Ultrasound algorithm outperforms JPEG at high compression rates

Today`s medical ultrasound scanners generate 0.25-Mbyte images at 30 frames/s. To transmit and store such images economically, researchers are turning to data-compression techniques to reduce their size. "Unfortunately," says Kevin Parker, director of the Rochester Center for Biomedical Ultrasound at the University of Rochester (Rochester, NY), "the nature of ultrasound images is very dissimilar to that of photographic and broadcast television images."

Ultrasound algorithm outperforms JPEG at high compression rates

Today`s medical ultrasound scanners generate 0.25-Mbyte images at 30 frames/s. To transmit and store such images economically, researchers are turning to data-compression techniques to reduce their size. "Unfortunately," says Kevin Parker, director of the Rochester Center for Biomedical Ultrasound at the University of Rochester (Rochester, NY), "the nature of ultrasound images is very dissimilar to that of photographic and broadcast television images."

Indeed, the small speckle patterns that dominate ultrasound images combined with fast-moving cardiac and blood structures result in very high spatial and temporal frequency components. "These strong high-frequency components strip away the compression efficiency that most existing algorithms, such as the Joint Photographic Experts Group (JPEG) and Motion Picture Experts Group (MPEG) standards, achieve in smoothly varying, nonspeckled reflectance images such as video," says Parker. Such specular patterns result from coherent interference between reflected acoustic pulses.

To compress such images, Parker has developed a model-based algorithm (MBA) that synthesizes the original RF ultrasound image from an array of point scatterers. "In operation, local peaks from the RF of an ultrasound scanner are assumed to be caused by a large-amplitude point scatterer at that location," says Parker. These scatterers` locations and amplitudes are then stored in a sparse matrix called a scatterer map, where the size of the local neighborhood in which a peak is identified is determined by the shape and extent of the system`s point-spread function (PSF).

To obtain image compression, the scatter map is quantized and compressed using Huffman encoding. The encoded map, along with information describing the PSF, can then be transmitted at low data rates in teleradiology systems. To reconstruct the image, points in the scattermap are deconvolved with the PSF to reconstruct the original image.

"Because further iterations of the image operate on the differences between an RF image and the synthesized image, the result is progressive image encoding and reconstruction," says Parker. Using a phantom scatterer image, Parker and his colleagues compared the MBA approach with the JPEG compression technique.

"Because JPEG quantizes the discrete-cosine-transform coefficients in 8 ¥ 8-pixel blocks, blocking artifacts appear," says Parker. "With the MBA technique, the images retain their specular nature and exhibit a potentially significant performance advantage over JPEG at high compression rates."

More in Life Sciences