基于群体智能的真假肺结节分类算法研究与实现

发布时间：2019-03-09 12:19

【摘要】：肺癌是目前已知类型肿瘤中死亡率最高的一种,肺结节是早期肺癌的表现形式,肺结节检测是利用计算机辅助肺癌诊断的重要方式。由于肺组织的复杂,肺结节的种类多种多样,导致了经过图像预处理之后仍然存在大量的假结节。本文针对检测过程中,较难区分真假结节的问题,引入群体智能优化方法,设计并实现了肺结节分类算法,从以下几个方面对肺结节分类进行了讨论： (1)肺结节的形态和纹理多样,单一特征不能取得较好的描述效果。本文对肺结节进行多特征提取,包括灰度特征、纹理特征、梯度特征以及形状特征,并且将二维与三维特征相结合,全面描述图像特性。 (2)针对肺结节数据不均衡与特征高维的问题,引入代价敏感支持向量机(Cost-sensitive SVM),利用其中的RBF核函数,将多维数据映射到高维空间,使原来在低维空间中不可分的数据变得可分,并提出用多分类器组合分类,进一步提高分类效果。 (3)将群体智能优化方法应用于结节分类问题中,利用遗传算法、粒子群算法、人工蜂群算法等方法实现特征选择与分类器参数调整,有效提高了分类准确率。本文设计和实现的真假肺结节分类算法,保证了肺结节检测中对真假结节的有效分类,具有良好的实用性。
[Abstract]:Lung cancer is the highest mortality among known types of tumors. Pulmonary nodules are the manifestations of early lung cancer. Detection of lung nodules is an important way of computer-aided diagnosis of lung cancer. Because of the complexity of lung tissue and the variety of pulmonary nodules, there are still a lot of false nodules after image preprocessing. In order to solve the problem that it is difficult to distinguish the true and false nodules in the detection process, this paper introduces the swarm intelligence optimization method, and designs and implements the lung nodule classification algorithm. The classification of pulmonary nodules is discussed from the following aspects: (1) the morphology and texture of pulmonary nodules are diverse and the single feature can not get a good description effect. In this paper, multi-feature extraction of pulmonary nodules is carried out, including gray-scale features, texture features, gradient features and shape features, and the two-dimensional and three-dimensional features are combined to describe the image characteristics in an all-round way. (2) to solve the problem of imbalance and high dimension of pulmonary nodule data, cost-sensitive support vector machine (Cost-sensitive SVM),) is introduced to map multi-dimensional data to high-dimensional space by using the RBF kernel function. In order to further improve the classification effect, the data which were not separable in the low dimensional space were made divisible, and the multi-classifier combination was proposed to further improve the classification effect. (3) the swarm intelligence optimization method is applied to the problem of node classification. Genetic algorithm, particle swarm algorithm and artificial bee swarm algorithm are used to realize feature selection and classifier parameter adjustment, which improves the classification accuracy effectively. The algorithm designed and implemented in this paper ensures the effective classification of true and false nodules in the detection of pulmonary nodules and has good practicability.
【学位授予单位】：东北大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：R563;TP18

【参考文献】