Research And Application Of Genetic Roved Particle Swarm Optimization For Feature Selection

In this thesis, a genetic improved binary particle swarm optimization for feature selection is proposed and a computer-aided diagnosis system for malignant risk estimation of thyroid nodules with our method for the selection of morphology and texture features extracted from ultrasound images is established. This thesis is organized as follows:In the first place, the background of optimization problem, particle swarm optimization, genetic algorithm and feature selection is introduced. Especially, a review of literature about several major improve algorithm for particle swarm optimization is described.Then, the core of genetic improved particle swarm optimization is achieved by using the the natural coding feature of binary particle swarm, and is conducted comparative test of its performance of other algorithm. According to the aim of our research, the proposed algorithm is applied to the problem of feature selection and compared with other ones. Experiments reveal that the proposed genetic improved particle swarm optimization can significantly enhance the ability of feature selection.Based on the above study, a computer-aided system for malignant risk estimation of thyroid nodules is developed. Before making the classification, several methods of digital image processing are used to feature extraction for thyroid nodules in Ultrasound image. And finally78morphology and texture features are achieved. Soon after, the proposed method is applied to these features in order to generate the best subset of features by considering some quality of classification which built with support vector machine and achieved88.20%accuracy. The result shows that the compress, smoothness and some other features of thyroid nodules play a key role to distinguish benign and malignant ones.At last, the study is expanded to stochastic process inspired by the random application of particle swarm optimization and genetic algorithm. In probability theory, a stochastic process, or sometimes random process is a collection of random variables, representing the evolution of some system of random values over time. From an instructive view, the determinacy and order which universally exist in nature probably are the result of atomic and hierarchical randomness. This will demonstrate the blueprint of my future research.At my practical work, thyroid nodules become the first choice for developing computer-aided system because of their multiple features in the Ultrasound images. In my second graduated year, I have extracted the morphology and texture features from ultrasound images based on my under-graduated research experience, and had once tried to seek an auto recognition algorithm for nodules’ location and segmentation. However it turned into a must overcome problem that feature selection in classification with the further research. Therefore the intelligent algorithm for feature selection have been introduced to the study inspired by the book Programming Collective Intelligence. In the Spring of the third year, I changed my research field from medical computer-oriented application research to bioinformatics which force me to focus on particle swarm optimization and get a primary improved algorithm with genetic.