Purpose: To compare agreement between mosaicked and seven field photographs for classification of the diabetic retinopathy (DR) severity.
Methods: Mosaic digital (MosD) images were compared with seven field stereo film (7FF) and stereo digital (7FD) photographs from a 152-eye cohort with full-spectrum Early Treatment of Diabetic Retinopathy severity levels for agreement on severity level, DR presence with ascending severity thresholds, DR index lesion presence, and classification repeatability.
Results: There was a substantial agreement classifying the Early Treatment Diabetic Retinopathy Study DR severity level between MosD and 7FF (kunweighted = 0.59, klinear weighted = 0.83), MosD and 7FD (κ = 0.62, κ weighted = 0.86), and 7FD and 7FF (κ = 0.62, κ weighted = 0.86) images. Marginal homogeneity analyses found no significant difference between MosD and 7FF (P = 0.44, Bhapkar's test). Kappa between MosD and 7FF ranged from 0.75 to 0.91 for the presence or absence of DR at 8 ascending severity thresholds. Repeatability among readers using MosD images was similar to repeatability among those using 7FF or 7FD. Repeatability among readers using MosD and 7FF images at various severity thresholds was similar. Kappa between MosD and 7FF grading for identifying DR lesions ranged from 0.61 to 1.00.
Conclusion: Mosaic images are generally comparable with standard seven-field photographs for classifying DR severity.