Corneal endothelium analysis by means of Bayesian shape modeling

A new method is proposed for the automatic detection of cell contour in images of corneal endothelium. The main drawback of the systems proposed so far in the literature, which mainly rely on classical image segmentation techniques, is the impossibility to embed prior information on the object being segmented, i.e., the regular hexagon pattern of endothelial cells. In our approach, the introduction and use of a-priori information is obtained by means of the Bayesian modelling framework, which allows to combine prior information on the applied model with experimental data through a statistical description of the model given the data. The resulting posterior probability density function is then optimized by a maximum a posteriori (MAP) estimator, using the simulated annealing maximization algorithm. Preliminary results on a few images showed that the developed system is capable of correctly recognizing the contour of cells. The developed theoretical framework is extremely flexible and can be easily adapted to different prior distributions or even to different object detection applications involving shape prior information.