RESEARCH INTERESTS
(Last update:
December 29, 2007)
Local blood flow patterns (hemodynamics) are thought to play an important key role in the development, diagnosis and treatment of vascular disease. For example, atherosclerotic plaques are known to develop primarily at artery bifurcations and bends, where complicated wall shear stresses (the force exerted by flowing blood on the arterial wall) are thought to occur. The high shear stresses and enhanced residence time of blood present in diseased vessels may promote the development of blood clots that ultimately cause many heart attacks and strokes. Finally, the inadvertent introduction of such hemodynamic features into the design of interventional devices and prostheses may be one cause of their ultimate failure. Despite decades of active research, many questions remain unanswered about the role of hemodynamics in vascular disease. This is largely due to the fact that these key hemodynamic factors — shear stresses and residence times — are notoriously difficult to measure directly.
With recent advances in medical imaging, it is possible to image vascular anatomy and disease non-invasively, with sub-millimeter resolution. Although in principle capable of measuring blood velocity as well, current imaging technology requires scan times that make routine extraction of hemodynamic factors difficult if not impossible. Computer modeling, however, has advanced to a stage where it is now possible to faithfully model pulsatile hemodynamics in realistic arterial models provided, for example, by non-invasive medical imaging. The focus of my research program is therefore to develop and apply what I have termed "computational imaging" — the integration of computer modeling and medical imaging technologies — in novel studies aimed at improving the understanding, diagnosis, and treatment of vascular disease.
After more than three decades of research, it is still unclear which hemodynamic factors favor the development of atherosclerosis. This is largely due to the fact the prior studies have tried to correlate hemodynamics from idealized flow models with measurement of disease from post-mortem specimens. Using high-resolution magnetic resonance imaging (MRI) to provide us with vascular anatomy and flow rates, we reconstruct the vascular hemodynamics using computer-assisted image processing and computational fluid dynamic (CFD) modeling techniques. We have recently applied these techniques to demonstrate the subject-specific nature of flow patterns in the normal human carotid bifurcation [Milner et al., 1998], and are now applying this approach in a prospective study of patients with early carotid artery disease with the aim of identifying the hemodynamic factors the cause plaques to develop and progress/regress. This "computational imaging" approach is also ideally suited to the study of aneurysm hemodynamics in vivo, since it is extremely difficult to resolve the inherently slow flow in the aneurysm sac by any other means.
Although it is well known that stroke risk is correlated with severity of stenosis at the carotid bifurcation, it remains that the majority of patients with carotid artery disease never have a stroke. Many strokes are thought to occur when the diseased (stenosed) carotid artery sheds small blood clots (thromboemboli) that occlude the smaller vessels downstream in the brain. Clot formation (thrombogenesis) is known to be promoted by the hemodynamic factors found near stenoses: elevated shear produced by the stenosis itself activates platelets that are then transported post-stenotic regions of stagnation or recirculation. We have recently shown that the post-stenotic hemodynamic environment is sensitive to stenosis geometry rather than stenosis severity alone, and are currently developing simple techniques for quantifying the thromboembolic potential at an individual diseased artery using the volumetric residence time approach [Kunov et al., 1996]. The ultimate goal is to provide more specific and clinically useful indicators for assessing risk of stroke.
As we develop improved imaging protocols and techniques, it often becomes necessary to identify potential sources of error or artifact and strategies to reduce them. For example, by coupling a model of MR physics to particles (representing spins) tracked through a computed velocity field, we can simulate the effects of flow on the evolution of the MR signal. We have used these techniques to understand the effect of displacement artifacts on the accuracy of MRI velocity measurements [Steinman et al., 1997], and to identify and ultimately resolve the presence of plaque-mimicking flow artifacts in black blood MRI [Steinman and Rutt, 1998]. We have also developed simple techniques for simulating the appearance of color Doppler imaging for arbitrarily oriented scan planes and Doppler angles. Such tools are being used to better understand the effects of complex flow on the appearance of color Doppler images and Doppler spectra, and to help identify potential sources of error in the reconstuction of the vector velocity field from 3-D Doppler ultrasound measurements.
To visualize the vast amounts of data produced by CFD simulations, we were inspired by an in vitro flow visualization technique: particle imaging velocimetry (PIV). With PIV, the trajectories of luminescent particles are recorded as they move through an illuminated transparent flow model; particle velocity is then determined from the particle displacement and the temporal resolution of the recorded images. Such images, particularly when played in a cine loop, often provide a readily comprehensible view of complicated flow patterns. In a reversal of the PIV experiment, we use the known velocity field to compute particle trajectories. In our "simulated PIV" (SPIV) approach, a user-specified distribution of particles is seeded at the inlet of the CFD model throughout the cardiac cycle. Each particle is then tracked by integrating the time-varying velocity field, and the resulting computed trajectories are displayed as progressing streaks in a series of successive frames. The user specifies a frame rate to control the spacing between streaks, and the shutter speed to control the streak length. Unlike the in vitro PIV (which typically visualizes the trajectories of monochromatic particles through a 2D plane), SPIV allows us to: (i) visualize particles throughout an arbitrarily-oriented 3D volume; and (ii) color-code particles according their velocity or (in the case of a bifurcation) the branch they ultimately enter. Many of the animations in the Image and Movie Gallery were constructed using variations of this apporach.
CFD modeling of arterial hemodynamics has historically been limited to simplistic or idealized representations of vascular geometries. The current trend in hemodynamic modeling is to eschew such idealized models in favor of patient-specific models derived from medical imaging of the vascular anatomy. We have developed a number of computer-assisted image processing tools to simplify the process of extracting the lumen surface and wall thickness for subsequent finite element meshing. Using a class of geometrically deformable models (GDM), we are now able to extract the wall thickness from in vivo MR images of human volunteers with a precision of less than 200 microns. By extending the GDM from a 2D "string" to a 3D "balloon", we are also able to directly extract the three-dimensional lumen boundary from a serial set of MR images, in effect combining the conventional segmenrtation and serial reconstruction phases. We are also currently investigating techniques that will allow us to generate the finite element volume mesh directly from a series of MR images.
To date virtually all computational hemodynamic studies have been carried out on normal arteries, where flow patterns are typically laminar. As we begin to consider the hemodynamics of diseased arteries, which typically have constrictions that produce high-speed jets, turbulence becomes an important factor. Standard turbulence models may not be appropriate under physiological flow conditions, so we are therefore testing their applicability by comparison with LDA measurements. We are also investigating novel turbulence models that take into account the time-scales associated with physiological pulsatile flow.