Design And Implementation Of Microwave Thermoacoustic Imaging Reconstruction Based On Heterogeneous Platform

With the development of medical imaging technology,biological tissue imaging technology based on different imaging mechanisms has been widely used in the detection of various diseases to assist physicians in disease diagnosis and treatment.The nondestructive microwave-induced thermoacoustic tomography(TAT)combines the characteristics of high resolution of ultrasonic imaging and high contrast of microwave imaging,which has become a new medical imaging method with great development potential.TAT reconstruction algorithm based on finite element method(FEM)by solving thermoacoustic equation,using an iterative approach to image reconstruction.It can not only provide biological information within the organization structure,showing the location and shape of diseased tissue,can also provide the physical parameters related to the thermoacoustic effect.But this kind of reconstruction algorithm based on FEM has large calculation.With the increase of finite element meshing units,the time needed for image reconstruction increases rapidly.It can take hours to reconstruct an image.This seriously affects the research and application of TAT reconstruction algorithm based on FEM.With the development of science and technology,people have realized the importance of computing more and more.Whether in scientific research,technological development or daily life,the demand for computing is growing rapidly.Due to the limitations of physical constraints,the method of improving computing performance only on a single processor has encountered a bottleneck.Therefore,the strategy of combining multiple processors to build high performance computing platform is proposed.Based on this strategy,there are many supercomputers in the world.Up to 40 Tflop/s of floatingpoint computing power,far beyond the capacity of any single processor.In order to solve the problem of long running time of TAT reconstruction algorithm based on FEM,this paper studies from the following aspects:1.The CPU is good at processing logical and complex tasks.GPU is good at processing computation-intensive and parallel tasks.Combining the characteristics of CPU and GPU,we build a heterogeneous architecture based on CPU and GPU.And deploy a development environment based on CUDA and the Python language.2.We research on programming techniques based on heterogeneous platform,including data storage optimization,memory access optimization,data transmission optimization,multithreaded parallel computing,block computing and other acceleration methods.3.TAT reconstruction algorithms based on FEM are deeply analyzed and redesigned.In this paper,we propose a design scheme of absorption distribution reconstruction and thermoelastic elastography based on CPU and GPU heterogeneous platform.We designed a simulation experiment to verify the algorithm based on heterogeneous platform and compare the computational performance.Compared with the CPU,the absorption distribution reconstruction based on heterogeneous platform achieves an acceleration of 67.549 times,and thermoelastic elastography achieves an acceleration of 16.36 times.