Segmentation of lung lobes in isotropic CT images using wavelet transformation

Advanced multi-slice CT scanners produce isotropic CT images, which have pixel dimensions equal to their image thicknesses of 0.6 mm. Comparing to clinical standard CT images with a thickness of 2.5 - 7.0 mm, isotropic CT images have clearly visible lobar fissures. This poses a challenge for developing automatic algorithms to identify the fissure locations and curvatures. This paper presents a wavelet algorithm that allows automatic identification of the left and right oblique fissures, as well semi-automatic identification of the horizontal fissures. This algorithm took a two-stage approach: (a) adaptive fissure sweeping to find fissure regions; and (b) wavelet transform to identify the fissure locations and curvatures within these fissure regions. Tested on 8, 6 and 6 stacks of isotropic CT images for the left oblique, right oblique and horizontal fissures, respectively, the algorithm yielded an accuracy of 77.1 - 93.6% with strict evaluation criteria. This provides promising potential for developing an automatic algorithm to segment lung lobes.