Diagnostic Accuracy of Van Herick Technique to Detect Pre-Disease States of Primary Angle Closure Glaucoma in a Resource Constraint Region

From 2 dimensions to 3rd dimension: Quantitative prediction of anterior chamber depth from anterior segment photographs via deep-learning

Anterior chamber depth (ACD) is a major risk factor of angle closure disease, and has been used in angle closure screening in various populations. However, ACD is measured from ocular biometer or anterior segment optical coherence tomography (AS-OCT), which are costly and may not be readily available in primary care and community settings. Thus, this proof-of-concept study aims to predict ACD from low-cost anterior segment photographs (ASPs) using deep-learning (DL). We included 2,311 pairs of ASPs and ACD measurements for algorithm development and validation, and 380 pairs for algorithm testing. We captured ASPs with a digital camera mounted on a slit-lamp biomicroscope. Anterior chamber depth was measured with ocular biometer (IOLMaster700 or Lenstar LS9000) in data used for algorithm development and validation, and with AS-OCT (Visante) in data used for testing. The DL algorithm was modified from the ResNet-50 architecture, and assessed using mean absolute error (MAE), coefficient-of-determination (R2), Bland-Altman plot and intraclass correlation coefficients (ICC). In validation, our algorithm predicted ACD with a MAE (standard deviation) of 0.18 (0.14) mm; R2 = 0.63. The MAE of predicted ACD was 0.18 (0.14) mm in eyes with open angles and 0.19 (0.14) mm in eyes with angle closure. The ICC between actual and predicted ACD measurements was 0.81 (95% CI 0.77, 0.84). In testing, our algorithm predicted ACD with a MAE of 0.23 (0.18) mm; R2 = 0.37. Saliency maps highlighted the pupil and its margin as the main structures used in ACD prediction. This study demonstrates the possibility of predicting ACD from ASPs via DL. This algorithm mimics an ocular biometer in making its prediction, and provides a foundation to predict other quantitative measurements that are relevant to angle closure screening.

Machine Learning-Guided Prediction of Central Anterior Chamber Depth Using Slit Lamp Images from a Portable Smartphone Device

There is currently no objective portable screening modality for narrow angles in the community. In this prospective, single-centre image validation study, we used machine learning on slit lamp images taken with a portable smartphone device (MIDAS) to predict the central anterior chamber depth (ACD) of phakic patients with undilated pupils. Patients 60 years or older with no history of laser or intraocular surgery were recruited. Slit lamp images were taken with MIDAS, followed by anterior segment optical coherence tomography (ASOCT; Casia SS-1000, Tomey, Nagoya, Japan). After manual annotation of the anatomical landmarks of the slit lamp photos, machine learning was applied after image processing and feature extraction to predict the ACD. These values were then compared with those acquired from the ASOCT. Sixty-six eyes (right = 39, 59.1%) were included for analysis. The predicted ACD values formed a strong positive correlation with the measured ACD values from ASOCT (R² = 0.91 for training data and R² = 0.73 for test data). This study suggests the possibility of estimating central ACD using slit lamp images taken from portable devices.

Non-contact tests for identifying people at risk of primary angle closure glaucoma

BACKGROUND

Primary angle closure glaucoma (PACG) accounts for 50% of glaucoma blindness worldwide. More than three-quarters of individuals with PACG reside in Asia. In these populations, PACG often develops insidiously leading to chronically raised intraocular pressure and optic nerve damage, which is often asymptomatic. Non-contact tests to identify people at risk of angle closure are relatively quick and can be carried out by appropriately trained healthcare professionals or technicians as a triage test. If the test is positive, the person will be referred for further specialist assessment.

OBJECTIVES

To determine the diagnostic accuracy of non-contact tests (limbal anterior chamber depth (LACD) (van Herick test); oblique flashlight test; scanning peripheral anterior chamber depth analyser (SPAC), Scheimpflug photography; anterior segment optical coherence tomography (AS-OCT), for identifying people with an occludable angle.

SEARCH METHODS

We searched the following bibliographic databases 3 October 2019: CENTRAL; MEDLINE; Embase; BIOSIS; OpenGrey; ARIF and clinical trials registries. The searches were limited to remove case reports. There were no date or language restrictions in the searches.

SELECTION CRITERIA

We included prospective and retrospective cross-sectional, cohort and case-control studies conducted in any setting that evaluated the accuracy of one or more index tests for identifying people with an occludable angle compared to a gonioscopic reference standard.

DATA COLLECTION AND ANALYSIS

Two review authors independently performed data extraction and quality assessment using QUADAS2 for each study. For each test, 2 x 2 tables were constructed and sensitivity and specificity were calculated. When four or more studies provided data at fixed thresholds for each test, we fitted a bivariate model using the METADAS macro in SAS to calculate pooled point estimates for sensitivity and specificity. For comparisons between index tests and subgroups, we performed a likelihood ratio test comparing the model with and without the covariate.

