Juri Minxha Medical Image Analysis Professor Benjamin Kimia Spring 2011 Brown University Problem Statement 2 Signal Sources 3D volumetric data CT scan MRI 2D images ex frame from fluoroscopy video ID: 626750
Download Presentation The PPT/PDF document "A Computationally Efficient Approach fo..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
A Computationally Efficient Approach for 2D-3D Image Registration
Juri Minxha Medical Image Analysis Professor Benjamin Kimia Spring 2011
Brown UniversitySlide2
Problem Statement
2 Signal Sources - 3D volumetric data (CT scan, MRI) - 2D images (ex. frame from fluoroscopy video)Project 3D data onto a 2D plane and compare it to existing 2D image. The projected image is also known as the digitally reconstructed radiograph (DRR)
Brown UniversitySlide3
Why do we need this?
Image guided surgery Pre-operative data (CT/MRI acquisitions)Good resolution3D dataSlow AcquisitionIntra-operative data (fluoroscopy images)Can be quickly acquiredPoor resolution, more noise (ex. temporal blurring)Brown UniversitySlide4
Typical Approach to Registration
Similarity Metric Optimization1. Similarity Metric Mutual Information, Cross-Correlation, Correlation Ratio, Cross Correlation Residual Entropy2. Optimization, Non-gradient vs. Gradient
Gauss-Newton,
s
teepest descent,
Levenberg-Larquardt, simplex method etc.
The main challenge is:
Minimize computation time
Brown UniversitySlide5
Approach Outlined in this paper
Similarity Measure: Sum of Conditional VariancesOptimization Algorithm: Gauss-NewtonRequires computation of gradient
Fast convergence
Brown UniversitySlide6
Similarity Metric: SoCV
I0Ro
R
0
=100·ln(256-I
0)-300
Quantize images to 64 possible values
Each pixel in the image on the left corresponds to
a bin in the histogram (64 x 64 bins)
Notice the non-linear relationship between
I and RSlide7
Similarity Metric: SoCV
What happens if I0 is translated to the right? For each value of R, we have a range of values in I’Slide8
Similarity Metric: SoCV
Compute the conditional expectation/mean of this distribution Slide9
Replace each value in R with the conditional mean
Similarity Metric: SoCVSlide10
Optimization: Gauss-Newton
Goal: Find values of 3D rigid-body transform that minimize SSlide11
Initial Testing (Matlab MRI data)
Rotation aboutz –axis (25o)Slide12
So far…
Similarity metric: Sum of Conditional VariancesOptimization StepThe optimization step only converges for certain casesOptimization over 1 variable only (needs to be debugged)Testing with MRI data built into MatlabSlide13
Plan of Action
Brown UniversityFix optimization over all 6 parameters [rx r
y
r
z
t
x
t
y
tz
]
Test on a real data set
Implement the computationally efficient approach to this algorithm from their follow up paper
Test on real data set and compare computation speed to original
April 28 - May 2
May 2 – May 10
May 10 - May 14Slide14
References
A computationally efficient approach for 2D-3D image registration Haque, M.N.; Pickering, M.R.; Biswas, M.; Frater, M.R.; Scarvell, J.M.; Smith, P.N.; 2010 Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), Issue Date: Aug. 31 2010-Sept. 4 2010, On page(s): 6268 – 6271M. Pickering, A. Muhit, J.
Scarvell
, and P. Smith, "A new multimodal
similarity measure for fast gradient-based 2D-3D image
registration," in Proc. IEEE Int. Conf. on Engineering in Medicine
and Biology (EMBC), Minneapolis, USA, 2009, pp. 5821-5824.
Brown University