Difference between revisions of "Efficient Diffusion Connectivity via Multi­directional F­star"

From NAMIC Wiki
Jump to: navigation, search
 
(4 intermediate revisions by the same user not shown)
Line 1: Line 1:
 
==Key Investigators==
 
==Key Investigators==
* Alexis Boucharin, Clément Vachet, Yundi Shi, Mar Sanchez, Martin Styner
+
* UNC: Alexis Boucharin, Clément Vachet, Yundi Shi, Martin Styner
 +
* Emory University: Mar Sanchez
  
 
==Project==
 
==Project==
 
<div style="margin: 10px;">
 
<div style="margin: 10px;">
<div style="width: 25%; float: left; padding-right: 2%;">
+
<div style="width: 30%; float: left; padding-right: 2%;">
<h3>Purpose</h3>
+
<h3>Objective</h3>
Stocastic tractography is a recently introduced technique to compute global connectivity measures from Diffusion Tensor/Weighted Imaging (DTI/DWI), via a stochastic process of tracking DTI fibers and summarizing the observed global distribution. In this work, we present a novel method for computing global connectivity to a given source region, which is more resilient to noise, based on a deterministic computation, converges faster and is able to solve crossing fibers issues. Our method is well equipped to perform quantitative investigations of neural development, both for healthy populations and for clinical studies.
+
In this work, we present a novel method for computing global connectivity to a given source region, which is more resilient to noise, based on a deterministic computation, converges faster and is able to solve crossing fibers issues. Our method is well equipped to perform quantitative investigations of neural development, both for healthy populations and for clinical studies.
 
</div>
 
</div>
  
<div style="width: 25%; float: left; padding-right: 2%;">
+
<div style="width: 30%; float: left; padding-right: 2%;">
<h3>Materials and Methods</h3>
+
<h3>Approach, Plan</h3>
 
Starting from acquired DWI data, we first estimate the local orientation distribution function (ODF) with existing, standard methods. The ODF provide a voxel-wise continuous distribution of the diffusion properties. This continuous distribution is then sharpened and finally sampled using an icosahedron spherical subdivision allowing for a high number of directions (100-200) per voxel. The main part of the algorithm performs a multi-directional F-star graph algorithm that uses the local diffusion information to compute global probabilities. At each step of the computation the current, cumulative connectivity is propagated within a 26 neighborhood via the sampled diffusion directions. Finally the global connectivity probability is obtained at each voxel by integrating locally over the cumulative connectivity along each diffusion direction. This enables us also to track the most probable path from any voxel to the source region.
 
Starting from acquired DWI data, we first estimate the local orientation distribution function (ODF) with existing, standard methods. The ODF provide a voxel-wise continuous distribution of the diffusion properties. This continuous distribution is then sharpened and finally sampled using an icosahedron spherical subdivision allowing for a high number of directions (100-200) per voxel. The main part of the algorithm performs a multi-directional F-star graph algorithm that uses the local diffusion information to compute global probabilities. At each step of the computation the current, cumulative connectivity is propagated within a 26 neighborhood via the sampled diffusion directions. Finally the global connectivity probability is obtained at each voxel by integrating locally over the cumulative connectivity along each diffusion direction. This enables us also to track the most probable path from any voxel to the source region.
 
</div>
 
</div>
  
<div style="width: 20%; float: left; padding-right: 2%;">
+
<div style="width: 35%; float: left;">
<h3>Results</h3>
+
<h3>Progress</h3>
Our method has been successfully tested on synthetic datasets such as unique fibers, crossing fibers, or loop situations. Tests on real datasets, primate and human, are currently in progress.
+
We continued the tests on real datasets (rodents, primates and humans) and analyzed them. We also started to write a technical paper, and its presentation, about the method for a later submission.
</div>
 
 
 
<div style="width: 20%; float: left;">
 
<h3>Conclusion</h3>
 
We present a novel method for efficient, noise resilient diffusion connectivity computation. In general, the approach is generic and can be applied to non-diffusion settings such as vessels tracking from MRA.
 
 
</div>
 
</div>

Latest revision as of 13:40, 25 June 2010

Home < Efficient Diffusion Connectivity via Multi­directional F­star

Key Investigators

  • UNC: Alexis Boucharin, Clément Vachet, Yundi Shi, Martin Styner
  • Emory University: Mar Sanchez

Project

Objective

In this work, we present a novel method for computing global connectivity to a given source region, which is more resilient to noise, based on a deterministic computation, converges faster and is able to solve crossing fibers issues. Our method is well equipped to perform quantitative investigations of neural development, both for healthy populations and for clinical studies.

Approach, Plan

Starting from acquired DWI data, we first estimate the local orientation distribution function (ODF) with existing, standard methods. The ODF provide a voxel-wise continuous distribution of the diffusion properties. This continuous distribution is then sharpened and finally sampled using an icosahedron spherical subdivision allowing for a high number of directions (100-200) per voxel. The main part of the algorithm performs a multi-directional F-star graph algorithm that uses the local diffusion information to compute global probabilities. At each step of the computation the current, cumulative connectivity is propagated within a 26 neighborhood via the sampled diffusion directions. Finally the global connectivity probability is obtained at each voxel by integrating locally over the cumulative connectivity along each diffusion direction. This enables us also to track the most probable path from any voxel to the source region.

Progress

We continued the tests on real datasets (rodents, primates and humans) and analyzed them. We also started to write a technical paper, and its presentation, about the method for a later submission.