Visual fields in glaucoma: Where are we now?

Visual fields are an integral part of glaucoma diagnosis and management. COVID has heightened the awareness of the potential for viral spread with the practice of visual fields modified. Mask artefacts can occur due to fogging of the inferior rim of the trail lens. Fortunately, the risk of airborne transmission when field testing is low. The 24‐2c may be useful to detect early disease and the 10‐2 more sensitive to detect advanced loss. The SITA faster test algorithm is able to reduce testing time thereby improving clinic efficiency, however, may show milder results for moderate or severe glaucoma. The technician has an important role of supervising the visual field performance to achieve reliable output. Home monitoring can provide earlier detection of progression and thus improve monitoring of glaucoma as well as reduce the burden of in‐clinic assessments. Artificial Intelligence has been found to have high sensitivity and specificity compared to expert observers in detecting field abnormalities and progression as well as integrating structure with function. Although these advances will improve efficiency and guide accuracy, there will remain a need for clinicians to interpret the results and instigate management.


| PRACTICE IN THE COVID ERA
Since early 2020, the COVID-19 pandemic has changed how we practice ophthalmology. To prevent the potential spread of pathogens, minimising direct contact between patients, staff and equipment has become of chief importance. Visual field testing has come under closer scrutiny. The patient is positioned in a confined, darkened space, with their chin on the chinrest and forehead on the bar, breathing into the Ganzfeld bowl with a corrective lens in a holder in front of the eye being tested and pressing the handheld patient response button for an extended period. The other eye is occluded with an eyepatch. All these surfaces are potential sources of contamination. Field analyser companies such as Zeiss (Oberkochen, Germany) and Haag-Streit AG (Köniz, Switzerland) have published guidelines for cleaning between patients. 3,4 It is recommended that in between patients, 70% isopropyl alcohol is used to wipe all patient and technician interface surfaces, excluding the bowl. If there is a desire to clean inside the bowl, the bowl surface may be treated by spraying it with a 70% isopropyl alcohol solution. An atomizing-type sprayer is necessary to avoid drips and achieve good coverage. It is advised to avoid rubbing anything on the bowl surface as it is easily damaged. The perimeter does not require complete darkness to operate and is designed to notify the user if the room is too bright. Thus, it is possible to leave testing room doors partially or even completely open to increase the fresh airflow into the testing area.
Patients can be asked to wear masks; however proper fitting of the mask is needed to ensure an adequate seal. Mask artefacts can occur, especially in the inferior field, and have a significantly increased rate of fixation losses. 5 It is proposed the most likely reason for artefacts is the breath exhaled through the superior border of the mask, resulting in fogging of the perimeter's trial lens. This can be reduced with adhesive tape covering the superior border of the mask. 5 It is also important to note real progression of the field due to glaucoma is not dismissed as a mask artefact and that a repeat field in combination with OCT of the nerve fibre and ganglion cell layers be performed. Face shields are impractical, may interfere with the testing, and make the test even more claustrophobic.
Fortunately, there are studies to suggest that airborne transmission of COVID-19 within the field analyser bowl was low. 6 Jain et al. showed that particulate count in suspension within the bowl of a Humphrey visual field analyser (HFA) was significantly reduced by the activation of the cooling fans present within the machine and, crucially, that this particulate count is unaffected following use of the machine by a patient who is not wearing a mask. This confirms the manufacturer's claim that the ventilation current of the HFA perimeters pushes air from within the bowl past the patient. It is unlikely that whether or not a patient wears a mask during testing itself has any bearing on airborne transmission. 6 It has been suggested that the number of visual fields performed should be minimised to reduce the potential for viral transmission. The recommended time interval between visual field tests should in principle, depend on the glaucoma's severity and the predicted rate of progression. In earlier stages of glaucoma, there are defects in the OCT of the nerve fibre layer but not detectable on visual field analysis i.e. pre-perimetric disease. It is often quoted that approximately 30%-50% of retinal ganglion cells are lost before visual field defects are detected with automated perimetry. 7 This has been questioned by several authors, with support for an alternative hypothesis that before significantly detectable retinal ganglion cell loss occurs, there is a detectable loss in visual field loss as measured with perimetry. The existing retinal ganglion cells could be "sick", but not missing; thus, these structural retinal ganglion cell changes would precede the 24-2 visual field loss. 8 This reinforces the value of visual field testing for early disease, particularly more recent algorithms such as the 24-2c that test points in the macular region.
