Introduction

It is well documented that playing video games results in improvements in a wide range of visual and cognitive tasks1,2,3,4,5,6; these include light sensitivity, contrast sensitivity, visual resolution, visual crowding, backward masking, visual attention, and useful visual field size. The discovery that playing these games generalizes broadly has led to the adoption of gaming in therapeutic applications for the recovery of vision in the disordered visual system. Indeed, this video-game-based approach has shown to be useful clinically for improving eyesight, or visual acuity, in patients with amblyopia (also known as lazy eye)7,8.

Intriguingly, we recently reported that playing flat two-dimensional (2D), video games does not enhance stereoacuity in individuals with normal vision9. Rather, playing the identical three-dimensional (3D) video games resulted in the substantially enhanced stereoacuity—i.e., precision of stereopsis. Additionally, under the conditions of our testing, essentially unlimited viewing time, no improvement in contrast sensitivity was found in either group.

How does 3D video game play enhance stereo thresholds? Stereopsis, the vivid impression of 3D depth based on the slightly different viewpoints of the two eyes is very acute. Binocular depth, stereo, thresholds are about a factor of ten better than monocular depth thresholds10. Fine stereopsis is critical for fine motor activities that require precise eye-hand coordination—for instance, threading a needle, catching a flying ball, pouring water in a glass, and even suturing a wound during endoscopic surgery, etc. When viewing is restricted to one eye, visually-guided hand reaching movements are greatly impaired11,12,13. Nevertheless, not everyone experiences normal stereopsis. It is estimated that approximately seven percent of the population is stereoblind14, and many more may have reduced stereopsis15, and may benefit from 3D video game play.

When measuring stereo depth discrimination, it is possible to derive two separate performance indices: (1) threshold, reflecting the precision of depth judgments based on the variance of the depth detection response, and (2) bias, also known as the point of subjective equality (PSE), reflecting the accuracy of depth judgments based on how close the subjective depth judgements are to the true physical depth. Stereo threshold and stereo bias can vary independently16,17; however, a large depth bias (i.e., significantly misjudging the depth of a plane) may act as a “depth pedestal” and it is well known that the precision of stereopsis is degraded by depth pedestals. A large depth bias can reduce depth precision systematically in one direction of depth; the increment disparity threshold function usually rises exponentially with disparity pedestal18. Indeed, it has been shown that if fixation is biased slightly closer or farther from the reference stimulus can result in stereothresholds being elevated by up to about a factor of 219.

Here we asked whether the improvements in the precision of stereo depth judgements (lowered or improved threshold) were related to improved accuracy (reduced bias, if any) of stereopsis. In addition, we re-examined the relationship between improved stereo thresholds and contrast thresholds on individual observers using correlation analysis.

Results

Twenty-one young non-gamers with normal vision participated. All had no previous experience with stereoscopic 3D video games. Stereo-discrimination performance (i.e., stereo bias and stereo threshold) and contrast sensitivity were measured before and after a total of 40 h of video game play (twenty 2-h sessions over 4–6 weeks) in the baseline and post-gaming assessment sessions, respectively. Off-the-shelf first-person shooter action games on the PlayStation 3 platform (PS3, Sony Computer Entertainment, Japan) were used. The participants were required to wear a pair of active liquid crystal shutter glasses (Panasonic, TY-EW3D3, Japan; resolution, 1920 × 1080 pixels) for viewing stereoscopic 3D content. All participants wore full optical correction for video gaming and all the visual testing.

Pre-gaming stereo bias

The random-dot stereo test developed by Patel et al.20 was used to measure stereo depth discrimination performance (Fig. 1A, inset; see Methods for detailed measurement protocols). In most participants (81% of all participants), the baseline pre-gaming PSE was behind the reference square (Fig. 1A, gray shaded area; negative stereo bias: uncrossed disparity); while in the rest of them (19% of all participants), the baseline pre-gaming PSE was in front of the reference square (positive stereo bias: crossed disparity).

