The Design And Implementation Of Software And Hardware Of Tomosynthesis Imaging System Based On Dynamic DR Model

Nowadays,the number of cancer patients is increasing rapidly,and cancer screening is the most effective means to reduce the incidence and mortality of cancer.As the most commonly used cancer screening methods,DR and CT are difficult to apply to large-scale and regular cancer screening scenarios because of their own limitations.As a feasible auxiliary means of CT,tomosynthesis can obtain arbitrary tomographic images of human body with low dose and low cost,and has higher image resolution.This makes tomosynthesis the easiest tomographic method to popularize in the population,and it also becomes a research hotspot in the field of medical imaging.However,at present,many commercial tomography systems only support the tomography of small-size human body region,which not only reduces the utilization rate of the system but also cannot image long-sized human body parts.In order to solve this problem,the research and development of tomography system for long-size human body region is carried out in this thesis from both hardware and software aspects.The hardware aspect mainly involves the system structure design and construction of hardware devices.Tomography of long-size human body region requires the support of long-distance linear scanning trajectory.In order to realize this scanning trajectory under the condition of limited detector size,this thesis designs a hardware system structure in which ray source and detector move in the same direction synchronously.This system structure can not only easily collect the projection data of long-size human body region through one scan,but also be realized by a slight modification of the company’s dynamic DR machine,which greatly saves the equipment purchase cost and avoids the complicated hardware construction process.Due to the inseparable relationship between the tomographic performance and the hardware stability,it is necessary to evaluate and correct the system stability of the modified dynamic DR machine to make it meet the requirements of tomography.The software aspect mainly involves the research work of tomography algorithm that meets the clinical requirements.Long-distance linear scanning trajectory usually truncates the projection data,in which case the iterative tomography algorithm is more suitable than the analytical tomography algorithm.Simultaneous iterative reconstruction technique(SIRT)is used as the main tomography algorithm in this thesis because of its low requirement on the completeness of projection data.The common problem of iterative tomography algorithm is that the tomographic time is too long to meet the clinical requirements for time.Using the parallel computing ability of GPU to accelerate the algorithm is a recognized and effective solution.Due to the wide application of CUDA technology,GPU parallel computing technology based on CUDA is used in this thesis to realize fast SIRT algorithm,which greatly improves the tomography speed.In addition,in order to further suppress stripe artifacts caused by truncation of projection data,regularization method is generally used.The total generalized variation(TGV)model based on compressed sensing theory has been widely used as the regularization term of iterative tomography algorithm,which can effectively improve the accuracy of tomography.In order to fully combine the characteristics of the system structure,this thesis improves the TGV model and finally realizes the SIRT algorithm based on unidirectional adaptive TGV regularization.Experiments show that the proposed algorithm can effectively suppress the perspective artifacts and stripe artifacts,and keep the image details while suppressing the noise,which meets the clinical requirements in terms of tomographic quality and tomographic time.