Lung Lobar Segmentation Using Tubular Tissue Density from Multidetector-row Ct Images

In evaluating thoracic function, it is effective to segment the five lung lobes from multidetector-row computed tomography (MDCT) images. Almost all of the conventional methods are based on extracting the lobar fissures; however, some parts of the fissures may not be observed from MDCT images due to CT artifacts and/or adhesions between the lung lobes. This article proposes an alternative method for segmenting the lung lobes. It is based on tubular tissue density, and is not based on lobar fissure extraction. The tubular tissues are the peripheral blood vessels and peripheral bronchi. Because tubular tissues do not exist on the boundary between the lung lobes, our method determines the boundary by finding a continuous three-dimensional space in which tubular tissues are absent. The boundary determination process is automatically performed using fuzzy control. The proposed method was applied to five normal subjects, one patient with chronic obstructive pulmonary disease, and one patient with emphysema. The absolute mean error of detecting lobar boundaries was 3.4 mm and that the volumetric accuracy for the proposed method was an absolute ratio of 3.8 and 5.9% for inspiration and expiration, respectively. The proposed method is also applicable to MDCT images in which the lobar fissures cannot be distinguished.