Understanding accommodative control in the clinic: Modeling latency and amplitude for uncorrected refractive error, presbyopia and cycloplegia by Read JCA, Maus G, Schor CM, ReadMausSchor2024.pdf (3.9 MiB) - Accommodation is the process of adjusting the eye's optical power so as to focus at different distances. Uncorrected refractive error and/or functional presbyopia mean that sharp focus may not be achievable for some distances, so observers experience sustained defocus. Here, we identify a problem with current models of accommodative control: They predict excessive internal responses to stimuli outside accommodative range, leading to unrealistic adaptation effects. Specifically, after prolonged exposure to stimuli outside range, current models predict long latencies in the accommodative response to stimuli within range, as well as unrealistic dynamics and amplitudes of accommodative vergence innervation driven by the accommodative neural controller. These behaviors are not observed empirically. To solve this issue, we propose that the input to blur-driven accommodation is not retinal defocus, but correctable defocus. Predictive models of accommodative control already estimate demand from sensed defocus, using a realistic “virtual plant” to estimate accommodation. Correctable defocus can be obtained by restricting this demand to values physically attainable by the eye. If we further postulate that correctable defocus is computed using an idealized virtual plant that retains a young accommodative range, we can explain why accommodative–convergence responses are observed for stimuli that are too near—but not too far—to focus on. We model cycloplegia as a change in gain, and postulate a form of neural myopia to explain the additional relaxation of accommodation often seen with cycloplegia. This model produces plausible predictions for the accommodative response and accommodative convergence signal in a wide range of clinically relevant situations
The Use of Eye-tracking Technology in Cleft Lip: A Literature Review by Plonkowski AT, Breakey RW, Read JCA, Sainsbury DCG, PlonkowskiBreakeyReadSainsbury2023.pdf (0.9 MiB) - Background: Eye-tracking has become an increasingly popular research tool within the field of cleft lip and/or palate (CL+/−P). Despite this, there are no standardized protocols for conducting research. Our objective was to conduct a literature review of the methodology and outcomes of previous publications using eye-tracking
in CL+/−P.
Methods: The PubMed, Google Scholar, and Cochrane databases were searched to identify all articles published up to August 2022. All articles were screened by two independent reviewers. Inclusion criteria included using eye-tracking, image stimuli of CL+/−P, and outcome reporting using areas of interest (AOIs). Exclusion criteria included non-English studies, conference articles, and image stimuli of conditions other than CL+/−P.
Results: Forty articles were identified, and 16 met the inclusion/exclusion criteria. Thirteen studies only displayed images of individuals following cleft lip surgery with three only displaying unrepaired cleft lips. Significant variation was found in study design, particularly in the AOIs used to report gaze outcomes. Ten studies asked participants to provide an outcome score alongside eye-tracking; however, only four compared outcome data to eye-tracking data. This review is primarily limited by the minimal number of publications in this area.
Conclusions: Eye-tracking can be a powerful tool in evaluating appearance outcomes following CL+/−P surgery. It is currently limited by the lack of standardized research methodology and varied study design. Before future work, a replicable protocol should be developed to maximize the potential of this technology.
