Image upgrade

New software and sensor arrays to achieve ‘super resolution’ images to aid the early detection of breast cancer is being developed by an international team led by Imperial College London.

Dr Francesco Simonetti of Imperial College is collaborating with the Detroit-based Karmanos Cancer Institute and Los Alamos National Laboratory in New Mexico, to investigate arrays that will improve the detection capabilities of ultrasound.

The measurements recorded by the sensors will be processed by software, developed by Simonetti with Boston’s Northeastern University, to produce high-resolution images that would rival those of CT (X-ray-based) scans.

‘At the moment in the field of cancer detection, the gold standard is X-ray, but X-ray has many limitations,’ said Simonetti. ‘One of these is that it does not work for the early detection of breast cancer in young women because X-ray sees density contrast between tumours and healthy tissue.

‘Tumours are usually denser than healthy tissue. However, in young women the healthy tissue also tends to be quite dense, so X-ray cannot see the tumours. That is why women are only usually advised to take X-ray mammography after they are 30 or 40.

‘Ultrasound, however, is ultra-sensitive to stiffness changes between healthy and unhealthy tissue. It has been shown that ultrasound is potentially more sensitive than mammography.’

The researchers are using variable array technology, with thousands of sensors, to develop a new probe for sensing.

‘Now we can play with different architectures of these arrays, which means we can arrange these sensors in different shapes. Traditionally we use linear rays, so the radiographer uses a rectangular-shaped probe. That probe contains 128 or 256 small sensors that are all lined up. We are using these types of sensors, but also a new generation of sensors that are deployed in a circle, like a toroidal array.’

A significant part of the project involves moving beyond a mathematical model typically used in imaging techniques, known as the Born approximation. It is one of the standard models used to describe the interaction of waves with matter.

‘The Born approximation is crucial to imaging problems because we exploit the interaction of waves with matter to reconstruct the properties of matter.

‘All the state-of-the-art imaging technology at the moment is based on the Born approximation, but this is not sufficient to describe completely the interaction of waves with matter, and we actually lose information using this,’ said Simonetti at Imperial’s Mechanical Engineering Department.

The loss Simonetti is referring to is the trade-off between resolution and imaging depth. To produce an image, waves are propagated through a matter and these are characterised by a wavelength that is the distance over which the wave does one oscillation. The smaller the wavelength, the higher the resolution of an image.

‘In principle, we would like to have a very small wavelength so that we could have very detailed images,’ he said. ‘The problem is that as the wavelength decreases, the attenuation of the wave propagating inside the material we want to image increases. So very short wavelengths can only penetrate a small distance inside, for example, the human body.

‘A typical problem is that if we were to image deep inside the liver, we can only put the sensors on the skin, so the wave has to travel several centimetres before reaching the area of interest.

‘If we propagate one megahertz, that corresponds to a wavelength of around a few millimetres. But if you wanted a resolution of 10 times smaller, we would need to propagate 10MHz and this does not go deep enough because the wavelength is completely killed before reaching the area of interest.’

The matter through which a wave is propagated determines how quickly the wave attenuates, but these are not considered in the Born approximation.

‘We have demonstrated that if we consider multiple scattering, which means we consider the fact that as a wave travels through an object it is distorted by the presence of the object, this actually adds more information. In order to implement multiple scattering you need a much higher computational power than what was available five or 10 years ago,’ said Simonetti.

Thanks to progress in computer processing and the researchers’ development of inversion algorithms, Simonetti said that the team has demonstrated a resolution in the far field of a sixth of the wavelength, or lambda/6.