At the other end of the spectrum, the OCT of the circumpapillary retinal nerve fibre layer in advanced glaucomatous nerve loss has a floor effect or lowest possible reading of 38 μm on spectral-domain OCT 9 and 45 μm on time-domain OCT. 10 Thus, glaucoma progression in advanced cases is better monitored by visual field testing, with the 10-2 algorithm particularly useful for examining the remaining central field. The 10-2 assesses 68 points in the central 10 and is thus much more sensitive being 2 apart compared to the 24-2 that tests only 12 points in the central 10 degrees. 11 The limiting factor then is the visual acuity and once falls below 6/60, patients find the central fixation difficult, hence visual field testing below less reliable. Generally, in patients who have been first diagnosed, it is recommended to do three fields in the first 12 months to establish a baseline and then six monthly. In some patients where progression needs to be monitored more closely, for example, high intraocular pressures, secondary glaucomas with fluctuating pressures, more frequent fields can be done. Other patients who are more stable, with consistently controlled intraocular pressures may only require annual fields, mainly if the OCT of the nerve fibre layer is stable. COVID-19 has impacted ophthalmology clinics worldwide, with followup of patients greatly limited. Glaucoma clinics and fields have shown significant reductions in clinic visits, visual field exams as well as surgical procedures and medications prescribed. 12 A study from Brazil, indicated a 72.4% reduction in the number of fields performed at their institution over 1 year. This loss of adequate monitoring leads to management delays, potentially resulting in an increased presentation of more advanced glaucoma.

| VISUAL FIELD ALGORITHMS
The gold standard for perimetry for many decades was a full threshold automated visual field examination, however, the long test durations introduced eye fatigue throughout the examination and limited the possibility of frequent examinations. 13 More recent strategies, including the Swedish Interactive Threshold Algorithm (SITA), were formulated to reduce the testing duration while maintaining accuracy. 14 However, due to the continual development of new algorithms and new perimeters, it has become increasingly difficult to decide which strategy to use in practice.
Compared to its predecessor, SITA Standard (SS), the duration of visual field tests using the SITA Faster (SFer) algorithm reduces from an average test duration of 6.15-6.23 min in the older algorithm to 2.81-2.87 min. This time reduction of 53.5%-60.4% is the main advantage of using the newer SFer algorithm. [14][15][16][17][18] The disadvantage of having a shorter examination is evident as SFer has a 30.0%-49.7% rate of unreliable visual field results which is significantly higher than the SS rate at 10.8%-16.6%. Most of these unreliable results stem from seeding point errors (low sensitivity measurements at one or more of the four primary test locations of the 24-2 test grid) which lead to more severe global indices and increased clusters of threshold sensitivity reduction. Despite the differences between the two algorithms, the overall visual field parameters show good agreement, where the visual field index (VFI), foveal threshold, glaucoma hemifield test, and the number of depressed points on the deviation probability maps are similar. 14,15,17,18 Along with having similar test-retest variabilities, both SFer and SS have comparable visual field deficit sensitivities of 92.8% and 95.1% and specificities of 68.0% and 61.0%, respectively. 14,17,19,20 Although the algorithms show no significant difference in their mean deviations (MD) for mild or suspect glaucoma patients, SFer tends to show milder results for moderate or severe glaucoma. This difference may lead to disease progression being concealed when transferring from SS to SFer. 16,21,22 Other differences between the two algorithms include a significantly higher pattern standard deviation in SS and a higher threshold sensitivity for the nasal sector in SFer. 15,17 Both algorithms seem to improve in accuracy as patients undergo multiple tests, which can be seen in their high false positive rates in perimetry-naïve patients. 17 Compared to SITA Fast (SF), which is the previous alteration of the SITA family, SFer reduces test durations by 30.4%-36.5%. 13,18 While using the same 24-2 pattern, this time reduction is achieved by seven modifications which include implementing age-corrected initial stimulus intensities, reducing reversals at primary test points, adopting SF's Prior model, testing perimetric blind points once, removing false negative catch trials, using a gaze tracker, and eliminating extra delay times. 13 The two algorithms show no significant difference when it comes to MD, VFI, or the number of depressed points in their perimetry results. 13,18 However, SFer consistently shows smaller visual field deficits, making the results between the two algorithms uninterchangeable. 13 Despite this, SF and SFer have similar test-retest variability and a similarly high sensitivity of 92.8% and 92.8%, respectively. 14,19,20 A more recent version of SFer is the SITA Faster 24-2c (SFer-c) which includes 10 additional points from the 10-2 pattern to detect changes in the central visual field. 23 Due to the additional test points, SFer-c takes around 20-30 s longer. 