Computer-aided diagnosis of pulmonary embolism in opacified CT images

• Main Author: Sebbe, Raphaël
• Other Authors: Gosselin, Bernard
• Format: Others
• Language: en
• Published: Faculte Polytechnique de Mons 2007
• Subjects: imagerie médicale; traitement d'images (techniques numériques)
• Online Access: http://theses.fpms.ac.be/ETD-db/collection/available/FPMSetd-02222007-171337/

Pulmonary embolism (PE) is an extremely common and highly lethal condition that is a leading cause of death in all age groups. PE is the third most common cause of death in hospitalized patients, with an estimated 0.24% annual death rate of the population. Its symptoms are often vague and its diagnosis is a major medical challenge, with 70% of missed diagnosis in people dying from PE in hospitals. If left untreated, approximately one third of patients who survive an initial PE subsequently die from a future embolic episode. Most patients succumb to PE within the first few hours of the event. However, when properly identified, an effective treatment consisting of anticoagulants or thrombolytics is administered that dramatically reduces the mortality rate of the disease. Typically, the diagnosis of PE is manually performed by radiologists on CT images. It is a time-consuming and error-prone process, in particular because of the huge amount of data and more specifically in the case of sub-segmental and peripheral clots, which are less visible. Indeed, a typical CT dataset that is used for PE diagnosis can have more that 600 slices, the size of the smallest visible volume being in the order of magnitude of one millimeter. The duration of that review process by radiologists, excluding acquisition time, is in the order of 5 to 10 minutes, depending on evidence of the clots. The duration of an exam may become a concern considering that there is currently a lack of radiologists, that is partly due to the availability of new modalities, the increasing complexity of existing ones, and the growing number of new applications of medical imaging (functional imaging, heart CT, etc.). In that context, a computer aid can be provided whose main goal would be to decrease the time required to perform an exam, by acting either as a safeguard for radiologists or even better, if sufficiently robust and conservative, as a preliminary detection step in a computer-aided diagnosis (CAD) system. The work presented in this thesis consists of a combination of a method for segmenting the pulmonary arteries (PA), two emboli detection methods as well as a scheme for the evaluation of the performance. The segmentation of the PA serves one of the clot detection methods, and is carried out through a region growing algorithm that makes use of a priori knowledge of vessels topology. Two different approaches for clot detection are introduced. The evaluation of the method is also discussed, and a scheme for measuring its performance in terms of sensitivity and specificity is proposed, including a practical approach to making reference detection data, or ground truths, by radiologists.