斑马鱼视网膜细胞图像分析研究

【作者】 张东杰；

【导师】 陆建峰；

【作者基本信息】 南京理工大学 ， 生物医学工程， 2008， 硕士

【摘要】 斑马鱼视网膜细胞的计数与分类是检查细胞变异的重要依据,具有重要的生物学意义。先前的视网膜细胞的计数和分类以人工操作为主,由于工作量太大,因此,必须研究自动的方法来帮助研究人员完成这两项工作。本文在细胞图像分割、细胞计数、特征提取与选择、细胞分类这四个方面进行了研究。在细胞图像的分割方面,本文采用了基于梯度流跟踪的分割方法。它包括以下三个步骤:梯度向量扩散、梯度流跟踪和自适应阈值分割。通过梯度向量扩散与梯度流跟踪,细胞图像被分割成很多个仅包含一个细胞和周围背景的小区域;然后使用自适应阈值法成功地从每一个小区域的背景中提取独立的细胞。该方法很好地解决了其他算法在粘连细胞图像分割过程中经常发生的较严重的过分割或欠分割问题。在细胞图像分割的基础上,本文参照八邻域边界跟踪技术和图像标号算法,采用了基于边界跟踪的计数算法实现了细胞计数。最后,本文研究了斑马鱼视网膜七类细胞的特征及分类规则,提取了细胞的彩色光密度特征和形态特征,并对特征的选择和分类器的设计进行了讨论和研究。选用最小距离分类器对斑马鱼视网膜细胞进行分类。更多还原

【Abstract】 Counting and classification of zebrafish retinal cells are important gist in the inspection of cell mutation, have important biological significance. Previous counting and classification of zebrafish retinal cells are mainly based on manual operation. Because of heavy workload, automatic methods should be developed to finish above task. In this thesis, in the view of image analysis of zebrafish retinal cell, following issues are researched: image segmentation, cell counting,feature extraction and selection,and cell classification.For image segmentation,gradient- flow-tracking based method is adopted, which is composed of three key steps: gradient vector diffusion,gradient flow tracking and adaptive thresholding. After gradient vector diffusion and gradient flow tracking, the cell image is separated into some small regions, each region containing one cell and its background. Local adaptive thresholding is performed for each small region, the single cell can be extracted from the background. Above method favorably overcomes the shortcoming of over-segmentation and under-segmentation by other algorithms for the segmentation of touching cells On the basis of above cell image segmentation, boundary tracking based algorithm is used for cell counting, which refers to image labeling algorithm and 8-connected boundary tracking.Finally, cell features and classification rules are studied, color density features and morphological features of zebrafish retinal cells are extracted. Then the problems of feature selection and the design of classifier are studied. The least distance classifier is used to classify the zebrafish retinal cells.更多还原