扫频光学相干层析角膜图像轮廓自动提取算法

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

在扫频光学相干层析系统中,远心扫描模式造成角膜图像中存在伪影、部分结构缺失及低信噪比区域,影响了角膜轮廓提取的精度.针对该问题,本文提出了一种针对低质量角膜图像的轮廓自动提取算法.该算法首先依据图像标准差分布将图像划分为高、低信噪比区域;针对高信噪比区域,通过峰值点定位法获取角膜轮廓;针对低信噪比区域,通过连续帧图像间配准叠加实现图像增强,为低信噪比区域提供参考轮廓点,再通过权衡参考轮廓点与局部直线拟合结果的优劣,实现角膜轮廓定位;最后,通过全局多项式拟合实现对全区域的角膜整体轮廓信息.对光学眼模型进行实验,结果表明,与已有算法相比,本文算法对角膜轮廓的提取精度平均提高了4.9%.

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.