Study Of GPU-based Phase Retrieval And CT Reconstruction On Grating Phase Contrast Imaging

The content of this thesis is the GPU based parallel computing method on gratingphase contrast imaging and CT reconstruction. There are mainly two aspects: one is theset-up of the whole grating phase contrast imaging system and the advantage of the gratingphase contrast imaging system in medical application and other fields; the other one is tofulfill the parallel computing acceleration using CUDA C programming modelimplemented on GPU for data processing.First of all, the set-up of the whole set of grating phase contrast imaging system hasbeen built in the thirteenth line station of Shanghai Synchrotron Radiation Facility (SSRF).The whole system mainly contains three important parts:1, the sample stage with controlpanel, the shift and rotation of which is controlled by motor Kohzu through theNport5610(serial port). Before the experiment, it would be efficient to utilize the levelgauge on checking sample whose height should have equal altitude with the X-ray;2, thegrating interferometer stage with control panel which contains two type of gratings: one isthe phase grating and the second one is the analyzer grating. The control panel of gratingsshould be qualified for each grating to have transverse, lengthways, or rotary movements.3,data acquisition system in which includes a CCD detector or even a SCMOS camera, a dataacquisition card and a computer.Secondly, compared with the conventional absorption imaging, the grating phasecontrast imaging system shows a higher spatial resolution and contrast resolution onreconstruction images especially when dealing with the samples with low attenuationcoefficient such as the vessels, soft tissues and so on. The reason is that the phase shift ismore apparent than the absorption when X-ray passes through the weak absorption tissues,so it’s more effective to utilize the phase rather than the absorption information. What’smore, the grating phase contrast system has more advantage on the wide range of sample selections and independence with the crystal compared with the propagation-based phasecontrast imaging and the diffraction enhancement imaging.Furthermore, large data acquired from the phase-stepping method in every projectionangle consumes the time in the grating phase contrast imaging experiment. What’s more,the Filtered Back Projection (FBP) algorithm requires a complete set of projection datafrom all the angles which means the data to be dealt with in CT reconstruction based on thesystem would be a big project. Graphic Processing Unit (GPU) has a wide advantage onparallel computing with multiprocessors and thousands of threads. Besides, the complexityof retrieving phase and back projection is suitable for parallel computing.In conclusion, this thesis utilized the CUDA C programming model based on GPU toaccelerate the phase retrieval and FBP algorithm on grating phase contrast system.According to the different size of data, the CUDA C program shows a different speed-upover the standard C program on the same Visual Studio2010platform. Meanwhile, thespeed-up ratio shows an increasing tendency when the size of data increases.