【作者】 滕奇志；

【导师】 袁支润；

【作者基本信息】 四川大学 ， 生物医学工程， 2002， 博士

【摘要】 通过对活体心肌细胞图象运动特征的定量分析，从细胞力学和分子生物学的角度来研究药物的作用机理，是目前生物医学工程学科的前沿性课题。开展心肌细胞的研究，对了解人体心血管系统的基本性质，探索心血管疾病的发病机理，研究药品的药理药效、毒副作用等，有着重要的作用。而心肌细胞在不受外力作用的情况下有自主运动特性，使得对于它的动态分析更有基础理论意义。用图象分析方法，研究心肌细胞的形态变化和运动特征，为我们从本质上研究药物的作用机理，提高药物筛选的准确性，科学地调整药物配方的比例，以期达到最佳治疗效果，提供了一个新的研究手段。 本文的目的是，针对活体心肌细胞的动态图象，研究合适的图象分析算法，实现心肌细胞特征的定量分析，特别是动态特性的研究。完成的主要工作为：心肌细胞形态变化的检测，心肌细胞运动矢量检测以及运动频率和幅度的检测，心肌细胞运动机理的模拟等，并从活体细胞的图象采集到图象处理、动态图象分析，建立了一套完整的图象识别与处理实验系统。 由于活体细胞图象目标与背景无灰度差，用基于灰度分布的二值分割算法或基于梯度的边缘检测算法难以将细胞边缘提取出来，无法检测细胞形态的变化。而基本的活动轮廓算法又存在不能解决图象凹点收敛问题、抗噪声能力弱等缺陷。为此，本文在基本活动轮廓算法基础上提出三项改进方法：第一，增加自适应外部约束力，使“蛇点”在图象平滑区，即内力和图象力均为零时仍能向目标边缘移动，并使活动轮廓能向凹点收敛。第二，采用均值差分和Gauss-Laplacian差分滤波器，计算图像能量，消除图象的随机高频噪声干扰，提高边沿检测精度，增强活动轮廓的抗噪能力。第三，提出新的活动轮廓停止中文摘要准则，使轮廓收敛到目标边缘，并提高了运算速度。实验证明，新算法适合于活体心肌细胞的边缘跟踪。 通过对活体心肌细胞运动的观察，发现心肌细胞呈弹性运动，其运动方式是形变而不是位移，在很多情况下其运动状态无法从边界反映出来。但对于某些特征点存在位移。鉴此，可通过研究细胞图象中这些特征点的移动过程，实现对心肌细胞运动特征的研究。本文采用图象块匹配算法，提出了归一化运动图象匹配准则，进行心肌细胞特征点的运动矢量检测，并在此基础上计算细胞运动频率和幅度。针对图象局部极值问题，提出了多重三步搜索及自适应选择候选点算法。在第一次匹配时不仅选择一个匹配点，而是保留最优和次优两个点，再以这两个点为中心进行下一次搜索。避免了第一次没有搜索到全局最优点从而导致以后的匹配不准确，提高了搜索精度。提出了优选帧策略，即根据当前匹配情况，选择初始帧或当前帧，确定下一步的基准匹配帧。以上改进算法提高了匹配精度和运算速度。从实验结果看，较之传统的三步搜索算法具有更好的检测效果。 为了解决图象多峰值性而引起的块匹配算法不能寻到最优解的问题，本文用遗传算法进行运动矢量检测。根据心肌细胞运动特点，采取期望值与赌轮相结合的选择算法，并提出图象块匹配的归一化均方差函数作为适应度函数。经典遗传算法的交叉概率和变异概率为固定值，在实际运用中取值不易确定，且不利于群体的繁衍。针对此问题，本文提出了自适应交叉算子和自适应变异算子，根据个体的适应度函数值动态地确定交叉概率和变异概率的值。实验证实，新算法综合性能评价高于经典的遗传算法。用于心肌细胞动态图象矢量检测，匹配精度高于块匹配算法，但运行速度较低。 心肌细胞的运动非常微弱，有些序列图象难于从特征点检测出运动矢量，针对这种情况，本文提出了两种基于图象整体灰度信息的解决方法。一是，将某个区域在时间轴上的灰度信息作为一维离散信号，利用傅立叶变换的频率谱求得心肌细胞的运动频率和周期，以及心肌细胞运动的其它频率分量。二是，用序列图象的相关性算法检测运动频率。 通过对心肌细胞运动频率和幅度的研究，结合心肌细胞生理电的基本性质，从心肌细胞内部离子流的运动机理出发，将离子流运动过程用张驰振荡电路和单稳态触发电路形式进行模拟，提出了心肌细胞运动机理的电路网络模四川大学博士学位论文型，将图象处理方法与生理电实验相结合，建立了心肌细胞运动机理的数学模型的雏形，为今后工作提出了新的研究内容。更多还原

