Neuroscience 2003 Abstract
| Presentation Number: | 863.24 |
|---|---|
| Abstract Title: | Meta-algorithm for automated brain extraction from a structural MRI. |
| Authors: |
Rex, D. E.*1
; Shattuck, D. W.1
; Woods, R. P.1
; Stoltzner, S. E.1
; Toga, A. W.1
1LONI, Neurol., UCLA, LA, CA |
| Primary Theme and Topics |
Techniques in Neuroscience - Data analysis, physiological methods, statistics |
| Secondary Theme and Topics | Motor Systems<br />- Cortex and Thalamus<br />-- Imaging |
| Session: |
863. Imaging Methods Poster |
| Presentation Time: | Wednesday, November 12, 2003 11:00 AM-12:00 PM |
| Location: | Morial Convention Center - Hall F-I, Board # VV60 |
| Keywords: | imaging, segmentation, analysis, human |
Computer algorithms exist to remove non-brain tissue from structural MRIs. Each algorithm possesses strengths and weaknesses dependent on scanner, protocol, and subjects, and produces different results for the same data. Simultaneous use of multiple algorithms with a combination step builds on the strengths of the individual extractors yielding a meta-algorithm that produces more accurate and robust results.
The meta-algorithm incorporates information from four different extractors, BET (Smith, 2002), BSE (Shattuck et al., 2001), 3dIntracranial (in AFNI; Ward, 1999), and Watershed (in FreeSurfer; Segonne et al., 2003), and a registration procedure, FLIRT (Jenkinson et al., 2002). A trainer program analyzes results from the four extractors aligned to an atlas space against manually traced gold-standards of brain. For each region in the atlas space it selects a four input boolean function that produces the best results across the gold-standards. It outputs a combination-key that stores the proper combination of extractors to use at each voxel in the atlas space. A combiner program applies the key, aligned to a subject, to the results of the four extractors to produce a final result volume for the brain extraction.
We built a combination-key using 30 whole-head T1-weighted scans. We processed 20 separate scans under the same protocol with each extractor and the meta-algorithm using the previously derived key. Dice coefficients measured the similarity between an extractor’s result and its gold-standard. The meta-algorithm was significantly closer (p<<0.001) to the gold-standards than any single extractor (means: meta=0.97, BET=0.96, BSE=0.96, 3dIntracranial=0.92, Watershed=0.95) and possessed a smaller standard deviation in its Dice coefficients suggesting a more robust result.
The meta-algorithm incorporates information from four different extractors, BET (Smith, 2002), BSE (Shattuck et al., 2001), 3dIntracranial (in AFNI; Ward, 1999), and Watershed (in FreeSurfer; Segonne et al., 2003), and a registration procedure, FLIRT (Jenkinson et al., 2002). A trainer program analyzes results from the four extractors aligned to an atlas space against manually traced gold-standards of brain. For each region in the atlas space it selects a four input boolean function that produces the best results across the gold-standards. It outputs a combination-key that stores the proper combination of extractors to use at each voxel in the atlas space. A combiner program applies the key, aligned to a subject, to the results of the four extractors to produce a final result volume for the brain extraction.
We built a combination-key using 30 whole-head T1-weighted scans. We processed 20 separate scans under the same protocol with each extractor and the meta-algorithm using the previously derived key. Dice coefficients measured the similarity between an extractor’s result and its gold-standard. The meta-algorithm was significantly closer (p<<0.001) to the gold-standards than any single extractor (means: meta=0.97, BET=0.96, BSE=0.96, 3dIntracranial=0.92, Watershed=0.95) and possessed a smaller standard deviation in its Dice coefficients suggesting a more robust result.
Supported by NIMH (1 P20 MH65166, 5 P01 MH 52176), NCRR (2 P41 RR13642, 2 M01 RR00865), supplement by Biomedical Informatics Research Network (2 P41 RR13642 [www.nbirn.net]). DER: ARCS Foundation, NIGMS MSTP (GM08042)
Sample Citation:
[Authors]. [Abstract Title]. Program No. XXX.XX. 2003 Neuroscience Meeting Planner. New Orleans, LA: Society for Neuroscience, 2003. Online.
Copyright © 2003-2025 Society for Neuroscience; all rights reserved. Permission to republish any abstract or part of any abstract in any form must be obtained in writing by SfN office prior to publication.
