Co-supervisor: Dr Nick Spadaccini, CSSE, UWA
External consultant: Dr Jay Ives, Radiologist, SKG Radiology

The extraction of information captured in sophisticated medical imaging techniques is of immense clinical interest. Digital signal processing, visualisation, and feature recognition techniques have the potential to disclose features and automate tasks of significance in medical practice. Despite more than a decade of work, only a few useful tools have become available to the clinician.

MRI slice, extracted white matter, grey matter, and resulting labelled map, from Barra, V. and Boire, J, "Tissue Segmentation on MR Images of the Brain by Possibilistic Clustering on a 3D Wavelet Representation", J. Magnetic Resonance Imaging, 11, 267-278, 2000.

These projects will concentrate on the development of novel methods of feature extraction and visualisation in neuroimaging in association with a senior neuroimaging specialist in the major imaging practice in Perth. Images will be available from state-of-the-art MRI, CT and PET scanners.

A number of projects are available with emphasis in three areas:

1. Image processing

   Before machine learning or pattern analysis techniques can be used to classify an image, the image must be processed to identify features that can be classified. Techniques such as multi-spectral analysis, parametric imaging, and wavelets have been used for this purpose.

2. Feature classification

   Once the images have been transformed into to a set of useable features, machine learning or pattern recognition can be used to attempt to automatically find objects or classify areas in the scans. Techniques such as Bayes' classifiers, k-nearest neighbours, neural networks and fuzzy clustering have been used for this purpose.

3. Software tool development for imaging specialists

   See BE(SE) Projects (to appear)

External consultant:
Dr Jenny Rodger, Neurobiology, UWA

Keywords: biotechnology, modelling complex systems, search, optimisation, visualisation, 3D graphics

Visual maps in the brain of the Fat-Tailed Dunnart. Regions of the brain corresponding to distinct regions of the eye are mapped using axonal tracer dyes. Courtesy Lisa Tee and Sarah Dunlop, Neurobiology Lab.

In order to function successfully the central nervous system (CNS) of humans and other animals must be able to form topographic connections. That is, patterns of neurons in one part of the CNS must connect up with corresponding patterns in another location. An example is the visual system. Behind the retina in the eye is a pattern of nerve cells (called ganglion cells). The long stems (or axons) of these cells form the optic nerve which carries signals from the retina to the visual region of the brain (the optic tectum). The ability of animals to correctly interpret the visual world relies on the retinal cells connecting up to corresponding tectal cells in the right topographic order, or pattern. The question is, when the nerve cells are growing (or regrowing after damage), how do they know where to go?

The mechanisms for topographic map formation are important to understand particularly when it comes to nerve regeneration (for example, after spinal injury). Many lower vertabrates, such as fish and reptiles, are able to regenerate CNS connections after injury. In humans regeneration is far more difficult, and even where some regrowth can be medically stimulated, it does not tend to form correct topographic projections and as a result function is not restored.

Neurobiologists, including those at UWA's Neurobiology Laboratory, have uncovered a number of candidate mechanisms for topographic projection. These include gradients of biochemical markers, activity-dependent mechanisms (cells that "fire together wire together") and timing considerations. As yet, however, no-one has found a way of combining these mechanisms so as to fully explain the experimental evidence.

Pathfinding along the optic nerve. These pictures show the bundling of optic nerves (running into the page) demonstrating that nerves from different eye regions "stick together" as they navigate to the visual cortex. Courtesy Lisa Tee and Sarah Dunlop, Neurobiology Lab.

Because of the complexity of the interacting mechanisms of nerve regrowth a number of researchers have turned to computer models as a way of trying to understand the processes involved. Computer models allow a "what-if" analysis for testing hypotheses that is faster, less expensive and less invasive than laboratory experiments.

The aim of these projects is to develop a computer model of retinotopic projection. Dr Rodger will act as a client for the purposes of requirements elicitation. If more than one student takes the projects the requirements analysis and high-level design may be done as a group. It is expect that a model-view-controller (MVC) framework will be used. Students will then focus on a distinct project which will require further research (for example, into global search algorithms, graphics, etc depending on interests) and implementation.

Further reading:

Goodhill, G.J. and Richards, L.J., Retinotectal Maps: Molecules, Models, and Misplaced Data. Trends in Neuroscience, 22:529-534, December 1999.

Students taking these projects will become members of the Adaptive Systems Group.