Model - based estimation of arterial diameter from x - ray

X-ray coronary angiography is a widely used imaging technique for visualizing the presence of coronary artery disease (CAD). The measurement of arterial diameter from these images reflects disease burden, and provides an assessment of treatment benefit from existing and experimental therapies. Unfortunately, existing derivative-based and model-based measurement methods provide poor estimates for small vessel diameters precisely the range that is associated with the presence of clinically significant stenoses. We have found that because of blurring and noise, the apparent widths of small vessels in angiographic images contain little information about the real diameters, leading to poor estimation performance by existing techniques. Information about vessel diameter does remain available in the intensity height, but this has not, until now, been exploited to improve diameter estimation. The use of intensity information is complicated by a variable, non-linear relationship between intensity and vessel diameter. Separate x-ray calibration procedures might be used to experimentally determine crude approximations to this nonlinear relationship. However to date, such procedures are clinically impractical to perform. To overcome this limitation, we have used a more accurate description of x-ray imaging function to pose this problem within a model-based estimation framework. With this approach, we can estimate the unknown imaging model parameters directly from an angiographic image, without the need for additional x-ray experiments. The intensity relationship thus determined forms the basis for a novel model-based estimator which uses the combined intensity and width information within an angiogram to measure arterial diameter. We have used computer simulation to analyze the performance and sensitivity of this method to imaging noise, blurring, angiographic backgrounds, imaging system operating points and estimation initial conditions. These simulations have shown that this approach is robust to variations in each of the conditions studied. Finally we have compared the diameter estimates from our technique against those from existing methods, using simulated arterial vessel images and real x-ray images of vessel phantoms. Our results suggest that diameter estimation using our new approach is significantly improved over estimation using the current forms of diameter measurement. Thesis Supervisor: Robert S. Lees Title: Professor of Health Sciences and Technology Thesis Supervisor: W. Clem Karl Title: Research Affiliate, Laboratory for Information and Decision Systems