24 Even though the two algorithms show similar global VFIs, the SFer-c algorithm has better central structure-function concordance and shows more clusters of central functional deficits. 25 When pairing perimetry with macular thickness scans, SFer-c has half the locations in the 24-2c pattern close to or outside the Ganglion Cell Analysis (GCA) grid used in most macular thickness scans. 25 There are several disadvantages when using SFer. With increased speed, SFer compromises some accuracy compared to SS mainly due to SFer not checking the blind spot and lacking false-negative catch trials. 23,26 This reduced reliability is more evident as the glaucoma state worsens and is partly due to the increase in false positives in severe glaucoma. 14,26 The 24-2 pattern applied in SFer is also not as accurate for the central visual field compared to the 10-2 pattern, which is partially implemented in SFer-c. 23,26 SFer has high false positive rates and seeding point errors, leading to a small but significant, increase in MD on intra-visit results. 27,28 Lastly, gaze tracker output in SFer provides no clinically meaningful interpretation for intra-visit visual fields. 29 Apart from the SITA strategies used in the Humphrey Fields Analyser (Zeiss, Oberkocken, Germany) different algorithms that are used in other perimeters including the Tendency Orientated Perimetry (TOP) in the Octopus (Haag-Streit USA, Mason, USA), SPARK Precision (SPARK) (Oculus, Inc, Arlington, USA), and ZETA Fast (Optopol Technology, Zawiercie. Poland). These algorithms are designed to reduce test duration and produce similar sensitivities and specificities to SF.
Developed for the Octopus perimeters, TOP uses mathematical extrapolation after testing each visual field point once. 30 This can reduce the mean test duration from 4.04 min on the SF to 2.38 min. 30 TOP and SF have high correlation coefficient values for MD measurements; however, this correlation diminishes when there are increased defects. 30 Even though TOP has been reported to underestimate focal visual loss, TOP and SF have similar sensitivities of 84.7%-85.2% and 86.4%-89.2%, and similar specificities of 76.5%-86.7% and 80.0%-93.8%, respectively. 30 SPARK is a strategy that can be implemented on the Oculus perimeters. 31 The 33% reduction in test duration compared to SS can be credited to its use of statistical relationships between neighbouring test points of the visual field, which is generated from 90 000 perimetry results using the strategy. 31 SPARK is reported to have a significantly smaller inter-visit variability and a similar sensitivity to SS. 31 ZETA Fast uses the 24-2 pattern and is performed on Optopol Technology perimeters. 32 Compared to SF, it is 23 and 49 s slower in healthy and glaucoma patients, respectively. 32 Pattern standard deviation values were higher and visual field index values were lower from ZETA Fast in glaucomatous subjects. ZETA Fast was 92.0% sensitive in capturing early glaucomatous defects with MD ≤2 when compared to SF.

| PATIENT INTOLERANCE OF FIELDS
Patients often comment on the visual field test as unfavourable. Patients find the test difficult, particularly for more advanced field loss. They comment that the lights presented are too dim or nothing seems to be happening for long periods, the chair may not be comfortable, the posturing is poor, eyes get dry from the internal fan, and claustrophobia-inducing due to the confined space. Patients have also stated how anxious they are before the test, including feeling the need to perform (performance anxiety) or indicating the amount of field they have lost (reality check). On rare occasions, patients comment they enjoy the test as it shows them how their glaucoma is going. The technician plays a key role in positioning the patient comfortably for optimal test conditions. They are present to allow the patient to pause during the test if they become fatigued or stressed, as well as to encourage the patient to maintain fixation. The appropriate near correction is critical as poor correction or misalignment results in artefacts. These need to be recognised and the test repeated to exclude actual peripheral field loss. Virtual reality goggles will potentially resolve some of these issues as outlined below.

| HOME MONITORING
In glaucoma, earlier detection of disease progression results in a more timely treatment escalation to preserve vision. Perimetry testing using standard automatic perimetry machines is the current gold standard in visual field testing, although has limitations. The test utilises significant clinic resources in terms of time and the need for dedicated staff to administer the test. In practice, this means most patients do not perform visual field tests often enough to allow for earlier detection of glaucoma, particularly given the subjective nature of visual field tests. Previous modelling showed that with a medium degree of variability in field tests, having visual fields performed six monthly requires 3 years to detect rapid visual field progression (>1 dB/year) based on the visual field alone. 33 However, if the testing frequency can be increased to once every month, rapid progression can be detected within 12 months. 34 Such a high frequency of testing is impractical in normal clinical settings with standard perimetry; however, may become a viable proposition if patients can perform the testing at home with portable devices.
