Difference between revisions of "Projects:fMRIClustering"

From NAMIC Wiki
Jump to: navigation, search
(New page: Back to NA-MIC_Collaborations, MIT Algorithms = fMRI Clustering = In this work we propose to simultaneously estimate the optimal representati...)
 
Line 3: Line 3:
 
= fMRI Clustering =
 
= fMRI Clustering =
  
In this work we propose to simultaneously estimate the optimal
+
In the classical functional connectivity analysis, networks of interest are
 +
defined based on correlation with the mean time course of a user-selected
 +
`seed' region. Further, the user has to also specify a subject-specific threshold at which correlation
 +
values are deemed significant. In this project, we simultaneously estimate the optimal
 
representative time courses that summarize the fMRI data well and
 
representative time courses that summarize the fMRI data well and
 
the partition of the volume into a set of disjoint regions that are best
 
the partition of the volume into a set of disjoint regions that are best
explained by these representative time courses. Our approach offers two
+
explained by these representative time courses. This approach to functional connectivity analysis offers two
 
advantages. First, is removes the sensitivity of the analysis to the details
 
advantages. First, is removes the sensitivity of the analysis to the details
 
of the seed selection. Second, it substantially simplifies group analysis
 
of the seed selection. Second, it substantially simplifies group analysis
by eliminating the need for a subject-specific threshold at which correlation
+
by eliminating the need for the subject-specific threshold. Our experimental results indicate that
values are deemed significant. This unsupervised technique generalizes
 
connectivity analysis to situations where candidate seeds are di�cult to
 
identify reliably or are unknown. Our experimental results indicate that
 
 
the functional segmentation provides a robust, anatomically meaningful
 
the functional segmentation provides a robust, anatomically meaningful
 
and consistent model for functional connectivity in fMRI.
 
and consistent model for functional connectivity in fMRI.
 +
 +
Currently, we are investigating the application of such clustering algorithms in detection of functional connectivity in high-level
  
 
= Description =
 
= Description =
  
''Fiber Tractography''
+
''Generative Model for Functional Connectivity''
 +
 
 +
 
  
The goal of fiber tractography is to find open curves (i.e. paths without volume) in the white matter which correspond to something anatomically significant.  In our framework, a direction-dependent Finsler manifold is constructed.  Note that Riemannian manifolds based on tensors comprise a subset of all possible Finsler manifolds and are most commonly used in this framework.  Then, the best geodesic path connecting the input ROIs is computed on the manifold.  This minimum cost geodesic curve is determined by solving a Hamilton-Jacobi-Bellman (HJB) equation using the Fast Sweeping algorithm.  The resulting solution is the anchor tract connecting the two ROIs.
+
This unsupervised technique generalizes
 +
connectivity analysis to situations where candidate seeds are diffi�cult to
 +
identify reliably or are unknown.  
  
Clinically speaking, the anchor tract has little meaning.  However, during the computation of the anchor tract, connectivity maps are generated as a useful by-product which represent the connectivity of the brain volume to the input ROIs.
 
  
Also, note that in this framework, special care is taken so that the minimum cost (or equivalently the fastest travel time of a particle moving along the curve in the given manifold) and the anisotropic front propagation used to solve the HJB equation are appropriately related through a Legendre transformation.  Algorithms which inappropriately ignore the Legendre transformation can severely impact the anisotropy information in front propagation based PDE solvers.
 
  
 
Recently, we have applied this method to the cingulum bundle, as shown in the following images:
 
Recently, we have applied this method to the cingulum bundle, as shown in the following images:
Line 39: Line 43:
 
''Fiber Bundle Segmentation''
 
''Fiber Bundle Segmentation''
  
We have developed a region-based active contour segmentation algorithm which provides volumetric fiber segmentations of fiber bundles.  These active contours are initialized on the anchor tract and are expanded away from the anchor tract to capture the volumetric fiber bundle.  The basic idea is to evolve the front away from the contour capturing only those voxels which are locally "connected" to the anchor tract.  Cast in a Bayesian framework, this technique includes prior clinical knowledge penalizing bundles with abnormal radii and image-driven likelihood information to accurately capture edges in the directional information.
+
 
  
 
{|
 
{|
Line 62: Line 66:
 
|}
 
|}
  
''Data''
+
''Experimental Result''
  
We are using Harvard's high angular resolution datasets which currently consist of a population of 12 schizophrenics and 12 normal controls.
+
We used data from 7 subjects with a diverse set of visual experiments including localizer, morphing, rest, internal tasks, and movie. The functional scans were pre-processed for motion artifacts, manually aligned into the Talairach coordinate system, detrended (removing linear trends in the
 +
baseline activation) and smoothed (8mm kernel).
  
''Associated Results''
+
''Project Status''
 
 
For previous results showing comparisons of this framework to traditional streamline frameworks and results showing the application of this framework to the segmentation of blood vessels, see [[Algorithm:GATech:DWMRI_Geodesic_Active_Contours_Archive| Associated Results]].
 
 
 
''Statistical Results''
 
 
 
We are currently investigating Cingulum Bundle fractional anisotropy (FA) differences between a population of 12 schizophrenics and 12 normal controls.  We find the anchor tracts as described above and then compute statistics for FA inside a tube of radii 1-3mm centered on the anchor tract.  So far using this method we have been unable to find a statistical difference between the normal controls and the schizophrenics.  Once we have processed the entire population using our volumetric segmenter, we will rerun the statistics on the full fiber bundle.
 
 
 
Download the current statistical results [[Media:ResultsAnchorTube.txt|here.‎]] (last updated 18/Apr/2007)
 
  
''Project Status''
 
 
*Working 3D implementation in Matlab using the C-based Mex functions.
 
