Skip to main content
  • Using AI to Characterize Central VF Loss in End-Stage Glaucoma

    By Lynda Seminara
    Selected and Reviewed By: Neil M. Bressler, MD, and Deputy Editors

    Journal Highlights

    JAMA Ophthalmology, February 2020

    Download PDF

    Wang et al. used artificial intelligence (AI) to quantitatively identify and clas­sify patterns of central visual field (VF) loss in chronic glaucoma. They found that the patterns in end-stage glaucoma may be subtype specific and that the initial sign of central VF deterioration is likely to be nasal loss. In most of the identified patterns, at least part of the inferotemporal region was preserved.

    The authors collected retrospec­tived data from five U.S. glaucoma services for a 15-year period. Central VF patterns were determined by an unsupervised AI algorithm that the authors termed “archetypal analysis.” An advantage of this form of AI is its ability to determine representative VF patterns lying at the corners of the data space. The identified VF patterns closely resemble those identified by physicians, and the findings are easy to interpret.

    The researchers also performed longitudinal analyses to investigate whether central VF loss affects certain vulnerability zones. Their cutoff point for end-stage glaucoma was 24-2 VF mean deviation (MD) of –22 dB.

    Altogether, there were 2,912 reliable 10-2 VFs (1,103 eyes of 1,010 patients), which were measured after end-stage 24-2 VFs. The mean (standard devi­ation) age of patients was 70.4 (14.3) years, and the 10-2 MD was –21.5 (5.6) dB. Fourteen central VF patterns were determined, including temporal-spar­ing patterns (27.5%), mostly nasal loss (25.4%), hemifield loss (15.3%), central island (10.9%), total VF loss (8.3%), nearly intact field (5.1%), and others that occurred less frequently.

    Location-specific analyses demon­strated two major zones of involve­ment: superonasal (more vulnerable) and inferotemporal (less vulnerable). Follow-up at one and two years showed that new defects were more common in the superonasal zone and that initial encroachments on an intact central VF were more likely nasal related. One pattern of nasal loss had a high risk of shifting to total VF loss by two years.

    The zone-related findings of this study corroborate those of the Hood model. The early-onset nasal loss is analogous to the known nasal step patterns in early glaucoma, as mea­sured by the 24-2 VF test. The authors emphasized that learning more about nasal loss may shed light on the patho­physiologic mechanism responsible for the onset of central VF deficiency. (Also see related commentary by Paolo Brusini, MD, in the same issue.)

    The original article can be found here.