Skip to content

Central Visual Field Patterns Identified by Artificial Intelligence Improve Prediction of Glaucoma Progression

Key findings

  • In this study, unsupervised artificial intelligence identified 17 distinct central visual field (VF) loss patterns for patients who had a range of severities of glaucomatous functional vision loss
  • The newly identified VF patterns corresponded well with stages of glaucoma determined by 24-2 VF
  • Some central VF defects were found across glaucoma stages; others were specific to a certain glaucoma stage, which may represent different subtypes of central VF loss
  • Consideration of central VF patterns strongly improved prediction of central VF mean deviation slope over time, compared with using global indices for two baseline VF results
  • It should be possible to use the central VF patterns to develop a computer algorithm for detecting progression of glaucoma

Preserving central vision is vital to the care of patients with glaucoma, yet no studies have systematically characterized central visual field (VF) loss patterns.

Using unsupervised artificial intelligence, Mengyu Wang, PhD, assistant scientist at Massachusetts Eye and Ear and assistant professor of Ophthalmology at Harvard Medical School, Tobias Elze, PhD, assistant scientist at Mass Eye and Ear and assistant professor of Ophthalmology, and colleagues determined numerous distinct central VF patterns in patients with glaucoma, and the patterns proved useful for following patients over time. The researchers published their results in Ophthalmology.

Study Methods

The team used VF data from the Glaucoma Research Network, a consortium of five leading eye institutes, for a two-part study:

  • Cross-sectional analysis—Central VF loss patterns were determined from 13,951 eyes with 10-2 VF test results; the 24-2 VF test result within a three-month window of the 10-2 test was used to stage eyes into mild, moderate or severe glaucomatous functional loss
  • Longitudinal analysis—1,191 eyes with at least five reliable 10-2 VF test results, at least six months apart, were monitored for VF changes over at least 24 months

Central VF Pattern Quantification

Artificial intelligence identified 17 representative central VF patterns from the multicenter dataset:

  • Intact fields (accounted for 38% of results)
  • Diffuse loss (20%)
  • Superior loss (17%)
  • Inferior loss (13%)
  • Central loss (5%)
  • Total loss (5%)
  • Temporal loss (2%)

Results by Stage

The newly identified central VF patterns corresponded well with stages of glaucoma; for example:

  • 3,529 eyes with mild glaucoma—11 patterns: 50% intact field, 2% diffuse loss, 17% superior loss and 15% inferior loss
  • 1,528 eyes with moderate glaucoma—11 patterns: 29% intact field, 4% diffuse loss, 40% superior loss and 11% inferior loss
  • 3,066 eyes with severe glaucoma—16 patterns: 9% intact field, 40% diffuse loss, 28% superior loss and 17% inferior loss

Superonasal loss and diffuse loss with a temporal island were found in more than one glaucoma stage. Conversely, some defects were unique to a single stage; for example, superior–peripheral loss was found only in mild glaucoma and inferotemporal loss only in severe glaucoma. These differences may represent different subtypes of central VF loss.

Assessing Change

Inclusion of central VF pattern features strongly improved prediction of central VF mean deviation slope over time, compared with using global indices for two baseline VF results. This was true even in patients with severe functional loss.

On the Horizon

It should be possible to use the central VF patterns to develop a computer algorithm for detecting progression of glaucoma. Researchers at Mass Eye and Ear developed a similar algorithm to detect progression based on 24-2 VF, and it proved to be more accurate than existing methods, as reported in Investigative Ophthalmology & Visual Science.

Visit the Ophthalmology at Mass Eye and Ear/Mass General

Refer a patient to the Mass Eye and Ear/Mass General Department of Ophthalmology

Related topics

Related

In the first study of its kind, Lucia Sobrin, MD, MPH, and John H. Kempen, MD, MPH, PhD, MHS, of the Massachusetts Eye and Ear/Mass General Hospital Department of Ophthalmology, and colleagues found a low risk of noninfectious uveitis among women using hormonal contraceptives or menopausal hormone replacement therapy.

Related

Milica A. Margeta, MD, PhD, and Janey L. Wiggs, MD, PhD, of the Mass Eye and Ear/Massachusetts General Hospital Department of Ophthalmology have demonstrated that the APOE ε4 allele, the strongest genetic risk factor for late-onset Alzheimer's disease, has a protective effect against primary open-angle glaucoma.