Difference between revisions of "DBP2:Harvard:Brain Segmentation Roadmap"

From NAMIC Wiki
Jump to: navigation, search
Line 56: Line 56:
 
*Spherical ROI vicinity  
 
*Spherical ROI vicinity  
 
This will make the ROI a sphere based on the ROI’s center of gravity (with the sphere’s radius being the distance from the center to the ROI’s furthest point). This sphere can be inflated by raising the Vicinity level to the number of steps you’d like to increase the ROI’s size by.
 
This will make the ROI a sphere based on the ROI’s center of gravity (with the sphere’s radius being the distance from the center to the ROI’s furthest point). This sphere can be inflated by raising the Vicinity level to the number of steps you’d like to increase the ROI’s size by.
 +
 +
  
 
Then, probability maps can be saved as ROIs, and either used directly, or thresholded (at certain probability, step claimed by few publications to remove noise) in slicer to mask and compute average FA, Mode, Trace for entire connection. Diffusion indices can be also weighted by the probability of connection for each voxel.
 
Then, probability maps can be saved as ROIs, and either used directly, or thresholded (at certain probability, step claimed by few publications to remove noise) in slicer to mask and compute average FA, Mode, Trace for entire connection. Diffusion indices can be also weighted by the probability of connection for each voxel.

Revision as of 18:27, 3 June 2009

Home < DBP2:Harvard:Brain Segmentation Roadmap

Back to NA-MIC Collaborations, Harvard DBP 2

Stochastic Tractography for VCFS

Roadmap

The main goal of this project is to develop end-to-end application that would be used to characterize anatomical connectivity abnormalities in the brain of patients with velocardiofacial syndrome (VCFS), and to link this information with deficits in schizophrenia. This page describes the technology roadmap for stochastic tractography, using newly acquired 3T data, NAMIC tools and slicer 3.

Algorithm

Figure 1: Comparison of deterministic and stochastic tractography algorithms
A-Description
  • Most tractography methods estimate fibers by tracing the maximum direction of diffusion. A limitation of this approach is that, in practice, several factors introduce uncertainty in the tracking procedure, including, noise, splitting and crossing fibers, head motion and image artifacts. To address this uncertainty, stochastic tractography methods have been developed to quantify the uncertainty associated with estimated fibers (Bjornemo et al., 2002). Method uses a propagation model based on stochastics and regularization, which allows paths originating at one point to branch and return a probability distribution of possible paths. The method utilizes principles of a statistical Monte Carlo method called Sequential Importance Sampling and Resampling (SISR). Based on probability functions, using a sequential importance sampling technique (Bjornemo et al., 2002), one can generate thousands of fibers starting in the same point by sequentially drawing random step directions. This gives a very rich model of the fiber distribution, as contrasted with single fibers produced by conventional tractography methods. Moreover, from a large number of sampled paths, probability maps can be generated, providing better estimates of connectivity between several anatomical locations. A comparison of the algorithms can be seen here. (Figure 1)
Figure 2: Stochastic tractography of uncinate fasciculis on anatomical data (left) and cingulum bungle on fMRI scan (right)
B-Possible Applications
  • Since diffusion direction uncertainty within the gray matter is quite significant; principal diffusion direction approaches usually do not work for tracking between two gray matter regions. Thus if one requires finding connections between a priori selected anatomical gray matter regions, defined either by anatomical segmentations (in case of using structural ROI data), or functional activations (in case of megring DTI with fMRI), stochastic tractography seems to be the method of choice. Here is an example of this application to anatomical data (Figure 2, left image) and to fMRI data (Figure 2, right image).
  • Stochastic Tractography is also comparable, if not better, in defining large white matter fiber bundles, especially those traveling through white matter regions characterized by increased diffusion uncertainty (fiber crossings). Example of such application to internal capsule. (Figure 3)
Figure 3: Streamline vs. stochastic tractography of the Internal Capsule
C-References
Figure 4: Python Stochastic Tractography GUI

Module

Can be found in: MODULES > PYTHON MODULES > PYTHON STOCHASTIC TRACTOGRAPHY

Functionality of Python Stochastic Tractography module in Slicer 3.0
  • IO:

Module reads files (DWI and ROIs) in nhdr format.

