Abstract: Hard exudation is one of symptoms in Diabetic Retinopathy（DR）. The bright speckles in Spectral Domain Optical Coherence Tomography（SD-OCT） have close relationship with exudation. In order to locate the lesion information in DR images, this paper proposes a method for extracting the bright speckles. The analysis includes four steps：denoise images with bilateral filtering, limit target regions by layer segmentation algorithm, grow regions using more large thresholds between the global threshold and single threshold by self-adaption threshold algorithm, grow again in the range of the seed images generated by other thresholds. 20 eyes with DR are selected to test, and the bright speckles coverage rate is increased by about 7% compared with the existing algorithm.

Key words: Spectral Domain Optical Coherence Tomography（SD-OCT）, Diabetic Retinopathy（DR）, hard exudation, self-adaption threshold, 3D region growing

摘要： 糖尿病视网膜病变是糖尿病严重的微血管并发症，硬性渗出是其病变表现之一，而频域光学相干断层（SD-OCT）视网膜图像中的高信号亮斑与硬性渗出有着紧密联系。为了能够准确定位糖尿病性视网膜病变图像中的这类病损信息，提出了一种自动分割亮斑的方法。提取亮斑分为四个步骤：使用双边滤波对图像去噪；使用基于图论的层分割算法来限制亮斑所在区域；在自适应阈值法产生的整体阈值及单帧阈值间选择大阈值帧数更多的作为基础进行第一次三维区域生长；将上一步得到的图像在另一阈值产生的种子图像范围内进行第二次区域生长。选取20只患有糖网的眼睛进行实验，自动分割的亮斑面积覆盖率较现有方法提高了约7%。

关键词: 频域光学相干断层（SD-OCT）, 糖尿病视网膜病变（DR）, 硬性渗出, 自适应阈值, 三维区域生长