Investigation Of Cuda-basedacoustic Radiation Force Imaging

Ultrasound Elastography is an imaging tomography using Youngs modulus, shearmodulus and stress-strain elastic parameters obtained by ultrasound echoes. Compared withother conventional ultrasound tomographies, Ultrasound Elastography is more intuitively toreveal mechanical property of tissues. Ultrasound Elastography is widely used in fatty liverdisease, hepatic cirrhosis, and breast tumor testing and achieves a good diagnosis rate.Elastography is divided into two categories: Quasi-Static Elastography, Acoustic RadiationForce Imaging（ARFI）.Quasi-Static Elastography depends on exterior force source to drivethe transducer to compress the tissues and can not offer quantitative information of tissues.ARFI is a new tomography using measure parameters related with displacement induced byacoustic radiation force produced by highly-focused acoustic waves generated by transducer.ARFI offers good performance in deep region and quantitative information.However, drawbacks such as long imaging time and intensive data exist in ARFI. LongImaging time has a great impact on results obtained from elastography for increasing anxietyof patients. On the other hand, probabilities of burning transducer due to temperature risearising from long imaging time. To resolve these problems, we advance a method ofimplementingARFI on GPU. The research is organized as follows.①The background and the significance of this paper are introduced, Overseas anddomestic research status of real-time ultrasound imaging is introduced and displacementestimation is selected to illustrate the difficulties the real-time ARFI facing as one module ofARFI. At the end of this part, the problems and challenges in ARFI is summed up and themethods discussed in the thesis is introduced.②Background of ARFI is first briefly introduced. And then research works ofdisplacement estimation, interpolation, shear wave speed estimation is analyzed and theadvantages of our methods is presented after comparing with these methods. In the end, Allthe modules ofARFI are formulated in detail.③Some fundamental knowledge of CUDA and GPU is introduced. To overcome theshortcomings of the traditional time-domain correlation algorithms such as poor data-reuseand high computing complexity, a parallel window-iterative correlation method is presented.Finally, details of implementation of allARFI modules is given.④The acceleration effects of ARFI using GPU is obtained by timing modules of ARFIrunning on GPU and CPU. Then, the image quality of elastography of GPU and CPU is compared.And no great loss of accuracy is observed on the elastography of GPU.