  • Smoothing:

One can smooth the DWI data (only Gausian smoothing is supported at this time). We recommend it if the data is noisy.

  • Brain Mask:

The Brain mask defines the volume in which the tensor will be computed and the tracts evaluated.

  • Diffusion Tensor:

This step calculates the tensor and can output anisotropy indices (FA/Mode/Trace)

  • Tractography:

Parameters that need to be adjusted:

1. The amount of tracts that will be seeded from each voxel (we recommend between 500 and 1000 tracts, depending on the workstation power- 1000 tracts per voxel seeded within the large ROI for high resolution DWI can take a long time to compute).
2. Maximum tract length (in mm), this can eliminate long, unwanted tracts if the regions for which connection is measured are located close to each other
3. Step Size: distance between each re-estimation of tensors, usually between 0.5 and 1 mm
4. Stopping criteria. This can be used on the top of WM mask, to terminate tracts (in case they really want to travel through the CSF, for example).
  • Connectivity Map:

This step creates output probability maps.

1. binary: each voxel is counted only once if at least one fiber pass through it
2. cumulative: tracts are summed by voxel independently
3. weighted: tracts are summed by voxel depending on their length
  • Length Based:

This step will output only either the shortest 1/3, middle 1/3, or longest 1/3 of the tracts.

  • Vicinity

This step traces back n number of steps from tract endpoint to check if track crosses target ROI. If so, tract is included.

  • Threshold

This step will reject tracts whose endpoints are lower than the threshold value).

  • Spherical ROI vicinity

This will make the ROI a sphere based on the ROI’s center of gravity (with the sphere’s radius being the distance from the center to the ROI’s furthest point). This sphere can be inflated by raising the Vicinity level to the number of steps you’d like to increase the ROI’s size by.


Then, probability maps can be saved as ROIs, and either used directly, or thresholded (at certain probability, step claimed by few publications to remove noise) in slicer to mask and compute average FA, Mode, Trace for entire connection. Diffusion indices can be also weighted by the probability of connection for each voxel.

Work Accomplished

Figure 6: Stochastic tractography from a single ROI on helix phantom
A - Optimization and testing of stochastic tractography algorithm 
  • Original methodological paper, as well as our first attempts to use the algorithm (CC+ and matlab scripts) have been done on old "NAMIC" 1.5T LSDI data (Structural MRI and DTI data).
  • Tri worked hard on making sure algorithm works on new high resolution 3T data (available here: 3T Data).
  • Tests have been done also on the spiral diffusion phantom, to make sure diffusion directions and scanner coordinates are handled properly by the algorithm (Figure 6).
B - Clinical Applications
  • Algorithm was used to trace and analyze anterior limb of the internal capsule on 1.5T data. It generated reacher representation of frontal fiber projections, it also turned out to be more sensitive to group differences in white matter integrity that conventional deterministic tractography (see Figure 6).
  • Algorithm was also used to trace smaller white matter fiber tracts, such as Cingulum, Fornix, Uncinate Fasciculus, Arcuate Fasciculus on 3T "Santa Fe" dataset (http://www.na-mic.org/Wiki/index.php/SanteFe.Tractography.Conference)
C - References

Work in Progress

A - Optimization and Testing of stochastic tractography module 
  • Julien is testing now separate components of work flow, including the masks, and impact of their precision on stochastic output, as well as impact of number of seeding points on tractography results.
  • At the same time, we are testing the module on a Max Plank dataset of Anna Rotarska-Jagiela by looking at the connections through the corpus callosum of the left auditory cortex to the right auditory cortex as defined by fMRI activation maps as well as running this data for the tractography comparison project.
  • We are discussing possibility of adding project specific functionality to module, such as third ROI to guide tractography, nonlinear registration button for merging fMRI and DTI data.
B - Related Clinical Projects
  • Arcuate Fasciculus Extraction Project
Figure 7: The arcuate fasciculus including seed, midpoint and target ROI's.

We have started the project of investigating Arcuate Fasciculus using Stochastic Tractography (Figure 7). This structure is especially important in both VCFS and schizophrenia, as it connects language related areas (Brocka and Wernicke's), and is involved in language processing quite disturbed in schizophrenia patients. It also can not be reliably traced using deterministic tractography.

