Supervised Automated Kinetic Perimetry (SAKP) Using Simulated Visual Field Data – Presentation of a New Examination Technique

Abstract

Purpose The aim of this study was to develop, optimise, train, and evaluate an algorithm for performing Supervised Automated Kinetic Perimetry (SAKP) using digitalised perimetric simulation data.

Methods The original SAKP algorithm was based on findings from a multicentre study to establish reference values by semi-automated kinetic perimetry (SKP) combined with an automated examination method with moving stimuli (“Program K”, developed in Japan). The algorithm evaluated the outer angles of isopter segments and responded to deviations from expected values by placing examination vectors to measure the outer boundaries of the visual field (VF). Specialised interpolation methods were also used to create individual 3D hills of vision and local “probing vectors” to optimise the eccentricity of the vector origins. This algorithm was trained iteratively on seven representative digitalised 3D VF results from five typical classes and optimised in each step: (1) Normal VF, (2) Central scotoma, (3) Concentric VF constriction, (4) Retinal nerve fibre layer defects in the visual field (VFDs), (5) VFDs with respect to the vertical meridian. The optimised SAKP algorithm was then applied to a new set of twenty 3D VF results of varying origin and severity. The primary targets were measured in agreement between actual calculated VF expressed as accuracy (A), that is, the ratio between the area containing correct predictions and total area of predictions measured between 0 = worst and 1 = best, and examination duration (T). The results are given as median (and interquartile range). We also verified the testʼs robustness by varying individual error rates (ERs) and error magnitudes (EMs).

Results The median and interquartile range (IQR, in brackets) for the total of representative VFs were 0.93 (0.02) for A and 7.0 min (5.2 min) for T, respectively. A gave the best result for altitudinal VFDs and VFDs with hemianopic character and macular sparing (0.98 each) and worst in superior wedge-shaped VFDs (0.78); T was shortest in blind spot displacement (3.9 min) and longest in hemianopic VFDs with hemianopic character and macular sparing with preserved temporal crescent (12.1 min). Error rate and magnitude (up to 30% each) only showed a comparatively low influence on A and T.

Conclusion The SAKP algorithm presented here achieves a comparatively high degree of accuracy and robustness for actual, simulated visual field data within acceptable examination times. This algorithm is currently being prepared for application in real patient examinations under clinical conditions.

Keywords

kinetic perimetry - automated kinetic perimetry - automation - quality - accuracy - examination times

General note: For brevity, the indefinite third-person pronoun they is used to refer to any person or persons of any number or gender.

Introduction

More than 90% of human sensory perception involves sight [1], [2]. This places outstanding importance on assessing the visual system regarding topographic and clinical diagnosis, monitoring, and evaluation of visual pathway lesions, especially in determining disabilities or disqualifications for certain activities. Visual field defects affect around 3 – 3.5% of the total population; this figure increases steeply with age to 13% of the population aged 65 and over [3]. The essential role of sensory-physiological functional diagnostics – especially visual field examinations, or perimetry – is becoming increasingly apparent as second-generation diagnostic tools even with the great success of morphometric procedures with anatomical structure assessment. Morphometric methods are likely to play a major role in early detection; their comparatively low dynamic range place these methods in a clearly inferior position to perimetric methods, especially in advanced lesions. According to current awareness, functional examination results on visual pathway lesions are the gold standard especially in official assessments in determining fitness to drive/driving licence assessment, setting degrees of disability and disability allowances.

The visual field describes the sum of all visual sensory impressions perceived when looking straight ahead; the measurement parameter used is usually the differential luminance sensitvity (DLS), that is, contrast between the luminance of a stimulus (measuring mark) and homogeneous surroundings at specified visual field locations.

Static perimetry methods use stationary light stimuli that ‘flash’ briefly in certain luminance levels at predetermined locations in the field of vision (the so-called examination grid). This is currently preferred as a procedure as it is easily automated and can be delegated to practice staff without extensive training. However, a disadvantage is the fact that in the context of static, threshold-determining perimetry, a large proportion of the stimuli presented cannot be perceived due to methodological reasons (bracketing procedure). This is both frustrating and tiring for patients. This phenomenon increases with increasing severity of the visual field loss.

Kinetic perimetry methods operate with moving stimuli on the principle of detecting scotoma edges; they work even on large, deep visual field defects and are clearly more efficient and patient-friendly than static procedures. Standardised kinetic procedures are also likely to play a major role in clinical trials, such as in gauging the success of molecular and gene therapy studies; this especially applies given that such studies usually involve comparatively large defects or small remaining visual field islands that need to be preserved as long as possible with documentation that is as reliable as possible over time. Moving stimuli also reflect the clinical relevance of visual impairment far more effectively than static stimuli – stationary objects relative to road users rarely pose an immediate danger in real life by comparison to moving objects.

Professional associations such as the German Society of Ophthalmology (DOG) and German ophthalmologistsʼ association (BVA) recommend kinetic examinations with “manually guided” moving stimuli on a Goldmann perimeter or equivalent instruments with Goldmann properties III4e (visual angle 26′, luminance 320 cd/m²) in the case of expert analyses [4], [5]. The original device has unfortunately not been manufactured for years, so the trained personnel required in manual kinetic perimetry are hard to come by in our region.

