Artificial intelligence-based screening for amblyopia and its risk factors: comparison with four classic stereovision tests.

Introduction: The development of costs-effective and sensitive screening solutions to prevent amblyopia and identify its risk factors (strabismus, refractive problems or mixed) is a significant priority of pediatric ophthalmology. The main objective of our study was to compare the classification performance of various vision screening tests, including classic, stereoacuity-based tests (Lang II, TNO, Stereo Fly, and Frisby), and non-stereoacuity-based, low-density static, dynamic, and noisy anaglyphic random dot stereograms. We determined whether the combination of non-stereoacuity-based tests integrated in the simplest artificial intelligence (AI) model could be an alternative method for vision screening. Methods: Our study, conducted in Spain and Hungary, is a non-experimental, cross-sectional diagnostic test assessment focused on pediatric eye conditions. Using convenience sampling, we enrolled 423 children aged 3.6-14 years, diagnosed with amblyopia, strabismus, or refractive errors, and compared them to age-matched emmetropic controls. Comprehensive pediatric ophthalmologic examinations ascertained diagnoses. Participants used filter glasses for stereovision tests and red-green goggles for an AI-based test over their prescribed glasses. Sensitivity, specificity, and the area under the ROC curve (AUC) were our metrics, with sensitivity being the primary endpoint. AUCs were analyzed using DeLong's method, and binary classifications (pathologic vs. normal) were evaluated using McNemar's matched pair and Fisher's nonparametric tests. Results: Four non-overlapping groups were studied: (1) amblyopia (n = 46), (2) amblyogenic (n = 55), (3) non-amblyogenic (n = 128), and (4) emmetropic (n = 194), and a fifth group that was a combination of the amblyopia and amblyogenic groups. Based on AUCs, the AI combination of non-stereoacuity-based tests showed significantly better performance 0.908, 95% CI: (0.829-0.958) for detecting amblyopia and its risk factors than most classical tests: Lang II: 0.704, (0.648-0.755), Stereo Fly: 0.780, (0.714-0.837), Frisby: 0.754 (0.688-0.812), p < 0.02, n = 91, DeLong's method). At the optimum ROC point, McNemar's test indicated significantly higher sensitivity in accord with AUCs. Moreover, the AI solution had significantly higher sensitivity than TNO (p = 0.046, N = 134, Fisher's test), as well, while the specificity did not differ. Discussion: The combination of multiple tests utilizing anaglyphic random dot stereograms with varying parameters (density, noise, dynamism) in AI leads to the most advanced and sensitive screening test for identifying amblyopia and amblyogenic conditions compared to all the other tests studied.

Validation of dynamic random dot stereotests in pediatric vision screening.

PURPOSE: Stereo vision tests are widely used in the clinical practice for screening amblyopia and amblyogenic conditions. According to literature, none of these tests seems to be suitable to be used alone as a simple and reliable tool. There has been a growing interest in developing new types of stereo vision tests, with sufficient sensitivity to detect amblyopia. This new generation of assessment tools should be computer based, and their reliability must be statistically warranted. The present study reports the clinical evaluation of a screening system based on random dot stereograms using a tablet as display. Specifically, a dynamic random dot stereotest with binocularly detectable Snellen-E optotype (DRDSE) was used and compared with the Lang II stereotest. METHODS: A total of 141 children (aged 4-14, mean age 8.9) were examined in a field study at the Department of Ophthalmology, Pécs, Hungary. Inclusion criteria consisted of diagnoses of amblyopia, anisometropia, convergent strabismus, and hyperopia. Children with no ophthalmic pathologies were also enrolled as controls. All subjects went through a regular pediatric ophthalmological examination before proceeding to the DRDSE and Lang II tests. RESULTS: DRDSE and Lang II tests were compared in terms of sensitivity and specificity for different conditions. DRDSE had a 100% sensitivity both for amblyopia (n = 11) and convergent strabismus (n = 21), as well as a 75% sensitivity for hyperopia (n = 36). However, the performance of DRDSE was not statistically significant when screening for anisometropia. On the other hand, Lang II proved to have 81.8% sensitivity for amblyopia, 80.9% for strabismus, and only 52.8% for hyperopia. The specificity of DRDSE was 61.2% for amblyopia, 67.3% for strabismus, and 68.6% for hyperopia, respectively. Conversely, Lang II showed about 10% better specificity, 73.8% for amblyopia, 79.2% for strabismus, and 77.9% for hyperopia. CONCLUSIONS: The DRDSE test has a better sensitivity for the detection of conditions such as amblyopia or convergent strabismus compared with Lang II, although with slightly lower specificity. If the specificity could be further improved by optimization of the stimulus parameters, while keeping the sensitivity high, DRDSE would be a promising stereo vision test for screening of amblyopia.

