| Ultrasound (US) imaging is widely used in a number of different clinical applications because it is non-ionizing, portable, low-cost and real-time. However, several characteristics of US images limit their clinical usefulness. First, the vast majority of currently available commercial systems, as well as most of the published research work use two-dimensional (2D) US images (formed by one-dimensional (1D) arrays of piezoelectric elements) despite the fact that human anatomy is three-dimensional (3D). Second, due to the hand-held nature of the US transducer, US 2D cross-sections are of arbitrary orientation with respect to each other and with respect to the anatomy. Third, a single US image provides a limited field-of-view that is often smaller than the anatomy of interest. Finally, US images have low quality and signal-to-noise ratio (SNR) because of several artifacts including speckle noise, refraction of the US beam, and shadowing. As a result, interpretation of US images becomes a significantly challenging task for the clinician, who has to move the US transducer appropriately and mentally integrate all the individual 2D US images together to form an understanding of the 3D anatomy. Registration of US images can address the previous limitations of US imaging and therefore increase the clinical usefulness of the modality.; In this thesis, we provide a systematic approach to image registration methods pertinent to US imaging. In particular: (i) we developed a system and architecture-specific software implementations for real-time 3D US and US panoramic imaging, from a series of 2D US images, using a digital signal processor called MAP-CA which is the backbone of a programmable US system we are currently developing in our laboratory, (ii) we invented and constructed a new phantom for fast and easy-to-use integration of a position sensor with a 2D US probe for tracking the position and orientation of an US probe and forming geometrically accurate 3D US data sets, and (iii) we developed an automatic 3D US deformable registration method for 3D US spatial compounding which results in US images with higher SNR and reduced artifacts. |