【Abstract】 Study of mechanism of medicine actions, by quantitative analysis of isolated cardiac myocyte, in myocyte dynamics and molecule biology, is one of the cutting edge researches. Cardiac myocyte research plays an important role in understanding the fundamental properties of human cardiovascular system, exploring the mechanism of cardiac disease, and studying the behaviors, effects, side effects and poison of medicine. The characteristics of cardiac myocyte auto-beating without foreign stimulation make the research sense. Research of the morphology and motion of cardiac myocyte using image analysis can reveal the fundamental mechanism of medical actions, increase the accuracy of medicine filtering, and design the optimal formula of medicine for best medical treatments.The aim of this paper is at developing theories and approaches for analysis of living cardiac myocyte motion images and implementing quantitative analysis of cardiac myocyte features. The studies been done include mainly cardiac myocyte morphology detecting, motion vector, motion amplitude and frequency measuring, and motion mechanism modeling. A system of hardware and software has been built with complete sets of functions includeing living cardiac myocyte image acquisition, image processing, motion image analysis, and image recognition.In living cardiac myocyte image, the difference of the intensities between objects and background is small. It is difficult to segment the myocytes from the background using methods of intensity based thresholding or intensity gradientbased edge detection. Active contour or snake algorithm in terms of energy function was proposed to approximate the boundary of an object with a moving snake under the criterion of energy minimization. Basic active contour algorithm has the disadvantages of inability in concave point solving and low capability in anti-noise. In this paper, three developments of active contour algorithm are proposed: An adaptive external constraint force is applied to drive the snake, which has zero image energy, i.e. lives in the areas of smooth image, and zero internal energy, to move towards the object boundary and the concave points. Mean differential filter and Gauss-Laplace differential filter are employed to calculate image energies so that the effects of high frequency random noises can be reduced, the accuracy of boundary detection can be increased, and the anti-noise ability of the snake can be improved. A new criterion is proposed for stopping the snake moving to gain better boundary detection and faster computation. Experiments done show that the new algorithms developed are better enough for living cardiac myocyte boundary detection.The action of a cardiac myocyte is elastic deformation, but not shifting movement. In many cases, the status of the action is not evident on the boundary and difficult to be recognized from the image intensity using edge detection. Observing the actions of living cardiac myocytes, it is found that the whole cell is not shifting, but some particular points are moving, and that the motion status can be studied through the analysis of the particular points’ movements. In this paper, ,a new block image matching method is developed for motion vector detection of the particular points and amplitude and frequency detection of a cardiac myocyte. In the method, schemes of multi three-step search, adaptive vote-point selection, and optimal frame selection are proposed to increase the matching accuracy and computation speed as high as possible. The experiments using the method show that the results of the detection are much better than those using traditional methods.Because of the multi-value of characteristic points, the result searched using block matching method may not the global optimal solution. This leads toinaccurate matching and becomes a disadvantage. In order to solve this problem, a genetic algorithm is developed in this thesis for motion vector detection. Considering the properties of the motion of a cardiac myocyte, a gene selection method comb更多还原