System Design Study Of Pulmonary Nodule Cad Integrating With Three-dimensional Information Of DICOM CT Sequence

Lung cancer is the highest and mortality of malignant tumor, the mortality of about91.6%, and0.5%trend growth each year, accounting for23.8%of all cancer deaths.Pulmonary nodules are the most common forms of early lung cancer. CT images candirectly display and observe the pathological changes, is the most effective way forearly lung cancer and lung nodules detection. Computer-aided diagnosis can overcomethe influence of subjective factors and automatically analyze patient CT image sequence,identify the suspected lung nodules, and provide the nodule shape, volume, grayinformation auxiliary parameters, reduces the risk of missed diagnosis andmisdiagnosis.In accordance with CT Imaging sequence information for DICOM format,combined with the clinical characteristics of pulmonary CT images, this paper make adetailed study on DICOM format and resolution process, image sequence of normalizedpretreatment, automatic lung parenchyma single image data extraction, serialization issplit,3D-space of pulmonary nodule detection, abstracting Characteristic andclassification. Specifically：First of all, analyses the international DICOM medicalimage format and the resolution process, summarizes the imaging characteristics havebeen founded in recent years’literature and clinical studies which was suitable for CADof pulmonary nodules, also have a study on pulmonary CT images and DICOMsequence normalization set window width and window level, noise reduction and otheraspects of research and experimentation. And then proposed a layer-by-layer algorithmabout eliminate backgrounds build lung parenchyma template to obtain comprehensiveand accurate data of the lung parenchyma, construct a circular structure data on therepair of the lung template borders, and also introduce Graph search algorithms basedon supervision function to gets the lung parenchyma borders, to achieve automaticserialize segmentation of lung region.Then to use①combine improved localthresholds and connecting area marking algorithm,②design two-and three-dimensionalGaussian template matching algorithm for nodule in lung parenchyma and③three-dimensional data space based on gray, gradient to obtain seeds of three-dimensional Adaptive region growing algorithm in parallel to ROI measurement,extract multiple two-dimensional feature and three-dimensional features to make up thematrix. And then take multiple sample training build linear classification criteria forclassification and discrimination.Through a large number of experiments, by using the images and expert annotationinformation provided by LIDC of The United States as subjects and evaluation criteria,use segmentation and detection algorithm presented in the text, after a great deal oftraining and validation feedback, algorithm of nodule detection sensitivity had reached90%, the false-positive rate of detection of nodules to2.3FPs/Scan.