A Study On Motion Artifacts Reduction And Diffusion Tensor Imaging Of Magnetic Resonance

Magnetic Resonance Imaging (MRI) is a non-invasive approach of mapping the internal structure of the human body. Compared with other modalities of medical imaging, MRI has many advantages, such as with no ionizing radiation, imaging with multiple parameters in arbitrary plane, and providing excellent soft-tissue contrast. All those advantages make MR a very important tool for both research and clinical applications.The phenomenon of nuclear magnetic resonance (NMR) was first discovered separately in 1946 by research groups led by Bloch and Purcell. With this discovery, they were jointly awarded the Nobel Prize for physics in 1952. Nearly 20 years later, the earliest MR images was produced by Lauterbur in 1972 and Mansfield in 1973, and the first whole human body MRI scan was reported by Damadican in 1977. Over the past three decades, there has been a rapid development of MRI from the laboratory prototypes to the widespread installation in hospitals all over the world as an important imaging modality for clinical examination.As a powerful imaging modality, MRI is based principally upon sensitivity to the presence and properties of water which makes up 70% to 90% of the most tissues. The properties and amount of water in tissue can alter dramatically with disease and injury which makes MRI a very sensitive technique for medical diagnosis. As its most advanced, MRI can be used not only to image anatomy and pathology but also to investigate organ functions, to probe in vivo chemistry and even to visualize the human thinking.Along with its development, MRI always encountered the problem of artifacts caused by chemical shift, truncation, wrap around and motion. With the development of technology and the improvement of both hardware and software techniques, some of the artifacts have been avoided or suppressed remarkably. But the motion artifact is still a problem due to the relatively long acquisition time of MRI. Motion artifacts can be produced by physiological motion or involuntary motion of patient and tent to appear as blurring and/or ghosts, which obscure vital anatomical and pathological details, and obviously degrade the quality of the image.To avoid or minimize the artifacts caused by motion, many approaches and techniques have been proposed, which can be broadly divided into two categories: pre-processing methods and post-processing methods. Techniques in the first category, sometimes called real-time artifact suppression, attempt to prevent the motion from corrupting the data. And those in the second category utilize signal processing techniques to restore the image from corrupted data.The real-time artifact suppression approaches are now widely used in clinical applications. These techniques depend basically on controlling the acquisition time and restraining patient's motion. Although effective in reducing artifacts caused by periodic motion, they are inadequate in general, especially to random involuntary motion of the patient. Practice showed that even with the methods of restraining patient's motion , tiny rotations are always occurring. To suppress this kind of artifacts, several approaches were presented depending basically on acquisition time reducing. But such approaches as PROPELLER scan turned out to be very sensitive to the inhomogeneity of magnetic field. Other fast acquisition methods, such as echo-plane imaging (EPI), can complete the acquisition in shorter time with the cost of resolution and signal-noise ratio degradation. In fact in those applications requiring high resolution and high SNR, the 2D Fourier Transform based spin-warp imaging is still being used.The post-processing methods are useful in restoring images corrupted by the motions immune to the suppression techniques mentioned above. Initial investigation work in this area, were instigated by Wood, Henkelman, and Haacke during 1985-1986. Since then considerable research efforts have been invested. Post-processing technique usually requires a fitting motion model which can approximate the actual patient motion. The parameters associated with the chosen model must be measured or calculated using the information extracted from the acquired data. Despite its prominence in research, post-processing methods have not been implemented for wide use. This is mainly due to the inadequateness of the existing models and the enormous computational cost required for parameter estimation and consequent restoration. However the study for sophisticated post-processing algorithms has been continued due to the promise that post-processing techniques might be used in conjunction with existing motion suppression techniques in order to achieve greater suppression of motion artifacts than either type of techniques can provide alone. Therefore the motion artifact suppression and restoration techniques should be viewed as complementary each other rather than excluding.The primary objective of this thesis is investigating the artifacts caused by in-plane rigid motion, including translation and rotation, developing post-processing algorithms for the suppression of these motion