Explore the vast range of titles available.

In this manuscript we develop a deep learning algorithm to improve estimation of rates of progression and prediction of future patterns of visual field loss in glaucoma. A generalized variational auto-encoder (VAE) was trained to learn a low-dimensional representation of standard automated perimetry (SAP) visual fields using 29,161 fields from 3,832 patients. The VAE was trained on a 90% sample of the data, with randomization at the patient level. Using the remaining 10%, rates of progression and predictions were generated, with comparisons to SAP mean deviation (MD) rates and point-wise (PW) regression predictions, respectively. The longitudinal rate of change through the VAE latent space (e.g., with eight dimensions) detected a significantly higher proportion of progression than MD at two (25% vs. 9%) and four (35% vs 15%) years from baseline. Early on, VAE improved prediction over PW, with significantly smaller mean absolute error in predicting the 4 th , 6 th and 8 th visits from the first three (e.g., visit eight: VAE8: 5.14 dB vs. PW: 8.07 dB; P < 0.001). A deep VAE can be used for assessing both rates and trajectories of progression in glaucoma, with the additional benefit of being a generative technique capable of predicting future patterns of visual field damage.

Artificial intelligence capabilities have, recently, greatly improved. In the past few years, one of the deep learning algorithms, the recurrent neural network (RNN), has shown an outstanding ability in sequence labeling and prediction tasks for sequential data. We built a reliable visual field prediction algorithm using RNN and evaluated its performance in comparison with the conventional pointwise ordinary linear regression (OLR) method. A total of 1,408 eyes were used as a training dataset and another dataset, comprising 281 eyes, was used as a test dataset. Five consecutive visual field tests were provided to the constructed RNN as input and a 6 th visual field test was compared with the output of the RNN. The performance of the RNN was compared with that of OLR by predicting the 6 th visual field in the test dataset. The overall prediction performance of RNN was significantly better than OLR. The pointwise prediction error of the RNN was significantly smaller than that of the OLR in most areas known to be vulnerable to glaucomatous damage. The RNN was also more robust and reliable regarding worsening in the visual field examination. In clinical practice, the RNN model can therefore assist in decision-making for further treatment of glaucoma.

BackgroundMacular Integrity Assessment (MAIA) microperimetry is increasingly used in clinical and research settings to assess point retinal sensitivity and fixation stability. Testing occurs under mesopic conditions, commonly after a period of dark adaptation. Our aim was to identify the minimum length of adaptation required to optimise microperimetry performance.MethodsMAIA microperimetry using the 10-2 grid was performed on 40 right eyes of 40 healthy participants aged 18–73 with no ocular pathology and vision of at least 0.1 logMAR after ambient light exposure, with 0, 5, 10, 15, 20 and 30 min of adaptation in mesopic settings. Ten right eyes of 10 participants with choroideremia were also tested following 0 and 20 min of adaptation. We further tested 10 right eyes of 10 healthy participants after bright light exposure, with 0, 10 and 20 min of adaptation. We compared changes in threshold sensitivity and fixation stability across time points.ResultsMicroperimetry performance did not improve with increasing adaptation time in healthy participants or patients with choroideremia after ambient light exposure. After bright light exposure, we found microperimetry thresholds improved after 10 min of adaptation, but did not improve further at 20 min.ConclusionMesopic adaptation is not required before MAIA microperimetry after exposure to ambient light. Ten minutes of adaptation is sufficient after exposure to a bright light stimulus, such as ophthalmoscopy or retinal imaging. The brief time of dark adaptation required corresponds to cone adaptation curves and provides further evidence for cone-mediated central retinal function under mesopic conditions.

To determine if deep learning networks could be trained to forecast future 24-2 Humphrey Visual Fields (HVFs). All data points from consecutive 24-2 HVFs from 1998 to 2018 were extracted from a university database. Ten-fold cross validation with a held out test set was used to develop the three main phases of model development: model architecture selection, dataset combination selection, and time-interval model training with transfer learning, to train a deep learning artificial neural network capable of generating a point-wise visual field prediction. The point-wise mean absolute error (PMAE) and difference in Mean Deviation (MD) between predicted and actual future HVF were calculated. More than 1.7 million perimetry points were extracted to the hundredth decibel from 32,443 24-2 HVFs. The best performing model with 20 million trainable parameters, CascadeNet-5, was selected. The overall point-wise PMAE for the test set was 2.47 dB (95% CI: 2.45 dB to 2.48 dB), and deep learning showed a statistically significant improvement over linear models. The 100 fully trained models successfully predicted future HVFs in glaucomatous eyes up to 5.5 years in the future with a correlation of 0.92 between the MD of predicted and actual future HVF and an average difference of 0.41 dB. Using unfiltered real-world datasets, deep learning networks show the ability to not only learn spatio-temporal HVF changes but also to generate predictions for future HVFs up to 5.5 years, given only a single HVF.

