Research On GPU Based Real-time Ultrasound Elasticity Imaging

It is known to all that tissue elasticity has a close relationship with its pathological state due to the significant difference in lesion tissue and normal tissue, which can been used for many disease such as tumor and so on. Ultrasound elastography is a new type of Ultrasound imaging, which identifies lesions through the hardness of tissue recovered from ultrasonic data. However, it needs huge data to construct a elastography image. To get real-time diagnosis information is quite hard. This paper focuses on two aspects:1. Improving the parallel programs of traditional displacement estimators with the GPU.2. Improving the algorithm itself to fit for the GPU Programming.This paper studies on frequently-used displacement estimators:1.time-based cross-correlation method;2.phase-based displacement estimators, including Doppler motion estimation and Phase Zero Estimation. Based on computer programming, I transplant the CPU program to GPU. Because of the raw data is quite huge, it is usually processed offline. Thus it is impossible to acquire real-time elastography. We anew design the three main displacement estimators programs based on GPU. And based on GPU programming, we analyze and compare the three estimators’ strong points and weak points. Then we propose a new algorithm which is better suitable for GPU implementing. This new algorithm is based on three steps. First, it uses Hilbert Transform to construct the analytic signal, and uses the cross-correlation method but only compute once. Then, cubic spline interpolation is used to make the loss of phase values which is caused by the sampling. To improve the whole computation speed we also transform the Hilbert algorithm and cubic spline interpolation algorithm into GPU. The whole speed is improved apparently.Last, I demonstrate the methods above in simulated data that are acquired in Field II and Phantom model data. After comparison, I do some discussions on the experiment results.