Summer Work

Brain Atlas

Dan Diner

Introduction

• PET is a nuclear imaging technique• A radiotracer is injected into subject, and the

signal that it sends off is captured and measured to find out how much activity occurs and where

• The information that is collected is used for brain study, and disorder analysis and diagnosis

QuickTime™ and a decompressor

are needed to see this picture.

ROI/VOI Mapping

• ROIs(Regions of Interest) are small regions mapped onto brains

• Most common way of mapping is manual• Manual mapping is very time-consuming and

requires highly-paid personnel (raters)• Raters have inter- and intra- subject variation• As PET scan quality increases, so does the

time it takes for raters to do their job

• PET has very poor spatial resolution

• MRI has very high spatial resolution

• ROIs are therefore mapped onto the MRI of a subject, onto which the PET scan is co-registered

QuickTime™ and a decompressor

are needed to see this picture.

MRI PET Co-Registered

Brain Atlases• An “average brain” (co-registered PET/MRI)• Has all VOIs labeled manually onto it• These automated methods have been developed to label

VOIs (Volumes of Interest)• Never have been as accurate as manual methods• Very precise - have no inter- or intra- subject variation• Take drastically less time to label a brain than manual

methods

Brain Atlases

• Many imaging labs have their own, but we could not use another labs’ because each lab has different MRI acquisition and labeling techniques

• Can be single- or multiple- subject template• Could be made for only one specific region or

many• Probabilistic atlases tell you the probability of

a certain voxel belonging to a certain ROI

Methods and M aterials: Subject and Image Acquisition

• MRIs and PETs from 176 subjects of past studies

• 123 from a 1.5 Tesla MRI

• 53 from a 3.0 Tesla

Image Analysis Setup

• Worked on a PowerMac G5 using Mac OS X 10.5.4 Leopard

Major Software used:• Matlab (R2007a)• FSLVIEW and FLIRT• ART• iView

Image Pre-Processing

• Raw MRI files were

copied to a common

folder QuickTime™ and a decompressor

are needed to see this picture.

• Previously made Gray Matter,

White Matter, and Cerebral Spinal Fluid Masks were multiplied(matrix multiplication) against the raw

files to get rid of all extraneous

material (neck, skull, dura matter)

QuickTime™ and a decompressor

are needed to see this picture.

QuickTime™ and a decompressor

are needed to see this picture.

QuickTime™ and a decompressor

are needed to see this picture.

QuickTime™ and a decompressor

are needed to see this picture.

QuickTime™ and a decompressor

are needed to see this picture.

• Used iView to compare new images against raw ones

• We did this to ensure that all extraneous material was gone and that no brain was accidentally segmented

• I used a scale of 0-3 to rate the new images (0= perfect, 1=some dura, 2=much dura, 3=segmented brain)

• Most images were 0s and 1s.

QuickTime™ and a decompressor

are needed to see this picture.

Normalization Procedure

• MNI template to which were were normalizing was reoriented towards raw MRI native space

• New MRI images also normalized to this template

• We examined the normalized images, and found that some were very well aligned, but others were misplaced or improperly warped

• ART uses a 2-step process to warp MRIs: linear registration, then warp

• Linear registration is rotations• Warping is dilations• Because the rotation was bad, we thought

that the linear registration had gone wrong• We tried linear registration(first 9 Degrees of

Freedom) with FLIRT, and then warping(up to 12 dof) with ART

• The rotations came out looking good, but close inspection revealed that some leftover dura got warped into the new brain

QuickTime™ and a decompressor

are needed to see this picture.

• An MRI with a particularly large amount of dura was manually cleaned, and then had FLIRT and ART run on it

• The resulting image was perfect

Conclusion

• ART is very sensitive to extraneous material

• all dura needs to be properly removed from all images before normalization

• A better automatic method for cleaning up dura needs to be developed

QuickTime™ and a decompressor

are needed to see this picture.