This study compared between TEMPO, a new binocular perimeter, with the Humphrey Field Analyzer (HFA). Patients were tested with both TEMPO 24–2 Ambient Interactive Zippy Estimated by Sequential Testing (AIZE)-Rapid and HFA 24–2 Swedish Interactive Threshold Algorithm (SITA)-Fast in a randomized sequence on the same day. Using a mixed-effects model, visual field (VF) parameters and reliability indices were compared. Retinal nerve fiber layer (RNFL) thickness was measured using Cirrus optical coherence tomography (OCT), and coefficient of determinations for VF and OCT parameters were calculated and compared using Akaike information criteria. 740 eyes (including 68 healthy, 262 glaucoma suspects, and 410 glaucoma) of 370 participants were evaluated. No significant differences were seen in mean deviation and visual field index between the two perimeters (P > 0.05). A stronger association between VF mean sensitivity (dB or 1/L) and circumpapillary RNFL was found for TEMPO (adjusted R 2  = 0.25; Akaike information criteria [AIC] = 5235.5 for dB, and adjusted R 2  = 0.29; AIC = 5200.8 for 1/L, respectively) compared to HFA (adjusted R 2  = 0.22; AIC = 5263.9 for dB, and adjusted R 2  = 0.22; AIC = 5262.7 for 1/L, respectively). Measurement time was faster for TEMPO compared to HFA (261 s vs. 429 s, P < 0.001). Further investigations are needed to assess the long-term monitoring potential of this binocular VF test.

The aims of this randomized observational case control study were to quantify fixation behavior during standard automated perimetry (SAP) with different fixation targets and to evaluate the relationship between fixation behavior and threshold variability at each test point in healthy young participants experienced with perimetry. SAP was performed on the right eyes of 29 participants using the Octopus 900 perimeter, program 32, dynamic strategy. The fixation targets of Point, Cross, and Ring were used for SAP. Fixation behavior was recorded using a wearable eye-tracking glass. All participants underwent SAP twice with each fixation target in a random fashion. Fixation behavior was quantified by calculating the bivariate contour ellipse area (BCEA) and the frequency of deviation from the fixation target. The BCEAs (deg2) of Point, Cross, and Ring targets were 1.11, 1.46, and 2.02, respectively. In all cases, BCEA increased significantly with increasing fixation target size (p < 0.05). The logarithmic value of BCEA demonstrated the same tendency (p < 0.05). A positive correlation was identified between fixation behavior and threshold variability for the Point and Cross targets (ρ = 0.413-0.534, p < 0.05). Fixation behavior increased with increasing fixation target size. Moreover, a larger fixation behavior tended to be associated with a higher threshold variability. A small fixation target is recommended during the visual field test.

To investigate the luminance capacity and achievable threshold levels of commercially employed virtual reality (VR) devices for visual field testing. This two-part study included (1) a literature review of VR headsets used for perimetry with luminance data extracted from technical specifications in publications and manufacturers; and (2) empirical evaluation of three most employed VR headsets in the literature using a custom virtual testing environment. Three most employed VR devices for visual field testing were Pico Neo, Oculus Quest, and HTC Vive. The maximum reported luminance was 250 cd/m2 for the HTC Vive Pro. Information on luminance measurement was not consistently available, reporting only handheld luminance meters. Empirical measurements show that handheld luminance meters significantly overestimate luminance compared to standard spectroradiometers. Measured luminance varies significantly across aperture size and decreases for peripheral stimuli up to 30 degrees peripherally. Assuming conventional background of 10 cd/m2, the best performance with lowest possible thresholding was with HTC Vive at 16dB, corresponding to luminance of 80 cd/m2 centrally. Oculus Quest 2 and Pico Neo 3 had minimum threshold of 20dB. Commercially available VR devices do not meet luminance requirements or threshold sensitivities for visual field testing. Current VR technology is not designed-nor has the capacity-to threshold at mid-to-low dB ranges, which limits accuracy in diagnosing and monitoring defects seen in glaucoma. Translational Relevance: This study highlights the technical limitations of current commercially available VR devices for visual field testing and significant variables in evaluating luminance performance in these devices.