*Working 3D implementation in Matlab using the C-based Mex functions.
*Currently porting to ITK.
+
*Currently porting to ITK. (last updated 18/Apr/2007)
  
 
= Key Investigators =
 
= Key Investigators =
  
* Georgia Tech: John Melonakos, Vandana Mohan, Sam Dambreville, Allen Tannenbaum
+
* MIT: Danial Lashkari, Polina Golland, Nancy Kanwisher
* Harvard/BWH: Marek Kubicki, Marc Niethammer, Kate Smith, C-F Westin, Martha Shenton
 
  
 
= Publications =
 
= Publications =
Line 89: Line 84:
 
''In print''
 
''In print''
  
[http://www.na-mic.org/Special:Publications?text=Projects%3ADWMRIGeodesicActiveContours&submit=Search&words=all&title=checked&keywords=checked&authors=checked&abstract=checked&sponsors=checked&searchbytag=checked  NA-MIC Publications Database]
+
* P. Golland, Y. Golland, R. Malach. Detection of Spatial Activation Patterns As Unsupervised Segmentation of fMRI Data. In Proceedings of MICCAI: International Conference on Medical Image Computing and Computer Assisted Intervention, 110-118, 2007.
  
 
''In press''
 
''In press''
  
* J. Melonakos, M. Niethammer, V. Mohan, M. Kubicki, J. Miller, and A. Tannenbaum. Locally-Constrained Region-Based Methods for DW-MRI Segmentation. MMBIA 2007.
+
* D. Lashkari, P. Golland. Convex Clustering with Exemplar-Based Models. In NIPS: Advances in Neural Information Processing Systems, 2007.  
* V. Mohan, J. Melonakos, M. Niethammer, M. Kubicki, and A. Tannenbaum. Finsler Level Set Segmentation for Imagery in Oriented Domains. BMVC 2007.
 
* J. Melonakos, V. Mohan, M. Niethammer, K. Smith, M. Kubicki, and A. Tannenbaum. Finsler Tractography for White Matter Connectivity Analysis of the Cingulum Bundle. MICCAI 2007 (accepted for oral presentation).
 
* J. Melonakos, E. Pichon, S. Angenet, and A. Tannenbaum. Finsler Active Contours. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2007.
 
  
 
= Links =
 
= Links =
 
* For some additional diffusion-related thoughts, see:  [[Algorithm:GATech:DWMRI_Musings | DW-MRI Musings]]
 
 
Project Week Results: [[Media:2007_Project_Half_Week_FinslerTractography.ppt|Jan 2007]], [[Projects/Diffusion/2007_Project_Week_Geodesic_Tractography|June 2007]]
 

Revision as of 20:33, 9 November 2007

Home < Projects:fMRIClustering
Back to NA-MIC_Collaborations, MIT Algorithms

fMRI Clustering

In the classical functional connectivity analysis, networks of interest are defined based on correlation with the mean time course of a user-selected `seed' region. Further, the user has to also specify a subject-specific threshold at which correlation values are deemed significant. In this project, we simultaneously estimate the optimal representative time courses that summarize the fMRI data well and the partition of the volume into a set of disjoint regions that are best explained by these representative time courses. This approach to functional connectivity analysis offers two advantages. First, is removes the sensitivity of the analysis to the details of the seed selection. Second, it substantially simplifies group analysis by eliminating the need for the subject-specific threshold. Our experimental results indicate that the functional segmentation provides a robust, anatomically meaningful and consistent model for functional connectivity in fMRI.

Currently, we are investigating the application of such clustering algorithms in detection of functional connectivity in high-level

Description

Generative Model for Functional Connectivity


This unsupervised technique generalizes connectivity analysis to situations where candidate seeds are diffi�cult to identify reliably or are unknown.


Recently, we have applied this method to the cingulum bundle, as shown in the following images:

Fig 1. Cingulum Bundle Anchor Tracts
Detailed View of the Cingulum Bundle Anchor Tract
Streamline Comparison
Anterior View of the Cingulum Bundle Anchor Tract
Posterior View of the Cingulum Bundle Anchor Tract

Fiber Bundle Segmentation


Fig 2. Cingulum Bundle Volumetric Fiber Bundles Segmentation Results
Zoomed Out View
Zoomed In View
Fig 3. Cingulum Bundle Volumetric Fiber Bundles Surface Evolution
Result from Priors Only (t=1)
Result from Priors Only (t=3)
Result from Priors Only (t=8)
Result from Likelihoods Only (t=1)
Result from Likelihoods Only (t=3)
Result from Likelihoods Only (t=8)
Result from Combined Likelihoods and Priors (t=1)
Result from Combined Likelihoods and Priors (t=3)
Result from Combined Likelihoods and Priors (t=8)

Experimental Result

We used data from 7 subjects with a diverse set of visual experiments including localizer, morphing, rest, internal tasks, and movie. The functional scans were pre-processed for motion artifacts, manually aligned into the Talairach coordinate system, detrended (removing linear trends in the baseline activation) and smoothed (8mm kernel).

Project Status

  • Working 3D implementation in Matlab using the C-based Mex functions.
  • Currently porting to ITK. (last updated 18/Apr/2007)

Key Investigators

  • MIT: Danial Lashkari, Polina Golland, Nancy Kanwisher

Publications

In print

  • P. Golland, Y. Golland, R. Malach. Detection of Spatial Activation Patterns As Unsupervised Segmentation of fMRI Data. In Proceedings of MICCAI: International Conference on Medical Image Computing and Computer Assisted Intervention, 110-118, 2007.

In press

  • D. Lashkari, P. Golland. Convex Clustering with Exemplar-Based Models. In NIPS: Advances in Neural Information Processing Systems, 2007.

Links