Published At: International journal of imaging systems and technology
Human airway tree segmentation from computed tomography (CT) images is a very important step for virtual bronchoscopic applications. Imaging artifacts or thin airway walls decrease the contrast between the air and airway wall and make the segmented region to leak from inside of the airway to the parenchyma. This in turn begins the leakage phenomenon to build and then large parts of the lung parenchyma might be erroneously marked as the airway tree instead. Unfortunately, existing methods typically do not sufficiently extract the necessary peripheral airways needed to plan a procedure. In this article, we propose a new shape based human airway segmentation scheme to suppress the leakage into surrounding area which is based on fuzzy connectivity (FC) method. Complex medical image features such as weak boundary edges in the CT images of the lung parenchyma have fuzzy properties and can be described by FC in many extents. Our method aims to embed a mathematical shape optimization approach in a FC algorithm. Using the partial derivatives of the image data that is minimized with respect to the polar angle and cylindrical axis direction, a proper cost function based on cylindrical features of the airway branches is proposed. This approach retains the cylindrical properties of the airway branches during the segmentation process. The proposed cost function includes two parts named cylindrical‐shape feature and smoothed final error term. The former term arranges the underlying voxels on a cylindrical shape and the latter term controls and smoothes the final error considering the local minima's problem. To evaluate the efficiency of our proposed optimization technique in term of segmentation accuracy, the cost function is first applied to the simulated data with the spongy shape of leakage and the leakage origin. The impact of each term of the proposed cost function on the final error and the convergence of the algorithm are also evaluated. Then, the cost function with best proper parameters is applied to real image dataset. Comparisons of the results on multidetector CT chest scans show that our segmentation algorithm outperforms the fuzzy region growing algorithm. Quantitative comparisons with manually segmented airway trees also indicate high sensitivity of our segmentation algorithm on peripheral airways. On the basis of the results, it is concluded that the proposed method is able to detect more branches up to the sixth generation with no leakage which provides 2–3 more generations of airways than several other methods do. The extracted airway trees enable image‐guided bronchoscopy to go deeper into the human lung periphery than past studies. The novelty of our proposed method is to apply a shape optimization approach embedded in an efficient FC segmentation algorithm. Hence, our method prevents leakage from its origination in contrast to most previously published works that just set their algorithms to repeat the segmentation steps to reduce leakage. As our results indicate leakage suppression in human airway segmentation instead of readjusting the segmentation parameters, more airway branches can be extracted with correct shape.