Pulmonary embolism (PE) is an obstruction within the pulmonary arterial tree and in the majority of cases arises from a thrombosis that has travelled to the lungs via the venous system. Pulmonary embolism (PE) is a fatal condition which affects all age groups and is the third most common cause of death in the US. Computed tomographic angiography (CTA) imaging has recently emerged as an accurate method in the diagnosis of pulmonary embolism. Each CTA scan contains hundreds of CT images, so the accuracy and efficiency of interpreting such a large image data set is complicated due to various PE look-alikes and human factors such as attention span and eye fatigue. Moreover, manual reading and interpreting a large number of slices is time consuming and it is difficult to find all the pulmonary embolisms (PE) in a data set. Consequently, it is highly desirable to have a computer aided detection (CAD) system to assist radiologists in detecting and characterizing emboli in an accurate, efficient and reproducible manner. A computer aided detection (CAD) system for detection of pulmonary embolism is proposed in CTA images. Our approach is performed in three stages: firstly the pulmonary artery tree is extracted in the region of the lung and heart in order to reduce the search area (PE occurs inside the pulmonary artery) and aims to reduce the false detection rate. The pulmonary artery is separated from the surrounding organs by analyzing the second derivative of the Hessian matrix and then a hybrid method based on region growing and a new customized level set is used to extract the pulmonary artery (PA). In the level set implementation algorithm, a new stopping criterion is applied, a consideration often neglected in many level set implementations. In the second stage, pulmonary embolism candidates are detected inside the segmented pulmonary artery, by an analysis of three dimensional features inside the segmented artery. PE detection in the pulmonary artery is implemented using five detectors. Each detector responds to different properties of PE. In the third stage, filtering is used to exclude false positive detections associated with the partial volume effect on the artery boundary, flow void, lymphoid tissue, noise and motion artifacts. Soft tissue between the bronchial wall and the pulmonary artery is a common cause of false positive detection in CAD systems. A new feature, based on location is used to reduce false positives caused by soft tissue. The method was tested on 55 data scans (20 training data scans and 35 additional data scans for evaluation containing a total of 195 emboli). The system provided a segmentation of the PA up to the 6th division, which includes the sub-segmental level. Resulting performance gave 94% detection sensitivity with an average 4.1 false positive detections per scan. We demonstrated that the proposed CAD system can improve the performance of a radiologist, detecting 19 (11 %) extra PE which were not annotated by the radiologist.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:592758 |
| Date | January 2012 |
| Creators | Ebrahimdoost, Yousef |
| Publisher | Kingston University |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://eprints.kingston.ac.uk/24027/ |
Page generated in 0.0022 seconds