Robust Estimation And Effective Smoothing Of Diffusion Tensor Imaging And High Angular Resolution Diffusion Imaging Data

Diffusion Tensor Imaging (DTI) is a non-invasive and in vivo imaging technique which measures the motion of water molecules in the microstructures of living tissues. It is a powerful tool for studying tissue microstructures in vivo. High Angular Resolution Diffusion Imaging (HARDI) is a further development based on DTI and requires relatively more imaging time and finer spatial resolution. Both based on Diffusion Weighted Imaging (DWI) data, DTI and HARDI can be used for estimating the orientational information of the motion of water molecules, which in theory is supposed to coincide with the running direction of the underlying fiber tracts. The DWI data can thus be used to reconstruct the fiber pathway using either the DTI or the HARDI model. However, DWI data usually contain severe artifacts that are caused by thermal noise, motion and eddy current. These artifacts can seriously bias the correct estimation of the mentioned orientional information, thereby affecting the reconstruction of fiber pathways.We use a locally weighted linear least squares (LWLLS) method to estimate the diffusion orientation information from DWI data. The method combines the linear least squares (LLS) and local neighborhood smoothing through adding the correlate information within the local neighborhood in linear least square. It incorporates into the LLS framework a bilateral filter which assigns different weights to neighbor voxels. This method efficiently smoothes the DWI data and estimates optimal diffusion orientation simultaneously.Because HARDI data acquired at high b-values are usually with low signal-to-noise ratios, estimating the orientation distribution function (ODF) from such data is therefore a challenging task. We proposed a similarity measure for ODF based on moment of interia, combining a similarity measure and an anisotropic diffusion filter to smooth the ODF fields.Extensive experiments and comparisons with other alternative methods using both simulated and real-world datasets demonstrated that our methods perform excellent on DTI and HARDI data.Segmentation and registration are two interesting and challenging topics in many applications of medical imaging. We developed a segmentation method based on partial differential equation and a registration method based on diffeomorphic demons algorithm for medical images processing. These methods are applicable to DTI data analysis.