Research Opportunities in the Medical Imaging Group

We are always looking for new students with knowledge and interest in one or more of the following areas: C/C++ programming and algorithm design, signal processing, computer graphics, image processing, ultrasound. The list below gives an overview of some of our current research interests.

• 3D freehand elastography. Some tumours can be felt as a lump beneath the skin, because of the difference between the elastic modulus in the tumour and that of the surrounding tissue. In elastography the goal is to map the elastic modulus at a fine resolution and thus enable the detection of tumours that are too small or deep to feel. The medical imaging group have recently demonstrated the first ever 3D elastogram based on freehand 3D ultrasound. A paper has been submitted to MICCAI04 reporting this result.
• Combining freehand 3D ultrasound with volumetric probes. There are currently two ways to acquire 3D ultrasound data: volumetric probe and freehand. In a volumetric probe either a 2D array of transducer elements, or some sort of mechanical scanning arrangement is used to sweep the ultrasound beam over the volume of interest. In a freehand system the clinician sweeps the 2D probe over the volume and the probe trajectory is measured to permit subsequent 3D visualisation and measurement. The volumetric probe is good for small volumes. The freehand system is good for larger volumes, when specialist probes are required, and when registration of the data to an external coordinate system is required (eg. radiotherapy planning). In the medical imaging group, we intend to combine both approaches by building a freehand system around a volumetric probe. This system will offer data of good homogeneity and local resolution, yet permit the scanning and measurement of structures that cannot fit into a single sweep of the probe.
• Freehand 3D ultrasound without an external position sensor. Freehand 3D ultrasound involves tracking the trajectory of the ultrasound probe as the clinician performs the scan so that the 2D ultrasound slices can be subsequently reconstructed in 3D by computer. Such systems are inconvenient to use in a clinical environment because of the need to have a position sensing system with a fixed datum (eg. magnetic transmitter or camera unit) external to the ultrasound probe. There are currently two image-based algorithms for performing freehand 3D ultrasound without any sensor at all; one developed by Siemens, and the other by ourselves. However, these techniques are of limited accuracy and prone to drift. We are currently developing combinations of other sensors, such as optical fibre tapes that can sense their own curvature and MEMS accelerometers and gyros to combine with the image-based algorithms. The result will be a system that is both accurate and easy to use. For the MEMS work we are collaborating with Dr Ashwin Seshia from the micromechanics research centre.
• Realtime radio frequency (RF) 3D ultrasound. Most freehand 3D ultrasound systems work by combining conventional 2D scan slices. Through a unique collaboration with Dynamic Imaging Ltd, the medical imaging group have built a system in which the RF data from the ultrasound machine is recorded live at 66.7 MHz. This enables 3D information to be computed directly from the raw data, without the loss of phase and dynamic range caused by conventional 2D pre-processing. The realtime RF system is a key first step to enable us to make progress on the two projects listed below.
• Determining material properties using ultrasound. Ultrasound images are formed using information from backscattered sound. Information about deep structures is therefore liable to be corrupted by the material properties of the superficial tissue through which the ultrasound beam has to pass during the imaging process. To overcome this limitation we will use high frequency short-range scans to determine the speed of sound and attenuation characteristics of the superficial tissue. This information can then be used to correct any distortion in the deeper scans. Data from each depth is used to correct the next level down. We therefore call the approach "Progressive establishment of tissue properties at depth".
• Increasing ultrasound resolution by deconvolution. Traditionally, ultrasound imaging has always been limited by an inability to resolve scatterers closer together than the wavelength of the sound used. The rate of attenuation of sound in the body is proportional to its frequency. This means that high frequency ultrasound (with a short wavelength and good resolution), has limited penetration range. We are studying techniques based on combining multiple scans to overcome this limitation. Some of this work is being performed in collaboration with Dr Nick Kingsbury and James Ng from the Signal Processing Laboratory.