MAIN RESULTS

We included 47 studies involving 26,151 participants and analysing data from 23,440. Most studies were conducted in Asia (36, 76.6%). Twenty-seven studies assessed AS-OCT (analysing 15,580 participants), 17 studies LACD (7385 participants), nine studies Scheimpflug photography (1616 participants), six studies SPAC (5239 participants) and five studies evaluated the oblique flashlight test (998 participants). Regarding study quality, 36 of the included studies (76.6%) were judged to have a high risk of bias in at least one domain.The use of a case-control design (13 studies) or inappropriate exclusions (6 studies) raised patient selection concerns in 40.4% of studies and concerns in the index test domain in 59.6% of studies were due to lack of masking or post-hoc determination of optimal thresholds. Among studies that did not use a case-control design, 16 studies (20,599 participants) were conducted in a primary care/community setting and 18 studies (2590 participants) in secondary care settings, of which 15 investigated LACD. Summary estimates were calculated for commonly reported parameters and thresholds for each test; LACD ≤ 25% (16 studies, 7540 eyes): sensitivity 0.83 (95% confidence interval (CI) 0.74, 0.90), specificity 0.88 (95% CI 0.84, 0.92) (moderate-certainty); flashlight (grade1) (5 studies, 1188 eyes): sensitivity 0.51 (95% CI 0.25, 0.76), specificity 0.92 (95% CI 0.70, 0.98) (low-certainty); SPAC (≤ 5 and/or S or P) (4 studies, 4677 eyes): sensitivity 0.83 (95% CI 0.70, 0.91), specificity 0.78 (95% CI 0.70, 0.83) (moderate-certainty); Scheimpflug photography (central ACD) (9 studies, 1698 eyes): sensitivity 0.92 (95% CI 0.84, 0.96), specificity 0.86 (95% CI 0.76, 0.93) (moderate-certainty); AS-OCT (subjective opinion of occludability) (13 studies, 9242 eyes): sensitivity 0.85 (95% CI 0.76, 0.91); specificity 0.71 (95% CI 0.62, 0.78) (moderate-certainty). For comparisons of sensitivity and specificity between index tests we used LACD (≤ 25%) as the reference category. The flashlight test (grade 1 threshold) showed a statistically significant lower sensitivity than LACD (≤ 25%), whereas AS-OCT (subjective judgement) had a statistically significant lower specificity. There were no statistically significant differences for the other index test comparisons. A subgroup analysis was conducted for LACD (≤ 25%), comparing community (7 studies, 14.4% prevalence) vs secondary care (7 studies, 42% prevalence) settings. We found no evidence of a statistically significant difference in test performance according to setting. Performing LACD on 1000 people at risk of angle closure with a prevalence of occludable angles of 10%, LACD would miss about 17 cases out of the 100 with occludable angles and incorrectly classify 108 out of 900 without angle closure.

AUTHORS' CONCLUSIONS

The finding that LACD performed as well as index tests that use sophisticated imaging technologies, confirms the potential for this test for case-detection of occludable angles in high-risk populations. However, methodological issues across studies may have led to our estimates of test accuracy being higher than would be expected in standard clinical practice. There is still a need for high-quality studies to evaluate the performance of non-invasive tests for angle assessment in both community-based and secondary care settings.

LVPEI Glaucoma Epidemiology and Molecular Genetic Study: teleophthalmology screening for angle-closure disease in an underserved region

OBJECTIVES

To assess the ability of teleophthalmoscopic grading of peripheral anterior chamber depth (PACD) using the van Herick (vH) technique in detecting gonioscopically occludable angle; and to determine whether combining results from vH grading and ocular biometry can improve the accuracy to diagnose gonioscopically occludable angle METHODS: This cross-sectional study was an offshoot of a rural population-based study, Glaucoma Epidemiology and Molecular Genetic Study (GLEAMS). A masked urban ophthalmologist graded digital slit lamp photographs of PACD by vH technique. Sussman four-mirror lens was used to perform dark room indentation gonioscopy. Cutoff values of the tests were, vH technique: grade ≤ 2, central anterior chamber depth (ACD), as well as axial length: ≤ 25th percentile and lens thickness ≥ 75th percentile value of the study population.

RESULTS

We studied 1965 eyes of 1029 adult participants. The vH grade was ≤2 in 188 (9.5%) eyes. The angle was occludable by gonioscopy in 101 (5.1%) eyes. The performance of the vH test to rule out gonioscopically occludable angle was good [negative predictive value (NPV): 97.3%], despite low sensitivity (52.5%), while its efficacy to rule in the condition was low [positive predictive value (PPV): 28.2%] despite high specificity (92.8%). However, test combination strategy increased the PPV nearly twofold (53.8%). The calculated PPV at 10% prevalence of gonioscopically occludable angle was even higher (70.5%).

CONCLUSIONS

Van Herick technique can be incorporated into a teleophthalmology program by means of slit lamp photographs of PACD. Combined vH grading and ocular biometry improved the predictability of a gonioscopically occludable angle.