DBP2:Harvard:Brain Segmentation Roadmap

Roadmap

The main goal of this application is to characterize anatomical 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 brain automatic segmentation, using newly acquired 3T data, NAMIC tools and slicer 3.

A - Optimization of slicer EM white/gray/csf segmentation 

• We have been using EM segmenter for our 1.5 Tesla data. This protocol needs to be optimizes for 3T data, adjusting for higher data resolution, different intensity profiles and bias field inhomogeneity. This protocol will be in Slicer 3 (Sylvain, Brad)
  • Since technology needed for this project already exists in Slicer 2, its implementation in Slicer 3 is a low risk project, and will be accomplished within the next couple of months.
• Since the EM segmentation uses brain atlas, we need to have the technology in place to generate new templates automatically. (Sylvain, Brad, Polina)
  • This technology should be developed in relatively short period of time. The fact that segmentation atlases will be generated for each study, will ultimately make the technology more robust to other brain diseases.
  • We can discuss the details/use cases at MICCAI. We should have at least a basic version prepared for the January AHM. One possibility is to prepare a command line version of the program first and then combine it with Slicer3/add additional functionality after the AHM. We need to keep track of the recent plan to maintain image orientation information within the ITK registration pipeline as this will effect our final product (Brad is checking into this).

B – Segmentation performance comparison, and validation 

• In order to make sure there is no systematic bias between segmentation results of newly acquired 3T data and old 1.5T data, we have chosen 15 control subjects and 15 schizophrenic subjects, which have both scans. We will run and compare results of segmentation, both between methods and within methods between groups.
  • Since the scanning protocol was established and tested on schizophrenia subjects, and thus data collection is much more advanced there, and since the ultimate goal is to compare anatomical abnormalities in VCFS with these in schizophrenia, this project has two benefits- it gives schizophrenia comparison data, as well as leads to establishing the segmentation protocol that will be easily applicable to VCFS, once more data is collected.

C - Analysis of small anatomical structures 

• After the protocol for whole brain segmentation is established, small anatomical structures, such as STG, hippocampus, cingulate gyrus, thalamus, caudate and dorsolateral prefrontal cortex will be segmented in both schizophrenia (first) and VCFS (later) (Sylvain, Brad).
  • Technology for segmentation of most of these regions is already in place in Slicer 2 (Sylvain, Kilian). DLPFC module is especially interesting for VCFS population, and this is the first module that will be optimized for our project.
  • This module is currently being implemented into Slicer3 (John, Brad)
    • According to John there is no current plan to transition the module to Slicer3.
    • A Slicer3 version of this module should combine the current Slicer2 rule-based ROI selection with grey matter segmentation from the EMSegmenter module as this segmentation method is more advanced than the current method. However, this would add to development time.
    • In order to adapt this method for new data the segmentation portion of the algorithm must be optimized to the new data. If we use the EMSegmenter then we can leverage our already-planned work of adapting it to the new data. The rule-based portion of this algorithm is essentially data independent as it only requires the selection of landmarks within images.
• Using segmented DLPFC ROIs, we will perform cortical thickness analysis (Marc, Sylvain)
  • Marc Niethammer has developed a cortical thickness algorithm, which will be put in Slicer 3 technique

D - Subject comparison 

• Local analysis requires techniques which are not currently in the NA-MIC Kit
  • Freesurfer could be used for the local analysis (but it is not in the NA-MIC Kit)
  • Ipek is developing local analysis tools and may have a tool available in Fall 2008.

To do

• Assign owners to tasks
• Define schedule

Staffing Plan

• Sylvain and Marc are the DBP resources charged with adapting the tools in the NA-MIC Kit to the DBP needs
• Polina is the algorithm core contact
• Brad is the engineering core contact

Schedule

• xx/xx/2007 - White matter/gray matter/CSF segmentation of the 3T schizophrenia brain using the Slicer 3 EM segment module
  • Brad: my guess is 1Dec07
• xx/xx/2007 - Automatic atlas generation module in Slicer 3
  • Brad: I would like to have a limited/prototype version by the AHM 10Jan08, but more advanced functionality (e.g., coordination with Martin's pipeline) will come after
• xx/xx/2007 - Comparison between 1.5T vs 3T segmentation performance
  • Brad: This will be straightforward; it is possible to finish by the AHM if item (1) is completed by 1Dec07. Maybe better is to say we will start it by the AHM and complete by 1Feb08.
• xx/xx/2007 - Automatic segmentation of DLPFC
  • Brad: The slicer2 module can be used now. Adding the module to Slicer3 and utilizing the EMSegment module code (with 3T imaging parameters) could be done by 1June08
• xx/xx/2007 - Cortical thickness measurement of DLPFC using Marc Niethammer's Slicer3 module
• xx/xx/2007 - Local analysis of cortical thickness as a Slicer3 module
• xx/xx/2007 - BatchMake workflow
  • Brad: This should start and stop in late summer/fall 08 after a detailed workflow plan is established
• xx/xx/200x - Groupwise local analysis of volumes and cortical thickness as a NA-MIC Workflow
  • Brad: Ready for the AHM 09?

Team and Institute

• PI: Marek Kubicki, PhD, (kubicki at bwh.harvard.edu)
• DBP2 Investigators: Sylvain Bouix, PhD, Marc Niethammer, PhD, Harvard
• NA-MIC Engineering Contact: Brad Davis, Kitware
• NA-MIC Algorithms Contact: Polina Gollard, MIT