Development, characterization and optimization of a deformable model based approach for carotid artery lumen and plaque surface segmentation.

摘要

Stroke is the most common serious neurological problem in the world and causes staggering personal, community and health care costs. Its prevalence clearly indicates a need for improved diagnostic methods, and better ways to treat and monitor patients, before and after a stroke.; X-ray and Doppler ultrasound imaging are both common, but inadequate methods for determining absolute stroke risk. Both methods overlook the importance of carotid plaque morphology, internal structure and composition in determining stroke risk. Three dimensional ultrasound imaging (3DUS) has been proposed to address some limitations of conventional stroke risk diagnostics, but has not replaced the standard diagnostics because of an inability to determine plaque and lumen boundaries with a high degree of precision. Improved information may be available from 3DUS if reliable post-processing methods were available for determining plaque and residual lumen boundaries. The hypothesis of this thesis is that a semi-automated method for identifying' carotid artery lumen and plaque boundaries from 3DUS images can be developed that has less variability, and comparable accuracy to an expert observer's measurements of residual lumen and plaque geometry.; To test this hypothesis, a unique semi-automatic method for segmenting 3D carotid vascular ultrasound images was developed. To characterize the algorithm, a novel method for analysis of geometric variability and accuracy was designed. From a study comparing manual segmentations of a 3DUS image of a carotid bifurcation to those generated using the semi-automatic algorithm, it was determined that the algorithm was less variable than the intra-observer manual segmentation variability, and had comparable accuracy.; Since the performance of the algorithm depends on a number of image-specific operating parameters, the feasibility of determining optimal model parameters, for a given class of image, was investigated. An optimization analysis was performed, which compared the semi-automatic segmentation results to a reference standard derived from repeated manual segmentations. The algorithm accuracy was shown to be well-behaved with respect to adjustments in model parameters, indicating that automatic calibration of the algorithm to distinct classes of images is feasible.; Finally, a 3D curvature-based morphological metric was developed as a possible means of characterizing plaque ulceration. The metric was applied to a number of 3D segmented carotid lumen and plaque surfaces, and compared to an expert observer's manual assessment of plaque morphology.