Premium
Assessment of different methods for inter‐individual registration of OCT topography for statistical analysis
Author(s) -
HALDAR S,
DAVIES N
Publication year - 2012
Publication title -
acta ophthalmologica
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.534
H-Index - 87
eISSN - 1755-3768
pISSN - 1755-375X
DOI - 10.1111/j.1755-3768.2012.2716.x
Subject(s) - standard deviation , foveal , gaussian , convolution (computer science) , computer science , centroid , artificial intelligence , pixel , optics , mathematics , physics , ophthalmology , medicine , statistics , retinal , quantum mechanics , artificial neural network
Purpose Assessment of OCT topography is subjective. A statistical method of analysis would be helpful to aid interpretation. This requires the generation of accurate normative topography which in turn requires accurate alignment of normal OCTs. This study assesses 9 methods of alignment. Methods Normal topography maps were exported from a spectral domain OCT system. Code was written to perform image registration using different methods: User selection of foveal centre, cross‐correlation to a difference of two Gaussian macula template, finding of the thinnest central point and automatic fovea‐finding using Gaussian convolution and centroiding for the red, green and blue channels of the image respectively. Results Data from 127 left and 110 right eyes were analysed. Mean and total standard deviation across the central 400x400 pixels of the aligned maps were calculated. The lowest standard deviation was achieved by the cross‐correlation method (38.9 microns), followed by the blue channel centroiding method (40.4 microns), the thinnest central point (40.5 microns) and the user‐selected foveal centre (40.6 microns). Convolution with a Gaussian to identify the fovea produced the worst results with mean SD of 49.9‐64.8 microns. Conclusion Cross‐correlation with a difference of two Gaussian macula template appears superior for inter‐individual topography registration in OCT in comparison with fovea‐finding methods. Blue centroid and thinnest point were the best other methods. The cross‐correlation technique will lead to the most accurate normal maps for statistical comparison with data from pathological OCT topography.