Dissertation > Excellent graduate degree dissertation topics show

Research on Parallel Processing of Sparse MRI and Direct Fourier Transform Reconstruction Algorithm

Author: WangCong
Tutor: ChenWuFan;FengYanQiu
School: Southern Medical University,
Course: Applied Computer Technology
Keywords: Magnetic resonance imaging Non-Cartesian sampling Sparse sampling Parallel Computing Multi-core processors GPGPU
CLC: TP391.41
Type: Master's thesis
Year: 2011
Downloads: 66
Quote: 0
Read: Download Dissertation

Abstract


Magnetic resonance imaging has no ionizing radiation, the advantages of multi-parameter imaging, functional imaging, can be in any direction tomography. Become one of the most important imaging method has been widely used in clinical magnetic resonance. Has many advantages at the same time, the slow scanning speed is a major bottleneck in the magnetic resonance imaging, not only a long time to wait for the pain caused to the patient, so that the magnetic resonance difficult to apply to the dynamic imaging require fast imaging. Since the magnetic resonance imaging appears to improve the imaging speed has been a research topic, non-Cartesian sampling (such as a spiral path, radial trajectory, propeller, etc.) has been a good application, emerging sparse magnetic resonance imaging methods are as a hot research. The data for non-Cartesian sampling to various interpolation methods commonly used resampling to the the homogeneous Cartesian coordinate points, and then the fft method reconstructed image classic. Interpolation method is relatively mature, but not interpolation and direct conjugate phase reconstruction algorithm is still considered the most accurate method for the discrete Fourier transform. Conjugated phase algorithm biggest problem is that the time complexity is too high, it is generally not used clinically only in research to evaluate the accuracy of the various interpolation algorithms used to generate a control image. The recently appeared sparse magnetic resonance imaging reconstruction algorithm to solve the problem is from the the pathological high degree of underdetermined infinite solutions of the equations to find the a sparse or transform sparsity solution, which is an iterative algorithm, if nonlinear conjugate gradient descent algorithm, the iterative process complex matrix operations to multiple spatial transform, gradient, and therefore is a very time-consuming algorithms. Solve compute-intensive time-consuming method of parallel computing, that is a large task into multiple sub-tasks can be carried out at the same time, at the same time perform subtasks and then distributed to multiple processors in the same computer or a different computer subtasks finished then consolidated the results of each sub-task to host the final result of the task. Classic parallel computing using multi-CPU computers or computer clusters, covers an area of ??and power consumption of these supercomputers are very large, and applied to clinical costs; desktop computers with multi-core CPU although there is some effect but not enough to significantly speed emerging GPGPU integrated into the Grid computing hardware required for a board, largely to reduce the size and power consumption. From another perspective, GPGPU is the improvement of traditional GPU traditional GPU can only be carried out in accordance with the fixed pipeline graphics processing pipeline every level of high-performance parallel computing power, but very easy to use, GPGPU the GPU improved processor available C / C or Fortran programming language, can be more convenient to be applied to a variety of compute-intensive tasks. Firstly relatively simple principle conjugate phase algorithms for two kinds of granularity task decomposition, and then multicore desktop computer easier to understand the shared storage and slightly complex GPGPU its acceleration, op-phase and single-core serial The ratio of 5-fold and 72-fold speedup. Sparse magnetic resonance reconstruction algorithm is more complex process, This article and its sub-processes were parallel decomposition, sub-process to handle by the GPU, a small amount of the overall iterative process control calculations, as a serial program executed by the CPU. Acceleration effect is obvious with Michael Lustig SparseMRI0.2 compared to 76 times the acceleration.

Related Dissertations

  1. MRI Guoded Experimental Cryoablation of Rabbit’s Sciatic Nerve for MRI and Pathology Contrast Study,R445.2
  2. The Clinic Value of Diffusion Weighted Whole Body Imaging with Background Body Signal Suppression (DWIBS) Compared with hPET/CT,R445.2
  3. Diffusion Tensor Imaging Study of Idiopathic Epilepsy at 3.0T High Field Magnetic Resonance Scaner,R445.2
  4. A Study of Wallerian Degeneration in Cortico Spinal Tract after Cerebral Infarction by fMRI,R445.2
  5. A Volumetric Study of External Capsule, Extreme Capsule and Inferior Frontal Gyrus White Matter in First-Episode Paranoid Schizophrenia,R749.3
  6. Healthy Chinese central olfactory structures related to MRI volume and olfactory function with age correlation,R445.2
  7. 3.0T BOLD MR techniques on acute unilateral ureteral obstruction of renal damage,R693.2
  8. Progressive supranuclear palsy clinical and magnetic resonance imaging characteristics of,R741
  9. 3.0T high-field magnetic resonance imaging of the central type lung cancer preliminary study,R734.2
  10. MR diffusion weighted imaging in Application Value of osteonecrosis,R681.8
  11. Application Study on Quantitative Analysis of Diffusion-weighted Imaging and Dynamic Contrast-enhanced MRI in Thediagnosis of Breast Lesions,R445.2
  12. Preliminary Study of the MRI Diagnosis and Differential Diagnosis in Multiple System Atrophy , Progressive Supranuclear Palsy and IPD,R445.2
  13. The Application of STIR-FLAIR and T2W-SPIR for Diagnosing the Orbital Lesions: a Contrast Study,R777
  14. Application of Functional MR Imaging in Diagnose for Prostate Cancer,R737.25
  15. Comparative Study of Clicking Temporomandibular Joint Before and after Being Cured by MRI,R445.2
  16. Research and Design of a High-Performance Scalable Public Key Cryptographic Coprocessor,TN918.1
  17. Research on Video Compression Algorithm Based on Multi-core Computing Platform,TN919.81
  18. Implementation and Research on Radio Link Control Protocol in TD-SCDMA,TN929.533
  19. Study of Binocular Stereo Matching Algorithm and Algorithm Implementation Based on Multi-core.,TP391.41
  20. Research of Finite Element Method on GPU,O241.82
  21. Numerical Simulation of Radiofrequency Waves in Magnetized Plasma,TL612

CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device
© 2012 www.DissertationTopic.Net  Mobile