Segmentation of connective tissue in the optic nerve head using an anisotropic Markov random field

The study of the biomechanical response of the optic nerve head (ONH) to different levels of intraocular pressure (IOP) is key in understanding the mechanisms that underlie the development and progression of glaucoma. Our goal is to study this behavior using the finite element method on serial sections of normal and glaucomatous monkey ONHs. Segmentation of the beamlike connective tissue of the ONH is particularly challenging due to the artifacts introduced in the acquisition procedure. We present a new algorithm for segmentation of beam-like structures, based on the expectation-maximization (EM) method. The main difference with previous EM-based segmentation algorithms is that we employ an anisotropic Markov random field, which incorporates preferred directions based on the underlying image structure and coherence. Results on synthetic and ONH images show the ability of the algorithm to accurately segment these structures in presence of significant artifacts, and suggest its suitability to other medical imaging segmentation tasks.