Automatic Choroid Layer Segmentation: A Slicing Approach

Optical coherence tomography (OCT) is a powerful and noninvasive imaging technique for eye ground imaging. This method is usually used to assist glaucoma diagnosis nowadays. We propose a novel automated method for the segmentation of choroid layer in OCT image. Our approach is based on normalized cut algorithm, which aims at extracting the global impression of an image and treat image segmentation as a graph partitioning problem. However, due to the complexity of the layer structure of eye ground and choroid layer itself, we employ a series of preprocessing algorithms to make the cut more deterministic and accurate. Furthermore, to make the cutting algorithm focused on cutting in a horizontal manner instead of meaningless vertical cuts, we slice the image into small slices plumb and run the normalized cut algorithm respectively on these slices. After that, we combine them together to get a global cut on the original image. We calculate the thickness of choroid layer, which is a key factor for diagnosis. We used the thickness as the major factor to measure our performance. The mean relative error of the algorithm is about 0.04 compared with manual segmentation performed by oculists. The proposed method was tested on a database including more than eight hundred images of fifty healthy people between 20 and 50 years old.