Study On Algorithms For Detection And Tracking Of The Subcellular Structures From Confocal Microscopy Images

The development of digital fluorescence imaging techniques and molecular-level biological coloring techniques greatly promotes the direct observation and research on subcellular structures. Researchers apply the green fluorescent proteins to mark subcellular structures and study on dynamic molecular biology of living cells. However, with the rapid development of digital microscopy technologies, the research on subcellular structures generates huge digital image data. Therefore, it is very necessary to analyze and process these image data automatically, as well as quantitatively analyze the motion of subcellular structures.Aiming at the quantitative analysis of the motion of subcellular structures, this thesis studies the algorithems for detection and tracking of subcellular structures marked by green fluorescent proteins inside cells, from the confocal microscopy image sequence of subcellular structures. The main research contents are as follows:(1) This thesis studies the commonly used algorithms for image denoising and segmentation, as well as analyzes and summarizes the advantages and disadvantages of existing methods. Aiming at the characteristics on image analysis and object tracking of subcellular structures, this thesis proposed an algorithm for the detection of subcellular structures based on the marked image model.(2) This thesis implements the algorithm modeling of multiple objects tracking of subcellular structures. The Minimum Bounding Box(MBB) is used to describe a single subcellular structure, and the joint state is formed to represent multiple subcellular structures using Sequential Monte Carlo(SMC) algorithm. Finally, the Reversible Jump Markov chain Monte Carlo(RJMCMC) algorithm is designed, including five jumps. It is used to do sampling on the joint state-space model of multiple subcellular structures. The proposed algorithm is capable of tracking multiple subcellular structures efficiently.(3) This thesis implements the modeling of the interaction of subcellular structures. By introducing the“auxiliary height variable”and the“height swap jump”based on RJMCMC sampling, the modeling of mutual occlusion of subcellular structures is implemented.(4) This thesis proposes an algorithm for marking the residual image, and implements the algorithm modeling of the appearance and disappearance of subcellular structures in the image scene at anywhere and any time. By comparing the two neighboring frames of the binary images of subcellular structures, this thesis detects the possible new subcellular structures, and introduces the new subcellular structures based on the “appearing jump”of RJMCMC sampling. By the“appearing jump”of RJMCMC sampling, the disappearing subcellular structures are deleted. Therefore, the tracking of the multiple subcellular structures with various numbers is implemented.(5) In order to verify the performance of the proposed algorithm for detection and tracking of subcellular structures, this thesis designs and implements a system for detecting and tracking of multiple subcellular structures, applying the software development technologies such as JSP, Javabean and Servlet. This system has the characteristics of being extensible and interactive. It provides flexible parameter setting for users and can be applied to related applications of image analysis on molecular biology. Extensive experimental results show that the proposed algorithm for detection and tracking of subcellular structures, can successfully track different types of moving activities of subcellular structures in the image sequence, such as moving, disappearing, appearing, overlapping and splitting.