Improved OCT Human Corneal segmentation Using Bayesian Residual Transform

Abstract

The inherent poor signal to noise ratio of Optical Coherent Tomography(OCT) is considered as a main limitation of OCT segmentation,particularly because images are sampled quickly, at high resolutions,and in-vivo. Furthermore, speckle noise is generated bythe reflections of the OCT LASER limits the ability of automaticallysegmenting OCT images. This paper presents a novel method toautomatically segment human corneal OCT images. The proposedmethod uses Bayesian Residual Transform (BRT) to build a noiserobust external force map, that guides active contours model to thecorneal data in OCT images. Experimental results show that theproposed method outperforms the classical as well as the state-ofthe-art methods.