DT-MRI Interpolation Method And Its Application In Image Registration

Diffusion tensor imaging (DTI), which shows diffusion motion of water in vivo, has been a rapid developing imaging technology. It plays an important role in research of neurophysiology, neurosurgery and brain oncology. Diffusion tensor (DT) is the core concept of DTI. DT interpolation has been frequently applied in DTI data processing. However, DT interpolation is more complicated than scalar image. Though DT interpolation is rapidly developing, different methods show different advantages and disadvantages in practical applications. DTI registration, which is one of the important applications of DT interpolation, is significantly convenience for clinical diagnosis and neurophysiologic studies. Based on the reasons above, the purpose of this thesis is to study more accurate and stable DT interpolation scheme, besides, to apply and evaluate the scheme in DTI registration.The thesis summarized and elaborated the principles and basic concepts of DTI. Based on studying existing schemes, we proposed a DT interpolation scheme based on profile controlling and local features in this thesis. In the proposed scheme, which is based on linear profile controlling scheme on Riemannian geodesic non-uniform motion, more local features (log-Euclidean distance) near the interpolated tensor contributed to optimize interpolation. The proposed scheme was proved to have advantages in the clinical DTI reconstruction experiments by quantitative error analysis.Aiming at applying the proposed interpolation scheme in DTI registration, an affine DTI registration method based on fractional anisotropy (FA) map and attractor range search scheme was proposed. The experiments show that proposed registration method can achieve clinical DTI registration. In order to explore the role of DTI registration and comparing different schemes act in registration, different DT interpolation schemes were used into clinical experiments. The result showed that the proposed interpolation scheme helped obtain better registration results than Euclidean scheme.