CNV-Net: Segmentation, Classification and Activity Score Measurement of Choroidal Neovascularization (CNV) Using Optical Coherence Tomography Angiography (OCTA) by Vali M, Nazari B, Sadri S, Khalili Pour E, Riazi-Esfahani H, Faghihi H, Ebrahimiadib N, Azizkhani M, Innes W, Steel DH, Hurlbert AC, Read JCA, Kafieh R , ValiNazariSadriEAReadKafieh2023.pdf (4.7 MiB) - This paper aims to present an artificial intelligence-based algorithm for the automated segmentation of Choroidal Neovascularization (CNV) areas and to identify the presence or absence of CNV activity criteria (branching, peripheral arcade, dark halo, shape, loop and anastomoses) in OCTA images. Methods: This retrospective and cross-sectional study includes 130 OCTA images from 101 patients with treatment-naïve CNV. At baseline, OCTA volumes of 6 × 6 mm2 were obtained to develop an AI-based algorithm to evaluate the CNV activity based on five activity criteria, including tiny branching vessels, anastomoses and loops, peripheral arcades, and perilesional hypointense halos. The proposed algorithm comprises two steps. The first block includes the pre-processing and segmentation of CNVs in OCTA images using a modified U-Net network. The second block consists of five binary classification networks, each implemented with various models from scratch, and using transfer learning from pre-trained networks. Results: The proposed segmentation network yielded an averaged Dice coefficient of 0.86. The individual classifiers corresponding to the five activity criteria (branch, peripheral arcade, dark halo, shape, loop, and anastomoses) showed accuracies of 0.84, 0.81, 0.86, 0.85, and 0.82, respectively. The AI-based algorithm potentially allows the reliable detection and segmentation of CNV from OCTA alone, without the need for imaging with contrast agents. The evaluation of the activity criteria in CNV lesions obtains acceptable results, and this algorithm could enable the objective, repeatable assessment of CNV features.
(This article belongs to the Special Issue Advances in Optical Coherence Tomography Angiography)
Predicting attitudes towards screening for neurodegenerative diseases using OCT and artificial intelligence: Findings from a literature review by Nichol BAB, Hurlbert AC, Read JCA, NicholHurlbertRead2022.pdf (0.2 MiB)
Synthetic OCT Data Generation to Enhance the Performance of Diagnostic Models for Neurodegenerative Diseases by Danesh H, Steel DH, Hogg J, Ashtari F, Innes WF, Bacardit, J, Hurlbert A, Read JCA, Kafieh R, DaneshMaghooliDehghaniKafieh2021.pdf (3.1 MiB) - Purpose: Optical coherence tomography (OCT) has recently emerged as a source for powerful biomarkers in neurodegenerative diseases such as multiple sclerosis (MS) and neuromyelitis optica (NMO). The application of machine learning techniques to the analysis of OCT data has enabled automatic extraction of information with potential to aid the timely diagnosis of neurodegenerative diseases. These algorithms require large amounts of labeled data, but few such OCT data sets are available now.
Methods: To address this challenge, here we propose a synthetic data generation method yielding a tailored augmentation of three-dimensional (3D) OCT data and preserving differences between control and disease data. A 3D active shape model is used to produce synthetic retinal layer boundaries, simulating data from healthy controls (HCs) as well as from patients with MS or NMO.
Results: To evaluate the generated data, retinal thickness maps are extracted and evaluated under a broad range of quality metrics. The results show that the proposed model can generate realistic-appearing synthetic maps. Quantitatively, the image histograms of the synthetic thickness maps agree with the real thickness maps, and the cross-correlations between synthetic and real maps are also high. Finally, we use the generated data as an augmentation technique to train stronger diagnostic models than those using only the real data.
Conclusions: This approach provides valuable data augmentation, which can help overcome key bottlenecks of limited data.
Translational Relevance: By addressing the challenge posed by limited data, the proposed method helps apply machine learning methods to diagnose neurodegenerative diseases from retinal imaging.
ASTEROID stereotest v1.0: lower stereo thresholds using smaller, denser and faster dots by Read JCA, Wong ZY, Yek X, Wong YX, Bachtoula O, Llamas-Cornejo I, Serrano-Pedraza, ReadWongYekWongBachtoulaLlamasCornejoSerranoPedraza2020_compressed.pdf (0.6 MiB) - Purpose: In 2019, we described ASTEROID, a new stereotest run on a 3D tablet
computer which involves a four-alternative disparity detection task on a dynamic
random-dot stereogram. Stereo thresholds measured with ASTEROID were well
correlated with, but systematically higher than (by a factor of around 1.5), thresholds
measured with previous laboratory stereotests or the Randot Preschool clinical
stereotest. We speculated that this might be due to the relatively large, sparse
dots used in ASTEROID v0.9. Here, we introduce and test the stereo thresholds
and test-repeatability of the new ASTEROID v1.0, which uses precomputed
images to allow stereograms made up of much smaller, denser dots.
