Evaluating the Progression of Retinal Sensitivity Loss in Geographic Atrophy Using Machine-Learning-Based...

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Geographic atrophy (GA) secondary to age-related macular degeneration (AMD) causes vision loss.
• Predicting retinal sensitivity loss in GA is crucial for monitoring disease progression and evaluating treatments.
• Current methods for assessing retinal sensitivity can be time-consuming and resource-intensive.

Purpose of the Study:

• To evaluate machine-learning models for predicting retinal sensitivity in GA.
• To compare the progression of sensitivity loss using observed versus inferred data.
• To assess the accuracy and variability reduction of different predictive models.

Main Methods:

• Thirty patients with GA (37 eyes) were studied, undergoing microperimetry and SD-OCT at multiple time points.
• A deep-learning algorithm segmented retinal layers.
• Random forest, LASSO regression, and MARS models predicted retinal sensitivity in three scenarios: unknown patients, known patients at later visits, and within-visit interpolation.

Main Results:

• The random forest model achieved the highest prediction accuracy across all scenarios (MAE: 3.67 dB for unknown, 2.96 dB for known, 3.10 dB for interpolation).
• Inferred sensitivity data significantly reduced variability compared to observed data (residual variance: 2.72 dB² vs. 8.67 dB²).
• Patient-specific baseline data improved model accuracy for follow-up visits.

Conclusions:

• Machine-learning models, particularly random forest, are effective for predicting retinal sensitivity in GA.
• Inferred sensitivity mapping serves as a reliable, high-resolution surrogate endpoint for clinical trials.
• This approach reduces the need for extensive psychophysical testing in AMD research.