面向医学图像分割的免疫模糊聚类改进研究

发布时间：2018-06-23 12:17

本文选题：医学图像分割 + 模糊C-均值聚类　；参考：《东华大学》2017年硕士论文

【摘要】：医学图像分割一直是近几年研究的热点问题,由于受到成像设备等外界因素的干扰,医学图像呈现边界模糊,强度不均匀的特点,影响了医生对病情的诊断。因此如何进行快速、有效、准确的图像分割对后续临床分析起着至关重要的作用。由于图像自身的模糊不确定性,一些学者将模糊理论引入图像处理中,利用模糊聚类进行图像分割。其中,模糊C-均值聚类算法(fuzzy C-means algorithm, FCM)应用最为广泛,但是该算法在运行过程中需要提前确定初始聚类中心和聚类数,并且对噪声比较敏感,容易陷入局部最优解。人工免疫算法(AIS)继承生物免疫系统优良特性,具有分布式并行处理,快速收敛性,全局寻优等特点,在诸多领域得到成功的应用。本文结合人工免疫系统和聚类问题进行研究,提出一种新的免疫聚类算法,主要改进如下：1、噪声会严重影响算法的执行效率和分割效果,在进行医学图像分割之前加入改进型的开关极值中值滤波,改进的滤波算法能够有效识别噪声点和有用数据点。通过实验证明,改进型的算法实现了既能去除噪声,同时可以很好的保护图像边缘细节。2、为能够提前确定较为准确的初始聚类中心和聚类数,对灰度直方图通过插值法进行平滑,能够有效过滤伪峰点,进而通过峰值检测得到优秀的初始聚类中心和图像的聚类数,通过多中心组合的方式提高算法准确性,避免陷入局部极值。通过实验证明,改进后的FCM算法更加稳定,分割精度更高。3、对基本克隆选择算法引入抗体种群浓度调节机制,既保证种群不断的向优良的特性发展,又能避免种群过度单一化,有效的保持抗体多样性。再结合高斯变异和柯西变异的特点,提出一种混合自适应变异——高斯-柯西混合自适应变异,能够动态调节变异步长,避免算法陷入局部最优解,进一步提高算法的全局寻优能力。最后充分利用免疫记忆机制,不断保存优秀抗体,替换差的抗体,使得算法不断向良性发展。通过实验测试,改进后的算法全局寻优能力和收敛速度得到有效改善。4、将改进的克隆选择算法优化改进后的FCM算法,再配合改进的滤波算法,与传统FCM算法比较。新算法的抗噪能力、收敛速度、全局寻优能力、分割精度都得到显著提升。
[Abstract]:Medical image segmentation has been a hot issue in recent years. Due to the interference of imaging equipment and other external factors, medical image presents the characteristics of blurred boundary and uneven intensity, which affects the doctor's diagnosis of the disease. Therefore, how to carry out fast, effective and accurate image segmentation plays an important role in subsequent clinical analysis. Due to the fuzzy uncertainty of image itself, some scholars introduce fuzzy theory into image processing and use fuzzy clustering to segment images. Among them, fuzzy C-means clustering algorithm (fuzzy C-means algorithm) is the most widely used, but it needs to determine the initial clustering center and the number of clusters in advance in the running process, and is sensitive to noise, so it is easy to fall into the local optimal solution. Artificial immune algorithm (AIS) inherits the excellent characteristics of biological immune system, and has the characteristics of distributed parallel processing, fast convergence, global optimization and so on, and has been successfully applied in many fields. In this paper, the artificial immune system and clustering problems are studied, and a new immune clustering algorithm is proposed. The main improvements are as follows: 1. Noise will seriously affect the efficiency and segmentation effect of the algorithm. An improved switching extremum median filter is added before medical image segmentation. The improved filtering algorithm can effectively identify noise points and useful data points. The experimental results show that the improved algorithm can not only remove noise, but also protect image edge detail. 2. In order to determine the accurate initial clustering center and clustering number in advance, the improved algorithm can not only remove the noise, but also protect the image edge details. The gray histogram is smoothed by interpolation method, which can filter pseudo peak points effectively, and then obtain excellent initial clustering center and image clustering number by peak detection, and improve the accuracy of the algorithm by multi-center combination. Avoid falling into local extremum. It is proved by experiments that the improved FCM algorithm is more stable and has higher segmentation accuracy. The introduction of antibody population concentration regulation mechanism to the basic clone selection algorithm can not only ensure the population to develop to excellent characteristics, but also avoid the excessive singularity of the population. Effectively maintain antibody diversity. Combined with the characteristics of Gao Si mutation and Cauchy mutation, a hybrid adaptive mutation-Gauss-Cauchy hybrid adaptive mutation is proposed, which can dynamically adjust the step size of the mutation and avoid the algorithm falling into a local optimal solution. The global optimization ability of the algorithm is further improved. Finally, we make full use of the immune memory mechanism, keep the excellent antibody and replace the bad antibody, so that the algorithm continues to develop benign. The global optimization ability and convergence speed of the improved algorithm are improved by experiments. The improved clone selection algorithm is optimized by the improved FCM algorithm, and the improved filter algorithm is combined with the traditional FCM algorithm to compare with the traditional FCM algorithm. The performance of the new algorithm, such as anti-noise, convergence speed, global optimization and segmentation accuracy, has been improved significantly.
【学位授予单位】：东华大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP18;R44

【参考文献】