The Open Perimetry Initiative: A framework for cross-platform development for the new generation of portable perimeters

The Open Perimetry Initiative was formed in 2010 with the aim of reducing barriers to clinical research with visual fields and perimetry. Our two principal tools are the Open Perimetry Interface (OPI) and the visualFields package with analytical tools. Both are fully open source. The OPI package contains a growing number of drivers for commercially available perimeters, head-mounted devices, and virtual reality headsets. The visualFields package contains tools for the analysis and visualization of visual field data, including methods to compute deviation values and probability maps. We introduce a new frontend, the opiApp, that provides tools for customization for visual field testing and can be used as a frontend to run the OPI. The app can be used on the Octopus 900 (Haag-Streit), the Compass (iCare), the AP 7000 (Kowa), and the IMO (CREWT) perimeters, with permission from the device manufacturers. The app can also be used on Android phones with virtual reality headsets via a new driver interface, the PhoneHMD, implemented on the OPI. The use of the tools provided by the OPI library is showcased with a custom static automated perimetry test for the full visual field (up to 50 degrees nasally and 80 degrees temporally) developed with the OPI driver for the Octopus 900 and using visualFields for statistical analysis. With more than 60 citations in clinical and translational science journals, this initiative has contributed significantly to expand research in perimetry. The continued support of researchers, clinicians, and industry are key in transforming perimetry research into an open science.

package contains tools for the analysis and visualization of visual field data, including methods to compute 23 deviation values and probability maps. The use of the OPI and visualFields is shown through a custom static 24 automated perimetry test for the full visual field (up to 50° nasally and 80° temporally) developed with the OPI 25 driver for the Octopus 900 and using visualFields for statistical analysis. Its potential for the development of cross-26 platform apps for driving and testing portable devices is demonstrated with an OPI driver for an Android-based 27 headset. With more than 55 citations in clinical and translational science as listed in Scopus, this initiative has 28 contributed significantly to expanding the knowledge base in perimetry and clinical vision research at large, and 29 with clinical translation. The continued support of researchers, clinicians, industry, and public institutions are key 30 in transforming perimetry research from closed to open science. The Open Perimetry Initiative provides framework 31 to achieve this. 32 . CC-BY-NC 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted November 10, 2021.

Introduction 33
The advent of new technologies and the development of cross-platform software are paving the way for a new 34 wave of portable devices for visual field testing. The transition from traditional projection perimetry to display-35 based perimetry requires analysis and adaption of conventional perimetry methods and provides an opportunity 36 to revise and improve. The key product of the Open Perimetry Initiative is the Open Perimetry Interface (OPI). 1 The OPI not only provides 45 drivers for an increasing number of ophthalmic devices, but it also sets standards and protocols for the 46 implementation of custom visual field tests so that they can be run seamlessly on different machines, with one 47 implementation for many devices. Furthermore, the OPI can be run in simulation mode, so that new perimetric 48 procedures can be implemented, debugged, and assessed before they are ported to the actual test device. The 49 second key product of the Open Perimetry Initiative is visualFields. 2 This R package 3 is a tool for the statistical 50 analysis and visualization of perimetry results. Until now, the OPI and visualFields solutions have been developed 51 largely independently from one another. Recent developments in R, in particular the shiny package 4 52 (https://shiny.rstudio.com), have made it possible to easily develop cross-platform applications with graphical 53 interfaces that integrate the OPI and visualFields software. 54 . CC-BY-NC 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted November 10, 2021. ; https://doi.org/10.1101/2021.11.08.467428 doi: bioRxiv preprint 4 The purpose of this paper is to describe the recent advances in both the OPI and visualFields and to showcase their 55 use in conventional perimeters as well as new devices such as tablets, phones, and virtual reality headsets. In 56 addition to the development of novel paradigms, methods, and software, the optical properties need to be 57 characterized on these new devices, including resolution and the effects of chromatic and achromatic aberrations. 58 It is likewise necessary to develop adequate methods for calibration and compensation of refractive errors. 59 Considerations of optical characterization, calibration, and limitations are beyond the scope of this manuscript. 60 The OPI implementations and drivers follow conventions and standards that not only accelerate software 68 development, but also enable the creation of custom tests, perimetric algorithms, and procedures. These can all 69 be used with different computer operating systems, programming languages, and with different commercial and 70 experimental perimeters. The OPI commands are described more fully elsewhere 1 but are listed here for 71 completeness: 72 (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made

