Quantitative Interpretation of Uncalibrated Fundus Images

Abstract

Abstract: : Purpose:A model of light transport in tissue can be used to predict the reflectance of tissue based on its composition. The inverse process whereby composition is deduced from reflectance measurements is investigated to aid interpretation of fundus images. Methods:A simple model of fundus reflectance (excluding the optic disc) was constructed using published tissue characteristics. The variable parameters were assumed to be the concentrations of macular pigment in the receptor layer; melanin in the RPE; melanin in the choroid; and haemoglobins in the choroid. Reflectance spectra were computed for a wide range of parameters and convolved with appropriate filters to generate "image values". The inverse relationship was established to enable tissue composition to be inferred from an image. Filters were selected using a computational optimisation technique according to the criterion that they should minimise error in parameter recovery due to known errors in the imaging process and tissue characterisation data. In fundus images, illumination intensity varies with position due to the curvature of the fundus and the illuminating aperture (the pupil). To compensate for this, an additional filter was selected, and all image values were divided by this value. The resulting "image ratios" are independent of intensity and allow the method to be used in both the central and peripheral fundus. We applied these filters to a set of experimental spectra and compared the predicted composition to that from a full spectral fit. Results:The parameters of the optimal filters are shown in table 1. Parameters recovered from measured spectra using these filters are in agreement (within error bounds) with values obtained from full spectral fits in foveal and perifoveal regions. Conclusions:Model based analysis of fundus images can be used to extract histological information. Careful choice of filters can minimise error in parameter recovery, and the use of "image ratios" can compensate for uneven illumination. Table 1: Optimal Filter Parameters 

Keywords: computational modeling • imaging/image analysis: non–clinical