Automated Brain Mapping

Subcortical Segmentation

Accurate and reliable subcortical segmentation is a requirement for volumetric and morphometric studies of neuro-degenerative diseases.

FS+LDDMM

We have developed a fully-automated pipeline for robust and accurate subcortical segmentation. We use large deformation diffeomorphic metric mapping (LDDMM) to find the high-dimensional transformation between an atlas brain region of interest (ROI) and a target brain ROI, then propagate expertly-defined segmentations from the atlas to the target brain.

Freesurfer (FS) subcortical segmentations are used to initialize and guide the ROI-based mapping, eliminating the need for manual landmarking or intervention.

Advantages

• Fully automated
• Inherently smooth segmentations - ideal for shape analysis
• Applicable to a multitude of sub-cortical and neo-cortical structures
• Can supply your own expertly defined surface model and atlas

Further improvements in performance for difficult structures, such as the Amygdala, have been realized using confidence-weighted multi-structure registration and multi-atlas strategies (MICCAI 2009).

These advances have been used to segment the caudate nucleus in the CAUSE07 caudate segmentation evaluation, and placed in the Top 2 out of 20 segmentation methods worldwide. Details of the results can be seen at the competition website here.

Segmented Databases

• FS+LDDMM
• Hippocampus, thalamus, and ventricles in Alzheimer's patients (Multi-atlas propagation)
• Caudate, Putamen and Hippocampus in IBSR dataset (Multi-atlas propagation)
• Caudate, Putamen and Hippocampus in LPBA dataset (Multi-atlas propagation)
• Multi-structure FS+LDDMM
• Hippocampi in healthy middle-aged subjects (Multi-atlas propagation)
• Caudate Segmentation Evaluation, CAUSE07 (Multi-atlas propagation)
• Amygdala segmentation in autism subjects (Supervised-atlas correction)

Relevant publications

Khan, A., Wang, L., and Beg, M. F. 2008.
Freesurfer-initiated fully-automated subcortical brain segmentation in MRI using large deformation diffeomorphic metric mapping.
NeuroImage. 41(3): 735-746.

Wang, L., Khan, A., Csernansky, J. G., Fischl, B., Miller, M. I., Morris, J. C., and Beg, M. F. 2008
Fully-automated, multi-stage hippocampus mapping in very mild Alzheimer Disease.
MICCAI 2008 Workshop on Computational Anatomy and Physiology of the Hippocampus (CAPH'08).

Khan, A., Chung, M. K., Beg, M. F. 2009
Robust atlas-based brain segmentation using multi-structure confidence-weighted registration
MICCAI 2009. LNCS 5762: 549-557. Poster (PDF) Poster of the Day: Neuro, Cell, and Multiscale Image Analysis

Chen, J., Palmer, S. J., Khan, A. R., McKeown, M. J., Beg, M. F. 2009
Freesurfer-initialized large deformation diffeomorphic metric mapping with application to Parkinson's Disease.
Proceedings of SPIE Medical Imaging 2009. 7259 725931 Poster (PDF)