A Complete Method for the 3D Reconstruction of Axonal Pathways from 2 Orthogonal 3D OCT Images of the Lamina Cribrosa

The lamina cribrosa is a 3D mesh-like structure within the optical nerve head, consisting of pores through which all the nerve fibers from the retina pass to join the brain. Alterations and damages in this structure have proven to be linked with glaucoma, the second leading global cause of blindness. Therefore, an automatic and accurate 3D pore reconstruction method from 3D images is crucial for carrying out clinical studies. In this article, we present a fully automatic method for the segmentation and reconstruction of axonal pathways from 3D OCT images. Our method has be designed to overcome the difficulties related to the poor image quality and partial annotations. It is based on the registration of two OCT volumes acquired in two orthogonal scanning directions and the design of a U-Net with attention mechanism to segment the pores of the registered 2D en-face images. Finally, the segmented pores are tracked using a parametric snake model to reconstruct the axonal pathways in 3D. Experiments show that 84.7% of the paths are truly detected with a RMSE around 2 pixels, below the pore diameters.