Fig. 1
Fig. 1
Full size image

3D video game experience modifies stereo acuity, but not stereo bias. (A) Stereo bias before and after 40 h of 3D video game play (stereo biaspre and stereo biaspost, respectively). Most data points are scattered around the gray 1:1 reference line. Inset figure, the visual task was to determine the stereoscopic depth of the central square, in front or behind, relative to the reference background. Positive stereo bias, in front of the reference background (toward the observer). Negative stereo bias, behind the reference background (away from the observer). Stereo bias was expressed in seconds of arc (”). Open black circle, mean pre-gaming and post-gaming stereo bias. Sample size, 21. (B) Psychometric curves before and after playing 3D video games (dashed and solid curves, respectively). After gaming, there was a statistically significant improvement in mean stereo threshold, but not in mean stereo bias (red curves). (C) Percent change in stereo threshold as a function of percent change in stereo bias after 3D video game play. Our correlation analysis suggests that the changes in threshold and bias were essentially uncorrelated (coefficient of determination, 0.007). Inset, stereo threshold before and after 40 h of 3D video game play (stereothresholdpre and stereothreshold post, respectively, replotted here from our previous study1).

Improved precision but not accuracy

There was a dissociation between stereo threshold and stereo bias with 3D video game play. We found that 3D video game experience significantly improved stereo threshold, but did not modify stereo bias in our participants. After playing 3D video games, there was a statistically significant decrease in stereo threshold (mean pre-gaming stereoacuity, 33.05 ± SE 3.23 arcsec; mean post-gaming stereoacuity, 22.19 ± SE 1.80 arcsec; improvement, 32.9%; paired t = 2.941, p = 0.005; note that those stereo threshold data were reported in our recent study1 and were replotted in the inset of Fig. 1C), but interestingly not in stereo bias (open red circle; mean pre-gaming stereo bias: − 4.42 ± SE 1.98 arcsec; mean post-gaming stereo bias, − 4.17 ± SE 2.11 arcsec; paired t = 0.134, p = 0.895).

The representative psychometric functions before and after 3D video game play are displayed in Fig. 1B (red dashed and solid curves, respectively; constructed by the mean stereo bias and mean stereo threshold). The PSEs of both pre- and post-gaming curves being slightly shifted to the left of zero for small negative, uncrossed disparities, showing that the pre- and post-gaming PSEs are roughly the same. Moreover, the slope of the post-gaming curve was steeper than that of the pre-gaming curve (solid curve versus dashed curve), indicating stereo threshold was lowered or improved after the video game intervention.

Stereo threshold versus stereo bias

As can be seen in Fig. 1C for each individual participant, playing 3D video games results in changes in both the accuracy (bias—on the ordinate) and precision (threshold—on the abscissa) of stereoscopic depth perception; however, while the group change in stereo threshold was highly significant, the group change in bias was not. It is possible that individual differences were obscured by considering the group as a whole—for example, it might be the case that only those observers with large changes in bias showed improvements in threshold. However, our correlation analysis suggests that this is highly unlikely since the changes in threshold and bias were essentially uncorrelated (coefficient of determination, r2 = 0.007). This finding clearly suggests that the improvement in the precision of stereo depth perception (lowered stereo threshold) was not the result of increased accuracy of stereo depth perception (reduced stereo bias), arguing against the disparity pedestal hypothesis.

Stereo threshold versus contrast threshold

Since the immersive 3D video game experience could modify other visual functions as well, we wondered whether the sharpened depth perception was simply the result of enhanced contrast perception. In order to assess a potential relationship between the improvement of stereo and contrast thresholds, the contrast threshold (Th) versus spatial frequency (SF) datasets were fitted with a 3-parameter function21 for individual participants:

$$Th=\left({Th}_{opt}*{\left(\frac{SF}{{SF}_{opt}}\right)}^{EXP}\right)*{e}^{\left(\left(\frac{-SF}{{SF}_{opt}}+1\right) * EXP\right)}$$

where Thopt is the optimal contrast threshold (i.e., the lowest contrast threshold across the spatial frequencies tested; also see Fig. 2A, inset), SFopt is the optimal spatial frequency (i.e., the spatial frequency at which the optimal contrast threshold was obtained), and EXP is an exponent. Note that we did not measure contrast sensitivity in two participants (therefore, n = 19 in this part).

Fig. 2
Fig. 2
Full size image

Dissociation between depth perception and contrast perception. (A) Characteristics of contrast sensitivity functions before and after 40 h of 3D video game play. The pre-gaming and post-gaming contrast thresholds versus spatial frequency datasets were fitted with a 3-parameter function (i.e., optimal contrast threshold, optimal spatial frequency and exponent) for individual participants. Inset figure, mean pre-gaming and post-gaming contrast sensitivity functions. Open symbols, individual data. Solid symbols, mean data. Sample size, 19. (B) Changes in the optimal contrast threshold, optimal spatial frequency and exponent as a function of change in stereo threshold. Open symbols, individual data. Solid symbols, mean data.