The aim of this project is to develop, optimise, and validate a supervised automated kinetic perimetry (SAKP) algorithm for automatic (bowl) perimetry with kinetic functionality available on the market for supervised, fully automated, and therefore examiner-independent perimetry using stimuli moving at a predefined constant angular velocity.

Methods

Own preparations

The authors have been engaged in optimising static and kinetic perimetric examination methods for more than thirty years. Developments over this period include accelerated threshold-based techniques for static perimetry, including individualised grid compression in regions of interest (SCotoma-Oriented PErimetry, SCOPE). These techniques have been validated in clinical studies and international multicentre normative studies with participating centres receiving training and certification [6], [7], [8], [9], [10], [11]. These findings have led to structure-function interactions being demonstrated and visualised using a mathematical model specifically designed for nerve fibre layer defects [12], [13], [14], [15]. Expertise from one perimetric reading centre (PerCent-A) has been used several times in multicentre studies, and will also be used in future SAKP validation [16], [17].

The authors standardised the visual field examination using moving stimuli by introducing vector-based semi-automated kinetic perimetry (SKP) that includes individual reaction times; multicentre normative studies and subsequent clinical studies are also available here [18], [19], [20], [21], [22].

A classification scheme and corresponding summary of topographic diagnostic criteria have been developed for perimetrically documented visual field defects [18], [23], [24].

Brief description of the SAKP algorithm

The algorithm used here, supervised automated kinetic perimetry (SAKP), uses the principle of self-developed, vector-based semi-automated kinetic perimetry (SKP) system; this technique has found its way into commercially available devices, and age-matched standard values are also available for it [20].

The algorithm also uses procedural steps from Program K, which was developed by the Matsumoto working group [25]. This algorithm evaluates deviations in the outer angles from expected local values in the isopter course and uses the result for automatic vector placement. However, this approach can be applied to isopters on the outer boundaries of the visual field only, and thus cannot detect internal scotomas within the visual field.

The newly developed SAKP algorithm mainly uses automated, moving stimuli relevant to expert opinion (Goldmann III4e, corresponding to 26′, 320 cd/m²; constant angular velocity 4°/s) whose direction of movement is intended in such a way that the presumed scotoma boundary is intersected approximately perpendicularly. Radial basis functions (RBFs) are used to generate individual three-dimensional visual field results [26]. “Probe vectors” are used to shift and optimise the origin of the kinetic stimuli towards the centre or periphery depending on the patientʼs individual response behaviour. This approach also aims towards keeping the distances between isopters as even as possible across the patientʼs visual field. The operator can preselect the number of isopters (minimum: three isopters). The software tool automatically selects the most reasonable combination of stimulus size and luminance [6], [7]. The aim here is to position the innermost isopter within the eccentricity of the blind spot. Static stimuli are used in circumscribed regions of interest usually within in the central visual field (< 30°) to detect small visual field defects within this region, or where automated moving stimuli might not deliver a conclusive result due to an atypical isopter course. A Voronoi diagram is used to estimate the extent of a suspected scotoma, which is then tested using kinetic vectors [27], [28].

The examination procedure halts for both moving and static stimuli if the SAKP algorithm anticipates a course completely within 30° eccentricity from the isopter course; this allows the supervising examination staff to place an adequate near-correction lens in front.

A modified Heijl-Krakau technique is used to detect the blind spot [29]; this procedure is also designed to capture translational and rotational shifts in the blind spot. The SAKP algorithm switches to the most suitable stimulus property, Goldmann I4e (corresponding to 6.5′, 320 cd/m²) at a reduced stimulus angular velocity (2°/s) to measure the blind spot position using automatically moving stimuli. All examination steps are stored with timestamps to allow for more precise optimisation as necessary.

The SAKP algorithm also interacts with the examining staff when measuring the patientʼs individual reaction time, that is, motivation to respond as quickly as possible while maintaining exact fixation.

Initial SAKP algorithm training

We challenged the new algorithm with seven representative high-resolution kinetic perimetric results (training cases) from the K-Train perimetry training software tool (see [Fig. 1], left third) in the initial phase of development. These involved a normal perimetric finding, a central scotoma, two different forms of concentric visual field defects with a complete ring scotoma and a concentric visual field restriction with preserved peripheral residual islands, two different degrees of nerve fibre layer defects, and a representative visual field defect respecting the vertical meridian (upper nasal quadrantanopia).

Fig. 1 Representation of initial training cases (n = 7, left third of the image) and the consecutive cases (n = 20, right part of the image); classification scheme from perimetric results and scotoma forms modified according to [24]; with regard to the colour coding of the height of the hill of vision, see legend [Fig. 3].

The algorithm was applied to a total in twenty cases in the consecutive phase (see [Fig. 1], right part of the image).

