training session. In other words, this study failed to find that tDCS influenced the learning effects at a later stage, but this does not mean that tDCS has no effect on VPL. Thus, our study is consistent with some previous studies in which the combination of tDCS and behavior training was able to benefit VPL (Karlaftis et al., 2021;Lisa, 2011;Pirulli et al., 2013;Sczesny-Kaiser et al., 2016). From this, our findings provide a possible explanation for the inconsistent results of previous studies: the effect of tDCS on VPL may be modified by the distinct stage of VPL.
Why did the tDCS effects disappear as the training continued? A possible reason is that the brain regions dynamically change across the time course of training; this occurrence has been confirmed in many studies. For example, texture identification performance increased with activation enhancement of V1 in an early stage of training. In a later phase, i.e., when performance reached a plateau, V1 activation returned to the baseline level while performance remained high (Yotsumoto et al., 2008). Similarly, Chen et al. (2016) found that the contribution of MT+ to noisy motion processing is substituted with V3A after perceptual training. Additionally, Chang et al. (2014) demonstrated that training results in striking changes: identifying a target in noise leads to activity in brain areas from the posterior parietal cortex to the lateral occipital cortex after feature training. These findings provide important evidence for the multistage mechanisms of VPL (Shibata et al., 2014;Watanabe and Sasaki, 2015). Here, we speculated that the contribution of the stimulated visual brain region to visual motion perceptual learning was larger at the early stage than at the later stage. Thus, tDCS effects existed at the early stage but gradually faded away at later stages. Future studies should explore the brain loci during the later stages of VPL and then stimulate these brain loci, which may boost the later learning effects and contribute to an increase in the amount of learning.
Another possibility for the noninfluence of anodal tDCS on the later learning effect is the method adopted in this study. We used the adaptive three-down/one-up staircase method to control the thresholds during training. Although this method is commonly used in current VPL research, it may not precisely gauge the tiny amount of threshold change induced by tDCS when participants' performances approached the asymptote.
As mentioned above, the later learning effects were measured when the plateau was reached. Thus, defining the plateau was important in the current study. To our knowledge, there are no clear definitions of the plateau in previous studies. Thus, observations were made to determine whether a plateau had been reached in this study. We conducted a pilot study and defined the plateau by observing the learning curve. Finally, prior to the study, 9 training sessions were needed for participants to reach the plateau. The ideal definition is that the performance does not continuously improve once the plateau has been reached. However, we found that there was still a tendency for continued improvement after the plateau had been reached. Although the improvement was significant, the amount of absolute improvement was very slight.
Notably, the left hMT+ rather than the right hMT+ was stimulated in this study. Only one study found a significant improvement in motion perception after cathodal tDCS over the right hMT+ (Zito et al., 2015). Most studies stimulated the left hMT+ and found a beneficial effect of tDCS on motion perception (Battaglini et al., 2017;Wu et al., 2020). These results suggest that both left and right hMT+ may be effective targets for stimulation to improve motion perception. We chose anodal tDCS to stimulate the left hMT+ since this cerebral hemisphere has been the subject of more investigations, especially those researching the effect of tDCS on perceptual learning, allowing a comparison of results with those of previous studies.
As mentioned above, some tDCS studies on VPL have reported varying results. The possible reason for this variance is that the learning tasks, time sequences and stimulation parameters were different among these previous studies. In one of our previous studies, for example, participants were trained with motion direction discrimination in which they determined the minimum angle of the two directions of moving dots (Wu et al., 2022). In contrast, this study used motion direction identification in which participants had to identify the direction of coherently moving dots. Except for the different tasks used, the other experimental conditions were consistent. The results showed a beneficial effect of anodal tDCS on VPL when training on motion direction identification rather than motion direction discrimination, indicating the influence of learning tasks. Additionally, we adopted the same experimental protocol as the current study and found that anodal tDCS of the left hMT+ immediately after completion of the daily training session significantly improved motion direction identification learning (Wu et al., 2023a). Moreover, we found that anodal tDCS resulted in greater performance improvement when applied during daily training but not when applied before training (Wu et al., 2023b). Thus, we are confident that tDCS can boost VPL, but there are some factors that need to be carefully considered, such as the learning tasks, time sequences and stimulation parameters.
One limitation of this study was the number of participants. Variability in tDCS effects has resulted in calls for greatly increased sample sizes (Minarik et al., 2016). Our sample size (n = 12-14 per group) was comparable to or greater than several tDCS studies in the VPL that found significant effects (Herpich et al., 2019;Sczesny-Kaiser et al., 2016), although it was smaller than some studies (Karlaftis et al., 2021;Pirulli et al., 2013). Future research should increase the credibility of the findings with a larger sample size.
Another limitation was the lack of a well-controlled group at the early stage of training, such as a group with sham control or stimulation position control. This study showed that anodal tDCS decreased the threshold of the trained task at the early stage of learning. Considering the cutaneous sensation caused by tDCS, a possible placebo effect on the modulatory effects of tDCS on VPL cannot be ruled out. However, we think that even though anodal tDCS induces stronger sensations than sham tDCS (or no stimulation), participants were not influenced by expectancy; more specifically, they did not exhibit signs of more serious and careful training. This is because the current study was conducted with a between-subjects design. Specifically, the type of tDCS was kept constant at the early stage of training. In other words, participants did not experience the sensation induced by other types of tDCS and therefore could not judge whether they were applied with stimulation or without stimulation.
Finally, this study did not identify individual left hMT+ using fMRI, which may have resulted in ineffective stimulation of anodal tDCS on the target area. However, this seems unlikely. It has been shown that area hMT+ varies by only approximately 2.7 cm in the left hemisphere (Watson et al., 1993) and is 0.3 cm 3 in average size (Malikovic et al., 2007). In the current study, the central electrode was placed at PO7, and four return electrodes were placed at a distance of approximately 5 cm from the central electrode. Thus, the tDCS electrode dimensions exceed the area hMT+ . In addition, Herpich et al. (2019) stimulated O1 and O2 instead of PO3 and PO4 or PO7 and PO8, which are closer to the hMT+ complex. They still found a significant influence of transcranial random noise stimulation (tRNS) on motion perception learning, further indicating the availability of the PO7 electrode. In brief, it is likely that anodal tDCS with a central electrode on PO7 at least partially covered hMT+ .

Conclusions
In conclusion, anodal tDCS over the left hMT+ facilitated multisession visual motion learning at the early stage of the entire training series. However, it did not improve the later learning effects that were measured by the performance and plateau level when the saturated performance was reached.

Ethics statement
Participants signed the notice indicating that they had been informed before the experiment. This study received ethics approval from the Air Force Medical University.

CRediT authorship contribution statement
Methodology (WD and ZP); software (WD and ZP); data curation and formal analysis (WD); writing-original draft (WD and WY); and writing-review & editing (WU, ZY, WY and LN).

Declaration of Competing Interest
The authors declare that they have no conflict of interest.

Data availability
Data will be made available on request.