University of Washington

Abstract

Three-Dimensional Spatial Compounding of Ultrasound Images Acquired by Freehand Scanning:
Volume Reconstruction of the Rotator Cuff

by Daniel F. Leotta

Chairperson of the Supervisory Committee: Professor Roy W. Martin
Department of Bioengineering

Spatial compounding of ultrasound images can enhance visualization by combining views from multiple look directions and making use of the differences in geometry to fill in regions of shadowing and signal drop out. While previous studies have mostly been restricted to two-dimensional (2D) images, this dissertation addresses spatial compounding of three-dimensional (3D) data sets. Since successful compounding is dependent on the accurate registration of overlapping data sets, the initial focus was on the development of a 3D ultrasound imaging system, based on a magnetic tracking device, with performance suitable for medical imaging applications. Rapid freehand acquisition of image data allows 3D scans to be completed in approximately 1-2 minutes, with the advantages of increased patient comfort, reduced risk of movement during data acquisition, and flexibility in imaging protocols. Through careful calibration, the system is able to locate points with precision of 1 mm for typical imaging protocols. Validation studies of 3D reconstructions indicated that the system provided accurate reproduction of both the volume and shape of realistic in vitro targets. The demonstrated accuracy of the system made it feasible for use in 3D spatial compounding of volume data sets. Software was implemented for reconstruction of 2D images in a 3D volume space, and for the combination of multiple overlapping volumes. In addition, a technique was developed which uses information about both the imaging configuration and the target of interest to produce a weighted combination based on the incidence angle of the acquired data. This technique showed promise in maintaining resolution and contrast as views were combined from multiple interrogation directions. Compound volume reconstructions of the shoulder rotator cuff in vivo included regional anatomical landmarks, which are not normally available through 2D imaging. These surrounding features provide a context which can enhance interpretation and comparison of data sets. Semi-automated measurements of the thickness of the rotator cuff tendons, based on surfaces extracted from the volume data, provide accurate and repeatable measurements which are independent of the orientation of the original 2D image planes.