Purchase this article with an account.
V. Grau, J.C. Downs, C.F. Burgoyne; Three–dimensional (3–D) Segmentation of the Connective Tissue Architecture of the Lamina Cribrosa . Invest. Ophthalmol. Vis. Sci. 2004;45(13):1117.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Purpose: To develop a robust, accurate and automated method to segment the connective tissues of the lamina cribrosa from 3–D reconstructions of the normal and early glaucomatous monkey optic nerve head, so as to model their constituent IOP–related stress and strain within finite element (FE) models. Methods: Monkey ONHs were embedded in paraffin and serial sectioned at 3 µm intervals from the vitreous interface through the orbital optic nerve. After each section was cut, an image of the stained tissue block face (stained section image) was acquired at 2.5x2.5 µm2/pixel resolution. The 300 to 600 serial stained section images were aligned then stacked to form a 3–D reconstruction. An iterative, 3–D segmentation algorithm based on the Expectation–Maximization framework was then employed with two classes (connective and neural tissues) modeled with normal distributions. Prior knowledge of the connective tissue topology was introduced in the segmentation process via an anisotropic Markov Random Field (MRF). The MRF increased the overall coherence of the segmented structures, significantly reducing connectivity gaps and background noise. Results: The algorithm was run on a dataset containing an entire monkey ONH and validated visually by overlaying serial sections of the segmented lamina onto the corresponding stained section image. Selected regions were extracted from the laminar microstructure and used in a FE micromodel to calculate the biomechanical behavior of the lamina. The obtained segmentation of the laminar architecture was directly imported into the micromodel with only minor manual editing. Manual segmentation was also performed on a selected region, however because it lacked adequate 3–D connectivity, it proved inadequate for finite element modeling due to the proliferation of spurious and unconnected beams. Conclusions: An automatic 3–D segmentation algorithm was developed to extract the architecture of the lamina cribrosa from serial section images of the monkey ONH. Accuracy and laminar connectivity proved adequate to build FE micromodels. This will provide us with a valuable tool to study the behavior of normal and glaucomatous monkey ONHs subjected to varying levels of IOP. Direct comparison of this algorithm with manual segmentation and several 2–D automated strategies within the optic nerve head reconstructions of both eyes of 22 monkeys will provide a complete validation.
This PDF is available to Subscribers Only