Improved qualification of vascular abnormalities in contrast enhanced magnetic resonance angiographic images

Abstract

Stroke is the third leading cause of death in the western world, and the primary cause of adult disability. There is a great need for methods to identify patients at risk of stroke and decide appropriate treatment. The main criterion for identifying patients at risk of stroke is the percentage of the narrowing in the carotid arteries, which lead blood from the heart to the
brain, which currently is quantified manually. This project is in collaboration with Blackpool Victoria Hospital and Christies Hospital in Manchester. The aim of this project is to develop software methods to improve computer-assisted carotid artery stenosis measurement based on Magnetic Resonance Images. A methodology is therefore presented, based on 3D
geometry extracted from Contract-Enhanced MR Angiograms, to identify and segment the internal carotid arteries for stenosis quantification.
The MRA data volume is initially automatically reduced, by locating the carotid arteries and creating two volumes of interest, each including a single set of carotids (either left or right). The artery of interest (Internal Carotid Artery - ICA) is identified in each sub-volume automatically, by tracking the carotid bifurcation and selecting the artery branch with no further arterial branches. The central axis of the ICA is consequently determined by calculating and connecting together the centres of gravity (centroids) of the 2D contours of the carotid in the axial plane. Segmentation of the ICA is carried out, perpendicular to the central axis, by applying adaptive thresholds along the ICA central axis based on local image characteristics.
Hence, the cross-sectional area of the segmented ICA is then measured at different points along the vessel. The most stenotic area is identified, and a reference region is manually selected. The degree of stenosis is then quantified based on the reference and stenosed area measurements, according to the NASCET criterion. This provides a fully automated
methodology to locate, identify, and measure the internal carotid stenosis. It is the first time that such complete methodology that covers the processing of the MRI data until the stenosis measurement is taken is developed and is fully automatic. The segmentation results are thoroughly evaluated against the manual delineations of two clinical experts (each performed
the delineations twice), and against two popular segmentation techniques. The results were found successful and perform better than manual measurements and other current techniques. They present smaller variability than manual measurements and are able to deal with irregularities in the arterial structure, where other computerised techniques fail. The suggested methodologies seem promising and able to improve considerably both current clinical practice and other existing methodologies.