Quality assessment

Quality assessment includes quotients from intersections and unions in correctly determined regions, that is, scotomas recorded by both SAKP and actual underlying scotomas (shown in blue) divided by areas that can be assigned to at least one of these two regions. Areas that can be assigned to both perimetrically identified and actual underlying normal visual field areas (shown in grey) are then added, and then divided by areas that can be assigned to at least one of these two aformentioned regions (see [Fig. 2]). The quotient (accuracy, A) in the above procedure will range in values between 0.0 and 1.0 for complete lack of spatial coincidence to perfect spatial agreement, respectively.

Fig. 2 Quality assessment in perimetric examination results: Sum of quotients (also referred to as accuracy, A) from intersections on either actual visual field defects or normal visual field regions each correctly identified by perimetry, with scotomas shown in blue and healthy visual field regions in grey, and the union area of the above-mentioned area proportions in the underlying technically analysable observation area marked by the dashed purple line showing localisation agreement.

In principle, this method can also be used to analyse eccentricity-related visual field areas (such as: 10° – 30° or 30° – 60°).

The examination is repeated as a further quality criterion to test for reproducibility in each of the two perimetry methods (see also [Fig. 6]).

[Fig. 3] shows an example of an original three-dimensionally reconstructed and two-dimensionally colour-coded visual field results with a near-central absolute superior arcuate scotoma including the blind spot using the K-Train software tool (right half of the image) in comparison with the SAKP finding result (left half of the image).

Fig. 3 Right: Semi-automated kinetic perimetry (SKP) finding showing near-central arcuate absolute glaucomatous visual field defect (white area within the paracentral superior hemifield above); the following colour coding indicates the height of the original three-dimensional hill of vision. Violet corresponds to the maximum differential luminance sensitivity (DLS) of 30 dB; white encodes absolute defects (LDS < 20 dB). A dashed blue line shows the isopter course for the stimulus III4e as relevant in official assessment. Left: Final perimetric findings using results from the SAKP algorithm (SAKP: supervised automated kinetic perimetry, as the end result of an examination procedure visualised in time-lapse mode); agreement ratio as accuracy (A) for both visual field as a whole and eccentricity-related sub-regions is shown in the upper section of this Figure; the lower right section of this Figure shows the number of stimuli, examination duration, age of patient examined (in years), and (coloured) line symbols for the various Goldmann stimuli. The two small framed images in the lower right show the icons displayed during the SAKP examination process; these icons prompt removal or insertion of near correction lenses; see https://www.vision-research.de/sakp/ for an animated presentation of the examination process.

[Fig. 4] visualises how error rate (ER, from 10% in the top row to 30% in the bottom row) and error magnitude (EM, from ± 5 s in the left column to ± 20 s in the right column) affect the agreement ratio (AR) between visual field results obtained using SAKP (right part of the image) and error-free results (large image, left half of the image): Remarkably, this automated examination technique was able to detect and display quadrantanopia even at maximum error rate and magnitude (small image bottom right).

Fig. 4 Effect of error rate (ER) and error magnitude (EM) on visual field results using SAKP. The left (green box) shows the original error-free result: Green and grey areas show areas of agreement between simulated visual field results and SAKP result; red areas visualise deviations. The right half of the Figure shows perimetric results at increasing error rate from top to bottom (ER: 10% → 20% → 30%) and error magnitude increasing from left to right (EM: ± 5 s → ± 10 s → ± 20 s).

[Fig. 5 a] depicts a summary for the twenty SAKP results with corresponding accuracy (A) shown as box-and-whisker plots depending on the extent of error; the accuracy is generally high (median > 0.8). We saw only a slight overall decrease in accuracy with increasing error rate and magnitude.

Fig. 5 a Effect of increasing error rate and magnitude on the accuracy (A) in visual field results using SAKP for twenty digitalised visual field results with colour-coded sensitivity representations, see right third of the image (see also [Fig. 4]). The far left shows the original results without any errors; the error rate and magnitude increase gradually towards the right. The results are presented as box-and-whisker plots: The individual measurements are shown as black dots connected by thin black lines for visual field results in each individual test patient; the median is shown by a red horizontal line; the upper and lower edges of the box represent the 75th and 25th percentiles, respectively, to show spread (and therefore also the interquartile range, IQR). The whiskers extend 1.5 times the IQR from the upper and lower distribution ends of the box and represent the minimum and maximum values except where outliers extend beyond the ends of the whiskers. Any outliers beyond the ends of the whiskers are correspondingly plotted as points. The corresponding visual field results are shown as examples overlaying the right third of the image as an example for three such outliers. [Fig. 5] b Effect of increasing error rate and magnitude on examination time (exam time, T) of visual field results determined using SAKP for twenty digitalised visual field results; the corresponding visual field results are shown as examples overlaying the right third of the image for four outliers; see also [Fig. 4], with the far left showing original error-free results, error rate and magnitude increasing gradually towards the right. See [Fig. 5] a for an explanation of the box-and-whisker plots.

[Fig. 5 b] demonstrates that with a few exceptions, the critical examination time of 15 min was not reached for the twenty SAKP results even at increasing error levels. Median exam time was less than 11 minutes in all conditions.

