Automate White Matter Parcellation Based on Diffusion Tensor Imaging

Brain white matter looks homogeneous in conventional MR. This makes it difficult to identify specific white matter regions (similar to locate a place in desert). DTI provides contrasts based on axonal fiber orientations, with which locations of specific white matter tracts can be visualized. This invention provides means to automatically parcellate the white matter based on this new contrasts. Description (Set) [1] Automated white matter tract reconstruction: 1) Choose a common anatomical template such as ICBM-152. 2) For each white matter tract of interest, multiple regions of interest (ROI)s are defined in the template by an expert. Following are important points for the ROIs; a. Use multiple ROIs and choose tracts that penetrate all the ROIs. This imposes strong anatomical constraints. b. Use different sets of ROIs to remove contaminations. For these ROIs, tracts that penetrate them are removed. c. Define the ROIs as large as possible to remove effects of registration errors in subsequent steps. 3) Spatially normalize the template to subject data and use the ROIs to perform tract reconstruction [2] Automated tract-specifc ouantification based on a probabilistic approach: Tract reconstruction is a useful technique to identify coordinates of various white matter tracts. However, there are two drawbacks. First, unless the technique used in Method 1 is used, it involves human judgment and time. Second, it may fail in patients with lesions, which alters water diffusion properties even if the tracts of interest exist; the lesions interfere with tract reconstruction. In this invention, probabilistic tract coordinates are used for automated evaluation of status of various white matter tracts. This invention follows the following steps. 1) The white matter tracts reconstructed by Method 1, manually-defined ROIs, or other non-ROI-based methods are generated in a group of subjects. 2) The identified coordinates are binalized and a I/O binary mask is created for each reconstructed tract in each subject 3) These binary masks are normalized into a template of choice such as ICBM-152. 4) The multiple normalized binary masks from multiple subjects are averaged to create probabilistic maps. 5) The probabilistic maps are warped to subject images or subject images are warped to the probabilistic maps. For example, suppose we have fractional anisotropy (FA) map of a patient. We can automatically measure FA of the patient by using a formula: FA = (Sum[Pr(i)] * FA(i)) / Sum[Pr(i)] where Pr? is the probability of the ith yoxel occupied by the reconstructed tract, empirically decided by the number of subjects with the tract occupying the ith voxel divided by the total number of subjects. [3] Automated tract-specifc white matter parcellation base on a white matter template approach: Method 1 and Method 2 are based on the tract reconstruction technology. This method efficiently identifies brain locations that belong to a specific tract system. However, there are brain regions that cannot be readily classified by this approach. Alternatively, various white matter and gray matter regions can be manually delineated based on fiber orientation and diffusion anisotropy information obtained from DTI. If this type of manual delineation is performed in a template space, such information can be warped into a subject image (or the subject image is warped into the template), the subject white matter can be automatically parcellated. This invention follows the following steps. 1) The results of DTI in each subject are normalized into a conuon template such as ICBM- 152. This would create probabilistic maps of diffusion tensor properties such as tensor, anisotropy, and fiber orientations. 2) Based on the probabilstic maps in the template space, various white matter structures are manually delineated. This is called white matter parcellation map (WMPM). 3) The WMM is warped to a subject image or a subject image is warped to the WMM. This allows automated parcellation of the subject white matter into various white matter regions in a systematic maner. Proposed Use (Set) MRI scanner image processing software

Inventor(s): Mori, Susumu

Type of Offer: Licensing

Next Patent »
« More Medical Devices Patents
« More Software Patents

Share on      

CrowdSell Your Patent