多细胞随机性方法运动分析研究

发布时间：2018-04-08 15:17

本文选题：细胞运动分析　切入点：多模式　出处：《南京理工大学》2016年博士论文

【摘要】：生物医学图像在医疗诊断和疾病治疗中的作用日益显著,细胞图像的研究是医学图像中一个重要的分支。从细胞图像序列中提取细胞的特征信息及运动轨迹,是医学分析中一项重要的基本工作。近年来,尽管相关领域的研究者取得了很多有益的成果,但因信息量大,细胞的形状复杂且易变,细胞运动的随机性以及受到获取技术、图像质量等因素的影响,要准确计量细胞数目及获取其运动轨迹,仍然存在许多理论与技术上的难点。针对以上问题,本论文从分析多细胞存在的不同运动模式着手,充分利用时空信息,提出了若干种具有一定工程应用前景的随机性多细胞运动分析方法,改进了检测与跟踪的准确率。论文的主要研究内容如下:1.针对所研究图像序列中多细胞粘连问题,给出了一种基于阈值的混合细胞检测算法,仿真结果表明,基于该图像序列的查全率(Recall)和准确率(Precision)可达98.32%和97.03%左右。针对密集情形下出现的多细胞重叠问题,给出了一种改进分水岭混合检测算法,仿真结果表明,该方法可以很好地将重叠细胞分割为单个细胞,且几乎不会出现细胞区域标定不正确或者区域分割不完整的现象,基于该图像序列的准确率可达到96%。2.针对多细胞跟踪中存在的目标分裂、碰撞等引起的状态耦合问题,给出了一种扩展多模型粒子滤波多细胞运动分析方法。首先结合阈值处理和孔洞填充技术设计了细胞混合检测方法。其次基于细胞相互作用存在的三种事件(独立、碰撞、分裂),构建细胞运动模型,并应用角速度和面积特征参数对目标状态进行增广。最后通过计算面积和距离特征信息的差异性测量,给出了一种细胞数据关联策略。仿真结果表明了设计方法的有效性。3.针对多细胞数目时变和动力学特性差异等问题,给出了一种基于图像背景提取的蚁群多细胞运动分析方法。利用非参数核密度估计方法产生先验蚁群分布,通过多蚁群重构建立多峰信息素场,利用蚁群快速聚类算法实现细胞身份管理与状态提取。仿真结果表明了设计方法的有效性。4.针对低信噪比图像序列中多细胞近邻问题,给出了一种多任务蚁群近邻细胞运动分析方法。利用近似平均法方法提取细胞的前景图像,结合K均值聚类方法产生初始子蚁群。设计了蚁群的协作与竞争模式,并构建多峰信息素场。通过合并相似子蚁群和去除虚假目标子蚁群进行多细胞状态估计。仿真结果表明,该方法与其他跟踪算法相比具有较高的准确性。5.针对低信噪比图像序列中细胞密度变化问题,给出了一种多模式蚁群变密度细胞运动分析方法。利用当前帧区域平均似然度,并结合前一帧细胞动力学特性产生蚁群初始分布。基于细胞的稀疏与密集事件,设计蚁群的协作模式及交互竞争模式。利用蚁群之间的交互信息设计了蚁群工作模式实时更新策略。仿真结果表明了设计方法的有效性。
[Abstract]:Biomedical images play an increasingly important role in medical diagnosis and disease treatment. The study of cell images is an important branch of medical images.It is an important work in medical analysis to extract the characteristic information and motion track of cells from cell image sequences.In recent years, although researchers in related fields have made a lot of useful achievements, because of the large amount of information, the complex and changeable shape of cells, the randomness of cell movement and the influence of acquisition technology, image quality and so on,There are still many theoretical and technical difficulties to accurately measure the number of cells and obtain their motion trajectories.In order to solve the above problems, this paper begins with the analysis of the different motion patterns of multicellular, makes full use of space-time information, and puts forward several stochastic multicellular motion analysis methods with a certain engineering application prospect.The accuracy of detection and tracking is improved.The main contents of this thesis are as follows: 1.In order to solve the problem of multi-cell adhesion, a threshold based hybrid cell detection algorithm is proposed. The simulation results show that the recall rate and accuracy rate of the image sequence can reach 98.32% and 97.03% respectively.An improved watershed hybrid detection algorithm is proposed to solve the multi-cell overlap problem in dense case. The simulation results show that the proposed method can divide overlapping cells into single cells.And there is almost no phenomenon that cell region calibration is incorrect or region segmentation is incomplete. The accuracy rate based on the image sequence can reach 96.2.Aiming at the state coupling problems caused by target splitting and collision in multi-cell tracking, an extended multi-model particle filter multi-cell motion analysis method is presented.At first, a method of cell mixed detection was designed by combining threshold processing and hole filling technique.Secondly, based on the three events of cell interaction (independence, collision, division), the model of cell motion is constructed, and the target state is augmented by the parameters of angular velocity and area characteristic.Finally, a cell data association strategy is proposed by measuring the difference between the calculated area and the distance characteristic information.The simulation results show that the design method is effective.In order to solve the problem of time-varying number of multi-cell and difference of dynamic characteristics, a multi-cell motion analysis method of ant colony based on image background extraction is presented.The nonparametric kernel density estimation method is used to generate the prior ant colony distribution, the multi-peak pheromone field is established by the multi-ant colony reconstruction, and the cell identity management and state extraction are realized by the ant colony fast clustering algorithm.Simulation results show the effectiveness of the design method.In order to solve the problem of multi-cell nearest neighbor in low signal-to-noise ratio (SNR) image sequence, a multi-task ant colony nearest neighbor cell motion analysis method is presented.The foreground image of cells was extracted by the approximate averaging method and the initial ant colony was generated by K-means clustering method.The cooperation and competition model of ant colony is designed, and the multi-peak pheromone field is constructed.Multi-cell state estimation was carried out by merging similar ant colonies and removing false target ant colonies.Simulation results show that the method has higher accuracy than other tracking algorithms.In order to solve the problem of cell density change in low SNR image sequence, a multi-mode ant colony variable density cell motion analysis method is presented.The initial distribution of ant colony is generated by using the average likelihood degree of the current frame region and the cellular dynamics of the previous frame.Based on the sparse and dense events of cells, the cooperation mode and the interactive competition mode of ant colony are designed.Based on the interaction information between ant colonies, the real-time updating strategy of ant colony working mode is designed.Simulation results show the effectiveness of the design method.
【学位授予单位】：南京理工大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：R318;TP391.41

【相似文献】