Research On Irregular Lung Nodule Automatic Segmentation And Spiculation Detection

Lung cancer is a malignant tumor in cancer incidence and mortality in the world’s highest. Pulmonary nodules are round or approximate circular opacities in the lung whose diameter are less than 3cm, edge’s smooth, lobulated or spiculated. It is very difficult to identify the nodule because of its uncertain locations, various shapes and sizes, similar density with other organizations. Research shows that if lung cancer can be detected and treated in early stage, the five-year survival rate of patients can be 9～14% to 60%～70%. Therefore, early diagnosis and treatment of pulmonary nodules is critical to improve the survival rate of lung cancer patients.CT plays an important role in the detection and characterization of lung nodules. The imaging features of pulmonary nodules in CT images include nodule size, growth rate, edge feature, calcification, bronchial symptoms, pleural indentation and the relationship with the surrounding tissue. These imaging features are important to differentiate benign nodules from malignant nodules. Study said that most cancer cases can be made a preliminary judgment according to the nature of lung nodules in CT images. The edge information of pulmonary nodules include the edge is smooth or not, with or without lobulation and spiculation. Malignant nodules usually have irregular edges with deep lobulations, thin and short spiculations, the periphery edge structures to the nodules mustered. In contrast, benign nodules mostly have clear and smooth edges without lobulations. Spiculation is an important edge character that indicates higher malignant. Spiculation is straight, strong and performance as a thin strip structure extending radially from the nodule edge to the surrounding with no branch.Computer aided diagnosis(CAD)system of pulmonary nodules applied advanced computer image processing technology automatically analyze the lung CT images and assisted radiology physicians to identify and diagnose pulmonary nodules. If the CAD system can accurately quantify the spiculation and provide intuitive quantitative data, we can help them identify the benign and malignant pulmonary nodules, improve efficiency and accuracy of the identification. At present, few domestic and foreign scholars study the detection and quantification methods of spiculation, development is not mature. On the one hand, the existing detection methods can’t obtain complete spiculation which lead to inaccurate quantification of spiculation level; on the other hand, reasonable and accurate quantitative indicator of spiculation is also crucial to quantify the spiculation levels. There is currently no unified and accurate quantitative indicators of spicuation to be proposed. Therefore, spiculation detection of pulmonary nodules is still a lot of work to do. This article focuses on depth research of spiculation of pulmonary nodules which includes a lot of experiments of lung nodule segmentation and spiculation detection. In CT images, gray value of spiculation compared to lung parenchyma is much lower with nodules to lung parenchyma, the shape of spiculation is narrow, so it is easy to miss spiculation when segment the nodule and inaccurately analyze the characteristics of spiculation. To solve this problem, first, we accurately segment the main part of the lung nodule. Then, we extract the spiculation completely. In addition, we present quantitative indicators of spiculation to achieve the precise quantification of spiculation level. The main work of our experiments include:1, Segmentation of the main part of the lung nodule in CT imagesAccurate segmentation of pulmonary nodules is a prerequisite in lung cancer’s diagnosis and treatment, and also important for assessment of tumor growth and discrimination of benign and malignant lesions. Accurate segmentation of the main part of pulmonary nodules is the basis of spiculation extraction and quantification, there exist lots of lung nodule segmentation methods, in which region growing method is classic, simple and fast, firstly choose region growing method to initial pulmonary nodule segmentation. Level set algorithm is of accuracy, flexibility and calculation stability, the basic idea is to convert the evolution problem of closed curve (curved surface) to the implicit solution of level set function curve (curved surface) in a higher-dimensional space. We select the distance regularized level set method with no need for re-initialization to precisely segment the nodule and the region growing segmentation result is as the initial contour for level set method. This will not only speed up the convergence process, but also close to the real subject of the nodule boundaries.2, Detection and quantification of lung nodules spiculationSpiculation detection and quantitative evaluation of the lung nodule can assist doctors distinguish between benign and malignant pulmonary nodules, early diagnosis and treatment of malignant nodules can help improve patients’ survival. At present, the main problems of the spiculation detection and quantitative evaluation methods are:1) The extracted spiculation quantitative indicator which reflect spiculation level is of great dependence on the segmentation of lung nodules and also need the human’s experience that cause the inaccurate classification.2) The extraction of spiculation isn’t complete yet and caused the quantification inaccuracies. A new method was proposed to accurately detect and quantitatively evaluate the lung nodule spiculation. The growth direction of spiculation is approximately perpendicular to the border of main body of the nodule or with a certain range, while vessel and other tissues do not have this special direction. According to the growth characteristics, spiculated lines connected to the nodule boundary were extracted using a line detector in polar coordinates system on the basis of the accurate segment of the main part of the lung nodule. In order to quantitatively describe the lung nodule spiculation, spiculation index (SI) was introduced to quantitatively evaluate the lung nodule spiculation. This method can extract the full spiculation and help the doctors observe and diagnose, SI value overcome the problems that traditional quantitative indicators vulnerable to influence of nodule segmentation accuracy and the parameters to be set artificially by experience. In order to further evaluate the effectiveness of the method, we compare the experiments result with the lung image database consortium (LIDC), distinguish between the spiculated and non-spiculated nodules and compare with the radiology doctor’s classification; the consistency and correlation of spiculation index between the method and Lung Image Database Consortium (LIDC) are evaluated in detail after normalized. The experiment results show that the method can effectively detect and quantitatively describe the lung nodule spiculation in CT images.3, Lung nodule segmentation, spiculation detection and spiculation level quantification in three-dimensional CT image sequencesIn order to provide doctors a more intuitive observation of lesions, we design the main pulmonary nodule segmentation, spiculation detection and quantification methods of three-dimensional CT image sequence. On the basis of the accurate 3D segmentation of lung nodule, a 3D special detector was constructed to detect the spiculation and obtain the complete spiculation of the nodule. To achieve a quantitatively evaluation of the lung nodule spiculation, spiculation index of 3D CT images was introduced as the quantitative measurement of spiculation features. In order to quantitatively evaluate the effectiveness of the proposed method, we compare the experiments result with the lung image database consortium (LIDC), distinguish between the spiculated and non-spiculated nodules, calculate the consistency and correlation of spiculation index between the method and Lung Image Database Consortium. The experiment results show that the method can effectively describe the lung nodule spiculation in CT images. Furthermore, to evaluate the effectiveness of 3D spiculation detection and quantification methods, we compare the 3D spiculation evaluation results with the 2D results. Results show that it is more conductive for doctors to observe and diagnose the lung nodule under the three-dimensional space. Doctors will obtain richer information and the spiculation quantification results are closer to the results in LIDC database. Through the investigation of computer-aided diagnosis system (CAD) and the field of pulmonary nodules at home and abroad, we combine lung knowledge of anatomy and CT imaging features, take the data of LIDC database for the study and focus on the detection and quantitatively evaluation of the lung nodule spiculation. Spiculation index was introduced as the quantitative measurement of spiculation features on the basis of accurate segmentation of pulmonary nodules and the detection of the spiculaion. Our experiments use two dimensional and three-dimensional CT image sequence respectively for pulmonary nodules segmentation and spiculation detection. Results show that this method can help doctors observe and diagnose the lesions and provide a more reliable evidence for computer-aided diagnosis of pulmonary nodules in CT images.