Yousef Ebrahimdoost, Salah D. Qanadli, Tim J. Ellis, Zahra F. Shojaee, and Jamshid Dehmeshki
Pulmonary Artery Segmentation, Level set, Pulmonary Embolism, False Positive, CTA images, Lung Segmentation
Pulmonary embolism (PE) is blood clot in lung region and its diagnosis remains a major clinical problem. Several CAD systems have been developed aiming at detection and distinction between the real PE and look-alikes. However, the number of false positive detection is still too large as an output of systems. We propose a new CAD system for detection of pulmonary embolism in computed tomography angiography (CTA) images aiming at reduction of false positive detection. Our approach is performed in two stages: firstly we extract the pulmonary artery tree in the region of the lung and heart to reduce the region of search. In the second stage, we detect pulmonary embolism inside the segmented pulmonary artery, by an analysis of three dimensional features inside the segmented artery. Filtering is used to exclude more false positive detections associated with parenchyma disease, the partial volume effect on artery boundary, lymphoid tissue, noise and motion artifacts. The method was trained on a dataset of 20 patient scans and evaluated with a further 31 scans, containing a total of 121 emboli. The resulting performance gave 100% detection sensitivity with an average 4.7 false positive detections per scan.
Important Links:
Go Back
