|Ph.D Student||Burstein Pablo|
|Subject||A Method of 3D Reconstruction of Flow Velocity Fields from|
|Department||Department of Biomedical Engineering||Supervisor||Professor Emeritus Dan Adam|
An integral automatic and self-contained solution to the problem of reconstructing the flow velocity field from standard Color Doppler Ultrasound (CDUS) images is presented. This includes other sub-tasks such as lumen boundary extraction, vessel geometry modeling, low-flow extrapolation, and angle correction. CDUS images are acquired using standard, hand-held transducers, with no restrictions on the choice of transducer position and orientation. A modified balloon algorithm is used for edge extraction, providing a mean radius error of 3.45%, compared to the 7.01% obtained by estimating the boundary from the color maps. The color-maps are, therefore, used for obtaining the initial guess which greatly reduces the search area (and processing time), and provides an automatic way for initializing the balloon algorithm close to the lumen boundary, and obtaining good estimates of the parameters governing the balloon evolution. The 3D-velocity flow field is reconstructed by means of a physical model of flow. The Navier-Stokes and Continuity equations are solved using Penalty Function Approach-Finite Elements Method (PFA-FEM). The penalty parameter regularizes the solution. We studied the effects on the reconstructed flow field of measurement noise, wall-motion filtering and poor geometry estimation. It is noticed that poor geometry estimation affects the flow field reconstruction more than measurement noise and wall motion filtering, but it is easily corrected by better spatial acquisition protocols. Measurement noise effects are corrected by means of the PFA-FEM, while the wall-motion filtering distortions are dealt with by prior parabolic extrapolation. Feasibility of the method is shown, using a simulated flow field and CDUS images acquired from a phantom. Even though the velocity measurements, used as boundary conditions, are highly distorted, the reconstruction error, MSE = 0.05, is two orders of magnitude smaller than the error at the inlet and outlet planes (boundary conditions), MSE = 4.92.