29-10-2010, 08:13 AM
FAST & AUTOMATIC METHOD FOR RIGID REGISTRATION OF MR IMAGES OF HUMAN BRAIN
SUBMITTED BY:
INDU LATA PREM
S7 AEI
College Of Engineering, Trivandrum
2007-11 batch
[attachment=6967]
Hi friends, If you like this presentation, don't forget to thank the author parofriendly[at]gmail.com
OVERVIEW
INTRODUCTION
IMAGE REGISTRATION
CLASSIFICATION
APPROACHES
METHOD
EXPERIMENTAL RESULTS
ADVANTAGES
DISADVANTAGES
CONCLUSION
INTRODUCTION
Method for 3d registration.
Fast and automatic method.
Evaluated on 3D MR images.
Based on few principles.
Combines previous approaches.
Evaluated on 200 image pairs.
IMAGE REGISTRATION
Application in medical imaging.
Process that aligns images.
Problems associated.
Goal- to register images from epilepsy patients.
Enables comparative analysis.
CLASSIFICATION
Based on:
Nature of registration basis
Transformation
Domain of transformation
User interaction level
Transformation search method
Image Modality
Transformation subject
APPROACHES
Some earlier approaches:
I.
Based on similar strategy.
Instead of MSP uses EMP.
Computed by PCA.
Search for transformation.
Brain alterations- failure in registration.
II.
Addresses similar problem.
Matching partial.
Probabilistic search algorithm.
Divided into candidate parameter sets.
Used to rank.
METHOD
3D MR image I’ = (Di , I)
I(v) = intensity of voxel
J’ = (Dj , J) – Source & target images of brain
I’ and J’ aligned by MSP
Segmentation by Automatic tree pruning
Greedy search algorithm to find transformation T.
Applying T to remaining voxels
(a) AUTOMATIC BRAIN SEGMENTATION
Brain segmentation- Automatic tree pruning.
Graph based approach.
Selection of markers.
Optimum path forest
computed.
Detect edges that cross
object’s border.
(b) MSP LOCATION
Matches longitudinal fissures.
CSF appear as low intensity voxels.
Locates MSP candidate.
Mean intensity voxel- score
Performs greedy search.
Applicable to patients with structural abnormalities.
© IMAGE ALIGMNENT
Transformation to
reformat images.
MSP along z-axis.
New images must be
registered.
(d) GREEDY TRANSFORMATION SEARCH
Let I’s = (Si,I) & J’s = (Sj,J) be sub images of I’ and J’.
To search for transformation T.
Evaluate border and band sets.
Apply T (Sj) = S’j
To compute mutual information.
(e) VISUALISATION
Transformation T applied to target image domain.
Compose visualization.
Inspect registration for correctness.
Figure shows
examples.
EXPERIMENTAL RESULTS
Performed tests with 2 sets of images.
1st- source and target differ by rigid transformation.
2nd- create synthetic lesions.
Based on 20 MR-TI volumes of brain.
For each source image 5 target images generated.
Each set has 100 image pairs.
ADVANTAGES
Very accurate
Fast and automatic
Applications in medical imaging
Fully 3D & simple
DISADVANTAGES
Chances of errors.
Limited to MR images of brain.
CONCLUSION
Image registration- important task that enables comparative analysis of images acquired at different occasions.
Rigid Registration is directly applicable to organs without significant shape change.
Performs registration in 90 sec for brain volumes with 1 cu.mm.
Future work- to test in clinical data with actual lesions.
REFERENCES
M Audette,F.Ferrie, and T.Peters. An algorithmic overview of surface registration techniques for medical imaging. Medical image Analysis,4(3):201-217,2000.
P.J.Besl and N.D.McKay. A method for registration of 3D shapes. IEEE Transactions on pattern analysis and machine intelligence,14(2):239-256,1992.
E.D.Castro and C. Morandi. Registration of translated and rotated images using finite fourier transform. IEEE Transactions on Pattern Analysis and Machine Intelligence,PAMI-9(5):700-703,1997.
