2-DE图像重叠蛋白点检测算法研究

发布时间：2018-03-21 19:15

本文选题：2-DE图像　切入点：重叠蛋白质点　出处：《山东师范大学》2017年硕士论文　论文类型：学位论文

【摘要】：蛋白质组学在揭示细胞生命过程规律中发挥越来越重要的作用。双向凝胶电泳(2-DE)技术是一种非常重要的蛋白质组学研究技术。2-DE图像分析处理的目标是快速识别单个蛋白质点,以及一组凝胶图像样本中的表达差异的蛋白质。蛋白点的检测是2-DE图像基于计算机处理过程的基础环节,由于2-DE图像的复杂性及制作过程中存在的干扰因素,使得蛋白质点的检测成为一个复杂又耗时的难题。本论文以2-DE图像中重叠蛋白质点的检测为研究重点,提出两种重叠蛋白质点的检测方法,主要研究内容如下:(1)研究了2-DE图像蛋白质点预检测方法。对比研究了均值滤波,中值滤波和高斯滤波等空间滤波算法,采用高斯滤波算法对凝胶图像去噪,降低噪声对凝胶图像后续蛋白质点检测的影响;对比研究了基于阈值、基于分水岭算法、基于区域生长的蛋白质点预分割方法,采用基于区域生长算法对蛋白质点进行预检测,得到清晰的蛋白质点边界。(2)提出了基于凹区匹配的重叠蛋白质点分割方法。首先获得重叠蛋白质点图像的凸闭包结构,并通过凸闭包与原二值图的差值获得重叠蛋白质点的凹区;其次根据重叠蛋白质点的重叠模型对凹区进行两两配对;然后根据匹配的凹区之间的最短欧式距离来确定凹点;最后通过连接凹点对实现重叠蛋白质点的分离。此方法不再从原始蛋白质点边界轮廓上提取凹点,而是从凹区的轮廓上提取凹点,明显减小了凹点搜寻的范围及处理的像素数量,提高了重叠蛋白质点分割的速度。实验表明,该算法在轻度重叠及重叠复杂度较小的情况下,分离效果较好。(3)提出了基于角点检测与多边形近似相结合的重叠蛋白质点分割方法。首先采用Harris角点检测法得到重叠蛋白质点边界的所有角点;其次为降低蛋白质点边界上干扰凹点的影响,将边界的角点作为特征点,利用基于角点的多边形近似算法对重叠蛋白质点边界进行多边形近似表示;然后通过判断多边形顶点的凹凸性来判断角点的凹凸性进而获得凹角点;最后提出凹点匹配准则,并从凹角点中选取真正的凹点并构造分离线,实现对重叠蛋白质点的精确分割。该算法实现简单,不需要多次腐蚀与膨胀运算,最大程度保持蛋白质点边缘。实验结果表明,与基于凹区匹配的算法相比,该算法提取的凹点的误差更小,对重叠蛋白质点的分离更加准确,同时对重度重叠及重叠复杂度较大的蛋白质点也有显著的分离效果。
[Abstract]:Proteomics plays a more and more important role in revealing the laws of cell life process. Two-dimensional gel electrophoresis (2-DE) is a very important proteomics research technique. The detection of protein spots in a group of gel image samples is the basic part of 2-DE image processing based on computer. Because of the complexity of 2-DE image and the interference factors in the process of making 2-DE image, The detection of protein spots has become a complicated and time-consuming problem. In this paper, two detection methods of overlapping protein spots in 2-DE images are proposed. The main research contents are as follows: (1) the protein spot pre-detection method of 2-DE image is studied. The spatial filtering algorithms such as mean filter, median filter and Gao Si filter are compared and analyzed. The effect of noise reduction on the detection of protein spots in gel images is studied. The pre-segmentation method of protein points based on threshold, watershed algorithm and region growth is compared, and the pre-detection of protein spots based on region growth algorithm is studied. A new method of overlapping protein point segmentation based on concave matching is proposed. Firstly, the convex closure structure of overlapping protein point image is obtained. The concave region of overlapping protein points is obtained by the difference between convex closure and original binary graph. Secondly, the concave region is pairwise paired according to the overlapping model of overlapping protein points, and the concave point is determined according to the shortest Euclidean distance between the matching concave regions. Finally, the overlapping protein points are separated by connecting concave pairs. Instead of extracting concave points from the boundary contours of the original protein points, the method extracts concave points from the concave contour. The range of concave search and the number of pixels processed are reduced obviously, and the speed of overlapping protein spot segmentation is improved. In this paper, an overlapping protein point segmentation method based on corner detection and polygon approximation is proposed. Firstly, all corners of overlapping protein point boundary are obtained by Harris corner detection method. Secondly, in order to reduce the effect of interference concave on the protein point boundary, the corner of the boundary is taken as the feature point, and the polygonal approximation algorithm based on the corner point is used to represent the overlapping protein point boundary by polygon approximation. Finally, the concave point matching criterion is proposed, and the real concave point is selected from the concave corner and the separation line is constructed. The algorithm is simple to implement, does not require multiple corrosion and expansion operations, and keeps the edge of protein points to the maximum extent. The experimental results show that compared with the algorithm based on concave matching, the proposed algorithm is more efficient than the one based on concave matching. The algorithm has less error in extracting concave points and more accurate separation of overlapped protein spots, and it also has a significant effect on separation of protein spots with heavy overlap and high overlap complexity.
【学位授予单位】：山东师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：Q51;TP391.41

【参考文献】