High-resolution CT images of computer-aided diagnosis of pulmonary lesions
|School||University of Science and Technology of China|
|Course||Pattern Recognition and Intelligent Systems|
|Keywords||Image Segmentation Boundary tracking Lung region Correlation Matrix Detection of pulmonary nodules Vascular testing HRCT|
Lung cancer is the leading cancer patients, the main reason for high mortality, its early diagnosis and early treatment is the primary means to improve patient survival. Early symptoms of lung cancer lung nodules form generally appears therefore correctly detect lung nodules, an early understanding of lung lesions, thus treated, in saving the lives of lung cancer patients has important significance. In recent years, China's major hospitals in the number of installed multi-slice CT increased significantly, they can provide the lungs high resolution CT (High Resolution Computerized Tomography, HRCT) images, can be used to better evaluate pulmonary nodules in lung tissue interface and the internal structure of nodules, correct diagnosis of lung disease provides a powerful tool. But the mass of CT data to provide more detailed and more accurate diagnostic information, but also to read the piece doctors increased heavy workload, in order to improve the efficiency of the doctor's diagnosis and reduce their labor intensity, computer-aided diagnosis (Computer- AidedDiagnosis, CAD) system came into being. To study computer-aided diagnosis of pulmonary disease, help doctors diagnose early stage lung cancer, we must first solve the problem is to face and chest CT images of the computer to automatically detect lung nodules. To achieve the automatic detection of pulmonary nodules, image segmentation need to involve a series of processing and analysis methods. In this paper, HRCT data for the study, a computer to automatically detect lung lesions targeted research status in this field done extensive research, through a combination of anatomical knowledge of human tissue for detection of lung lesions and its related medical image processing methods for in-depth study and completed the following new idea work: (1) proposed a lung area automatic segmentation algorithm. Lung segmentation result is good or bad will directly affect the efficiency and effectiveness of subsequent treatment. The algorithm for lung CT images with gray values ??of other organizations within the human body there are significant differences in gray value features, the use of iterative calculations to strike the optimal threshold segmentation method, reducing the segmentation threshold selection effect . CT study of two adjacent levels of the tracheal / bronchial position relationship between the image in the front layer using tracheal / bronchial automatically determines the position of the image in the lower trachea / bronchus region growing seed point raised trachea / bronchus automatic area growth method to remove the tracheal / bronchial region of the lungs boundary extraction of interference. Use Based on 8 - neighborhood search boundary tracking algorithm to remove the background interference CT images and access to the lung area boundaries, to avoid a number of morphological operations, saving processing time and according to the smoothness of the boundary torso and lungs sexual characteristics in 8 - to improve neighborhood search strategies to improve the boundary tracking speed. The algorithm can quickly and accurately complete the CT images automatic segmentation of lung regions. (2) Early lung cancer is generally in the form of pulmonary nodules on CT images of the lung nodule enhancement can improve the accuracy of detection of pulmonary nodules. In this thesis, on the assumption that pulmonary nodules are spherical, cylindrical vessel is proposed based on the correlation matrix based pulmonary nodule enhancement algorithms. First, according to the positive and negative eigenvalues ??of the Hessian matrix from a relatively large number of gray value pixel filter out points to be enhanced, and then calculate the points need to enhance the correlation matrix, using the correlation matrix of the point value of the relationship between the characteristics of the design pulmonary nodule enhancement filter. Algorithm uses the image of a first-order partial differential information on pulmonary nodules can be effectively enhanced while reducing the sensitivity to noise of traditional filters. (3) Considering the fast two-dimensional and three-dimensional detection characteristics of high detection accuracy is proposed based on three-dimensional structure of pulmonary nodules lung nodule detection algorithm. The algorithm used in the on-chip two-dimensional layer convergence index filter to generate a grayscale candidate lung nodules, and then calculate the three-dimensional characteristics of the candidate lung nodule candidate lung nodules to remove false positive pulmonary nodules. Convergence index filter to quickly find the CT layer chip round and oval-shaped region, generating candidate lung nodules; take full advantage of the spatial structure of nodules information to remove false-positive pulmonary nodules, improve the detection accuracy. During execution of the algorithm does not require human-computer interaction, with high sensitivity and low false positives. (4) for lung nodule detection process due to vascular effects of false-positive pulmonary nodules present problem, a blood-based detection of pulmonary nodules culling algorithms. The vessel has a continuous cross-section as a two-dimensional Gaussian density distribution of the tubular body, the tubular model was designed based on vascular detection filter, use it to do further screening lung nodule candidate to reduce the false-positive rate. The blood detection filter design can also be used for other applications to detect blood vessels.