基于多特征融合和随机森林的视网膜血管分割

发布时间：2018-08-13 10:39

【摘要】：为了进行眼底疾病辅助诊断,提出一种基于多特征融合和随机森林的视网膜血管分割方法.首先为彩色眼底图中的每个像素点提取一个23维特征向量(包括图像不变矩、灰度共生矩阵、LoG结合高斯二阶导、梯度法、相位一致性和Hessian矩阵特征);然后选取一定数量的像素点,提取其特征共同构造一个特征矩阵作为输入数据,并采用随机森林算法训练分类器;再用训练好的分类器对待分割图像中的像素点进行分类,判断其是否为血管点;最后在初步分割基础上进行基于连通区域补足血管的后处理,得到优化后的血管分割结果.在DRIVE公共数据库上进行实验的结果表明,该方法平均精确度达0.9606,平均灵敏度达0.7447,平均特异性达0.9838,比已有方法性能更优.
[Abstract]:A method of retinal vascular segmentation based on multi-feature fusion and random forest was proposed to assist the diagnosis of ocular fundus diseases. Firstly, a 23-dimensional eigenvector (including image moment invariant, grayscale co-occurrence matrix log and Gao Si second-order derivative, gradient method) is extracted for each pixel in the color fundus image. Phase consistency and Hessian matrix feature), and then select a certain number of pixels, extract the feature matrix to construct a feature matrix as input data, and use the random forest algorithm to train the classifier. Then the trained classifier is used to classify the pixel points in the segmented image to determine whether they are vascular points. Finally, on the basis of the initial segmentation, the post-processing based on the complementary vessels in the connected region is carried out, and the optimized segmentation results are obtained. The experimental results on DRIVE public database show that the average accuracy of the method is 0.9606, the average sensitivity is 0.7447, the average specificity is 0.9838, and the performance of this method is better than that of the existing methods.
【作者单位】：中南大学文学与新闻传播学院;移动医疗教育部-中国移动联合实验室;中南大学信息科学与工程学院;
【基金】：国家自然科学基金(61573380,61562029)
【分类号】：R770.4;TP391.41

【相似文献】