Research On Algorithm Of Lung Nodule Detection Based On The Image Processing Technology

The study of the computer-aided detection is always seen as the research hotspot in the medical image processing field. It involves the knowledge areas of medical anatomy, digital image processing and computer graphics, etc. It combines the experience knowledge and machine processing method, and worths great significance and value in the aspect of medical diagnosis, surgical planning simulation and teaching. This thesis is about the research of computer-aided detection algorithm focusing on the lung nodules detection. It aims to present an effective lung nodule detection algorithm to detect the lung nodules as fully as possible and to minimize the number of false positive nodules. It also hopes to reduce the workload of doctors and scholars and do a good job to locate the lesions quickly and accurately.The thesis makes a detailed study on the development of current computer-aided detection system of lung nodules at home and abroad, concludes a variety algorithm of prevalent medical image segmentation and analysis. After the study of the computer-aided detection system and related detection theory and technology profiled home and abroad, the thesis mainly does the following work about the computer-aided detection of lung nodule.First, it mainly studies four methods on the lung nodule detection, like region growing, mathematical matrix calculation, level-set active contour extraction and multi-threshold segmentation. After comparison of the four algorithms, the multi-threshold segmentation is selected as the extraction method.Second, it achieves an optimized multi-threshold method and is applied to scan lung nodules in CT images to obtain the candidate primary lung nodules. The results show that the method can keep the information well and the detection rate is reliable.Third, it processes the target feature extraction and feature selection for lung nodules, and forms a feature space to descript the lung nodules. Moreover select the optimal feature set as the classification feature space.Finally, it uses the statistical learning theory to classify the feature space, and delete the non-nodule targets. The trained support vector machine is used in the classification. It can separate the blood vessels and highlights interference from the real nodules. This thesis integrates the content into an algorithm. After input the CT source images, it can finally output images which are labeled nodules through detection and reduction method...