Image-based registration for two-dimensional and three-dimensional ultrasound imaging.

Abstract

Image based registration techniques were developed, evaluated, and applied to 2D and 3D ultrasound (US) imaging in the context of deformation and aberration detection and correction. The specific applications demonstrated here include 3D compounding, generation of extended fields of view, and sound speed estimation. Despite the enormous clinical importance that diagnostic US has gained over more than four decades, and despite the fact that advances in software development and computer technology have made image registration a widely studied and moderately applied technique in other medical imaging modalities, US and image registration have rarely been combined in research or clinical application. We will show that not only can some image registration methods be transferred from other imaging modalities and adjusted to operate on US images, but also that registration can overcome or greatly ameliorate some of the existing limitations of US imaging. A nonlinear registration algorithm developed specifically for ultrasound showed registration accuracy of 0.2 mm in volumes with synthetic deformations, 0.3 mm in phantom experiments, and 0.6 mm in vivo. Extended high-resolution ultrasound volumes with lateral extents of over 10 cm were created by fusing together 3 or 4 individual volumes, using image registration in the areas of overlap. 3D compounding in the out-of-plane direction was achieved by registration of US volumes obtained from different look directions. Examples of compounding in phantoms and in vivo show increased contrast/noise and better visualization of specular reflectors. Image-based estimates of the average sound speed in the field of view were obtained using registration of steered 2D US images. The accuracy of the estimates was improved by including simulations of the sound field generated by the array. Evaluated over a range of sound speeds from 1490 to 1560 m/s in a custom-made phantom, the simulation results reduced the RMS deviation between the estimates and reference measurements from 0.3% to 0.1%. Using accurate estimates of the average sound speed in the beamforming process can effectively eliminate the contribution of gross sound speed errors to the overall phase aberrations.