The Feature Extraction Of Pulmonary Nodules In CT Images

CT images are widely used in the diagnosis and treatment of lung cancer. In recent years, the incidence rate and mortality rate of lung cancer has been both dramatically growing, currently ranks first beyond all other cancers. Hence,there are more and more patients that need to be remedied ,which increases the workload and depresses the efficiency that causing misdiagnosis. For the purpose of helping doctors minimize the workload, many researchers have proposed computer-aided diagnosis on CT images. CT image feature extraction can help doctors quickly diagnose the illness, which not only reduces the burden of repetitive work and diagnose time, but also brings more accurate and concise information for clinical treatment.The main achivements are as follows:By deeply analyzing and studying papers both at home and broad on medical image segmentation and feature extraction, two different image segmentation algorithms are proposed according to the features of CT image. The first one is the improvement of optimal thresholding segmentation algorithm. The second one is a local Otsu thresholding segmentation algorithm on the basis of global Otsu thresholding segmentation algorithm , is a adaptive thresholding segmentation algorithm. The experiment results show that the adaptive thresholding segmentation algorithm is obviously better than the improved and original optimal thresholding segmentation algorithm. Hence it lies a good foundation to the subsequent feature extraction. There are three steps to the pulmonary image extraction in this paper. First, pulmonary image extraction is preprocessed. Second step is the extraction of pulmonary contour and thoracic cavity. Finally we get the accurate and complete pulmonary parenchyma image by contour tracking.This paper purposed an improved 2DPCA with adaptive judgment of threshold value.2DPCA is widely used in face recognition and feature extraction, its performance is directly influenced by accumulative contribution ratio. The improved algorithm this paper proposed utilized statistical histogram to examine the grey value density of some segmented pulmonary images, and calculated the proportion of grey value with high density to all grey value from all segmented pulmonary images,and calculated the percentage of mean, with this mean to limit the accumulative contribution ratio.This mean may be called 2DPCA of the threshold. Experiments show that the proposed algorithm performs a better extraction result and recognition rate than the original one with only a little more time.