In anticipation of future study projects, [Fig. 6] shows a comparison between visual field results produced manually using the Goldmann perimeter method (MGKP, upper half of the image) and the automated method (SAKP) (left: baseline finding, right: follow-up) in an ophthalmologically healthy subject. All results for accuracy (A) are above the critical threshold (A = 0.8). There is a clear learning effect on the part of the operator for manual-kinetic Goldmann examinations as reflected in the significant decrease in examination time T from the initial 31 min to the subsequent 17 min. In contrast, no notable learning effect was seen in the automated kinetic perimetry (SAKP) previously performed at 13 min vs. 12 min.

Fig. 6 Exemplary comparison of baseline (left half of the image) and follow-up (right half of the image) visual field examinations; first using the SAKP algorithm (SAKP 1 or SAKP 2, lower half of the image), then using classic manual-kinetic Goldmann perimetry (MGKP 1 or MGKP 2, upper half of the image) on a ophthalmologically healthy person. The corresponding agreement results in terms of accuracy (A), see also [Fig. 2], are shown in relation to comparisons between manual (MGKP) and automated (SAKP) perimetric results from the information in the green ovals and method-specific test-retest reliability (reproducibility) results in orange ovals as well as graphically from the squares with the green and orange edges; red areas visualise differences between the visual field results. The respective examination duration (T) is shown in blue lettering and the corresponding time of examination (begin, to determine the examination sequence) in grey lettering.

Discussion

Functional diagnostics vs. morphometry

Morphometric testing is increasingly gaining focus; these examinations give comparatively rapid results, are easy to delegate, and do not require the patient to keep focused over extended periods. The ophthalmological field has a long history of attempting to establish direct quantifiable relationships between morphological and functional testing [30]. However, morphometric methods mainly measure and record variables with linear scaling such as thickness and distance. However, reduction by e. g. a factor of 10 may transcend the measurement limits in the morphometric devices used. In contrast, results from sensory-physiological procedures allow direct conclusions on activities of daily living due to their comparatively high dynamic range, and are therefore predominantly scaled logarithmically like almost all psychometric tests: Perimetry usually covers a differential luminance sensitivity range of at least 30 dB, which corresponds to a factor of at least a thousand based on corresponding stimulus luminance values. It comes as no surprise that functional testing is seen as authoritative in official assessments for social, administrative, and legal purposes, at least in the German-speaking world. Perimetric examination methods using moving stimuli are also considered as the gold standard in assessing hereditary retinal diseases in English-language literature [31]. These considerations are set to become increasingly important in complex ophthalmological treatment procedures in the future such as biological, stem-cell, and gene therapy as well as surgical procedures – provided that these perimetric methods produce coherent standardised results independent of the examiner. This provides a compelling reason for more research and development in largely automated – and also kinetic – perimetry techniques.

Disadvantages of conventional Goldmann kinetic perimetry

As already discussed at the beginning, the original instrument was technically complex and has not been in production for years. The pantograph mechanism leads to a comparatively poor spatial resolution in the important central visual field while also impeding sufficient documentation of the examination procedure. The incandescent light bulbs used in these devices were subject to ageing and dimming over time, needing regular recalibration. There was no way of automating or documenting this process. General consensus exists on minimum examination standards, which include: Background luminance of 10 cd/m², mandatory use of III4e stimuli (26′, stimulus luminance 320 cd/m²) suitable for expert examinations, and at least three additional stimuli that should be distributed as equal as possible across the hill of vision with the innermost stimulus lying within an eccentricity angle of 15°. The stimulus movements should always be directed from non-sighted to sighted areas and as perpendicular as possible to the expected scotoma boundary at a constant angular velocity (minimum 2°/s to maximum 5°/s) [31], [32]. There is no way of documenting compliance with these specifications and standards using a Goldmann perimeter. Any results from using such a device will therefore be completely subject to the expertise and arbitrary discretion of the examiner. The examination result itself is only available as a diagram and can at best be stored digitally by scanning the image. This poses challenges in meeting the explicit requirements on perimetric examinations in legal contexts according to Bernhard Lachenmayr, spokesperson for the German DOG and BVA traffic commission; e-mail correspondence from 20 June 2023. Well-trained and experienced perimetry personnel able to meet the complex requirements of everyday clinical practice, including examinations on patients with disabilities, is now becoming increasingly rare.

Further development in kinetic perimetry methods

There were initial attempts to control the pantograph with a motorised or pneumatically driven robot arm as early as the 1980s – the Perikon Optikon (OPTIKON 2000 S.p.A., Rome, Italy). Applying kinetic examination techniques as automated digital perimeters has enabled standardised movement of stimuli at specified intensity along predetermined or individually placed vectors at constant angular velocities in semi-automated kinetic perimetry (SKP) [20], [33]. However, SKP also requires considerable perimetric expertise, which is becoming increasingly rare (see above). Continuing these developments towards a largely examiner-independent, automated perimetry technique with moving stimulus markers would therefore appear to be essential.

