Estimation Of Diffusion Tensor And Its Relevant Technical Research Based On DT-MRI

Diffusion Tensor Imaging, the sole medical imaging technology capable of non-invasively examining the diffusion of water molecules in vivo, is a new nuclear magnetic resonance imaging technique arising in1990s. DTI has its unique advantages in the study of cerebral cortex, white matter tractography and that of various brain diseases as tumor, infarct, cognition, and etc., and thus has quite an extensive application prospect in medical imaging fields.This dissertation presents a thorough study, of the estimation of diffusion tensor, the denoising of DTI (Diffusion Tensor Images), the characters of its noisy field, the effect of noise on diffusion tensor. Firstly, the distributing characteristics of noise fields in DTI and the effect of the noise on diffusion tensor have been analyzed. The model of Rician noise has been built based on the analysis of the noise in DTI. And then the method of moment estimation has been proposed to be applied in the estimation of noise field, which is more effective. Meanwhile, based on this noise model, a Monte Carlo simulation experiment has been performed to make a thorough analysis of the bias caused by the noise in DTI and its indices (such as Fractional Anisotropy and Trace). The experiment leads to summarizing the variation laws of the features of diffusion tensor varying according to the noise variation, which are significant in analyzing and diagnosing medical images.Secondly, based on the character of diffusion technology, a model of tensor estimation, the features of the outlier caused by noise and its effect on diffusion tensor estimation have been analyzed. Meanwhile, after a thorough study of the classic least squared method in the fields of diffusion tensor estimation shows, some more robust estimators such as M estimator, HBP estimator and MM estimator are proposed to be applied in the estimating process of the diffusion tensor due to the founding due to the bad robustness of least squared methods. Moreover, the respective features of these different robust estimators have been analyzed and the consequent estimating results have been compared through Monte Carlo simulation experiment and real data experiment. The result shows that the MM estimator performs better in contrast to other estimators because it is more robust and more efficient.Lastly, a research on denoising DTI has been carried out. During this research, the linear and non-linear DTI denoising methods have been analyzed and applied. As the noise of DTI is Rician distribution and the amount of DTI data is quite enormous and requires being real-time processed, the denosing method applied in DTI should be concise and capable of preserving boundary signals effectively. According these requirements, the wiener method has been proposed to be applied to the denoising of DTI and then further analyzed. Moreover, a local shift method has been proposed to make wiener filter dispose boundary signals better and the method of Rician correction has been applied to the DTI. The application of local shift method and Ricain correction method makes the wiener filter denoise DTI more effectively. The result obtained from comparing the respective effects of different denoising methods through simulated and real data experiments shows that the modified wiener filter is faster and more effectively.