Published in the proceedings of the 4-th International Conference on Functional Mapping of the Human Brain

Improved automatic gross cerebral structure segmentation $^\dagger$

DL Collins, AP Zijdenbos and AC Evans

McConnell Brain Imaging Centre, Montréal Neurological Institute,
McGill University, Montréal, Canada

Introduction

We have developed two complementary segmentation strategies for automatic identification of tissue types and gross anatomical structures from volumetric MRI of the human brain. The first, named INSECT (Intensity Normalized Stereotaxic Environment for the Classification of Tissue), has been shown to successfully classify GM, WM, CSF and has been applied to the problem of automatically segmenting MS lesions from a large number of subjects in the context of a third phase clinical trial [1]. The second, named ANIMAL [2] (Automatic Nonlinear Image Matching and Anatomical Labeling), was designed to compute a non-linear warp to register a given subject's MRI to a pre-labeled target MRI that serves as an atlas. The inverse of the recovered transformation is used to warp the atlas labels back onto the subject's MRI, thus achieving segmentation.

Methods

ANIMAL has been shown to successfully segment basal ganglia structures [2] but it has not been able to segment cortical structures satisfactorially (overlap indexes have been typically around 40%). This is due to the fact that the deformation field estimated by ANIMAL does not have the power to unfold the cortex of one brain and then refold it back onto a target brain. The deformation field is bandlimited and therefore does not have high enough frequencies to introduce (or remove) cortical folds where needed. However, ANIMAL is able to identify the approximate region where the structure is located. While classification techniques such as INSECT are able to separate GM, WM and CSF and in so doing, extract fine detail from the MRI volume, they cannot differentiate between two structures with the same tissue type.

We have merged the complementary information from ANIMAL's non-linear deformation (i.e., low resolution region identification) with the output of INSECT's classification technique (ie, voxel class labels) in order to accurately identify specific structures in a subjects MRI. In short, the grey matter (extracted by INSECT) is masked by a structural region mask (defined by ANIMAL) in order to define specific brain structures. Since INSECT yields high resolution structure information, it is no longer necessary to run ANIMAL to fine resolutions, thus providing a considerable improvement in speed.

Results

$\parbox{6in}{ \small{{\bf Fig. 1:} (Left to right) Transverse slice... ...\end{sc}~ result improves segmentation at the cortex and at the ventricles.} }$

Conclusion

The method presented is completely automatic, therefore fully objective and applicable to large ensembles of brain volumes. This segmentation method has been applied to 150 data sets acquired as part of the ICBM project to estimate structure volumes (see Collins et al, 98) and to create a SPAM (statistical probability anatomy map) atlas [3] that represents normal anatomical variability of both structure shape and position.

Bibliography

  1. Zijdenbos, A.P., Evans, A.C., Riahi, et al. Proc Vis Biomed Comp. p 439-448 1996.

  2. D.L. Collins, C.J. Holmes, T.M. Peters and A.C. Evans.
Human Brain Mapping. 3(3):190-208, 1996.

  3. A.C. Evans, D.L. Collins, C.J. Holmes, et al. ``Towards a probabilistic atlas of human neuroanatomy'' in Brain Mapping: The Methods. 1996

$^\dagger$Supported by the U.S. Human Brain Map Project and the International Consortium for Brain Mapping.