Sparsity-based retinal layer segmentation of optical coherence tomography images

A novel method for optical coherence tomography retinal image segmentation utilizing sparsity constraints is demonstrated. Retinal images are sparse in the layer domain. The algorithm thus transforms an input retinal image into a layer-like domain, and then uses graph theory and dynamic programming to extract the retinal layers from the sparse representation. The number of identified boundaries is not fixed and is determined by the algorithm at run-time. Results show that this method can segment up to nine layer boundaries without making overly restrictive assumptions about anatomic structure.