Figure 2A displays the optimal contrast threshold, optimal spatial frequency, and shape before and after 3D video game play for each individual observer (open symbols) and averaged across observers (solid symbols). We found no statistical differences before and after the video game intervention for any of the three model parameters (Thopt: paired t = 0.8, p = 0.432; SFopt: paired t = 0.273, p = 0.79; EXP: paired t = 0.28, p = 0.784). The inset figure shows the mean contrast threshold as a function of spatial frequency (contrast threshold function, reciprocal of contrast sensitivity function), and the two curves show the 3-parameter fits to the pre-gaming and post-gaming datasets.

Figure 2B depicts the relationship between changes in fitting parameters (i.e., Thopt, SFopt and EXP) and changes in stereo thresholds for each observer (open symbols) and averaged across observers (solid symbols). Table 1 summarizes the slopes and correlation analyses of the linear regression functions. Note that only the optimal contrast threshold parameter exceeded the 95% confidence interval, but just barely, and it only accounted for roughly twenty percent of the variance, consistent with the dissociation between depth perception and contrast perception that we previously reported1. All these analyses support that although the participants had enhanced stereoacuity, they did not have enhanced contrast sensitivity after gaming in 3D.

Table 1 Correlation analysis.
Full size table

Discussion

Our recent work reported that video game experience enhances stereo depth perception in young adults with healthy vision. Specifically, those participants showed improved stereoacuity (i.e., stereo threshold) after playing stereoscopic 3D video games for a relatively short period of time1. However, the improvement in stereoscopic acuity could have been simply the result of a reduced disparity pedestal as indicated by a decrease in subjective equality bias, or changes in the contrast threshold versus spatial frequency function.

Our present findings suggest that the enhanced stereo precision (improved thresholds) was unlikely to result from increased accuracy of stereoscopic depth perception (reduced bias), arguing against reducing the effects of a disparity pedestal. In our experiment, the participants were not given any feedback to their responses in the stereoacuity measurements—in each trial they were not informed of whether the responses were correct or not—which may at least in part explain why the subjective bias remained unchanged in the post-gaming measurement. Consistent with our variance analysis (ANOVA) of the contrast threshold data averaged across spatial frequencies and observers1, the new correlation analysis shows at best a subtle improvement in the optimal contrast threshold, accounting for only approximately 20% of the variance.

Previous work has shown that playing 2D action video games may result in improvements in contrast sensitivity3. However, they used extremely brief stimulus durations (30 ms), while our stimuli were presented with free viewing, effectively unlimited duration. In their second experiment, they varied the stimulus duration and reported no improvement at the longest duration (180 ms), concluding that there was a causal effect of action game playing in the reduction in integration time. If that was the only effect of video game play, then no improvement would be expected with the long stimulus durations adopted in the current study for both contrast sensitivity and stereoacuity. In the current study, we did not test contrast sensitivity at shorter viewing durations, it is not clear whether playing 3D video games, similar to 2D video games, can modify contrast sensitivity at a very brief presentation duration. It is worth noting that in the current study most of the game scenes were salient and displayed in high contrasts, there was relatively little challenge to practice near-threshold contrast detection during gaming.

Why does playing 3D, but not 2D videogames improve stereopsis? One possible explanation is that the 3DVG group learned to adapt to the accommodation-convergence conflict—when focus is at the screen distance, but convergence is on a closer or farther object. This unnatural decoupling is believed to be a major source of blurry vision, visual discomfort and visual fatigue when viewing 3D displays with the latest technologies22. However, that seems unlikely to explain the enhanced stereoacuity in the 3DVG group, since our random dot test was conducted using a freely adjustable mirror stereoscope and therefore should not have introduced any conflict between accommodation and vergence.

In the natural environment, the binocular disparities encountered are usually not very large (Fig. 3). In the central 3 degrees, (i.e., roughly the size of our stereogram), 90% of retinal disparities are within the range between − 532 and + 1419 arcsec (uncrossed and crossed, respectively), and even smaller in the central 0.5 degree (i.e., the size of our depth target at the center of the stereogram). In contrast, the stereoscopic content present in the video games is substantially more dramatic; the largest screen disparities in the video games used in this experiment were on the order of 5000 arcsec (uncrossed; viewing distance, 1 m).