Quality assurance

The authors implemented an algorithm for the quality assessment and certification of the trainees towards the introduction of an interactive training software tool – K-Train – for SKP examinations on virtual patients. The quality criterion used was the agreement ratio (AR) – the quotient of scotoma intersections and unions from the given (K-Train) visual field results and the respective perimetric results – as already introduced in the K-Train software tool. The agreement ratio covered the overall visual field results as well as eccentricity-related sub-regions.

Only comparing areas with scotomas individually detected by perimetry and actual underlying scotomas is not sufficient without some kind of coordinates to localise the scotoma. This means that scotomas identified using different methods may appear congruent but rotated against each other or located at completely different places within the visual field.

Instead, the authors calculated the quotient of the intersection and union of scotomas determined by different methods in an earlier publication to describe the quality (for a specific stimulus property) and referred to it as the agreement ratio (AR) [34]. This approach is one-sided as it only evaluates the sensitivity of a procedure – in this case, the ability to detect visual field defects as a functional result of a visual pathway lesion. So specificity, another important quality criterion for a diagnostic procedure, is missing. Specificity signifies the ability to identify normal, healthy age-appropriate visual field areas. Accuracy (A) as implemented in the present contribution combines these two quality criteria (see also [Fig. 2]), allowing an adequate assessment of the SAKP algorithm quality. This approach allows different perimetry methods to be compared against each other in an immediately comprehensible form and quantified as a measurement, regardless of shape, extent, and location of defects detected by either method. This approach should also prove highly beneficial to quality assessment as required in future perimetry procedures, including AI-supported methods. Projection-related differences should also be taken into account as appropriate in documenting the perimetric results [31], [35].

Outlook

The next study phase should test the SAKP algorithm on a total of forty-eight patients with visual field defects of different shapes and degrees of severity under clinical examination conditions; the results should then be compared against results from conventional manual-kinetic Goldmann perimetry. Using the term “validation” would be euphemistic in this context, as classifying visual field results from a Goldmann perimeter as ground truth or gold standard does not seem justified in view of the considerable limitations in this device (see also the introductory passage in this Discussion). To make matters worse, manual kinetic perimetry also involves examiner-related errors that cannot be fully accounted for due to the lack of documentation possibilities available using this device.

Repetitions of both SAKP and conventional Goldmann examinations would allow conclusions on reproducibility and therefore spread attributable to each of the two procedures; however, this approach cannot provide any information on precision due to the above problems intrinsic to manual-kinetic Goldmann perimetry – inadequate ground truth quality, therefore lack of exact target criterion (cf. DIN ISO 5725 – 1) [36].

The order of paired examinations using Goldmann perimetry and SAKP should be randomised to cancel out sequence effects such as learning effect and fatigue as far as possible. Eliminating the pairing and completely randomising all four examinations would have been even more thorough. However, this approach would involve more frequent and logistically complex device changes inevitably extending exam duration, which might adversely affect examination quality.

Another clinical study on semi-automated kinetic perimetry (SKP) using modern, self-calibrating automatic perimeters with manually positioned vectors may mitigate the above problems, at least the insufficient suitability for documentation and standardisation [31], [37].

The operatorʼs level of qualification and experience plays a key role in any of the (manual-kinetic) perimetric reference methods listed so far, unlike fully automated kinetic perimetry. This issue alongside unavoidable patient-related factors such as fatigue, learning effect, and attention and concentration deficits interferes with an experimental design that lays claim to an undisputed “ground truth.”

A differentiated result assessment regarding primary target and quality parameters (accuracy, A and examination duration, T) appears beneficial in quantitatively evaluating kinetic visual field results on an automatic perimeter using the SAKP algorithm introduced here:

1. Results without any intervention by the examiner (that is, fully automated kinetic perimetry, FAKP)

2. Results with subsequent examiner-assisted manual isopter course correction without perimetric follow-up examination (SAKP_I)

3. Results with subsequent examiner-assisted manual local follow-up examination using kinetic, vector-based or static stimuli in a follow-up examination, including any resulting subsequent isopter course correction (SAKP_S)

A further development on SAKP may independently indicate local ambiguities or uncertainties in generating continuous closed isopter courses, giving the supervising examination staff an opportunity to intervene by closing the isopter courses themselves (see point 2) or resolving the above ambiguities and uncertainties by manually positioning additional vectors or stimuli (see point 3).

At another development stage, SAKP should be optimised for perimetric monitoring in extensive visual field defects. It would make follow-up checks more efficient if the algorithm did not always assume age-corrected normal hills of vision, instead generating findings based on individual kinetic, static and, if necessary, morphometric preliminary examination results towards optimised vector positioning.

A three-dimensional view of visual field results would seem beneficial as it would enable a comparative, quantitative analysis of both static and kinetic examination results in the form of scotoma volumetry at certain points in time and as it develops over time [38]. Weleber et al. laid the foundations for a three-dimensional approach as early as in 1986 [35]. These authors especially emphasised that observing scotomas in the planar visual field image would lead to distortions compared to the corresponding areas on a perimeter bowl. The “solid angles” they described could therefore be used in the future instead of areas detected on planar surfaces when quantifying agreement between Goldmann and automated results.

