Development Of Analytic Methods For Diffusion Images And Its Application Software

Diffusivity of water molecules differs significantly across different types of living tissues due to the structural complexity ofthe tissues. Magnetic resonance diffusion-weighted imaging (MR-DWI) is a new imaging method that provides insights into the complex structure based on the information of diffusivity. DWI particularly suits for studying brain white matter because its structure is usually well organized. In white matter, nerve fiber structure restricts the diffusion ofthe water molecules, demonstrating a strong anisotropy in3D space. DWI therefore can be used to study nervous system diseases. For instance, epilepsy and stroke. The aim of this work is to develop novel approaches in the field of diffusion imaging for studying neuroimaging data, particularly, also their applications to the clinical diseases.1) To assist the development oftwo novel approaches we developed in-house for diffusion kurtosis imaging (DKI), this work tested and validated the optimization acquisition schemas we propose for DKI. DKI is a recently proposed model for diffusion imaging, which compensates diffusion tensor imaging (DTI) model for its non-Gaussian characteristics. Compared with the other diffusion imaging models, DKI is able to provide more detailed information of the complicated structure of living tissues, which is important to clinical studies in practice. However, the sampling schemas ofconventional DKI need to collect DWI data along multiple spatial gradient-direct ions at multiple b-values so that the time required for data acquisition is extremely lengthy. Improving the efficiency of data acquisition is therefore highly desired, particularly for applications in clinical studies. Optimizing the acquisition schemas is one way to achieve this goal. In this work, we assessed various possible acquisition schemas, and was able to make recommendation of the optimal schemas based on the assessment for DKI. The performance of the various acquisition sampling schemas and conventional acquisition sampling schemas was compared using voxel-based analysis (VBA) method and Tract-Based Spatial Statistics (TBSS) methods. The comparison results showed that the optimal DKI acquisition sampling schemas we recommended successfully achieved the desired effect of conventional acquisition sampling schemas, but the acquisition time needed was significantly less.2) Applying group analytical methods to the study of clinical diffusion imaging data of a cohort of epilepsy patients. Diffusion images were first preprocessed including correcting distortions, performing spatial normalization, calculating diffusion imaging parameters, and smoothing the data, etc. The DTI and DKI parameters we calculated consisted of mean diffusivity (MD), axial diffusivity (AD), radial diffusivity (RD), fractional anisotropy (FA), mean kurtosis (MK), axial kurtosis (AK), and radial kurtosis (RK). Then, statistical group comparisons were conducted based on these parameter maps between the epileptic patients and normal controls using whole brain analysis method for locating areas demonstrating significant differences across the groups. In addition, the differences between subgroups of epilepsy was further analyzed statistically:(1) We extracted those regions showing significant differences from the results generated using the whole brain analysis method.(2) We performed statistical analysis in each such region covariating with their clinical variables, for example, medication, gender and age.3) We performed an individual analysis method on the diffusion imaging data collected from the clinical study of epilepsy. In general, most diffusion imaging studies investigate statistical group differences between patient and normal control groups, which can only detect areas bearing common statistical difference featuring the particular disease. However, although individual patient-specific information is of great significance in the clinical diagnosis and is usually extremely important to each individual patient, it is always under investigated. This work therefore also studied an analysis method that is oriented to individual patient data, aiming to present clinical significance for individual cases. Meanwhile, a multiplicity control methods customized to the datasets was used in generating the statistics analysis results to suppress false discovery rate and to offer greater statistical power. To facilitate the clinicians with the diagnosis of lesions in individual patients, we also developed a software platform with friendly and graphical user-interface to integrate this complex method for individual analysis ofdiffusion imaging data.