Fig. 3
Fig. 3
Full size image

Binocular disparities in central vision. In the central 3 degrees, 50% of retinal disparities are within the range between − 197 and + 845 arcsec (dotted lines; 25th–75th percentile;), and 90% of retinal disparities are within the range between − 532 and + 1419 arcsec (dashed lines; 5th–95th percentile). Negative disparity, uncrossed disparity. Positive disparity, crossed disparity. Data courtesy of Gibaldi A. and Banks M. (University of California, Berkeley).

Another notable difference is that the entire game scenes are rendered in virtual 3D space, no matter whether the visual objects are positioned closer to or farther away from the viewer, are sharply displayed on the screen. But in real life, the disparity cues of out-of-focus objects, in front of or behind the plane of focus, are somewhat blurred. We speculate playing 3D video games that contain exaggerated binocular disparities, unusually larger than those generally encountered in natural scenes, can increase awareness of relative depth localizations between visual objects for efficient stereoscopic processing.

While low-level vision may be more difficult to train indirectly in a generalized way with video game training, those visual functions can be enhanced more effectively using a direct perceptual learning approach. We previously showed that stereoacuity, over a wide range of target spatial frequencies, can be enhanced by perceptual learning, and the learning effect transfers across the spatial frequency spectrum23. The improvement in stereoacuity with perceptual learning (approximately 70% and 80% at low and high spatial frequencies, respectively; instead of random-dot stereograms, those participants were tested with narrow-band Gabor stimuli) appears to be considerably greater than that with 3D video game play observed in the present study (33%), but that is not very surprising as in a typical perceptual training protocol, participants are repetitiously practicing a visual task at near-threshold limits for a large number of response trials (typically, 6–10 kilo trials or more).

In the present study, we chose to measure contrast sensitivity binocularly, instead of monocularly. An obvious reason is that a larger number of response trials would be needed to obtain reliable measurements of five spatial frequencies in each eye. Another advantage is to rule out the potential effect of enhanced binocular summation, if any, of monocular contrast inputs after gaming. We should also note that our stereo acuity results are based on an improved ability to discern the sign of the disparity. This is important for at least two reasons: (1) stereo tests based on depth magnitude may fail to detect stereovision anomalies14,15, and (2) disparity magnitude and disparity sign may be computed by separate, possibly independent, processing mechanisms15,24. On a separate note, most of our participants were female, because of difficulties in recruiting male participants who were non-gamers and did not have much previous video game experience. Some caution should be taken before generalizing our findings.

Importantly, we have previously shown that this type of stereoscopic 3D video game play has potential therapeutic applications in restoring visual acuity and stereoacuity in adult patients with amblyopia25 (commonly known as lazy eye26,27; also see8,9). It appears that those amblyopic adults showed a modestly greater amount of improvement in stereoacuity than the healthy participants in the current study (amblyopic adults, 34–39% vs. healthy adults, 27%; the mean calculations were based on the improvement percentage of individual participants; identical time course of gaming, a total of 40 h: twenty 2-h sessions over 4–5 weeks), and we expect an even faster and probably more substantial recovery in child patients28. Our preliminary findings reveal that 3D video games may improve contrast sensitivity in the amblyopic eye, resulting in a smaller interocular difference in the contrast sensitivity function that can in turn facilitate depth perception29. The long-term maintenance of improved visual acuity with perceptual learning has been shown to be essentially stable during a period of 3 months to a year30,31. Further study may be needed to evaluate the longevity of benefits from this video-game training approach on improving stereopsis8,25, presumably it should be dependent on the type and depth of amblyopia.

In this study, all participants recruited were non-gamers and they had no, or very limited, video game experience. It is reasonable to expect a smaller amount of stereoacuity improvement for those gamers, as their stereoacuity might have been already boosted with previous 3D video game experience. However, that may not be the case for moderate and severe amblyopia. Those amblyopic people may mostly rely on the stronger sound eye because of monocular suppression, without working together with the weaker amblyopic eye to achieve binocular vision.

