Automatic extraction of three dimensional lung texture tree from HRCT images

Accurate segmentation of lung texture tree is an essential step for diagnosing pulmonary diseases, including pulmonary emboli and nodules detection, which provides powerful information for research of automatic computer-aided diagnostic (CAD) systems. It still remains a challenging problem because of partial volume effects, high density airway walls and no difference on CT values between arteries and veins. In this paper, we present a novel approach to automatically extract lung tissue textures which contain bronchus and pulmonary veins and arteries. Firstly, we extract the bronchus branch by branch with an adaptive region growing approach. Secondly, a new technique based on selective marking and depth constrained (SMDC)-connection cost is proposed to segment the lung blood vessels. At last, we present a new method to separate the lung blood vessels into pulmonary veins and arteries by using an anatomical feature between each vessel and bronchus. About 91% of arteries and 92% of veins are correctly extracted. The results show that the proposed algorithm provides an automatic and efficient method to extract pulmonary veins and arteries and bronchus.