Detection of Calcification in Abdominal Aortic Aneurysms

Abstract

Abdominal aortic aneurysm (AAA) is an expansion in the abdominal aorta having a diameter of 3 cm or greater. This cardiovascular disease may cause death if the aneurysm were to rupture. In several cases, varying amounts of calcification have been observed in AAA. Calcification leads to an increased rupture risk. These calcifications are often not included though in rupture risk assessment models due to the fact that calcifications are small, which makes their inclusion a difficult task. Therefore, it is important to develop a tool capable of detecting these calcified deposits. This is expected to lead to a better rupture risk prediction, reducing AAA rupture fatalities. At first, a thorough literature survey was conducted to identify existing solutions to segment different components (lumen, thrombus, calcification, and aortic wall) of the aneurysm. In order to detect the calcification from abdominal aortic aneurysm, a two staged method was developed in this thesis. In the first stage, the aneurysm is segmented using a newly proposed statistical topology prior based method. In the second stage, the calcifications are detected using the Bayesian classifier. The thrombus segmentation accuracy (in the first stage) achieved a mean hausdorff distance of 3.5703±3.1941 mm, and a mean dice similarity coefficient of 0.9303±0.0499. These results were found to be competitive with the highest reported values in the literature. The testing was performed on a total of 48 CTA slices equally taken from 6 different patients. Based on the literature reviewed, it was observed that the calcification detection problem from AAA hasn't been addressed before, so we suggested using the Bayesian classifier to detect calcifications from the previously segmented thrombotic volume. In the calcification detection stage, a sensitivity of 0.9580, specificity of 0.9428, and accuracy of 0.9433 were acquired. This testing was performed using 25 slices from 7 different patients. Ground truth data of the thrombotic volume and calcification deposits were manually labeled by a radiologist from the radiology department of Louisville University.

Date of Award	May 2017
Original language	American English
Supervisor	Tim McGloughlin (Supervisor)