Characristics Detection Of Solitary Pulmonary Nodules Based On Time Series

Lung cancer is one of the most rapidly growing and mortality rate, which is one of the most important malignant tumors. Therefore, the early diagnosis and treatment of lung cancer is significant for patients,pulmonary nodules are the early manifestation of lung cancer. The early diagnosis of pulmonary nodules by using computer aided detection method has gradually became a hot spot in recent years. In this paper, We propseded a new method to identify the morphological features of pulmonary nodules, which is based on the point of view of machine learning and times series algorithm.And we detected the good or evil of pulmonary nodules with the morphological features and other medical signs which contain semantic information.the contents and the corresponding achievement of this thesis include:(1) Using the threshold algorithm and region growing algorithm to extract lung nodules from the human lung CT images in the US public database, and extract the. external shape of the pulmonary nodules by edge detection and edge tracking algorithm.(2) Expand the edge of the lung nodules according to the angle of the order of 0 to 360 degrees,and then we carry out the cubic spline interpolation and uniform sampling algorithm to make the lung nodule series homogenization,so that we can regard it as special time series.(3)we propse to make use of time series similarity detection algorithm on detecting the lung nodule morphological features,which contains the spiculation and lobulation.The DTW and PRCD algorithm were used to detect the characteristics metioned above respectively, and the recognition accuracy rates are both more than 80%,it performs better than the traditional mathematical description algorithm.(4) Four medical signs,which is spiculation, lobulation, empty sign and calcification, are used for the classification of benign or malignant pulmonary nodules.The four featues have a common character is that they are all have a high correlation with the severity of the pulmonary nodules,and they are all contain the high level semantic information.As the input of these four characters, SVM is used to classify the lung nodules.Both theoretical and experimental results show that the proposed method can provide a new way for the computer aided detection of pulmonary nodules. It is proved by the practice that this method is more accurate and robust than the traditional method.