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

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Deep learning (DL) models can predict chronological age from retinal images.
• The choice of image fixation (macula- vs. disc-centered) may influence DL-based age prediction accuracy and its clinical relevance.
• Understanding these differences is crucial for applying DL in ophthalmology and oculomics research.

Purpose of the Study:

• To compare the performance and agreement of DL-based age prediction using macula-centered versus disc-centered color fundus photographs (CFPs).
• To investigate whether image fixation type modifies the association between retinal age gap (RAG) and ocular diseases.

Main Methods:

• Retrospective analysis of 12,230 same-day macula- and disc-centered CFP pairs from 5,895 patients (≥40 years).
• Utilized a validated DL model for age prediction from both image types.
• Assessed prediction accuracy using mean absolute error (MAE) and intraclass correlation coefficients (ICCs).
• Examined RAG-disease associations using generalized linear mixed-effects models.

Main Results:

• Macula-centered CFPs showed a slightly smaller MAE (7.09 years) compared to disc-centered CFPs (7.77 years).
• High overall agreement was observed between macula- and disc-centered age predictions (ICC = 0.83).
• Retinal age gap (RAG) was significantly associated with age-related macular degeneration (AMD; OR = 3.57).
• No significant interaction was found between RAG and fixation type for disease association.

Conclusions:

• Image fixation type influences absolute DL-based age predictions but not the association between RAG and disease.
• Both macula- and disc-centered CFPs can be utilized in oculomics research, offering imaging flexibility.
• Formal interchangeability of fixation types is not supported due to identified biases; prospective validation is needed for clinical use.