Search Contact information
University of Cambridge Home Department of Engineering
University of Cambridge > Engineering Department > Machine Intelligence Lab

Abstract for lindop_thesis

PhD Thesis, University of Cambridge


Joel E. Lindop.

March 2008

Medical imaging is vital to modern clinical practice, enabling clinicians to examine tissues inside the human body non-invasively. Its value depends on accuracy, resolution, and the imaged property (e.g., density). Various new scanning techniques are aimed at producing elasticity images related to mechanical properties (e.g., stiffness) to which conventional forms of ultrasound, X-ray and magnetic resonance imaging are insensitive.

Elastography, palpography or strain imaging has been under development for almost two decades. Elasticity images are produced by estimating and analysing quasistatic deformations that occur between the acquisition of multiple ultrasound images. Likely applications include improved diagnosis of breast cancer (which often presents as a stiff lump), but the technique can be unreliable and difficult to perform. Practical imaging is based on freehand scanning, i.e., the ultrasound probe is moved manually over the surface of the tissue. This requires that elasticity images are calculated fast to provide a live display, and the images need to present meaningful elasticity data despite the poorly controlled properties of the deformations.

This thesis presents technical developments towards clinically practical elasticity imaging. First, deformation estimation is examined to devise algorithms that are both computationally efficient and accurate. Second, the entire image formation process is considered, providing strain data accompanied by indications of accuracy, which are then appropriately scaled and displayed in elasticity images representing the value and reliability of elasticity data.

Displacements are estimated by matching windows of radio-frequency data between pre- and post-deformation ultrasound frames. Robust tracking ensures that displacement estimates can be found by searching over small ranges without introducing large errors, location estimation corrects a well-known amplitude modulation artefact, and a ``weighted phase separation'' framework illuminates the scope for optimising the speed and accuracy of deformation estimators.

Strain estimates derived from each estimated deformation provide a form of elasticity image. A method is devised for predicting the accuracy of each strain estimate, which is first applied for dynamic resolution selection: parameters are automatically modulated to produce images with fixed precision at variable resolution. This indicates the scope for using accuracy indicators, which are applied to greater practical advantage in an interface concept: Nonuniform normalisation of strain data leads to ``pseudo-strain'' images. Values from multiple images are blended adaptively to produce a final display that is reliable, while indicating the level of uncertainty where data are less accurate.

This has made it possible to produce good 2D and 3D elasticity images by freehand scanning. Indeed, a clinical trial has recently been set up to evaluate the utility of this system in various clinical scenarios.

(ftp:) lindop_thesis.pdf (http:) lindop_thesis.pdf

If you have difficulty viewing files that end '.gz', which are gzip compressed, then you may be able to find tools to uncompress them at the gzip web site.

If you have difficulty viewing files that are in PostScript, (ending '.ps' or '.ps.gz'), then you may be able to find tools to view them at the gsview web site.

We have attempted to provide automatically generated PDF copies of documents for which only PostScript versions have previously been available. These are clearly marked in the database - due to the nature of the automatic conversion process, they are likely to be badly aliased when viewed at default resolution on screen by acroread.

© 2005 Cambridge University Engineering Dept
Information provided by milab-maintainer