Macula- Versus Disc-Centered Fundus Photography: Performance in Age-Prediction and Disease Associations

The purpose of this study was to compare deep learning (DL)-based age prediction performance and inter-method agreement using macula- versus disc-centered color fundus photographs (CFPs), and to determine whether image fixation modifies the association between retinal age gap (RAG) and disease.

This retrospective cohort study analyzed 12,230 same-day macula- and disc-centered CFP pairs from 5895 patients (≥40 years) attending Moorfields Eye Hospital (2008-2018). We used a previously validated model that predicts chronological age using either macula- or disc-centered CFPs. DL-predicted age was compared via mean absolute error (MAE), intraclass correlation coefficients (ICCs), and Bland-Altman analysis. Associations between RAG and clinical conditions were assessed using generalized linear mixed-effects models with a patient-level random intercept.

Macula-centered CFPs yielded a smaller MAE than disc-centered images (7.09 vs. 7.77 years), and age predictions differed significantly between fixation types (P < 0.01). Overall agreement between macula- and disc-centered predictions was high, with an ICC of 0.83 (95% confidence interval [CI] = 0.82-0.84). RAG was significantly associated with age-related macular degeneration (AMD; odds ratio [OR] = 3.57, 95% CI = 1.76-7.24, P < 0.001). No statistically significant interaction between RAG and fixation type was observed.

Although fixation type influenced absolute age predictions, it did not significantly modify RAG-disease associations. These findings support the use of either fixation type in oculomics research where imaging flexibility is required, although formal interchangeability cannot be claimed given the systematic bias, heteroscedasticity, and subgroup variation in agreement identified. Prospective validation in population-based cohorts is required before clinical application can be achieved.