Research On A GPU-based Real-time Ultrasound Imaging System

Ultrasound imaging, due to its real time, portability, low cost and harmless to human body, has been widely utilized in the prevention and diagnosis of various disease, which takes an important role in modern medical imaging systems. Delay and sum (DAS) beamforming is one of the most traditional and simplest ultrasound imaging algorithms. Beam control methods which are used in DAS beamforming, such as dynamic focusing, apodization and dynamic aperture, are regarded as effective methods to control the width of the main lobe and the amplitude of side lobes. Nevertheless, the imaging mode of the conventional DAS algorithm is too simple to further improve the spatial resolution, even if DAS brings in some beam control methods.In order to further improve the image quality, a kind of synthetic aperture (SA) imaging method is proposed. Initially, it divides the transducer array into multiple receiver subarrays and emission subarrays. Each subarray consists of one or more successive elements. Then, when transmitting ultrasound, each emission subarray sequentially transmits ultrasound on the basis of a certain transmission focusing pattern. And the transmission beam covers the corresponding region of tissue. Additionally, a receiving operation is carried out using the corresponding receiver subarray after each transmission. Afterward the system stores the received echo signal. Eventually, the focusing value of the target pixel is calculated according to the delays between the pixel position and each element of the receiver subarray. Compared with the traditional DAS beamforming, the SA imaging can achieve a higher resolution, but this kind of transmitting and receiving mode costs a lot of time to produce the calculation. Therefore, a common hardware platform is very difficult to realize real-time imaging.Given the increasing computation complexity of modern ultrasound imaging systems, a novel imaging method based on GPU acceleration is presented. This method makes CPU responsible for memory management, logical control and image display. Meanwhile, GPU is in charge of a large amount of parallel computing, including signal preprocessing, gain control, delay-and-sum, envelope detection, dynamic compression, noise reduction and image interpolation. Furthermore, image display function can be accomplished by calling Matlab engine to get data and compare with Matlab simulation, or using OpenCV for fast display.Finally, this article verifies the validity of DAS beamforming and multi-element SA imaging by reconstructing images with simulated data with a multi-point template and a multi-circle template and real data. And the result indicates that SA imaging attains a higher contrast, spatial resolution and robustness than DAS beamforming. Under the same experimental environment, the use of GPU obtains200to300times speedup than Matlab, and quadruple speedup than CPU multi-thread. Certainly, using better CPU and GPU devices, and allocating sources and adjusting the architecture more reasonable, can increase imaging rate further.