Extracting histological parameters from multi-spectral retinal images: a Bayesian inverse problem approach

Extracting histological parameters, especially macular pigment, from multispectral images of the ocular fundus is a potential technique for the assessment of age-related macular degeneration. Such approaches make use of a Monte Carlo radiation transport model relating spectral reflectance of the tissue to tissue histology. We develop a probabilistic surrogate for this computationally expensive physicalmodel using Gaussian processes (GP). Further, we present a Bayesian inversion algorithm that uses the surrogate model to recover model input parameters. This methodology is tested both on synthetic
data generated from the Monte Carlo model and on real image data. It is shown that our inversion methods can recover macular pigment concentrations in human retina with good accuracy and the spatial distribution is consistent with known physiology.