New AI Model Predicts Course of Geographic Atrophy

A better understanding of GA lesion growth patterns could refine trial design and therapeutic decisions for late-stage dry AMD.

By: Jim Gallagher, senior managing editor

Retinal Physician August 2, 2025 Vol 22, Issue July/August 2025

Clinical Report: New AI Model Predicts Course of Geographic Atrophy

Overview

A new AI-driven mathematical model developed by researchers at the University of Colorado predicts the growth patterns of geographic atrophy (GA) lesions. This model, validated against extensive patient data, reveals that GA growth is nonlinear and bounded, which has significant implications for clinical trial design and patient management.

Background

Geographic atrophy (GA) is a severe form of age-related macular degeneration (AMD) that leads to irreversible vision loss. Understanding the growth patterns of GA lesions is crucial for improving clinical trial designs and therapeutic strategies. The development of predictive models can enhance patient outcomes by allowing for more tailored treatment approaches.

Data Highlights

The AI model was validated using data from over 2,000 patients followed for up to 12 years, demonstrating a DICE coefficient of 0.91 for segmentation accuracy and a correlation of r=0.99 with manual measurements.

Key Findings

The AI model predicts GA lesion growth as a multiphase process: initial acceleration, linear growth, and eventual deceleration to a plateau.
Lesion growth is bounded, with many patients reaching an asymptotic limit, challenging the assumption of indefinite growth.
Over one-third of patients in clinical trials may already be at or near their asymptotic limit, potentially skewing treatment effect results.
The Gompertz curve best represents GA progression, outperforming traditional linear models in long-term predictions.
Using fellow eyes as controls in trials may introduce bias if they are at different phases of GA progression.

Clinical Implications

Clinicians should consider the nonlinear growth patterns of GA when designing clinical trials and interpreting outcomes. Understanding that many patients may already be at their growth plateau can inform better patient selection for trials and treatment strategies.

Conclusion

The development of this AI-driven model marks a significant advancement in predicting GA progression, with important implications for clinical practice and research in AMD.

PentaVision

Retinal Physician