Research Of Pulmonary Airway Tree Segmentation Based On Thoracic3D CT Images

Background:In modern society, air pollution is getting more and more serious, which result in high incidence of lung diseases such as lung cancer, pneumonia, emphysema and asthma. These diseases will seriously harm people’s health. With the development of multi-slice spiral CT, doctor could obtain high resolution images by one scan, combined with advanced image processing technology to diagnosis diseases. Based on the above advantages, CT imaging is considered the gold standard for diagnosis of lung disease. But even in CT image, the airway structure is complex and poor contrast, and affected by noise and volume effect. How to accurately segment the airway tree structure in the CT data still face challenges.Objective:This thesis proposes a new approach to segment the human airway structure based on the airway morphological structure, and provide reference for the clinical diagnosis.Meterials and methods:The CT image datasets used in this study are made in Shengjing Hospital using Philips and Toshiba CT with reconstructed matrix-size of512by512pixels resolution, The algorithm first based on region growing, wave propagation, morphological refinement and hole filling to segment the main airway, then segment the Low CT number voxels in lung, the next is using several morphological methods to get the lost airway, following connect the lost airway with the main airway, and finally to iteration several times to optimize the result.Results:The algorithm has been tested on28datasets, and invited radiologist of Shengjing Hospital to evaluate the segmentation results compared with the anatomical characteristics of the airway structure. The algorithm could successfully segment the fifth generation bronchi.Conclusion:The segmentation result of human airway tree could provide clinical reference for physicians and also provide data for the research of airway skeletonization and airway virtual endoscopy.