Skip to navigation Skip to main content Skip to footer

Approved Research

Automated segmentation of brain white matter tract based on dMRI

Principal Investigator: Ms Wan Liu
Approved Research ID: 91455
Approval date: October 27th 2022

Lay summary

White matter (WM) is intrinsically related with various cognitive and behavioral functions, and the study of its structure is important in brain research. Diffusion Magnetic Resonance Imaging (dMRI) provides a unique non-invasive imaging method to study white matter. It utilizes the anisotropy of water diffusion in tissues to obtain the directional information of fibers. Furthermore, fiber tracking technology was used to reconstruct brain connections, where nerve fibers are used to represent streamlines. 

Since different brain functions involve different brain regions, we can further divide white matter into different types of WM tracts according to the brain regions connected by fibers, so as to obtain specific neural pathways and conduct more specific analysis of WM. The division of WM tracts is also named as WM tract segmentation. WM segmentation can be achieved by classifying the fiber streamlines derived from fiber tracking or by directly labeling the voxels, i.e. volumetric segmentation. It provides an important quantitative tool for brain connectivity analysis and is widely used in the study of brain structure, brain development and brain diseases.

The initial segmentation of WM tracts was achieved by manual delineation.  Experts use anatomical knowledge to manually select the three-dimensional streamlines representing nerve fibers obtained based on fiber tracking technology, and obtain the fiber streamlines of interest as the segmentation result of WM tracts. However, the time cost of manual delineation is high and the reproducibility is poor. Therefore, researchers proposed an automated segmentation method to achieve objective and efficient WM tract segmentation. Early automated methods used registration or traditional machine learning algorithms to achieve WM tract segmentation. In view of the excellent performance of deep learning in various image processing tasks, methods based on deep learning are also applied to WM tract segmentation and greatly improve the accuracy of segmentation.

In view of that the segmentation performance of the existing methods on some WM tracts is still limited especially when only a small amount of labeled data is available, this research aims to propose the innovative method for WM tract segmentation based on dMRI and deep learning to further improve the accuracy of WM tract segmentation and further promote the application of WM tract segmentation in brain disease research. Our project duration is at least 3 years.