Visual field testing on portable devices is not a new concept. One of the earliest forms was the Moorfields motion displacement test software utilising a laptop computer 35 and the VisualField easy app on iPad (Apple, Cupertino, USA) device, 36 both of which are suprathreshold devices with reasonable performance in terms of sensitivity and specificity in identifying manifest glaucoma. However, the lack of thresholding limits their ability to detect changes in the visual field over time. A newer testing software design has been developed to perform threshold testing on various portable devices, including tablet devices [37][38][39] and virtual reality goggles. 40 The Melbourne Rapid Fields (MRF) is software designed initially for iPad and, more recently on other flat-screen devices. It uses Bayes logic for fast threshold detection and uses variable fixation points to allow testing up to 30 of eccentricity. 37 The software produces outputs with a high correlation with HFA, with an intraclass coefficient of 0.93 for mean deviation (MD) and 0.86 for pattern deviation in clinic-based testing. 37 Test-retest reliability of MRF was also high in a longitudinal clinicbased study, suggesting that MRF is suitable for detecting a change in visual field thresholds over time. 41 A shortterm pilot study over 6 weeks was conducted to determine whether patients could perform self-directed visual field testing following short training in the clinic. 42,43 The study found that without the need for a technician administering the test and the potential variability in the home environment, there was high concordance in the testing results performed at home and those obtained from HFA in the clinic (MD R = 0.85). Furthermore, the coefficient of repeatability was better for the MRF (4.3 dB) than for the HFA (6.2 dB), indicating that the more significant number of home tests reduces the between-test variability. 42,43 A group of 47 patients performed home monitoring extended for 12 months, again with high test-retest reliability of home-based testing. 41 Analysis of the mean absolute error (MAE) suggested that a slow progressor (À0.8 dB/year) can be detected after 10 home tests, given that the MAE reduces after the initial 'learning phase'. The study also showed for the first time that home monitoring of visual fields can identify clinical progression, with two eyes having clinical progression detected by home visual field monitoring after 16 weeks of testing. 41 The study also examined compliance to home monitoring tests and showed 72% compliance to requested weekly home testing with an 87% retention rate (active testing within 28 days).
In a study conducted in the United Kingdom, 20 patients with glaucoma were recruited in a home monitoring study. Patients were asked to perform visual field testing using a thresholding visual field test on a tablet computer (Eyecatcher) once every month for the duration of 6 months. 44 The study showed visual field MD measured at home was strongly correlated with HFA in the clinic (r = 0.94, p < 0.001). Furthermore, adding the six home-monitoring data to the 2 HFA tests made 6 months apart resulted in less between-test measurement variability, with a halving of MAE in 90% of eyes. This suggests that information obtained from homebased visual field testing can provide important information that increases confidence in determining the rate of progression.
A small short-term study using a virtual reality goggle-based visual field test (Virtual Field) (VF) over 1 week was conducted by Hu et al. 40 The study examined the acceptability and feasibility of the home-based test. They found that all 20 patients in the study group felt the VF device was easy to use, and most (94.8%) would use the VF as part of future telehealth visits. They also found most patients (94.8%) felt confident at the end of the trial to use the VF without assistance. However, it is of note that patients in the study were mostly younger, with an average age of participants being 55.4 years (25-83 years), and were motivated subjects. While the study was too small to assess the validity of the visual field test performed at home, the qualitative analysis showed patients responded positively to the accessibility, comfort, and convenience benefits of home monitoring. Patients reported that home monitoring allowed them to have enhanced empowerment and foster better bonds with their physicians.
In addition to home monitoring of visual fields for glaucoma, it has also been shown to have the potential to monitor other diseases. In a short-term study in patients with age-related macular degeneration, macular field tests performed at home using tablet perimetry showed similar outcomes as clinic-based microperimetry measurements, with 55% compliance to weekly testing. 45 Home monitoring of the visual field also identified visual field defects in patients with migraine. 46 Tablet perimetry performed at the bedside could also identify undetected visual field defects in hospital stroke patients, 47 with the potential that changes in visual field defects from neural plasticity in these patients could be monitored at home.
In summary, the home monitoring of visual field tests is an exciting new development in perimetry that could reduce the burden on clinic-based resources, lead to earlier detection of visual field progression, and improvement in patient engagement with their glaucoma.

