Automating the Analysis of Aortic Calcifications in DXA Images

Abdominal aortic calcifications (AAC) are build ups of calcium, fat and other material accumulated in the walls of the aorta. Their location and severity have been shown to be an important predictor of death through vascular disease, myocardial infarction and stroke thus their detection in a patient is important for diagnosing problems with the cardiovascular system. It has been shown that AACs can be visible on low dose DXA images of the spine. Large numbers of DXA images of the spine are taken at the Manchester Royal Infirmary for identifying vertebral fractures. The presents an opportunity for early detection of cardio-vascular disease at the same time as vertebral screening. Current scoring systems for measuring AACs are time-consuming for radiologists to perform.

In this project we propose to develop a computer system capable of automatically detecting signs of AACs on DXA images, with minimal human intervention, in order to speed up the analysis of such images.

Objectives:
To write a tool to allow manual annotation of AAC regions in DXA images
To build a large annotated set of DXA images of the spine suitable for training and testing
To develop both semi- and fully-automatic systems for identifying AACs in DXA images
To evaluate the performance of the systems against standard radiologist scoring schemes

The project will build on the Centre's world-class algorithms for accurately locating the vertebrae in images of the spine to identify the regions of the image likely to contain the aorta. Machine learning techniques will be used to develop classifiers to determine regions associated with AACs.

The ability to automatically detect AACs would be of great benefit in early diagnosis of cardio-vascular disease, particularly as it would allow opportunistic application to existing images, potentially identifying the disease on people who might not otherwise be examined.