The SAKP algorithm should use as many visual field results from as many people as possible in everyday clinical practice while respecting data privacy and security to generate a continuously improving, self-learning, self-optimising system [39], [40].

To this end, there are plans to transfer the SAKP results obtained using this method into a database in anonymised form with the consent of the patients examined and participating ophthalmological institutions. This would enable improvement in the algorithm itself and the evaluation and visualisation of results while also analysing and optimising weak points. The e-mail addresses of institutions submitting the results should be deleted to prevent the data records transmitted from being traced back. In addition, personal data would need to be obfuscated for data minimisation [40]) by reducing granularity in relevant transmitted information as far as possible; this would include transforming refraction information from 0.25 (original) to 1 dpt increments, cylinder axis positions from 1° (original) to 10° increments, visual acuity information to 3 dB intervals (visual acuity data in steps of 3 visual acuity lines), exact examination date into quarters, and age information into age groups spanning ten years each.

Participating patients could be given information material and commented printouts of the results as an incentive, and the license fees for the SAKP software could be reduced for participating ophthalmological institutions.

A future AI-supported automatic scotoma classification system could use this extended dataset in a subsequent development to quantify the probability of assignment to (7) scotoma classes: normal findings, nerve fibre layer defects, central/centrocoecal scotomas, changes around the blind spot, wedge/sector-shaped defects, concentric constrictions, and visual field defects respecting the vertical meridian; see also [Fig. 1] [41], [42]. The prerequisite for this is individually adequate vector positioning that intersects the scotoma boundaries as perpendicularly as possible (see above) and is not limited to exclusively centripetal vector orientations [43], [44].

The algorithm could be developed further towards efficient binocular SAKP examination based on previous monocular SAKP results. The union (Boolean OR) in previously determined outer boundaries of the visual field in right and left eyes could be used as a guide for positioning vectors running towards the centre for coherent automated vector placement. The intersection (Boolean AND) in the previously determined circumscribed scotomas in the right and left eyes within the outer boundaries could be used as a guide for positioning the vector origins for stimuli to moving out from the scotoma regions to estimate the expected positions of circumscribed binocular scotomas located within the outer boundaries.

Conclusion

The SAKP algorithm presented here achieves a comparatively high degree of accuracy and robustness with respect to the actual simulated visual field data while remaining within acceptable examination times. This algorithm is currently being prepared for application in real patient examinations under clinical conditions.

Conclusion

• Automation and therefore standardisation in kinetic perimetry procedures is possible and can be applied successfully to representative simulated visual field defects.

• The accuracy of the procedure may be quantified in a comprehensible manner by using intersection and union sets from (determined and predefined/simulated) visual field results.

• A comparison between an automated perimetry algorithm under clinical conditions against conventional manual kinetic methods is in preparation.

Conflict of Interest

Ulrich SCHIEFER acted as a consultant for HAAG-STREIT, Köniz/CH and SERVIER, Paris/FR. Ulrich SCHIEFER's research team received an IIT (Investigator-initiated Trial/Unrestricted Research Grant) from JOHNSON & JOHNSON Vision/AMO, Dublin/IRE and a research grant from the German Social Accident Insurance (DGUV). This perimetric field of activity is currently supported by the Pro Retina - Stiftung Deutschland eV, Bonn/D. Ulrich SCHIEFER holds several patents, preferably in the field of sensory physiology. He regularly holds courses in the field of refraction determination and perimetry and has been a speaker for many years at events supported by well-known pharmaceutical companies (including PharmAllergan, Ettlingen/Germany, Alcon, Freiburg/Germany, MSD, Munich/Germany, Pfizer, Berlin/Germany).
Michael WÖRNER is Managing Director of Blickshift GmbH, Stuttgart/Germany. He is co-owner of patents, preferably in the field of sensory physiology. His work in the current perimetric field is currently funded by the Pro Retina - Stiftung Deutschland eV, Bonn/Germany.
Ditta ZOBOR declares no conflicts of interest in connection with this publication.

Acknowledgements

This project is supported by the foundation Pro Retina – Stiftung Deutschland eV, Bonn, Germany. The company Haag-Streit AG, Köniz/CH has supported a preliminary project in connection with this topic. We would like to take this opportunity to thank Judith Ungewiß from the Centre for Optical Technologies (ZOT), Aalen University of Applied Sciences, Germany, for her support in this project.
We would also like to thank Prof. Dr. Peter Martus, Director of the Institute for Clinical Epidemiology and Applied Biometry (IKEaB) at the University of Tübingen, for his advice during the design phase of this project.
The authors would further like to thank Mr. Rainer Schönen (MSc), BfDI (Federal Commissioner for Data Protection and Information Security), Bonn, for his valuable advice and suggestions regarding aspects related to data privacy and security.