Methods: Stereo thresholds and test/retest repeatability were tested and compared
between the old and new versions of ASTEROID (n = 75) and the Randot Circles
(n = 31) stereotest, in healthy young adults.
Results: Thresholds on ASTEROID v1.0 are lower (better) than on ASTEROID
v0.9 by a factor of 1.4, and do not differ significantly from thresholds on the Randot
Circles. Thresholds were roughly log-normally distributed with a mean of
1.54 log10 arcsec (35 arcsec) on ASTEROID v1.0 compared to 1.70 log10 arcsec
(50 arcsec) on ASTEROID v0.9. The standard deviation between observers was
the same for both versions, 0.32 log10 arcsec, corresponding to a factor of 2 above
and below the mean. There was no difference between the versions in their test/
retest repeatability, with 95% coefficient of repeatability = 0.46 log10 arcsec (a
factor of 2.9 or 1.5 octaves) and a Pearson correlation of 0.8 (comparable to other
clinical stereotests).
Conclusion: The poorer stereo thresholds previously reported with ASTEROID
v0.9 appear to have been due to the relatively large, coarse dots and low density
used, rather than to some other aspect of the technology. Employing the small
dots and high density used in ASTEROID v1.0, thresholds and test/retest repeatability
are similar to other clinical stereotests.
Stereotest Comparison: Efficacy, Reliability, and Variability of a New Glasses-Free Stereotest by McCaslin AG, Vancleef K, Hubert L, Read JCA, Port NL, McCaslinVancleefHubertReadPort2020_compressed.pdf (0.6 MiB) - Purpose: To test the validity of the ASTEROID stereotest as a clinical test of depth perception by comparing it to clinical and research standard tests.
Methods: Thirty-nine subjects completed four stereotests twice: the ASTEROID test on an autostereo 3D tablet, a research standard on a VPixx PROPixx 3D projector, Randot Circles, and Randot Preschool. Within 14 days, subjects completed each test for a third time.
Results: ASTEROID stereo thresholds correlated well with research standard thresholds (r = 0.87, P < 0.001), although ASTEROID underestimated standard threshold (mean difference = 11 arcsec). ASTEROID results correlated less strongly with Randot Circles (r = 0.54, P < 0.001) and Randot Preschool (r = 0.64, P < 0.001), due to the greater measurement range of ASTEROID (1–1000 arcsec) compared to Randot Circles or Randot Preschool. Stereo threshold variability was low for all three clinical stereotests (Bland–Altman 95% limits of agreement between test and retest: ASTEROID, ±0.37; Randot Circles, ±0.24; Randot Preschool, ±0.23). ASTEROID captured the largest range of stereo in a normal population with test–retest reliability comparable to research standards (immediate r = 0.86 for ASTEROID vs. 0.90 for PROPixx; follow-up r = 0.68 for ASTEROID vs. 0.88 for PROPixx).
Conclusions: Compared to clinical and research standards for assessing depth perception, ASTEROID is highly accurate, has good test–retest reliability, and measures a wider range of stereo threshold.
Translational Relevance: The ASTEROID stereotest is a better clinical tool for determining baseline stereopsis and tracking changes during treatment for amblyopia and strabismus compared to current clinical tests.
Characterizing the Randot Preschool stereotest: Testability, norms, reliability, specificity and sensitivity in children aged 2-11 years by Read JCA, Rafiq S, Hugill J, Casanova T, Black C, O'Neill A, Puyat V, Haggerty H, Smart K, Powell C, Taylor K, Clarke MP, Vancleef K, ReadEARandotPreschool2019e.pdf (1.8 MiB) - Purpose
To comprehensively assess the Randot Preschool stereo test in young children, including testability, normative values, test/retest reliability and sensitivity and specificity for detecting binocular vision disorders.
