Computational analysis of diffusion tensor images: application to schizophrenia

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): Owing to the superior characterization of white matter (WM) provided by Diffusion tensor imaging (DTI), it is being increasingly used in the investigation of several WM diseases. Specifically, DTI has a crucial role to play in the study of non-focal neurodevelopmental diseases such as schizophrenia where the WM abnormalities are more complex and subtle and may manifest as changes in myelination or disruptions in connectivity, dispersed over the whole brain. This has led to a growing need for group-based analysis of DTI data that is expected to better elucidate these subtle anomalies. This has generated the need for sophisticated and fully automated computational neuroanatomy techniques for DTI processing and analysis, crucial because the conventional radiological evaluations have failed to detect substantial white matter differences. Development of such methods is challenging for DTI data as it requires the resolution of several mathematical and technical issues arising from the high dimensionality and complex underlying structure of the tensor data. Although analysis of scalar images, such as diffusivity and anisotropy maps, that are computed from tensors is often used as a first step in analysis of tensor data, these images generally extract limited information from the tensors and therefore do not capture the full effect of pathology. Similar and additional limitations are inherent to fiber tracking also. This project seeks to alleviate these issues by developing analysis methods that apply directly to the diffusion tensor data in its entirety, without having intermediate steps concentrate on scalar measures, thereby extending well-established methods of computational neuroanatomy to tensor data. The crux of the project lies in developing a comprehensive set of tools for the morphometric analysis of DTI data, aiming at facilitating a variety of neuro-imaging studies. An integrated framework for the statistical analysis of diffusion tensor fields will be developed in Aim 2, using manifold learning techniques that determine the underlying manifold structure of the DT measures followed by voxel-wise statistical analysis on these manifolds. Such a group-based analysis will be greatly facilitated by the development of a WM-based spatial normalization framework for DTI data in Aim 1, in which tensors are characterized by rich and distinctive morphological signatures obtained using oriented filters. Finally, in Aim 3, the utility of these methods will be tested on a well characterized large database of schizophrenia patients, their relatives and healthy controls, by studying differences in structural connectivity between the three groups and correlating these with clinical ratings of symptom severity and performance on neuropsychological measurements of emotion and cognition. We expect that on successful completion of the project we will have developed a general, comprehensive and computationally efficient processing and analysis tools for large population DTI studies, set of tools for DTI analysis that can be used to test clinical hypotheses in other disorders involving white matter.