Ultrasound image compression exploiting image formation models

The diagnostic benefits of teleradiology may not be available in locations which lack high-bandwidth transmission channels. In these cases, data compression is required in order to use low bit-rate digital channels. Standard compression algorithms, such as JPEG, are not ideally suited to ultrasound images, in which speckle structure plays an important role. We have developed an algorithm for progressively encoding ultrasound images which incorporates knowledge of both the system parameters and the image formation process, and which preserves the speckle structure in the image. The algorithm operates on the digitized RF output of an ultrasound scanner and identifies pixels in the image giving rise to the largest output in a local neighborhood. It iteratively allocates point scatterers to these pixels and transmits their amplitudes and locations using Huffman encoding. The image is reconstructed by convolving an RF pulse with the received sparse scatterer matrix. We describe the basic algorithm and compare its performance, using rate-distortion curves, to the JPEG standard. Distortion is measured by both simple root-mean-squared (RMS) error and human visual system (HVS)-weighted RMS error. The HVS distortions are included to better account for the perceived differences in image quality seen by the diagnostician.