In brief, our experiments show that playing 3D video games boosts the precision, but not the accuracy of depth perception, indicating that the enhanced stereoacuity did not result from reducing the effects of a disparity pedestal. Additionally, our further correlation analyses confirmed that the enhancement in stereoacuity was not simply a consequence of improved contrast processing. Engaging in an immersive stereoscopic 3D environment rich in disparity cues is possibly the key to inducing binocular plasticity. In clinical situations, this novel 3D video game training approach may have special benefit for enhancing stereopsis in people with subnormal binocular vision.

Methods

The methods are described in detail elsewhere1, and are briefly reviewed here.

Participants

Twenty-one healthy young university students with normal visual acuity and stereoacuity of 30 arcsec or better (Randot® Stereotest, Stereo Optical Company, Inc., USA) were recruited to play 3D video games. None had ever played 3D video games, and none had played 2D video games in the five years prior to this study (self-reported). All but one of the participants were female. Note that our original study included a matched group who played 2D video games; however here we focus on just the 3D game group, since stereo vision did not improve in the 2D group.

General procedures

Participants wore active liquid crystal shutter glasses (Panasonic, TY-EW3D3, Japan; resolution, 1920 × 1080 pixels) while playing 3D first-person shooter action games on a PlayStation 3 platform (PS3, Sony Computer Entertainment, Japan) for twenty 2-h sessions, a total of 40 h, over 4–6 weeks in the laboratory. Typically, the participants started with playing “Killzone 3” and “SOCOM 4: U.S. Navy Seals” (Sony Computer Entertainment, Japan); when they finished those games, they could choose to play either Ratchet & Clank (Sony Computer Entertainment) or Crysis 3 (Electronic Arts, USA). The viewing distance was roughly 1 m. Stereoacuity and contrast sensitivity were measured before and after video game play in the baseline and post-gaming assessment sessions, respectively. All participants wore full optical correction for both video gaming and visual testing.

Stereo threshold and bias

The random-dot stereogram, programming based on MATLAB (MathWorks, USA) and described in detail by Patel et al.20, was comprised of a pair of individually constructed images for the left and right eyes. The outline of the two central squares subtended zero binocular disparity, providing no monocular cue for the direction of the perceived depth. The stereogram was presented on a gamma-corrected 22-inch flat Mitsubishi Diamond Pro 2040u monitor screen (1280 × 960 resolution and 75 Hz refresh rate) and viewed in a custom-built 4-mirror stereoscope. The random-dot stimuli consisted of a 0.56° central square of random dots embedded in a 1.67° reference square of random dots. The pixel size was 1 arc min at the viewing optical distance of 1 m. Sub-pixel binocular disparities were introduced by shifting the phases of all spatial frequencies in the two central squares in opposite horizontal directions. The mean luminance was 50 cd/m2, and the maximum displayable stimulus contrast was 100%.

We used the method of constant stimuli to measure stereoacuity. The observers’ task was to determine whether the stereoscopic depth of the central square was in front or behind the outer reference square. The stereogram remained on the screen until the observer responded. No feedback was provided for each response. The stereograms were presented with a range of binocular disparities (both crossed and uncrossed), enabling us to construct a psychometric curve by fitting the response data with a Gaussian cumulative distribution function. Stereo threshold (precision) was defined as the standard deviation of the Gaussian function (proportions of “in front” responses = 0.84 [+ 1 SD] and 0.16 [− 1 SD]). Stereo bias (accuracy) was defined as the difference between the true zero disparity and the PSE at which a 50% proportion of “in front” responses was obtained. Both stereo thresholds and biases are based on the average of 4 separate runs (90 trials per run).

Contrast threshold

Contrast thresholds were measured binocularly for vertical Gabor patterns (spatial frequencies 1, 2, 5, 10 and 20 cpd) displayed on a gamma-corrected 22-inch flat NEC FP2141SB monitor screen (1312 × 983 resolution and 80 Hz refresh rate) using a four-alternative spatial forced-choice, 4AFC combined with a logarithmic staircase procedure. The stimulus exposure time was virtually unlimited (20 s); the target Gabor patch remained on the screen until the observer responded. The background luminance was 50 cd/m2 and the maximum displayable luminance was 100 cd/m2.

The observers’ task was to indicate the target grating location (top-left/right and bottom-left/right). On average, each run consisted of roughly 85 response trials. The contrast threshold was defined as the average contrast over the last 8 staircase reversals. Contrast thresholds reported are based on the average of 3 runs (≈ 85 trials for each of the five interleaved spatial frequency staircases per run).