Methods
The copyright holder for this preprint this version posted November 10, 2021. ; https://doi.org/10.1101/2021.11.08.467428 doi: bioRxiv preprint There are three distinct types of visual field stimuli that can be presented with the opiPresent() command: static, 78 temporal, and kinetic. The static type can be used for conventional static automated perimetry. 5,17,18 The temporal 79 type can be used to generate stimuli that vary over time and that are supported by the underlying hardware, such 80 as the frequency-doubling illusion 19,20 . The kinetic type can be used to present moving stimuli specified according 81 The implementation of the aforementioned OPI commands and drivers are necessarily different for each different 93 perimeter; however, the specific implementation can be selected with the command opiChoose(). Thus, once a 94 specific perimetry driver -Octopus900, Compass, IMO, Daydream, etc.-has been selected, a dispatcher is set in place 95 so that the same OPI commands listed earlier can be used without change with all supported hardware. 96 Using the same approach, it is possible to generate graphical user interfaces that are platform-independent (e.g., 97 the R package shiny 4 ), which provides a framework for developing interactive web apps. Shiny was used to develop 98 applications to drive the Octopus 900 perimeter and the Google Daydream headset (shown in Results in this 99 manuscript). The Daydream shiny app allows one to configure the device and perimetry test, obtain the luminance 100 . CC-BY-NC 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted November 10, 2021. ; https://doi.org/10.1101/2021.11.08.467428 doi: bioRxiv preprint 6 profile (i.e., the correspondence between pixel value and physical luminance, of compatible Android phones), 101 manage datasets, run a threshold test on conventional and custom grids of test locations, and generate graphical 102 reports from the results. 103 The visualFields analytical tool 104 The visualFields package has undergone a major revision moving from version 0.6 2 to version 1.0 introduced here. Its 105 core functionality is the same but the code has been simplified; some dated ideas and methods have been 106 deprecated, and a number of conventions have been adopted for clarity and simplicity. The most visually striking 107 change has been the adoption of Voronoi tessellation 26-28 for the graphical representation of visual field data and 108 statistics. In addition, an effort has been made to improve its transparency and the reproducibility of its methods. 109 For instance, the SUNY-IU dataset of healthy subjects that was used in the previous version to generate normative 110 values has been incorporated into the package, vfctrSunyiu24d2, along with a function that generates the normative 111 reference values. Thus, the command nvgenerate(vfctrSunyiu24d2) generates pointwise normative values and the 112 command nvgenerate(vfctrSunyiu24d2, method = "smooth") generates the normative values used in the visualFields 0.6 2 using 113 the smoothing techniques as those introduced by Heijl and colleagues for the Statpac 2. 29,30 Normative datasets, 114 vfctrIowaPC26 and vfctrIowaPeri, and reference values generated with the function nvgenerate for the custom tests used 115 to study the advantages of exploring the full visual field are also included in the package. (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted November 10, 2021. ; https://doi.org/10.1101/2021.11.08.467428 doi: bioRxiv preprint the OPI and visualFields packages and presented in the manuscript, as well as the datasets are available from the 124 corresponding author on reasonable request. 125 Results 126 Figure 1 illustrates the OPI architecture. Once the hardware is selected, the R OPI client dispatches the commands 127 to connect, disconnect, set background and other settings, and present stimuli to the corresponding OPI server 128 (Octopus 900, Compass, IMO, Cardboard, etc.), which ultimately communicates the commands to the hardware. 129 The server then waits for the hardware to send its state or respond (machine initialized, background lit, pupil  . CC-BY-NC 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted November 10, 2021. ; https://doi.org/10.1101/2021.11.08.467428 doi: bioRxiv preprint 9 The OPI was used to run a series of tests of the full visual field. 8-11 A publicly available dataset of 98 eyes of 98 145 healthy subjects 8 was used to derive normative values with the visualFields R package. 2 Figure 2 shows a statistical 146 analysis of the results for a specific full visual field, which consists of the combined analysis of two tests taken on 147 the same day, one for the central visual field and another for the far periphery. 8-11 The tiles involving each visual 148 field location were obtained using Voronoi tessellation to achieve an efficient representation for both highly 149 irregular grids. Voronoi tessellations are a partitioning of a surface into regions so that the center of each cell is its 150 mean (center of mass). Every point in a given Voronoi polygon is closer to its generating point than to any other 151 cell. We believe this type of graphic best represents the visual area tested, without interpolating between different 152 points The dataset of ocular healthy eyes 8 is incorporated in the visualFields 1.0 for the central and peripheral tests, 153 vfctrIowaPC26 and vfctrIowaPeri. A script that generates figures for all subjects in those datasets is provided as 154 supplemental material with name vfPlotFullField.r. 155  . CC-BY-NC 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted November 10, 2021. ; https://doi.org/10.1101/2021.11.08.467428 doi: bioRxiv preprint In addition to these single field analyses, the visualFields package also has tools to analyze longitudinal data, 165 including pointwise linear regression, and the permutation of pointwise linear regression, or PoPLR. 12 Figure 3  166 shows a brief report (generated with the script vfPoPLRAnalysis.r provided as supplemental material and with the 167 vfpwgSunyiu24d2 dataset 13 ). The normative values to obtain total deviation values and probability maps were 168 generated using the dataset, vfctrSunyiu24d2, from a prospective longitudinal study conducted Bloomington, Indiana 169 University (IU) and New York City, State University of New York (SUNY). 2 The normative values were obtained with 170 the command nvgenerate(vfctrSunyiu24d2). 171   The ZEST algorithm parametrization used to measure the full visual field 8-11 has been updated and improved for its 182 use with an experimental Daydream OPI perimeter. The specific implementation of the ZEST algorithm has a 183 bimodal prior probability mass function with one peak centered at 0 dB to model sensitivities of damaged locations 184 and another peak that depends on sensitivity estimates in neighboring locations. 14 Neighboring locations are now 185  . CC-BY-NC 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted November 10, 2021. ; https://doi.org/10.1101/2021.11.08.467428 doi: bioRxiv preprint 11 defined as locations that share line segments which formed the polygons obtained via Voronoi tessellation. Figure  186 4 shows the Voronoi tessellation pattern for the 24-2 grid of test locations, with arrows indicating how locations 187 are sampled following a growth algorithm. A conventional 24-2 grid was used instead of the irregular one in Figure  188 2 for clarity of illustration. 189 . CC-BY-NC 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted November 10, 2021. ; https://doi.org/10.1101/2021.11.08.467428 doi: bioRxiv preprint through the midline since the ganglion cells on the upper and lower hemifields follow different trajectories 15 and 203 therefore there is no structural correlation between those neighboring visual field locations. 204