• References/Literatur

• 1 Taylor JF. Vision and driving. Practitioner 1982; 226: 885-889
• 2 Enoch J, McDonald L, Jones L. et al. Evaluating Whether Sight Is the Most Valued Sense. JAMA Ophthalmol 2019; 137: 1317-1320
• 3 Johnson CA, Keltner JL. Incidence of visual field loss in 20,000 eyes and its relationship to driving performance. Arch Ophthalmol 1983; 101: 371-375
• 4 Deutsche Ophthalmologische Gesellschaft (DOG), Berufsverband der Augenärzte Deutschlands (BVA). Fahreignungsbegutachtung 2019. 7. Aufl.. München: Deutsche Ophthalmologische Gesellschaft; 2019
• 5 Rohrschneider K, Tost F. Verkehrstauglichkeit beim Glaukom. Ophthalmologie 2023; 120: 1071-1084
• 6 Nevalainen J, Paetzold J, Papageorgiou E. et al. Specification of progression in glaucomatous visual field loss, applying locally condensed stimulus arrangements. Graefes Arch Clin Exp Ophthalmol 2009; 247: 1659-1669
• 7 Schiefer U, Pascual JP, Edmunds B. et al. Comparison of the new perimetric “German Adaptive Threshold Estimation” (GATE) strategy with conventional full-threshold and SITA Standard strategies. Invest Ophthalmol Vis Sci 2009; 50: 488-494
• 8 Schiefer U, Papageorgiou E, Sample PA. et al. Spatial pattern of glaucomatous visual field loss obtained with regionally condensed stimulus arrangements. Invest Ophthalmol Vis Sci 2010; 51: 5685-5689
• 9 Weleber RG, Smith TB, Peters D. et al. VFMA: Topographic Analysis of Sensitivity Data From Full-Field Static Perimetry. Transl Vis Sci Technol 2015; 4: 14
• 10 Zrenner E, Holder GE, Schiefer U. et al. Quality Control Procedures and Baseline Values for Electroretinography, Perimetry, Color Vision, and Visual Acuity in an International Multicenter Study: Observations from a Safety Trial in Chronic Stable Angina Pectoris. Transl Vis Sci Technol 2020; 9: 38
• 11 Zrenner E, Holder GE, Schiefer U. et al. Three-Year Changes in Visual Function in the Placebo Group of a Randomized Double-Blind International Multicenter Safety Study: Analysis of Electroretinography, Perimetry, Color Vision, and Visual Acuity in Individuals With Chronic Stable Angina Pectoris. Transl Vis Sci Technol 2022; 11: 2
• 12 Jansonius NM, Nevalainen J, Selig B. et al. A mathematical description of nerve fiber bundle trajectories and their variability in the human retina. Vision Res 2009; 49: 2157-2163
• 13 Jansonius NM, Schiefer J, Nevalainen J. et al. A mathematical model for describing the retinal nerve fiber bundle trajectories in the human eye: Average course, variability, and influence of refraction, optic disc size and optic disc position. Exp Eye Res 2012; 105: 70-78
• 14 Hood DC, Nguyen M, Ehrlich AC. et al. A Test of a Model of Glaucomatous Damage of the Macula With High-Density Perimetry: Implications for the Locations of Visual Field Test Points. Transl Vis Sci Technol 2014; 3: 5
• 15 Qiu K, Schiefer J, Nevalainen J. et al. Influence of the Retinal Blood Vessel Topography on the Variability of the Retinal Nerve Fiber Bundle Trajectories in the Human Retina. Invest Ophthalmol Vis Sci 2015; 56: 6320-6325
• 16 Diem R, Molnar F, Beisse F. et al. Treatment of optic neuritis with erythropoietin (TONE): a randomised, double-blind, placebo-controlled trial–study protocol. BMJ Open 2016; 6: e010956
• 17 Lagrèze WA, Küchlin S, Ihorst G. et al. Safety and efficacy of erythropoietin for the treatment of patients with optic neuritis (TONE): a randomised, double-blind, multicentre, placebo-controlled study. Lancet Neurol 2021; 20: 991-1000
• 18 Schiefer U, Wilhelm H. [Visual field compendium. Interpretation of perimetry findings. Comprehensive diagnostic measures]. Klin Monbl Augenheilkd 1995; 206: 206-238
• 19 Dolderer J, Vonthein R, Johnson CA. et al. Scotoma mapping by semi-automated kinetic perimetry: the effects of stimulus properties and the speed of subjectsʼ responses. Acta Ophthalmol Scand 2006; 84: 338-344
• 20 Vonthein R, Rauscher S, Paetzold J. et al. The normal age-corrected and reaction time-corrected isopter derived by semi-automated kinetic perimetry. Ophthalmology 2007; 114: 1065-1072
• 21 Nevalainen J, Paetzold J, Krapp E. et al. The use of semi-automated kinetic perimetry (SKP) to monitor advanced glaucomatous visual field loss. Graefes Arch Clin Exp Ophthalmol 2008; 246: 1331-1339