Correction: Simple reaction times to cyclopean stimuli reveal that the binocular system is tuned to react faster to near than to far objects.

[This corrects the article DOI: 10.1371/journal.pone.0188895.].

Simple reaction times to cyclopean stimuli reveal that the binocular system is tuned to react faster to near than to far objects.

Binocular depth perception is an important mechanism to segregate the visual scene for mapping relevant objects in our environment. Convergent evidence from psychophysical and neurophysiological studies have revealed asymmetries between the processing of near (crossed) and far (uncrossed) binocular disparities. The aim of the present study was to test if near or far objects are processed faster and with higher contrast sensitivity in the visual system. We therefore measured the relationship between binocular disparity and simple reaction time (RT) as well as contrast gain based on the contrast-RT function in young healthy adults. RTs were measured to suddenly appearing cyclopean target stimuli, which were checkerboard patterns encoded by depth in dynamic random dot stereograms (DRDS). The DRDS technique allowed us to selectively study the stereoscopic processing system by eliminating all monocular cues. The results showed that disparity and contrast had significant effects on RTs. RTs as a function of disparity followed a U-shaped tuning curve indicating an optimum at around 15 arc min, where RTs were minimal. Surprisingly, the disparity tuning of RT was much less pronounced for far disparities. At the optimal disparity, we measured advantages of about 80 ms and 30 ms for near disparities at low (10%) and high (90%) contrasts, respectively. High contrast always reduced RTs as well as the disparity dependent differences. Furthermore, RT-based contrast gains were higher for near disparities in the range of disparities where RTs were the shortest. These results show that the sensitivity of the human visual system is biased for near versus far disparities and near stimuli can result in faster motor responses, probably because they bear higher biological relevance.

The retention and disruption of color information in human short-term visual memory.

Previous studies have demonstrated that the retention of information in short-term visual perceptual memory can be disrupted by the presentation of masking stimuli during interstimulus intervals (ISIs) in delayed discrimination tasks (S. Magnussen & W. W. Greenlee, 1999). We have exploited this effect in order to determine to what extent short-term perceptual memory is selective for stimulus color. We employed a delayed hue discrimination paradigm to measure the fidelity with which color information was retained in short-term memory. The task required 5 color normal observers to discriminate between spatially non-overlapping colored reference and test stimuli that were temporally separated by an ISI of 5 s. The points of subjective equality (PSEs) on the resultant psychometric matching functions provided an index of performance. Measurements were made in the presence and absence of mask stimuli presented during the ISI, which varied in hue around the equiluminant plane in DKL color space. For all reference stimuli, we found a consistent mask-induced, hue-dependent shift in PSE compared to the "no mask" conditions. These shifts were found to be tuned in color space, only occurring for a range of mask hues that fell within bandwidths of 29-37 deg. Outside this range, masking stimuli had little or no effect on measured PSEs. The results demonstrate that memory masking for color exhibits selectivity similar to that which has already been demonstrated for other visual attributes. The relatively narrow tuning of these interference effects suggests that short-term perceptual memory for color is based on higher order, non-linear color coding.