FreeSurfer image processing pipeline for infant clinical MRI images
Presented During:
Tuesday, June 27, 2017: 10:43 AM - 10:55 AM
Vancouver Convention Centre
Room:
Ballroom AB
Submission No:
1703
Submission Type:
Abstract Submission
On Display:
Monday, June 26 & Tuesday, June 27
Authors:
Lilla Zöllei1, Yangming Ou2, Juan Iglesias3, P. Ellen Grant4, Bruce Fischl5
Institutions:
1Athinoula A Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, MA, 2Boston Children's Hospital, Boston, United States, 3University College London, London, United Kingdom, 4Boston Children's Hospital, Harvard Medical School, Boston, MA, 5MGH/HMS, Charlestown, MA
First Author:
Lilla Zöllei
-
Lecture Information
|
Contact Me
Athinoula A Martinos Center for Biomedical Imaging, Massachusetts General Hospital
Charlestown, MA
Athinoula A Martinos Center for Biomedical Imaging, Massachusetts General Hospital
Charlestown, MA
Introduction:
The targeted age group, 0-2 years, has been underserved with respect to specialized computational tools that can robustly and accurately analyze brain MRI images. Several segmentation solutions exist for specific ages within this range, for example, for newborns (Prastawa, Gilmore et al. 2005, Wang, Shi et al. 2011, Gui, Lisowski et al. 2012, Gousias, Hammers et al. 2013, Wang, Gao et al. 2015), for 1-year olds (Wang, Gao et al. 2015), and for 2-year olds(Gousias, Rueckert et al. 2008),), some relying on access to longitudinal time points from the same subject to handle segmentation of the more challenging ages (Wang, Shi et al. 2011), but none can handle this relatively wide age range exclusively on clinical T1-weighted images.
Methods:
Our image processing pipeline consists of three main components: (a) skullstripping (b) segmentation and (c) surface extraction.
Skullstripping: We relied on a novel double-consensus framework involving multiple atlases and multiple skull-strippers (Ou, Gollub et al. 2016). The first consensus established solutions from multi-atlas skull-stripping and the second fused carefully selected and optimized masks into the final result.
Segmentation: We developed a unified procedure that can handle scans from subjects at any age in the 0-2 year age range. It selects a subset of the atlas data that is most similar to the target subject and uses information derived from these to guide the segmentation and surface extraction steps. We implemented a prior-based Bayesian approach (Iglesias, Sabuncu et al. 2012), where we rely on ground-truth information from a manually annotated training data set (de Macedo Rodrigues, Ben-Avi et al. 2015). We used the deformable registration tool Deformable Registration via Attribute Matching and Mutual-Saliency Weighting (DRAMMS) (Ou, Sotiras et al. 2011) in order to ensure that the atlas images is in the same spatial coordinate system as the test image.
Surface reconstruction: We used the above-described segmentation results to initialize the construction of white matter and pial surfaces. As the intensity contrast in the 0-2 year range varies by region and age, our algorithms rely more heavily on the segmentation outcomes in the case of younger subjects and increase the weighting of the intensity contrast information for older ones. After the initial estimation of the surfaces, we ran spatial surface-based smoothing, and a robust genetic-algorithm-based topology correction algorithm (Ségonne, Pacheco et al. 2007).
Skullstripping: We relied on a novel double-consensus framework involving multiple atlases and multiple skull-strippers (Ou, Gollub et al. 2016). The first consensus established solutions from multi-atlas skull-stripping and the second fused carefully selected and optimized masks into the final result.
Segmentation: We developed a unified procedure that can handle scans from subjects at any age in the 0-2 year age range. It selects a subset of the atlas data that is most similar to the target subject and uses information derived from these to guide the segmentation and surface extraction steps. We implemented a prior-based Bayesian approach (Iglesias, Sabuncu et al. 2012), where we rely on ground-truth information from a manually annotated training data set (de Macedo Rodrigues, Ben-Avi et al. 2015). We used the deformable registration tool Deformable Registration via Attribute Matching and Mutual-Saliency Weighting (DRAMMS) (Ou, Sotiras et al. 2011) in order to ensure that the atlas images is in the same spatial coordinate system as the test image.
Surface reconstruction: We used the above-described segmentation results to initialize the construction of white matter and pial surfaces. As the intensity contrast in the 0-2 year range varies by region and age, our algorithms rely more heavily on the segmentation outcomes in the case of younger subjects and increase the weighting of the intensity contrast information for older ones. After the initial estimation of the surfaces, we ran spatial surface-based smoothing, and a robust genetic-algorithm-based topology correction algorithm (Ségonne, Pacheco et al. 2007).
