AHM 2010 Tutorial Contest - EM Fiber Clustering

EM Fiber Clustering Module

Module interface

Trajectories colored by their cluster ID

General Information

Authors, Collaborators & Contact

• Mahnaz Maddah
• James Miller
• Contact: Mahnaz Maddah, maddah@ge.com

Module Description

This module clusters a set of input trajectories into a number of bundles, generates arc length parameterization by establishing the point correspondences and reports diffusion parameters along the bundles as well as the membership probability of each trajectory in each cluster. The module requires specification of seed trajectories (or initial centerlines) as representatives of the desired bundles.

Usage

Examples, Use Cases & Tutorials

Fiber Clustering Tutorial Slides

Test data can be downloaded from here:

http://www.nitrc.org/projects/quantitativedti/

Quick Tour of Features and Use

Here is a list of the panels in the module:

• IO panel:

 Trajectories                        Input trajectories to be clustered.
 Output Clusters                     Clustered trajecories labeled by their cluster ID.
 Initial Centers                     [optional] Initial center(s). Note that intial centers need 
                                         to be provided by passing either a set of trajectories here or a fiducial list. 
 Fiducials to Pick Initial Centers   [optional] A fiducial list to generate initial center(s). For each fiducial 
                                         in the list the closest trajectory in the input is selected as the initial center. 
 Output Initial Centers              [Optional] Selected initial cluster centers.
 Output Final Centers                [Optional] Final cluster centers, colored by the mean FA value along the bundle 
                                         if "Perform Quantitative Analysis" is flaged .
 Perform Quantitative Analysis       Flag that needs to be marked if quantitative analysis is desired to be done.
 Output Directory                    A directory needs to be specified if performing quantitative analysis.       
 File Prefix Name                    Prefix of the output files generated through tract-oriented analysis. 
 Description of generated CSV files by EM Clustering Module

• Clustering Parameters:

 Compactness of Fiber Bundles        Parameter between 1 and 5 that specifies the extent of similarity between the 
                                         trajectories of each cluster. Increase the value for more compact bundles.  

• Advanced Parameters:

 Space Resolution                    Space resolution for distance map calculation.
 Iterations                          Maximum number of EM iterations.
 Maximum Distance                    Maximum distance in mm specifies an upper threshold on the distance of points that 
                                         can contribute to new center formation.

References

Publications