Automatic contour extraction algorithm for swept-source optical coherence tomography cornea image

Abstract

In a swept source-optical coherence tomography system, the telecentric scanning mode gives rise to central saturation artifacts,partial structural loss, and low SNR (signal-to-noise ratio) area in the corneal image, which affects the accuracy of corneal contour extraction. In order to solve this problem, in this paper we propose an automatic extraction algorithm for corneal image of low quality. This algorithm divides the image into high and low SNR region according to the standard deviation distribution of the cornea image. For the high SNR region, we localize the peak point to extract the contour. For the low SNR region, image enhancement is achieved by the registration and superposition of successive frames, which provides reference contour points for low SNR areas. Then corneal contour localization is achieved by weighing the advantages and disadvantages of reference contour points and local line fitting results. Finally, global polynomial fitting is used to achieve the whole corneal contour information. Experiments on the optical eye model show that comparing with the existing algorithms, the accuracy of corneal contour extraction is improved by 4.9% on average.