206
The web app sends commands to an Android Pixel 3 phone (shown in the right panel) to generate white visual stimuli at different intensities 207 and at different distances from the fixation point (green cross). At each presentation, the web app updates and shows the interim results of the 208 test. This screenshot was possible thanks to scrcpy (https://github.com/Genymobile/scrcpy). The advent of new technologies and the development of cross-platform software are preparing the way for a new 216 wave of portable devices for visual field testing. To lay the groundwork for this new era of perimetry, it is important 217 . CC-BY-NC 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted November 10, 2021. ; https://doi.org/10.1101/2021.11.08.467428 doi: bioRxiv preprint 13 to build a knowledge base on the strengths and weaknesses of the coming generation of perimeters, as well as 218 revise and accommodate conventional methods and analyses to unveil their full potential. It is essential to do this 219 in a manner that is as transparent and as accessible as

233
. CC-BY-NC 4.0 International license available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint this version posted November 10, 2021. ; https://doi.org/10.1101/2021.11.08.467428 doi: bioRxiv preprint

14
The Open Perimetry Initiative would not have evolved had it not been for the generosity and active support of key 234 researchers, clinicians, industry, and public institutions. It now needs the involvement of new researchers, 235 clinicians, and entrepreneurs to continue its development. 236

Acknowledgements 237
Veterans Administration Merit Review (I01RX-001821-01A1) and This work was supported by Computational 238 Optometry (Atarfe, Spain, URL: www.optocom.es). We thank Jize Dong for writing the initial versions of the Java 239 code for the OPI Daydream/Cardboard server. We also thank Zachary Heinzman for reviewing the manuscript. 240