Pulmonary Nodule Detection Based On EM Algorithm And Region Growing By Mathematical Morphology

Lung cancer is one of the most common visceral malignancy. Nearly half a century,the incidence and mortality rate of lung cancer have risen sharply, which become theleading cause of the cancer death. Lung cancer is considered to be the number one killer ofendangering human life and health. If it can be timely detection and treatment for the earlydiagnosis of lung cancer, the patient’s five-year survival rate will increase from14%to49%. The early symptom of lung cancer is pulmonary nodules, so the early detection anddiagnosis of pulmonary nodules is particularly important to improve the survival rate ofpatients with lung cancer.With the rapid development of computer technology,computer aided detection (CAD)is a new technology used in the field of diagnostic imaging in recent years. It automaticallyanalyzes the medical imaging by specialized computer algorithm and promtes the doctorsuspicious pulmonary nodules, so that physician can more safely and effectively analyzedata which makes mistaken sanctions and missed sanctions due to subjective factorsovercome. How to correctly detect pulmonary nodules from the image is the key ofcomputer aided detection and diagnosis.The relevant research for pulmonary nodule detection algorithm is carried out in thispaper. And automatic detection algorithm of pulmonary nodules based on the CT images isintroduced, whose respective effective algorithm is presented in the following steps. First,according to the characteristics of the lung CT images, the image is done by segmentationthreshold and anti-operation to achieve a complete extraction of the lung parenchymaregion. On this basis, the morphological filtering method for lung CT images is discussedto obtain a better result. And then the histogram specification and TopHat method are usedto enhance the filtered image, lays the foundation for the subsequent segmentation, then aimage segmentation algorithm combined with EM algorithm and region growing isproposed. The image by the EM segmentation as the seed point is to do region growing, which is the more noise immunity than the original EM method and does not easily sinkinto local optma to achieve good segmentation. At last, according to area of characteristicquantities combined with mathematical morphology, pulmonary nodules in the segmentedimage are detected.The above algorithm has been validated by the simulate test. The proposed algorithmhas achieved good test results, which indicates that the proposed detection algorithm willhave broad application prospects through continuous efforts and improvements.