Results:
We have run 1300 experiments, where for the segmentation step we varied the number of atlases used and the criterion for atlas selection. Overall, the infant brain MRI segmentation experiments were quite promising (Figure 1). The most challenging subjects, as expected, were the neonates and subjects of around 6-8 months of age, where the non-linear registration step to the atlases often does not perform to its best. The quantitative evaluation of the segmentation process was done by computing the Dice coefficient between the manually delineated and the automatically outlined ROIs. We compared such measurements across the whole data set and also across the different atlas groups. We assessed statistical significance with one-tailed, paired t-tests.
Conclusions:
We have developed an automated segmentation and surface extraction pipeline that is designed to accommodate pediatric brain MRI images from a population 0-2 year-olds relying on clinical T1-weighted MR images. The advantage of our algorithm is that it provides files consistent with the FreeSurfer image analysis package. Upon publication, the resulting pipeline and the necessary atlases will be made available to the research community via FreeSurfer and the NITRC website.
Imaging Methods:
Anatomical MRI
Informatics:
Workflows 1
Modeling and Analysis Methods:
Methods Development
Segmentation and Parcellation 2
Neuroanatomy:
Normal Development
Keywords:
Segmentation
STRUCTURAL MRI
Workflows
1|2Indicates the priority used for review
Would you accept an oral presentation if your abstract is selected for an oral session?
I would be willing to discuss my abstract with members of the press should my abstract be marked newsworthy:
Please indicate below if your study was a "resting state" or "task-activation” study.
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute the presentation in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels or other electronic media and on the OHBM website.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Internal Review Board (IRB) or Animal Use and Care Committee (AUCC) Approval. Please indicate approval below. Please note: Failure to have IRB or AUCC approval, if applicable will lead to automatic rejection of abstract.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references in author date format
Gousias, I. S., A. Hammers, S. J. Counsell, L. Srinivasan, M. A. Rutherford, R. A. Heckemann, J. V. Hajnal, D. Rueckert and A. D. Edwards (2013). "Magnetic resonance imaging of the newborn brain: automatic segmentation of brain images into 50 anatomical regions." PLoS One 8(4): e59990.
Gousias, I. S., D. Rueckert, R. A. Heckemann, L. E. Dyet, J. P. Boardman, A. D. Edwards and A. Hammers (2008). "Automatic segmentation of brain MRIs of 2-year-olds into 83 regions of interest " NeuroImage 40(2): 672-684.
Gui, L., R. Lisowski, T. Faundez, P. S. Huppi, F. Lazeyras and M. Kocher (2012). "Morphology-driven automatic segmentation of MR images of the neonatal brain." Med Image Anal 16(8): 1565-1579.
Iglesias, J. E., M. R. Sabuncu and K. Van Leemput (2012). A generative model for multi-atlas segmentation across modalities. 9th IEEE International Symposium on Biomedical Imaging (ISBI).
Ou, Y., R. L. Gollub, K. Retzepi, X. Da, J. Kalpathy-Cramer, S. N. Murphy, P. E. Grant and L. Zollei (2016). "PICASSO Skull Stripping: I. Algorithm and Validations in Multi-Site, Multi-Vendor and Multi-Platform Pediatric MRI of Health and Disease." under review.
Ou, Y., A. Sotiras, N. Paragios and C. Davatzikos (2011). "DRAMMS: Deformable registration via attribute matching and mutual-saliency weighting." Medical Image Analysis. 15(4): 622-639.
Prastawa, M., J. H. Gilmore, W. Lin and G. Gerig (2005). "Automatic segmentation of MR images of the developing newborn brain." Med Image Anal 9(5): 457-466.
Ségonne, F., J. Pacheco and B. Fischl (2007). "Geometrically Accurate Topology-Correction of Cortical Surfaces Using Nonseparating Loops." IEEE Transactions on Medical Imaging 26(4): 518-529.
Wang, L., Y. Gao, F. Shi, G. Li, J. H. Gilmore, W. Lin and D. Shen (2015). "LINKS: learning-based multi-source IntegratioN frameworK for Segmentation of infant brain images." Neuroimage 108: 160-172.
Wang, L., F. Shi, P.-T. Yap, W. Lin, J. H. Gilmore and D. Shen (2011). "Longitudinally guided level sets for consistent tissue segmentation of neonates. ." Hum Brain Mapp.