Methods
We tested 1005 children aged 2–11 years with the Randot Preschool stereo test, plus a cover/uncover test to detect heterotropia. Monocular visual acuity was assessed in both eyes using Keeler Crowded LogMAR visual acuity test for children aged 4 and over.
Results
Testability was very high: 65% in two-year-olds, 92% in three-year-olds and ~100% in older children. Normative values: In 389 children aged 2–5 with apparently normal vision, 6% of children scored nil (stereoblind). In those who obtained a threshold, the mean log threshold was 2.06 log10 arcsec, corresponding to 114 arcsec, and the median threshold was 100 arcsec. Most older children score 40 arcsec, the best available score. We found a small sex difference, with girls scoring slightly but significantly better. Test/retest reliability: ~99% for obtaining any score vs nil. Agreement between stereo thresholds is poor in children aged 2–5; 95% limit of agreement = 0.7 log10 arcsec: five-fold change in stereo threshold may occur without any change in vision. In children over 5, the test essentially acts only as a binary classifier since almost all non-stereoblind children score 40 arcsec. Specificity (true negative rate): >95%. Sensitivity (true positive rate): poor, <50%, i.e. around half of children with a demonstrable binocular vision abnormality score well on the Randot Preschool.
Conclusions
The Randot Preschool is extremely accessible for even very young children, and is very reliable at classifying children into those who have any stereo vision vs those who are stereoblind. However, its ability to quantify stereo vision is limited by poor repeatability in children aged 5 and under, and a very limited range of scores relevant to children aged over 5.
Visual Perception: Monovision Can Bias the Apparent Depth of Moving Objects by Read JCA, ReadMonovision2019.pdf (0.6 MiB) - ‘Monovision’ — using one eye for near work and one for distance — is a common alternative to reading
glasses. New work shows that monovision can cause the distance of moving objects to be misestimated,
with potentially serious consequences
Two choices good, four choices better: For measuring stereoacuity in children, a four-alternative forced-choice paradigm is more efficient than two by Vancleef K, Read JCA, Herbert W, Goodship N, Woodhouse M, Serrano-Pedraza I, VancleefReadHerbertGoodshipWoodhouseSerranoPedraza2018.pdf (4.5 MiB) - Purpose
Measuring accurate thresholds in children can be challenging. A typical psychophysical
experiment is usually too long to keep children engaged. However, a reduction in the number of trials decreases the precision of the threshold estimate. We evaluated the efficiency
of forced-choice paradigms with 2 or 4 alternatives (2-AFC, 4-AFC) in a disparity detection
experiment. 4-AFC paradigms are statistically more efficient, but also more cognitively
demanding, which might offset their theoretical advantage in young children.
Methods
We ran simulations evaluating bias and precision of threshold estimates of 2-AFC and 4-
AFC paradigms. In addition, we measured disparity thresholds in 43 children (aged 6 to 17
years) with a 4-AFC paradigm and in 49 children (aged 4 to 17 years) with a 2-AFC paradigm, both using an adaptive weighted one-up one-down staircase.
Results
Simulations indicated a similar bias and precision for a 2-AFC paradigm with double the number of trials as a 4-AFC paradigm. On average, estimated threshold of the simulated data was equal to the model threshold, indicating no bias. The precision was improved with an increasing number of trials. Likewise, our data showed a similar bias and precision for a 2-AFC paradigm with 60 trials as for a 4-AFC paradigm with 30 trials. Trials in the 4-AFC paradigm took slightly longer as participants scanned more alternatives. However, the 4-AFC task still ended up faster for a given precision.
Conclusion
Bias and precision were similar in a 4-AFC task compared to a 2-AFC task with double the number of trials. However, a 4-AFC paradigm was more time efficient and is therefore recommended
