Purchase this article with an account.
Jorryt Gerlof Tichelaar, Mengyu Wang, Lucy Q Shen, Louis R Pasquale, Michael V Boland, Sarah R Wellik, Carlos Gustavo de Moraes, Jonathan S Myers, Peter Bex, Osamah Saeedi, Neda Baniasadi, Dian Li, Hui Wang, Tobias Elze; Defect Classes of Visual Field Measurement in Glaucoma. Invest. Ophthalmol. Vis. Sci. 2019;60(9):2858.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Glaucomatous visual field (VF) loss is diagnosed and quantified by summary measures that do not provide information about specific glaucomatous patterns. Here, we computationally trace the evolution of VF loss patterns over time to develop a quantitative model of functional glaucoma subtypes (defect classes).
To determine representative patterns of vision loss, total and mean deviations (MD) of the most recent reliable VFs (SITA Standard 24-2) were retrospectively selected among all patients from five clinical glaucoma practices. VFs were grouped by severity into MD bins of 1dB width. Separately for each MD bin, archetypal analysis, an unsupervised machine learning technique, was applied to autonomously determine severity specific patterns (archetypes [ATs]). To trace their evolution over time, all eyes with multiple VFs were selected. Each VF was quantitatively decomposed into the ATs of each MD bin by weighting the AT coefficients according to the measurement uncertainty distribution of the respective measured MD (Fig. 1A). For each baseline-follow up VF pair, an evolution matrix of the respective AT coefficients is calculated (Fig. 1B). Pooling over all evolution matrices, for each AT within each MD bin, the most probable follow-up AT for each next more severe stage was calculated, which yielded pattern trajectories that define defect classes.
Based on 179,936 VFs from 179,936 eyes, between 7 and 14 ATs were identified for each MD bin. Based on baseline follow-up evolution matrices from 278,445 VFs from 74,806 eyes, for moderate severity (-6dB≥MD>-12dB), seven distinct defect classes could be determined that contain features which are compatible with nerve fiber defects (Fig. 2A): A single upper peripheral class as well as upper and lower hemifield pairs of paracentral, nasal, and temporal patterns. Nasal, but not temporal to the blind spot, the classes respect the horizontal midline. When evolution patterns over the entire severity range are assessed, distinctive features like presence/absence of central VF loss are preserved within the respective defect classes (e.g. Fig. 2B).
We introduce VF defect classes which allow quantitative defect classifications of arbitrary VF measurements, which may help to detect glaucoma progression and enhance subtype-specific structure-function modeling.
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.
This PDF is available to Subscribers Only