Automatic labeling of brain anatomy and FMRI brain activity

Automatic labeling of brain anatomy and FMRI brain activity

NemoImage 13, Number 6, 2001, Part 2 of 2 Parts lDEklW METHODS - ANALYSIS Automatic labeling of brain anatomy and FMRI brain activity Arno Klein...

75KB Sizes 0 Downloads 73 Views

NemoImage

13, Number

6, 2001, Part 2 of 2 Parts

lDEklW

METHODS

- ANALYSIS

Automatic labeling of brain anatomy and FMRI brain activity Arno Klein, Joy Hirsch FMRI Laboratory of the Memorial Sloan-Kettering Cancer Center, Weill Medical College of Cornell University, New York, USA Almost any comparative structural or functional brain study necessitates the identification of corresponding anatomical structures. Manually ascribing anatomical labels to individual brain image data is tedious, difficult, and very inconsistent. The motivation to automatize the process of labeling anatomical (and therefore functional) image data is therefore to provide a consistent and convenient method for preparation of data for analysis. This abstract briefly describes a new, fully automatic approach to labeling the anatomy and functional activity of a subject’s brain. Methods Co-registration

of brain

atlas

and subject

MRI

volume

The MRI (Tl) volumes of the Harvard brain atlas [l] and a subject are co-registered with the Montreal Neurological Institute’s average of 305 MRI volumes set in Talairach space [2],[3]. MNI’s auto-registration package is also used to register the functional (T2*) with the anatomical (Tl) MRI images. Image

pre-processing

Morphological points from pre-processed Self-organized

operations are performed to extract skeletons the skeletons. The composite 3-D skeletons in the same manner. mapping

of brain

mappings

with

neural network in 3-D.

the subject

and branch selected is

[4], also called

a self-organizing

map, is used to map combinations

a cost function

A cost function is used to determine which set of atlas-subject function penalizes mapping distortion, resulting dissimilarity, atlas and subject regions. Labeling

extract endpoints The brain atlas

regions

A modified form of an unsupervised Kohonen of atlas brain regions to subject brain regions Evaluating

of the (Tl) brain sections. 2-D filters are segmented into smaller regions.

mappings makes for a reasonable and differences in normalized

piece-to-piece correspondence. position, size, and branching

The of the

brain

The labels for the atlas regions color-labeled) activity data.

matched

to each subject

region

are used to label the subject

region

as well as any co-registered

(T2*

Conclusion In this study, anatomical labels are automatically assigned to brain image data with the use of a labeled brain atlas. The automatic labels are evaluated by comparison with labels assigned by experts, and with functional data of known regions acquired with standard tasks. Results are given as measures of correspondence between these independent labeling methods, in percent shared voxel labels. References Kikinis, et al (1996). A digital brain atlas for surgical planning, model driven segmentation and teaching. IEEE Transactions on Visualization and Computer Graphics, 2. Evans, A.C., et al (1994). Three-dimensional correlative imaging: applications in human brain mapping, in: Functional Neuroimaging: Technical Foundations (Thatcher, R.W., et al, eds), 145162. Collins, D.L., et al (1994). Automatic 3D intersubject registration of MR volumetric data in standardized Talairach space. Jountal of Computer Assisted Tomography, 18, 192-205. Kohonen, T. (1997). Self-organizing maps, 2nd Ed. New York: Springer-Verlag.

s174