Artificial intelligence may improve glaucoma management

This study used an unsupervised artificial intelligence (AI) model to quantify central visual field (VF) loss patterns in eyes with glaucoma.

Study design

In this multicenter retrospective study, researchers performed a cross-sectional analysis of the central VF patterns of 13,951 Humphrey 10-2 visual field tests using total deviation plots by archetypal analysis—an unsupervised AI method—in eyes with all glaucoma severities. The clinical usefulness of these patterns of central VF loss was demonstrated by using the patterns to track central VF changes longitudinally (1,191 eyes with at least 5 reliable 10-2 VF results obtained at follow-up intervals of at least 6 months) and to predict the central VF mean deviation (MD) slope from 2 baseline VF results with follow up of at least 24 months.

Outcomes

Analysis identified 17 distinct central VF patterns; these patterns could be divided into isolated superior loss, isolated inferior loss, diffuse loss, and other loss patterns. Within the diffuse loss pattern, 4 of the 5 patterns preserved the less vulnerable inferotemporal zone, whereas they lost most of the remaining more vulnerable zones. Features predicting a more negative 10-2 MD slope were identified as older age, decreased MD and pattern standard deviation of baseline VF. Inclusion of coefficients from central VF archetypal patterns improved the prediction of central VF MD slope.

Limitations

Detailed demographic and diagnostic information, treatment history, and structural measurements were not included in the analysis. Also, the central VF result samples could be skewed toward more severe glaucoma stages; these patients typically undergo more frequent central VF testing than do patients with more moderate cases of glaucoma. Additionally, archetypal analysis (unsupervised AI) is theoretically more prone to data outliers compared with other methods of analysis.

Clinical significance

Use of AI has the potential to improve glaucoma diagnosis and prognosis, especially unsupervised AI, which can help clinicians track patterns of glaucomatous damage and changes over time. This study confirmed previous observational findings of arcuate defects and less-vulnerable zones in the central VF. The ability to trace and detect focal changes of central VF over time can be used to develop new progression detection algorithms, which can better identify potential quality-of-life impairment in glaucoma patients than the use of central VF MD alone.