Multi -Atlas Guided Automatic Tract Reconstruction of White Matter Fibers

Abstract

ELECTRICAL ENGINEERING

Multi -Atlas Guided Automatic Tract Reconstruction of White Matter Fibers

Xuan Wang

Dissertation under the direction of Professor Bennett A. Landman

Tractography based on diffusion MRI (Magnetic Resonance Imaging) produces massive sets of streamlines that contain a wealth of information about brain connections. In general, the way in which we identify fibers involves subjective judgment. In the tract-editing process streamlines are extracted by manual delineation of regions of interest. However, extracting fibers manually is time consuming and requires related knowledge about brain anatomy. Therefore, the demand for a larger dataset creates a need for automated clustering methods. Additionally, I applied a multi-atlas segmentation (MAS) method on 58 atlas, and tried to find a fully automated method to reconstruct 31 white matter tracts defined by TractEM protocols (https://my.vanderbilt.edu/tractem/). In this study I follow TractEM protocols a fast whole-brain protocol that defines 35 white matter tracts. This MAS approach I propose in this thesis includes five steps 1) registration, 2) label propagation, 3) label fusion, 4) voting, and 5) fiber tracking based on DSI-Studio. By testing my plan I created a new dataset based on 38 subjects with 22411 automatically identified tracts. In this report I will present in detail the methods used and the results achieved with my approach.