| ARTIFICIAL INTELLIGENCE AND BIG DATA
In recent years, artificial intelligence (AI) is increasingly important in ophthalmology, both in research and clinical practice. Many aspects of ophthalmology are driven by the analysis and interpretation of data and imagebased investigations, which can be automated and enhanced by AI algorithms. AI encompasses many algorithms, including machine learning, artificial neural networks, and deep learning. 48 The general principle of AI algorithms involves training the algorithm with large quantities of data in a supervised or unsupervised fashion, then testing the algorithm on a separate validation dataset.
Most of the current AI research in glaucoma involves the analysis of structural information such as fundus photography or OCT measurements. However, AI could be applied to perimetry in several ways. It could assist in the rapid interpretation of the visual field in differentiating between disease and normality, and collate large datasets of visual fields to classify patterns of defects as well as patterns of progression. AI could be used to integrate perimetry with structural measurements (from OCT), with improvements in visual field testing and disease progression prediction.
In terms of the application of AI for visual field interpretation, previous groups showed neural networks can be used to classify visual fields into glaucomatous and non-glaucomatous fields with a high degree of accuracy. It can sometimes be better than human experts and standard visual field classification measures. 49,50 Interestingly, Asaoka et al. showed in pre-perimetric glaucoma eyes where the visual field defect did not meet Anderson-Patella's criteria for manifest glaucoma, a feed-forward neural network was able to differentiate between glaucoma and normal field with the area under the receiver operating characteristic curve (AUC) of 92.6%. 51 This suggests the early features of glaucomatous visual field change can be detectable by AI at the pre-perimetric stage and thus may assist clinicians in making better predictive decisions.
In the area of classification of large datasets, AI could provide researchers with a better understanding of disease progression over a longer time. [52][53][54] The analysis of this type of big data, if performed manually, would be time and labour-intensive. Wang et al. described an unsupervised AI technique to analyse the progression of the visual field based on 16 recognised patterns of visual field defect (archetype analysis). 53 A total of 11 817 eyes were analysed with this method, which showed 99% of progression occurred with three or fewer archetypal patterns. The most common archetypal pattern was decreased normal pattern (63.7%), with other patterns being increased nasal step (16.4%), altitudinal loss (15.9%), superior-peripheral defect (12.1%), paracentral/ central defect (10.5%), and near total loss (10.4%). The archetypal analysis correlates more with clinical progression from expert assessment than to traditional criteria or trend analysis. Another study using a spatial-ordinal convolutional neural network model was applied to 9212 eyes of 6047 patients who underwent regular, reliable visual field examinations for >4 years. 55 The neural network model outperformed the linear regression model in detecting progression with higher sensitivity and specificity. Similarly, Yousefi et al. found a machine learningbased index for glaucoma progression detection was able to identify visual field progression earlier than using trend-based methods using global, region-wise, and point-wise indices. 52 Another exciting application of AI is in integrating perimetry with structural analysis. Visual field global indices such as MD and VFI could be predicted using convolutional neural networks. 56,57 Other groups showed convolutional neural network architecture could be trained to predict areas of visual field defect using OCT images from peripapillary retinal nerve fibre layer thicknesses (RNFL) 58 or combined OCT images from macular ganglion cell-inner plexiform layer and RNFL thicknesses. 59 One application would be to feed the predicted visual field into perimetry testing as inputs, resulting in increased testing accuracy and speeds. Furthermore, studies have shown that by combining perimetry and structural measurements, AI could be used to diagnose glaucoma with a high level of diagnostic accuracy; this could assist clinicians in screening large numbers of patients. 60,61 7 | CONCLUSION Assessment of the visual field and progression detection has been an ongoing challenge. Experienced clinicians on a daily basis, study the visual field test, looking at quality indices such as fixation, false positives, and negatives and comparing them to previous tests to decide whether the patient's glaucoma is stable or if there are signs of progression and therefore requiring a change in management. This needs to be done in combination with the other clinical indicators of intraocular pressure, nerve fibre layer thickness, and compliance factors. Current visual field algorithms are improving the efficiency of visual field data acquisition without sacrificing accuracy. This also enables better patient tolerance of the test and the ability to perform more tests to detect change. Home monitoring of visual fields, to the standard of in-clinic field analysis, is a significant innovation, especially in this COVID-19 era of reduced access to regular clinic visits and late presentation of more advanced diseases. Artificial intelligence and big data have the potential to detect more cases of glaucoma at an earlier stage as well as monitor progression in existing patients. This will generate the need for even more clinicians to interpret the results and instigate management, albeit with improved efficiency and guided accuracy.