| Lung cancer is the most common malignant cancer. And lung nodules are mainindications of early stage of lung cancer and the lesion margin is a key characteristic todistinguish between benign and malignant. With medical imaging technology greatlydeveloping, computer assisted detection (CAD) is widely used to lung cancer to assistradiologist improving diagnosis efficiency and accuracy. Therefore, distinguishing lungnodules accurately is crucial for effective treatments and increasing survivors.Algorithms for lung nodule segmentation and detection based on CT Images areproposed in this thesis. Firstly, the paper introduces the imaging features of lung CTimages. To overcome the difficulty of repairing the lung parenchyma results inover-segmentation or under-segmentation, an algorithm based on matched filter isproposed. Experimental results show that the algorithm can repair the lung parenchymamore completely，which will lay a good foundation for the following detection ofpulmonary nodules. Considering huge amounts of CT data need to handle and the lungnodule is difficult segmented precisely using common segmentation algorithms, averageintensity projection (AIP) method is employed to reduce the amount of data, withrelatively accurate segmentation. Further, the translation Gaussian model is establishedto fit the lung nodules. With these improvements, the algorithm can achieve higheraccuracy segmentation and reduce detection time, significantly better than the currentmajor pulmonary nodule segmentation algorithm. Thirdly, for the purpose of reducingthe false positive rates in current major detection, the double-assisted detectionalgorithm is presented. Finally, the experimental results of9CT sequences containing150lung nodules (30vessels adhered nodules) show that our proposed algorithms candeal with various types of lung nodules, and get high accuracy (area overlapmeasure(AOM) achieves93%) with low false positive rates, which can satisfy theclinical application. |
PDF Full Text Request