Project involves:
  • Whole brain segmentation, and automatic extraction of regions interconnected by Arcuate Fasciculus (Inferior frontal and Superior Temporal Gyri).
  • White matter segmentation, in order to prevent algorithm from traveling through the ventricles, where diffusivity is high.
  • Non-linear registration of labelmaps to the DTI space.
  • Seeding tracts. We have piloted it using 5000 seeds per voxel, however it is quite time consuming running it on even most powerful computers in the lab, so we have experimented with smaller number of seeds per voxel. We tested 1000 seeds, which gave virtually identical results.
  • Extracting path of interest, and calculating FA along the path for group comparison. Presentation of previous results for 7 schizophrenics and 12 control subjects, can be found here: Progress Report Presentation.
  • Results presentation and paper submission. Abstract was submitted and accepted for presentation at World Biological Psychiatry Symposium in Venice, Italy. Paper is in preparation.
  • Semantic Network Connectivity Project

We use combination of fMRI and DTI data to define and characterize functional and anatomical connectivity within the semantic processing network in schizophrenia.

Project involves:
  • fMRI data analysis and identification of functional nodes involved in semantic processing in healthy controls and seubjects with schizophrenia
  • Analysis of functional connectivity (using FSL) between nodes of semantic network
  • Whole brain Voxel Based analysis of DTI data in same population
  • Use of stochastic tractography to identify connections between functional nodes
  • Correlational analysis involving anatomical and functional connectivity data.
  • Results presentation and paper submission. Data was presented at HBM conference in 2007, paper has been submitted to HBM.
Figure 8: Max Plank data showing tracts through the corpus connecting 2 cortical ROIs defined by fMRI activations.
  • Study of Default Network

We have started collaboration with Department of Neurophysiology Max Planck Institute in Frankfurt (Anna Rotarska contact person). They have dataset containing DTI and resting state fMRI in schizophrenia, and want to use stochastic tractography to measure integrity of anatomical connections within the default network in schizophrenia. Their fMRI data has been co-registered with anatomical scans, and has been put it into DTI space. Figure 8 shows pilot data of the white matter connections through the corpus callosum between the left and the right auditory cortex ROIs as defined by fMRI data. This data will be also used to test robustness of our module, since data was collected on a different scanner (3T Philips).

  • Tractography Comparison Project

We are also working on a tractography comparison projectdataset, where we apply stochastic tractography to phantom, as well as test dataset.

Staffing Plan

  • Sylvain and Yogesh are the DBP resources charged with adapting the tools in the NA-MIC Kit to the DBP needs
  • Doug Markant, our NAMIC RA has left the lab, and now Doug Terry, is a new NAMIC RA.
  • Julien is our new NAMIC software engineer. He is responsible for improving the STM (Stochastic Tractography Module), and making sure software works with STM compliant datasets.
 Development Progress
  • Polina is the algorithm core contact
  • Brad is the engineering core contact


Schedule

  • 10/2007 - Optimization of Stochastic Tractography algorythm for 1.5T data.
  • 10/2007 - Algorythm testing on Santa Fe data set and diffusion phantom.
  • 06/2008 - Optimization of Stochastic Tractography algorythm for 3T data.
  • 11/2008 - Slicer 3 module prototype using python.
  • 12/2008 - Slicer 3 module official release
  • 12/2008 - Documentation and packaging for dissemination.
  • 12/2008 - Arcuate Fasciculus results.
  • 01/2009 - Arcuate Fasciuclus first draft of the paper.
  • 05/2009 - Distortion correction and nonlinear registration added to the module
  • 05/2009 - Symposium on tractography, including stochastic methods at World Biological Psychiatry Symposium in Florence, Italy.
  • 05/2009 - Presentation of Arcuate Fasciculus findings at World Biological Psychiatry Symposium in Florence, Italy.

Team and Institute

  • PI: Marek Kubicki (kubicki at bwh.harvard.edu)
  • DBP2 Investigators: Sylvain Bouix, Yogesh Rathi, Julien de Siebenthal
  • NA-MIC Engineering Contact: Brad Davis, Kitware
  • NA-MIC Algorithms Contact: Polina Gollard, MIT

Publications

In print

Retrieved from "https://www.na-mic.org/w/index.php?title=DBP2:Harvard:Brain_Segmentation_Roadmap&oldid=38048"