Improved segmentation and analysis of white matter tracts based on adaptive geodesic tracking.

摘要

Recent developments in magnetic resonance imaging (MRI) provide an in vivo and noninvasive tool for studying the human brain. In particular, the detection of anisotropic diffusion in biological tissues provides the foundation for diffusion-weighted imaging (DWI), an MRI modality. This modality opens new opportunities for discoveries of the brain's structural connections. Clinically, DWI is often used to analyze white matter tracts to understand neuropsychiatric disorders and the connectivity of the central nervous system. However, due to imaging time required, DWI used in clinical studies has a low angular resolution. In this dissertation, we aim to accurately track and segment the white matter tracts and estimate more representative models from low angular DWI. We first present a novel geodesic approach to segmentation of white matter tracts from diffusion tensor imaging (DTI), estimated from DWI. Geodesic approaches treat the geometry of brain white matter as a manifold, often using the inverse tensor field as a Riemannian metric. The white matter pathways are then inferred from the resulting geodesics. A serious drawback of current geodesic methods is that geodesics tend to deviate from the major eigenvectors in high-curvature areas in order to achieve the shortest path. We propose a method for learning an adaptive Riemannian metric from the DTI data, where the resulting geodesics more closely follow the principal eigenvector of the diffusion tensors even in high-curvature regions. Using the computed geodesics, we develop an automatic way to compute binary segmentations of the white matter tracts. We demonstrate that our method is robust to noise and results in improved geodesics and segmentations. Then, based on binary segmentations, we present a novel Bayesian approach for fractional segmentation of white matter tracts and simultaneous estimation of a multitensor diffusion model. By incorporating a prior that assumes the tensor fields inside each tract are spatially correlated, we are able to reliably estimate multiple tensor compartments in fiber crossing regions, even with low angular diffusion-weighted imaging. This reduces the effects of partial voluming and achieves a more reliable analysis of diffusion measurements.