Segmentation of macular fluorescein angiographies. A statistical approach

摘要

This paper is concerned with the use of Bayesian methods in the segmentation of macular fluorescein angiographies. Fluorescein angiography is used in ophthalmic practice to evaluate vascular retinopathies and choroidopathies: Sodium fluorescein is injected in the arm's cubital vein of the patient and its distribution is observed along retinal vessels at certain times. In this task a previous and essential step is the segmentation of the image into its relevant components. In order to obtain this segmentation Bayesian methods can be used because a previous knowledge about the spatial structure of the scene to be segmented is available in this kind of images. The stochastic model assumed for the observed intensities is a simple model with a Gaussian noise process which is statiscally independent between pixels. The process of labels x is modelled as a Markov random field with a space-dependent external field expressing the anatomy of the ocular fundus and higher order interactions encouraging blood vessels to be thin and large. This procedure is applied to different cases of diabetic retinophaty and vein occlusions. Two algorithms have been used to estimate x, simulated annealing and iterated conditional modes. In order to evaluate the accuracy of the estimation several error measures have been calculated.