Research Of Segmentation And Reconstruction Algorithms For Complex Type Pulmonary Nodules Based On Multimodal Data

In the world today,lung cancer is one of the diseases with the highest morbidity and mortality.With the rise and flourishing of medical imaging technology,people can get the human lung's two-dimensional digital image sequence.PET/CT technology fully combines the PET technology's functional images and CT technology's structure images,and ultimately to get more information more than a single scan technology,providing more accurate positioning of the lesion area,which improve the detection efficiency significantly.The clinical application of these medical imaging techniques has led a great advance in medical diagnosis and treatment techniques.However,the twodimensional tomographic image can only expresses a certain cross-section information,physicians do not always make the best use of the data.Only from the obtained cross-sectional 2D images,it is difficult for people to think of the spatial structure of the lesion or organ and whether there is a spatial relationship with other tissues.Therefore,the conversion of two-dimensional lung sequence images into a three-dimensional image which more directly visualizes the appearance,size or relationship with the surrounding tissue of the lesion area can greatly improve the diagnostic accuracy of the physician.It can provide threedimensional topological structure,geometric information,anatomical structure and other key information of the lesions,and provide the technical foundation and guarantee for doctors to simulate surgical operations and human-computer interaction.Three-dimensional segmentation and reconstruction technology is proposed in this context and has been widely researched and used.The research of three-dimensional segmentation and reconstruction of pulmonary nodules is a hot spot in medical imaging research.The major work of this paper includes two parts below:(1)Existing 3D segmentation methods for pulmonary nodules cannot fully segment all nodule images and have high time complexity.For this situation,We proposed a supervoxel-based three-dimensional pulmonary nodule images segmentation method.Using lung parenchyma image sequences,registration PET/CT images by maximum Mutual Information to obtain nodule area,precisely locate nodules by multi-scale flexible circular template matching method,utilizing supervoxel's characteristics and SUV information,reconstructed the nodules by supervoxel region growing method.The experimental shows that the average processing time of 2.44 s for a slice and can achieve an average volume pixel overlap ratio of 94.97±4.00%.Compared with the 3D region growing method,our method can reconstruct the complex type nodules more accurately.(2)The blurred boundaries of the GGO,the ambiguity of the vascular and nodular junction area of the Juxta-vascular nodules and the inherent ambiguity of the medical imaging device,all this make it very difficult to segment the two complex nodules precisely.Based on this,we proposed a new method of for complex pulmonary nodules based on multimodal data and fuzzy-supervoxel.The algorithm first separates the ROI image by the Ostu method,fills the holes,automatically finds the seed points and generates a 3D mask.Then calculate the fuzzy map on the mask,and finally do the improved 3D supervoxel region growing on the fuzzy map.Obtain the three-dimensional pulmonary nodule segmentation results.The experimental shows that the algorithm can accurately segment the two types nodules in short time and has high robustness.It has certain reference value for the further study of complex pulmonary nodules' diagnosis.