• 22 Wetzel D, Ungewiss J, Wörner M. et al. Dissociation between red and white stimulus perception: A perimetric quantification of protanopic color vision deficiencies. PLoS One 2021; 16: e0260362
• 23 Schiefer U, Wilhelm H, Hart W. Clinical Neuro-Ophthalmology. Berlin: Springer; 2007
• 24 Ungewiss J, Schiefer U. Perimetrie in der neuroophthalmologischen Funktionsdiagnostik: Indikation – Methoden – Topodiagnostik. Klin Monbl Augenheilkd 2018; 235: 1218-1228
• 25 Hashimoto S, Matsumoto C, Okuyama S. et al. Development of a New Fully Automated Kinetic Algorithm (Program K) for Detection of Glaucomatous Visual Field Loss. Invest Ophthalmol Vis Sci 2015; 56: 2092-2099
• 26 Buhmann MD. Radial basis functions. Acta Numer 2000; 9: 1-38
• 27 Voronoi MG. Nouvelles applications des paramètres continus à la théorie des formes quadratiques. J Reine Angew Math 1908; 134: 198-287
• 28 Aurenhammer F. Voronoi diagrams – a survey of a fundamental geometric data structure. ACM Comput Surv 1991; 23: 345-405
• 29 Heijl A, Krakau CET. A Note on Fixation During Perimetry. Acta Ophthalmol (Copenh) 1977; 55: 854-861
• 30 Harwerth RS, Carter-Dawson L, Shen F. et al. Ganglion cell losses underlying visual field defects from experimental glaucoma. Invest Ophthalmol Vis Sci 1999; 40: 2242-2250
• 31 Barnes CS, Schuchard RA, Birch DG. et al. Reliability of Semiautomated Kinetic Perimetry (SKP) and Goldmann Kinetic Perimetry in Children and Adults With Retinal Dystrophies. Transl Vis Sci Technol 2019; 8: 36
• 32 Johnson CA, Wall M, Thompson HS. A history of perimetry and visual field testing. Optom Vis Sci 2011; 88: E8
• 33 Schiefer U, Rauscher S, Hermann A. et al. Realization of semi-automated kinetic perimetry (SKP) with Interzeag 101 instrument. In: Perimetry Update 2002/2003. The Hague, The Netherlands: Kugler Publications; 2003: 233-238
• 34 Schiefer U, Nowomiejska K, Krapp E. et al. K-Train – a computer-based, interactive training program with an incorporated certification system for practicing kinetic perimetry: evaluation of acceptance and success rate. Graefes Arch Clin Exp Ophthalmol 2006; 244: 1300-1309 Faraj
• 35 Weleber RG, Tobler WR. Computerized quantitative analysis of kinetic visual fields. Am J Ophthalmol 1986; 101: 461-468
• 36 Deutsches Institut für Normung. DIN ISO 5725-1: 1997-11, Genauigkeit (Richtigkeit und Präzision) von Meßverfahren und Meßergebnissen – Teil 1: Allgemeine Grundlagen und Begriffe (ISO 5725-1: 1994). https://www.dinmedia.de/de/norm/din-iso-5725-1/3050093 Stand: 2024-10-04
• 37 Nowomiejska KE, Vonthein R, Paetzold J. et al. Comparison between semiautomated kinetic perimetry and conventional Goldmann manual kinetic perimetry in advanced visual field loss. Ophthalmology 2005; 112: 1343-1354
• 38 Weleber RG, Smith TB, Peters D. et al. VFMA: Topographic Analysis of Sensitivity Data From Full-Field Static Perimetry. Transl Vis Sci Technol 2015; 4: 14
• 39 Aichroth P, Battis V, Dewes A. et al. Anonymisierung und Pseudonymisierung von Daten für Projekte des maschinellen Lernens. 2020. Accessed October 04, 2024 at: https://www.bitkom.org/sites/default/files/2020-10/201002_lf_anonymisierung-und-pseudonymisierung-von-daten.pdf
• 40 Toma M, Bönisch C, Löhnhardt B. et al. Research collaboration data platform ensuring general data protection. Sci Rep 2024; 14: 11887
• 41 Yadav SS, Jadhav SM. Deep convolutional neural network based medical image classification for disease diagnosis. J Big Data 2019; 6: 113
• 42 Cai L, Gao J, Zhao D. A review of the application of deep learning in medical image classification and segmentation. Ann Transl Med 2020; 8: 713
• 43 Mönter VM, Crabb DP, Artes B. Reclaiming the periphery: Automated kinetic perimetry for measuring peripheral visual fields in patients with glaucoma. Invest Ophthalmol Vis Sci 2017; 58: 868-875
• 44 Nowomiejska J, Wrobel-Dudzinska K, Ksaizek K, Ksziazek P, Rejdak K, Maciejewski R, Juenemann AG, Rejdak R. Semi-automated kinetic perimetry provides additional information to static automated perimetry in the assessment of the remaining visual field in end-stage glaucoma. Ophthalmic Physiol Opt 2015; 35: 147-154

Publication History

Received: 13 June 2024

Accepted: 23 September 2024

Article published online:
06 December 2024