Research On Pulmonary Nodule Detection And Segmentation Using CT Images

In recent years,because lung cancer has long been the leading cause of cancer mortality and morbidity in China,clinicians pay special attention to the early detection of lung cancer.As a significant marker of early lung cancer,the detection of pulmonary nodules has become the focus of clinical and scientific research.However,each lung CT data contains multiple layers and the distribution of blood vessels in the lungs is complicated,so that blood vessels and nodules on a single slice are both round and difficult to distinguish,which increase the diagnostic burden of the doctor and may cause misdiagnosis and missed diagnosis easily.At this time,they need the participation of computer-aided diagnosis.We combine the data in the international open database with the actual clinical data in this article to ensure the correctness of the research results and ensure the clinical feasibility.In this paper,the pulmonary nodule detection rate is mainly improved by enhancing lung image contrast and reducing the range of suspected pulmonary nodules during the processes before lung nodule detection.Firstly,in the segmentation of lung parenchyma,an advanced method is proposed for the repair of the depression in the lung parenchymal edge,which combined with the convexity of lung parenchyma and the continuity between layers in the nodule region.Then,during the process of image enhancement,a novel blob filter is constructed to select the suspected nodule region and achieve local enhancement,which is derivatived by vascular enhancement method based on the Hessian matrix eigenvalues.In the meantime,we use binary image for structural filtering instead of the original image in the traditional image enhancement operation,which avoid the effect of the grayscale in the image on the structure recognition.After that,the filtering results of binary images are used with the original image information to determine the final pulmonary nodule screening area.Finally,the pulmonary nodules detection step is achieved using the prior information of the pulmonary nodules,which finally get the detection results.The purpose of this research is to achieve automated lung parenchymal extraction and improve the accuracy and speed of lung nodule detection.Most of the existing lung parenchymal extraction methods cannot get rid of the artificial participation during the segmentation process.In this paper,automatic segmentation of lung parenchyma is achieved by using convexity and continuity between layers.In addition,the experiment results show that the use of spherical filters to process binary images can significantly improve the recognition rate of nodules.Traditional pulmonary nodule detection methods may have high detection rates in the experiment,but their applicability is not good due to the complexity of patient conditions in clinical data.The method proposed in this paper,after the verification of clinical complex data,is more practical than the previous method.