The IMOvifa™ perimeter with a 24plus (1-2) testing mode has additional measurement points within the central 10 degrees, which may help evaluate the visual field within this area. Here, we comparatively evaluated the IMOvifa™ 24plus (1-2) and HFA 10-2 for the first time. We included 30 patients (48 eyes) who underwent HFA 24-2 Swedish Interactive Threshold Algorithm Standard and IMOvifa™ 24plus (1-2) Ambient Interactive Zippy Estimated tests on the same day and HFA 10-2 within six months. We used Spearman's rank correlation coefficient to analyze the mean deviation (MD) and pattern standard deviation (PSD) between HFA 10-2 and IMOvifa™. The central 10-degree visual field was divided into four sectors, and concordance of visual field defects between IMOvifa™ 24plus (1-2) and HFA 10-2 was evaluated using kappa analysis. Additionally, all sectors showing a sensitivity of 0 dB on the HFA 24-2 were assessed for the presence and agreement of residual visual field in HFA 10-2 and IMOvifa™ 24plus (1-2). The MD (0.843/0.804) and PSD (0.852/0.763) of IMOvifa™ 24plus (1-2) and HFA 24-2 correlated strongly with those of HFA 10-2. Regarding the ability to detect visual field defects within the central 10 degrees, agreement with HFA 10-2 was κ = 0.715 (0.611,0.819) and 0.754 (0.654,0.854) for IMOvifa™ 24plus (1-2) and HFA 24-2, respectively. In the evaluation of residual visual field, IMOvifa™ 24plus (1-2) detected residual visual function in 100% of cases where HFA 10-2 indicated residual function. The IMOvifa™ 24plus (1-2) may have a higher ability to detect defects in certain areas of the visual field, compared with HFA 24-2, and may also detect residual visual function. However, the IMOvifa™ 24plus (1-2) is difficult to substitute for the 10-2 test, as the 10-2 test is necessary for evaluating visual field defects within the central 10 degrees.

We developed a new portable head-mounted perimeter, \"imo\", which performs visual field (VF) testing under flexible conditions without a dark room. Besides the monocular eye test, imo can present a test target randomly to either eye without occlusion (a binocular random single eye test). The performance of imo was evaluated. Using full HD transmissive LCD and high intensity LED backlights, imo can display a test target under the same test conditions as the Humphrey Field Analyzer (HFA). The monocular and binocular random single eye tests by imo and the HFA test were performed on 40 eyes of 20 subjects with glaucoma. VF sensitivity results by the monocular and binocular random single eye tests were compared, and these test results were further compared to those by the HFA. The subjects were asked whether they noticed which eye was being tested during the test. The mean sensitivity (MS) obtained with the HFA highly correlated with the MS by the imo monocular test (R: r = 0.96, L: r = 0.94, P < 0.001) and the binocular random single eye test (R: r = 0.97, L: r = 0.98, P < 0.001). The MS values by the monocular and binocular random single eye tests also highly correlated (R: r = 0.96, L: r = 0.95, P < 0.001). No subject could detect which eye was being tested during the examination. The perimeter imo can obtain VF sensitivity highly compatible to that by the standard automated perimeter. The binocular random single eye test provides a non-occlusion test condition without the examinee being aware of the tested eye.

To assess whether machine learning algorithms (MLA) can predict eyes that will undergo rapid glaucoma progression based on an initial visual field (VF) test. Retrospective analysis of longitudinal data. 175,786 VFs (22,925 initial VFs) from 14,217 patients who completed ≥5 reliable VFs at academic glaucoma centers were included. Summary measures and reliability metrics from the initial VF and age were used to train MLA designed to predict the likelihood of rapid progression. Additionally, the neural network model was trained with point-wise threshold data in addition to summary measures, reliability metrics and age. 80% of eyes were used for a training set and 20% were used as a test set. MLA test set performance was assessed using the area under the receiver operating curve (AUC). Performance of models trained on initial VF data alone was compared to performance of models trained on data from the first two VFs. Accuracy in predicting future rapid progression defined as MD worsening more than 1 dB/year. 1,968 eyes (8.6%) underwent rapid progression. The support vector machine model (AUC 0.72 [95% CI 0.70-0.75]) most accurately predicted rapid progression when trained on initial VF data. Artificial neural network, random forest, logistic regression and naïve Bayes classifiers produced AUC of 0.72, 0.70, 0.69, 0.68 respectively. Models trained on data from the first two VFs performed no better than top models trained on the initial VF alone. Based on the odds ratio (OR) from logistic regression and variable importance plots from the random forest model, older age (OR: 1.41 per 10 year increment [95% CI: 1.34 to 1.08]) and higher pattern standard deviation (OR: 1.31 per 5-dB increment [95% CI: 1.18 to 1.46]) were the variables in the initial VF most strongly associated with rapid progression. MLA can be used to predict eyes at risk for rapid progression with modest accuracy based on an initial VF test. Incorporating additional clinical data to the current model may offer opportunities to predict patients most likely to rapidly progress with even greater accuracy.