Pulmonary Artery Visualization for Computed Tomography Angiography Data of Pulmonary Embolism

Pulmonary embolism (PE) is a preventable life-threatening disease that is among the top three most common causes of cardiovascular deaths. Producing an accurate diagnosis can be challenging. Nowadays, computer-aided diagnosis has proven itself to be a useful tool for physicians. However, computers need to recognize the relevant human anatomy as accurately as possible. In case of PE, pulmonary artery is the structure in which the lesion manifests. Segmentation of the structure is required to define the area to search for emboli. In this study, we proposed a segmentation algorithm that accurately identifies voxels occupied by pulmonary artery in computed tomography angiography (CTA) images. The output could directly be used to create the 3D visualization of the pulmonary artery network for the PE diagnosis. The algorithm consists of three parts: lung mask extraction, pulmonary artery detection, and pulmonary artery connection. The technique involves several conventional image processing methods such as morphological operations and thresholding to separate the vessels from the background. The pulmonary artery connection further refined the preliminary vessel contours and improved the accuracy. We evaluated our method with the dataset from a publicly available FUMPE (Ferdowsi University of Mashhad's PE) dataset. The resulting Dice similarity coefficients against the ground truth created by human experts was about 81% ± 1%. The visualizations created by the automatic algorithm was also very similar to that created by human experts. Future works building upon our study may contribute to the better diagnosis of PE.