Post-processing multiple-frame super-resolution in ultrasound imaging

High resolution medical ultrasound imaging is an ongoing challenge in many diagnosis applications and can be achieved by instrumentation. Very few works have investigated ultrasound image resolution enhancement whereas many works regarded general purpose optical image or video fields. Many algorithms were proposed within these fields to achieve the "super-resolution" (SR), which consists in merging several low resolution images to create a higher resolution image. However, the straightforward implementation of such techniques for ultrasound imaging is unsuccessful, due to the intrinsic nature of ultrasound motions and speckle. We show how to overcome the intrinsic limit of super-resolution in this framework by refining the registration part of common multi-frame techniques. Classic super-resolution algorithms were implemented and evaluated using sequences of ultrasound images. Such methods not only fail to estimate the true elastic motion but also break the speckle characteristics, resulting in a degradation of the original image. Knowing that a registration error of only 1 pixel leads to a high-resolution image worse than an interpolation, the registration must be adapted to the framework of ultrasound imaging. For this purpose, we investigate different motion estimations. The process described above was evaluated on ultrasound sequences containing up to 15 phantom images with an inclusion scanned with a 7.5 MHz linear probe. Qualitative improvements were observable as soon as at least 5 low-resolution images were used. Ultrasound B-mode profiles of radio-frequency lines were studied and the inclusion was more accurately identified. The Contrast-to-Noise Ratio was increased by approximately 13%.