RESUMO
BACKGROUND: One of the difficulties and hot topics in the field of computer vision and image processing is extraction of the high-level pulmonary trachea from patients' lung CT images. Current, common bronchial extraction methods are limited by the phenomenon of bronchial loss and leakage, and cannot extract the higher-level pulmonary trachea, which does not meet the requirements of guiding lung puncture procedures. METHODS: Based on the characteristic "tubular structure" (ring or semi-closed ring) of the pulmonary trachea in CT images, an algorithm based on dynamic tubular edge contour is proposed. In axial, coronal and sagittal CT images, the algorithm could extract the skeletal line of the pulmonary trachea and vessel-connecting region, perform elliptical fitting, extract the pulmonary trachea by the ratio of the ellipse's long and short axes, and obtain point cloud data of the pulmonary trachea in three directions. The point cloud data was fused to obtain a complete three-dimensional model of the pulmonary trachea. RESULTS: The algorithm was verified using CT data from "EXACT09", and could extract the pulmonary trachea to the 10-11 level, which effectively solves the problems of leakage and loss of the trachea. CONCLUSIONS: We have constructed a novel extraction algorithm of pulmonary trachea that can guide the doctors to decide the puncture path and avoid the large trachea, which has important theoretical and practical significance for reducing puncture complications and the mortality rate.