Graphic illustration

Philips Medical Systems is supporting a joint project between two London universities to further develop faster magnetic resonance imaging (MRI) techniques.

The benefits of the research carried out by King's College London (KCL) and University College London (UCL) range from making the system easier to use to making the examination more comfortable for patients.

'The basic idea is to go very fast. At the moment cardiac MR is recognised as a nice tool but the main problem is that the scan times are quite long — and we have two major challenges: cardiac and respiratory motion,' said KCL's Prof Tobias Schaeffter.

In 2006, the scientists tried to increase the speed of image reconstruction by running the algorithms on a computing cluster with 60 computer nodes, each of which is a 64-bit machine. The cost of using that platform, around £15,000, proved to be too much however, and the scientists have now developed reconstruction algorithms that run on graphics cards — a much cheaper option at £500 a card.

'The reconstructions are really advanced and take up a lot of computer processing power. At the moment, the processing power of computers is in saturation at something like 2.5GHz.

'If you look at the graphics card, each year a new one comes out, these have something like 128 processors. The reason for this is the games industry. It is amazing how much the power of the cards increases annually by putting on more processors and more memory,' said Schaeffter.

UCL's Dr David Atkinson added: 'The reconstruction of data into an image has historically been time-consuming and for some schemes, it takes longer than the acquisition itself. By using graphics cards, we have been able to get the image reconstruction time down to less than the acquisition time.'

Before an image is acquired, MRI patients have to hold their breath for around 15 seconds while being scanned. This often uncomfortable process can be repeated several times so that clear image slices of the heart can be obtained.

The researchers hope to simplify this process by exploring ways to overcome the motion problem. The motion caused by a heart contracting is considered more predictable and easier to compensate for compared with breathing motion, which is unpredictable.

In one approach, known as motion-compensated reconstruction, a patient would be allowed to breathe freely during a scan, with motion being measured during the scan and then fed into the reconstruction stage. Schaeffter suggested measuring motion using techniques in the temporal spatial or the frequency spatial domains.

Frequency spatial techniques would involve assigning a frequency to pixels in an image, depending on how much they moved, and only measuring the pixels moving at a certain frequency. The scientists have already started looking at temporal correlations, distinguishing when and where there is motion in an image from parts of an image that are not moving, to try and reduce the number of data points to be acquired.

'At the moment we are doing stupid acquisition in MR — we are measuring all the data points we like to measure. But in principle we only have to measure certain data points once,' said Schaeffter.

'For instance, the liver only has to be measured once because it is not moving, whereas you have to measure the heart with a higher temporal resolution because it is moving and you have to decide which parts have to be sampled at different sampling rates.'

In preparation for an MR scan, the different sampling rates or the orientation of a scan are some of the details that doctors have to plan before a scan can take place, which can be time-consuming. The researchers are trying to do 3D whole-heart acquisition.

'That means I just put a volume [3D datasets] on the whole thorax and acquire everything. Then, after the scan the doctor can visualise every scan by reformatting the data. This is something that is done in CT at the moment,' said Schaeffter.

In terms of hardware, the scientists still plan to use a 32-channel coil (or antenna) to acquire multiple chunks of data simultaneously.

'We are trying parallel imaging techniques, where we use a number of different antennas,' said Schaeffter. 'Usually the normal systems have something like six element channel coils, but with our MR scanner we can have 32 channels which means we can parallelise the acquisition and make it faster,' he said. Philips Medical Systems is helping with the development of the 32-channel coil.