Better images with lower radiation doses

By harnessing the processing capacity of ordinary graphics cards, doctors can now access clear film sequences of a beating heart in a matter of minutes.

Today most doctors are forced to use two-dimensional images to detect and diagnose heart disease. But the technology to create 3D images exists, and a concrete product reached the market in 2009 from the LiU spin-off company ContextVision.

A group of researchers headed by Professor Hans Knutsson at the Department of Biomedical Engineering at LiU, has continued this work in collaboration with CMIV, the Centre for Medical Image Science and Visualization, and made it possible to create clear and sharp images of beating hearts in four dimensions – in a matter of minutes.

“We want to be able to create as good an image as possible with as low a radiation dose as possible. This is especially important when examining small children,” says Hans Knutsson.

The higher the radiation dose used, the better the images are. Lowering the radiation dose produces more noise – i.e. signals produced by other sources than the beating heart. The trick is to filter out as much noise as possible.

“We take pictures in four dimensions and also have filters in all four dimensions,” says Mats Andersson, first research engineer at Medical Informatics.

The heart is divided into parts, where each part is filtered individually: a cardiac wall can be seen as a moving plane. As long as you know which direction the movement takes place in, you can filter out the signal in the other directions. A blood vessel has two dimensions, so you can filter out any unnecessary information in the other two dimensions.

Often a slightly higher radiation dose is used when the heart is almost still and a lower dose when the heart is moving. This is another way to obtain the best possible images while keeping the radiation dose down. One major advantage of the method developed by the group of researchers is that it can handle the differences in noise level that result from the varying radiation dose.

Huge amounts of calculations are needed to filter the noise and obtain clear pictures. Computational power is used to analyse and describe the signal, and to determine the orientation of the cardiac wall in space and the direction in which it is moving. Computational power is also used to filter out noise in the other directions. Each tiny point in the volume – a “voxel” – must be analysed. The filters, which are used for every pixel, contain around 10,000 values. The volumes are 512 points high, wide and deep and the camera takes 20 images per heartbeat. This means that almost a peta (a 1 followed by 15 zeros) of calculations are needed for each heartbeat.

“It used to take an entire night to calculate, and that meant that the technology wasn’t clinically and practically useful,” says Mats Andersson (pictured at left above, with Anders Eklund and Hans Knutsson on the right)

Anders Eklund recently defended his doctoral thesis in Medical Informatics, and for his dissertation work he used commercial graphics cards as a more efficient means to compute images taken in a magnetic resonance imager. He used the same method here.

“You can program graphics cards to calculate several voxels in parallel and the calculation time has now been reduced from eight hours to eight minutes,” he says.

This makes the filtering technology useful in practice.

“But it is also important that we don’t filter too much out. In order for the images to be clinically useful, the doctors must be able to use their experience from the normal unfiltered images,” says Hans Knutsson.

The video below clearly shows how the noise in the left image has been almost completely eliminated in the right image.

 

 

The research project, Adaptive Filtering of 4D Heart for Image Denoising and Patient Safety, is financed by the Swedish Research Council. Part of the project is also run as part of the regional medical technology initiative NovaMedTech, and it is hoped that 4D imaging will lead to a commercial product in collaboration with ContextVision.

“The 3D technology exists and is used; the next step is to introduce 4D. In a workstation you can rotate and flip the images if you want to take a closer look at something,” says Mats Andersson.

This innovative use of ordinary commercial graphics cards, developed for computer games, has attracted a lot of international attention. Eklund will soon attend a major graphics card conference in San José in California, where he will present the research group’s work.

 

Related links

• Adaptive Filtering of 4D Heart for Image Denoising and Patient Safety
• Anders Eklund’s thesis; Magnetic imagers capture our thoughts
• Department of Medical Informatics
• NovaMedTech
• Read more about NovaMedTech at LiU News
  

Monica Westman Svenselius 2012-05-11