Clinical bottom line
Artificial intelligence (AI) models show promise in predicting visual field progression in patients with normal-tension glaucoma (NTG), a subtype characterized by glaucomatous optic neuropathy despite intraocular pressures (IOP) within normal ranges. Incorporating structural optic nerve head parameters such as optic disc rotation and size, alongside clinical factors, may enhance risk stratification beyond traditional IOP-based metrics. However, current evidence remains preliminary, and AI-driven prediction tools require further validation before routine clinical application.
What the evidence shows
Normal-tension glaucoma presents unique challenges in forecasting visual field deterioration due to its multifactorial pathophysiology and less direct correlation with elevated IOP. Several studies have identified structural optic nerve features as potential predictors of progression. For example, Sung Mi Sun et al. (2016) demonstrated that optic disc rotation, particularly in myopic NTG patients, correlated with faster visual field progression, suggesting that AI algorithms incorporating such morphologic parameters could improve prognostication.
Similarly, Yamazaki Yoshio et al. (2013) found that smaller optic disc size was associated with more rapid visual field damage in NTG, highlighting the importance of anatomical features in risk modeling. These structural markers may be integrated into AI frameworks to refine individualized risk assessment.
Clinical factors beyond IOP also influence progression. Lee Jeongmin et al. (2014) compared visual field progression rates between relatively low and high IOP groups within NTG patients, finding that even those with lower IOP experienced progression, underscoring the need for multifactorial models. Additionally, corneal biomechanical properties have been linked to progression speed (International Journal of Ophthalmology, 2016), which could be incorporated into AI models.
Comparative studies such as Ballae Ganeshrao Shonraj et al. (2019) emphasize that NTG progression rates differ from other glaucoma types, reinforcing the need for NTG-specific predictive tools. While some adjunctive therapies like Ginkgo biloba extract have been explored for their potential to slow progression (Lee Jeongmin, 2013), their role in modifying AI-predicted risk remains unclear.
Caveats and uncertainty
The current evidence base is limited by relatively small cohorts, retrospective designs, and heterogeneity in progression definitions and follow-up durations. Many studies focus on isolated structural or clinical parameters rather than integrated AI models combining multimodal data. Moreover, the generalizability of findings across diverse populations and varying degrees of myopia or optic nerve morphology is uncertain.
AI models for predicting visual field progression in NTG are largely in developmental or pilot stages, with few externally validated algorithms. The complexity of NTG pathogenesis, including vascular and neurodegenerative components, may limit prediction accuracy based solely on optic nerve morphology and IOP-related factors.
Additionally, the clinical utility of AI predictions depends on actionable thresholds and integration into decision-making workflows, which remain to be established. Ethical considerations around AI transparency and patient communication also warrant attention.
How this may change practice
As AI tools mature, incorporating optic disc morphology and clinical risk factors into predictive algorithms could enable earlier identification of NTG patients at high risk for visual field loss, facilitating tailored monitoring intervals and therapeutic intensification. This may be particularly valuable in myopic patients where optic disc rotation is prevalent.
Clinicians should remain cautious, recognizing that current AI predictions are hypothesis-generating rather than definitive. Continued research and prospective validation are needed before widespread adoption. In the interim, comprehensive clinical assessment including optic nerve imaging and functional testing remains essential.
Future AI-driven risk stratification may complement existing clinical judgment, potentially improving outcomes through personalized management strategies in NTG.