基于2DGE图像先验的蛋白质点检测方法  

2DGE Image Priors Based Protein Spots Segmentation Method

作　　者：欧巧凤[1,2] 张会生[1] 熊邦书[2] 李立欣[1] 

机构地区：[1]西北工业大学电子信息学院,西安710072 [2]南昌航空大学图像处理与模式识别江西省重点实验室,南昌330063

出　　处：《半导体光电》2015年第1期145-149,共5页Semiconductor Optoelectronics

基　　金：国家自然科学基金项目(61163047);江西省自然科学基金项目(20114BAB201036);江西省工业支撑计划项目(2010GB00405)

摘　　要：各种改进的分水岭变换是目前应用最广的二维凝胶电泳(2DGE)图像蛋白质点分割算法。其中标记控制分水岭分割效果较好,但由于忽略了蛋白质点大小和形状,导致外标记落在部分蛋白质点区域内而无法正确分割。针对该问题,提出一种结合2DGE先验的分水岭分割方法。首先,根据2DGE图像蛋白质点中心灰度极小先验提取中心标记,并通过中心标记膨胀提取距离标记;然后,根据2DGE图像背景灰度极大先验,在距离标记划分的区块内提取原始2DGE图的局部自然分区标记,并对二者进行融合优化,获得最佳背景标记;最后在梯度图像各区块内进行分水岭变换,提取蛋白质点边缘。通过对四组真实扫描2DGE图像进行实验,结果表明,该方法明显改善了蛋白质点边缘检测正确率。Watershed transform （WST） based methods are applied most widely in two- dimensional gel electrophoresis （2DGE） images segmentation. Mark controlled WST can accurately detect most protein spots except a few dense spots with various volumes and shapes. Without consideration of the spot shape, the external mark sometimes locates in the spot area and leads to incorrect segmentation. To overcome the problem, an improved WST algorithm based on 2DGE image priors was proposed in this paper. Firstly, it was to extract spot center marks according to the gray minimum prior and to dilate the center marks iteratively to get distance mark. Secondly, it was extract local nature mark in blocks divided by distance mark according to background gray maximum prior. Then, the distance mark and nature mark are fused to generate background mark. At last, WST was applied on the grads image to detect protein spot edges. Four experiments were carried on 4 real scanned 2DGE images and the results show that the proposed segmentation algorithm enhances the accuracy of the algorithm effectively.

关 键 词：二维凝胶电泳 图像先验 图像分割 分水岭变换 数据融合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]