artifacts, improving the quality of MR images for both research and clinical applications. These works include:The mechanisms and the influences of ghost artifacts are discussed, mathematical descriptions of motion artifacts are given, and the existing methods to suppress motion artifact are introduced and compared.A translation motion reduction method named EC (Inverse Iterative Correction) is proposed. The simulated inverse motion is used to compensate the acquired data, and is iteratively refined by minimizing the image histogram entropy. The histogram-based criterion favors alteration to data that tend to increase the number of dark pixels, suppress the vibration around the middle grayscale. And finally we achieve the goal of artifact suppression while minimum energy of the image is reached. The correction speeds up through optimal calculation of entropy which uses the grayscale levels instead of the number of image pixels.A rotation motion estimation algorithm is proposed, which is based on narrow-band level set boundary extraction and X-directional inverse Fourier transform of the acquired K-space views. This method is efficient and effective by transforming the problem from N-dimensional optimization to NX 1-dimensional optimization. The experiments showed that the proposed methods was capable of estimating rotation angles according to each view and performed better especially for large angle estimation.Based on the bilinear superposition algorithm, a rotation motion correction involving management of data overlap and void regions is presented. At the data overlaps, a more accurate weighting average is used. While to the void regions, data-filling is executed according to conjugate symmetry prior to POCS iterative optimization by definition of two convex sets.Fast acquisition techniques usually utilized two approaches to suppress motion artifacts. One is by reducing the acquisition time, and the other is by reducing the data acquired. The former techniques utilize new and fast sequences (such as EPI) or fast hardware. The later techniques reduce the acquired data by reducing the number of scanning lines in K-space, such as half Fourier transform, and key-hole imaging methods, etc. Although these approaches can reduce the acquisition time, but the data acquired are not complete. So the reconstruction by Fourier transform with those incomplete K-space data would introduce the Gibbs ringing artifacts which degrade the resolution and SNR of the image, blur the details of anatomy, and even make theimage useless for diagnosis. An approach named IPRM is proposed which uses partial K-space data to effectively accelerate the imaging and suppress the Gibbs artifacts at the same time. The new method reduces the construction error and computational cost effectively with no need to select the parameters.In the middle of twenty century, investigators used the pulsed gradient to encode the motion effect of water molecular and produced the diffusion weighted image(DWI).A new MR imaging method called diffusion tensor imaging (DTI or DT-MRI) has emerged, which utilizes the diffusion tensor of each voxel calculated from multiple direction DWI (at least 6 directions). As a non-invasive and very important tool for study of the internal structure of the brain, DTI has been widely used in the research of neurophysiology, neurosurgery, and diagnosis of brain cancer, etc. There is a second order tensor corresponding to each voxel in the DTI image, and the representation of those tensor data and digging of the meaningful information in those data become the main issues for the research of DTI and its applications, including the processing of diffusion tensor data, fiber tracking, seeking the real and intuitive representation or display of these results.The calculated tensor data might not satisfy to trace the fibers due to noise in both the diffusivity and orientation. The PDEs based method for the regularization of the tensor field is studied, by working directly on the spectrum decomposition of the noisy tensor field while preserving the constrains of positive semi-definition and symmetry in the spectral space. And further a fiber tracking algorithm using variable step in primitive direction of diffusion is presented which can achieve continuing paths of white matter fibers.The thesis consists of two parts:The first part is the study of post-processing techniques of suppression of motion artifacts in MRI. There are six chapters in this part. Chapter 1 introduced the fundamentals of MRI. Chapter 2 analyzed the causes of motion artifacts, along with methods to avoid or minimize them. Chapter 3 to Chapter 6 discussed translation motion correction, rotation estimation, rotation motion correction and reduction of Gibbs ringing in MR reconstruction with partial K-space data respectively.The second part of the thesis mainly concerns diffusion tensor imaging, including Chapter 7 and Chapter 8. Chapter 7 introduced in detail the fundamentals of DT-MRI, its application and visualization. Chapter 8 discussed regularization of tensor field and fiber tracking, and a prototype platform for DTI data processing and visualization was also established.