Design Of Parallel Platform For Optical Coherence Tomography

As a new type of biomedical optical imaging technology,optical coherence tomography(OCT)can not only provide intact,non-contact,high-resolution and highsensitivity images of the microstructures of living tissues,but also present functional distributions like blood oxygen,blood flow velocities.and different optical properties.Thus,from the beginning as a research method in ophthalmology and other clinical applications,OCT has gradually become an indispensable imaging aid in guiding interventional treatment of various diseases.With the development of OCT technology,intraoperative OCT navigation,rapid diagnosis and other technologies have been applied in clinical applications,and people’s requirements for real-time OCT imaging are becoming more and more urgent.The imaging speed of the OCT system mainly consists of two parts,one is the data acquisition speed of the system,and the other is the data processing speed.With the development of acquisition-related hardware and software technology and digital signal technology,the acquisition speed of FD-OCT has been achieved up to 300 k A-line/second,which provides the premise for real-time imaging of OCT.However,the data processing speeds of traditional OCT imaging systems based on central processing unit(CPU)platforms are low,which cannot meet the real-time imaging requirements of the OCT system.Therefore,how to improve the data processing speed has become the bottleneck of OCT system realtime imaging.Using graphics processing unit(GPU)high-speed parallel computing capabilities to reconstruct OCT images is the best solution for real-time imaging of OCT systems.On the basis of our laboratory’s existing frequency-domain OCT experimental system,the thesis adopts GPU graphics cards NVIDIA GeForce GTX TITAN X and other devices with powerful parallel computing capabilities and floating-point computing capabilities and sufficient memory,combined with the compute unified device architecture(CUDA)parallel development architecture,which is configured for CPUGPU heterogeneous computing,and builds a parallel computing platform suitable for data processing of OCT imaging,and carries out the parallel design and optimization of all the computing modules based on CUDA C in the process of FD-OCT data processing.In order to verify the accuracy and effectiveness of the parallelized design,the FD-OCT series of animal and human experiments were carried out based on the CPU mode and the CPU-GPU heterogeneous mode,respectively,two-dimensional OCT scanning imaging of mouse ears,abdomen and human fingers were performed.The series of FDOCT experimental results confirmed that under the same quality imaging effect,the OCT data processing efficiency based on the parallel computing platform developed by this work has increased by at least several tens of times,which meets the requirements of twodimensional FD-OCT real-time imaging,and it also has great potential in real-time threedimensional FD-OCT imaging,and the parallel design is easy to transplant into most FDOCT real-time imaging systems.At present,there are many foreign studies on GPU accelerated real-time OCT imaging,but domestic research in this area is still relatively backward and is in its infancy.Therefore,this work and subsequent research will focus on the development of domestic real-time OCT imaging technology and should have important academic significance and reference value for researchers in this area.