Comparative Study On The Performance Of Ultrasonic Shear Wave Velocity Calculation Based On GP

As an emerging technology,ultrasound shear wave elastography has solved the problem that early elastic elastography cannot quantitatively calculate the elastic modulus.The basic principle of this technology is to excite the acoustic radiation force through the ultrasonic transducer,so that the shear wave propagates in the tissue,then calculate the wave velocity of the shear wave propagating in the tissue,and finally calculate the elastic modulus through the inversion formula.Shear wave elastography is divided into two main steps:motion displacement estimation and shear wave velocity estimation.Motion displacement estimation consumes a lot of time due to the calculation of ultrasonic RF signals of multiple frames.When medical staff perform actual diagnosis,too long imaging time will not only increase the computational burden of the diagnostic equipment,but also cause the patient to have a certain sense of anxiety and affect the diagnosis result,so a technology is needed to reduce the imaging time overhead.For the motion displacement estimation of two-dimensional ultrasonic radio frequency signals,the GPU(graphics processing unit)platform can solve the problem of time overhead.At the same time,in order to reduce the time overhead of data copying,the shear wave velocity estimation algorithm needs to be implemented on the GPU platform.Although there have been some studies applying the relevant algorithms of shear wave elastic imaging on GPUs,there have been no relevant studies that have made systematic performance comparisons of these algorithms on GPU platforms.The contribution of this paper is to compare the computational performance of the relevant algorithms of shear wave elastic imaging on the GPU platform,and gives the parallel strategy of the algorithms mentioned in this article on the GPU platform for the reference of subsequent researchers.In order to generate experimental data for research,this paper first starts with the simulation of shear wave elastic imaging,discusses the parameter selection in the simulation scheme,and introduces the simulation process of acoustic radiation force and shear wave propagation in tissues in detail.Then compare the calculation performance of the fast motion displacement estimation method based on the sum table method on the GPU platform when using different similarity/correlation evaluation methods.At the same time,the parallel strategy and memory access optimization method adopted by the algorithm on the GPU platform are given.Finally,the calculation performance,anti-noise performance and algorithm accuracy of the TTP algorithm based on the RANSAC filtering based on the Time of Flight algorithm,the Radon algorithm and the time domain Fourier domain speed matching algorithm on the GPU platform are compared,and The theoretical basis,calculation flow,pseudocode implementation and parallel strategy on GPU platform of these three algorithms are given.The experimental results show that the fast motion displacement estimation method on the GPU platform will not be affected by the size of the tracking window,but as the search range increases,the computing performance will decline,including unexpanded SSD,SAD and CC The method is less affected by the search range,and the NCC and expanded SSD are more affected.On the GPU platform,the shear wave velocity estimation algorithm can have certain performance improvements,among which the performance of the TTP algorithm based on RANSAC filtering and the time domain Fourier domain speed matching algorithm are greatly improved.The time domain Fourier domain speed matching algorithm and the Radon sum algorithm can still estimate the modulus value relatively accurately when the signal quality is reduced(signal-to-noise ratio is at least greater than 10dB),and the TTP algorithm based on RANSAC filtering clearly deviates The true modulus value(signal-to-noise ratio is less than 10dB).In addition,the time domain Fourier domain speed matching algorithm has an obvious advantage in that it can improve the spatial resolution of the modulus estimation.