Reproducibility in the identification of retinal positions on serial fundus photographs: Theory and limitations

BACKGROUND: Imaging of the retina is accompanied by distortions so that positions in fundus photographs cannot be linearly translated to positions in the retinal fundus. The purpose of this study was to describe an algorithm and evaluate its reproducibility in identifying points on the retina from their representation on serial fundus photographs. METHODS: A mathematical formalism was derived to describe how the location of the fovea and the optic disc in fundus photographs, together with the centre-nodal point distance (d) and the curvature radius (r) of the eye, can be used to calculate spatial coordinates on the retina that correspond to given locations in the photograph. The effect of manual markings of the fovea and the optic disc as reference points was evaluated, and the formalism was tested on 105 photographs recorded with different image modalities and centring from 10 different patients. RESULTS: Retinal locations could be identified in serial fundus photographs with a reproducibility of at least 50 μm when the centre-nodal point distance and the curvature radius were set to those in Gullstrand's standard eye. A main source of the variability was the manual identification of the fovea and optic disc used as reference points. CONCLUSIONS: The reproducibility in the identification of locations on the retina from serial fundus photographs depends on the accuracy in the definition of the fovea and optic disc used as reference points. This affects the potential for describing locations, distances, areas and changes in retinal lesions on serial photographs over time.