Motion estimation in ultrasound imaging.
详细信息
文摘
This dissertation focuses on devising algorithms for motion estimation in ultrasound images. Tracking motion can be categorized into two groups: out-of-plane and in-plane motion estimation. We use out-of-plane motion estimation mainly for generating 3D volumes from a 2D ultrasound probe by sweeping the probe across the volume of interest. We perform in-plane motion estimation for imaging the mechanical properties of tissue by deforming the tissue and tracking the tissue motion. However, both out-of-plane and in-plane motion estimation have several other applications which are not the focus of this work. We validate our algorithms using simulation, phantom, ex-vivo and in-vivo experiments. Medical ultrasound is inexpensive, non-ionizing, real-time and easy to use. Therefore, it is the preferred modality in many diagnostic and surgical procedures in both equipped and unequipped clinics and hospitals. The algorithms that we developed do not require any additional hardware and are real-time. Therefore, while they add to the capabilities of ultrasound, the system still remains inexpensive, non-ionizing, real-time and easy to use. We perform out-of-plane motion estimation by analyzing the correlation between parts of image known as fully developed speckles FDS). Therefore, we first optimize an existing well-known speckle characterization function by minimizing its false positive and false negative in simulated ultrasound images. We then develop a two-step meshing algorithm that finds irregularly shaped FDS patches of ultrasound images. We also explore the possibility of calibrating ultrasound probe using real tissue, as opposed to the current state of the art that uses FDS phantoms. We then propose using beam steering for speckle classification and sensorless 3D ultrasound and show that data obtained from different steering angles contain independent information. Therefore, we propose a method for combining data from images with different steering angles and show significant improvement in both speckle classification and out-of-plane motion estimation. For in-plane motion estimation, we optimize cost functions that incorporate amplitude similarity and spatial motion smoothness. We use dynamic programming and analytic minimization techniques to optimize the cost function in real-time. We show that spatial motion continuity reduces the sensitivity of motion estimation to signal decorrelation. We also use techniques from robust statistics to limit the influence of outliers. We introduce using Kalman filter for calculating a strain map from the motion field. We finally impose constraints on temporal motion of tissue using mechanical properties of tissue. We use these constraints to obtain tissue displacement from more than two images in an expectation maximization EM) framework. We show that the displacement estimation obtained from three images contains less noise.