CT and MR Imaging of Abdominal Aortic Aneurysm

R O C I O C A B R E R A G U I L L A U M E L E M A I T R E

M O J D E H R A S T G O O

CT and MR Imaging of Abdominal Aortic Aneurysm

2

Presentation Outline

Introduction to Abdominal Aortic AneruysmsComputed Tomography of AAA

Imaging Technique Image Processing

Level Set Methods Active Shape Models

Magnetic Resonance of AAA Imaging Techniques

Galodinium-enhancement Diffusion weighted

Image Processing Markovian Method Graph-Theoretic Approach

Conclusions

1/13/2011Biological Basis of Medical Imaging - CT and MR Imaging for Abdominal Aortic Aneurysms

3


Aneurysm Vascular pathology consisting of an

irreversible dilation of a segment of a blood vessel

Abdominal aorta Continuation of the thoracic aorta

and begins at the level of the diaphragm

Largest artery in the abdominal cavity

Abdominal Aortic Aneurysm Accepted criterion: 50% increase in

vessel diameter

IntroductionWhat is an Abdominal Aortic Aneurysm?

4

CT Imaging

CT has been used widely in AAA CT Imaging speedCT provides detailed quality for better analysis of aneurysm and

adjacent arteriesDetailed information of aorta and its branches for 3D

reconstruction Flexible to different post processing methods Appreciated in surgical planning


5


Level Set Method (LSM) is a numerical method for tracking interface a shapes . More specifically for shape varying objects

Level set function plane

CT Image ProcessingAAA Segmentation through level set method

6



Defining the mesh on the object (Level Set function )

Updating the mesh values using the Speed function Multi resolution Analysis – LR( half) volume

Top – Down , Narrow band update restricted to the zero level set

Scaling up the result

Repeating the algorithm on full data set 2 - 3D reconstruction – Using Marching cubes

1 - Segmentation the volumes

Force at mesh point (x,y)

Advection term = 1 Curvature force (x’ ,y’) Image force based

on the Gaussian derivative filter

),(,,0 )),((),( yxFyxcyx

ieyxFFyxF

7


Advantage: Level set method has the

advantage to provide more accurate results specially in segmenting the small details

Disadvantage It has a very high computational

cost Suggestions

Combination of the methods , while level set can be used to improve the initial segmentation results


8


CT Image ProcessingAAA Segmentation using Active Shape Model

Active shape model (Smart Snake) was developed by Cootes et al. in order to over come the problems with snake segmentation Active counter model (Snake) segmentation depends on the initial snake Active counter model is not capable to deal with the occluded objects

In Medical Imaging ASM is applied on the combination of shapes and gray level sets

9



Shape Modeling

Shape alignment

Statistical Computations PCA

Aligning the images in the same reference axes , using Procrustes Analysis Procrustes Analysis minimize the distance between reference shape and each shape in the

dataset

Modeling the shape variations Computation of the mean shape Computation of the scatter matrix Sorting the eigenvectors and keeping the first k eigenvectors , based on the largest

eigenvalues Eigen decomposition of the shapes where ,

Value of k is based on

bxx

N

iixN

x1

1

N

i

Tii xxxx

Ns

1

))((1

1

i

},...,{ 21 k

k

ii

N

ivi f

11

10 vf

10



Shape Modeling

bxx

13 23 33

43 53 63

M. Bruijne ,B. van Ginneken, M. A. Viergever, W. J. Nieesen, « Interactive Segmentation of Abdominal Aortic Aneurysms in CTA Images », 2004

11



Grey level Appearance Modeling

Sum of absolute difference between the reference and sample image over several resolutions

The sum is performed for each landmarks for a defined window size

nt kk

irs xIxIAD

1

)()(


12



Model Fitting

First slice manually Initialized while for the others previous counter was considered as initialization

Performing the multiresolution analysis for higher accuracy


13


Gadolinium injection (paramagnetic CA) Shortens the T1 relaxation time of blood,

distinguishing it from its surroundings No known side effects nor nephrotoxicity

Prince et al. [2] reported an agreement in measured AAA size in CT, MR and US

Orta et al. [1] reported its use to diagnose inflammatory AAA Hyper-intensity surrounding the aorta Region ADC = 1.24 x 10-2 mm2/s

ADC consistent with a restricted diffusion due to inflammation

Gadolinium-enhanced MRI Diffusion Weighted MRI

MR Imaging of AAAAlternate MR Imaging Approaches

14

X – 2D [MxN] random field that models the segmentation labels Y – 2D [MxN] random field that models the input image s – site (pixel) located at position (i,j) MAP – Maximum A Posteriori Probability, through the Bayes Rule

Assuming Gibbsian distributions

A pixel s will switch classes if and only if at least one of its neighbors has already been assigned the new class label


MR Image ProcessingMarkovian-Active Contour Segmentation

Likelihood Energy.Natural Logarithm of a Gaussian

15


Direct extension to 3D and 4D 3D – Neighbouring sites in k+1 and k-

1 images 4D – Neighbouring sites in t+1 and t-1

time frames

AAA reconstruction from MRI Initialization done by expert hand Seed growth until convergence

4D segmentation results

MR Image ProcessingMarkovian-Active Contour Segmentation

16


Aortic Surface Pre-segmentation Fast marching level set method used to

compute {appSt}tє[0,N-1]

Centerline Extraction Centerline determined from each approximate

surface by skeletonization

Accurate Surface Segmentation Weighted graph G = (V; E)

V – node set of image pixels E – arc set of neighbourhood system Every arc ‹vi, vj› є E has a cost

Graph-cuts aim to partition a weighted graph into 2 disjoint subsets

Minimize the cost function ε(f)

Appropriate design of a energy function, a minimum s-t cut can segment a region of interest in an image.

MR Image ProcessingGraph-Theoretic Segmentation

17

Conclusions

Strong interest in exploiting the capabilities of the medical imaging modalities to diagnose AAA Choice of the imaging modality

CT Modality of choice in most institutions Appropriate for emergency patients Use of an iodinated contrast medium and ionizing radiation

MR Appropriate to detect inflammatory AAA Does not employ ionizing radiation Contrast agent is appropriate for patients with renal insufficiency

Image processing techniques 2 CT and 2 MR methods have been presented Most methods rely on segmentation of the aorta and measurement of the vessel diameter Each methods could be extended to the other modality, but not much research has been done on it

It would be interesting to perform a study in which the several segmentation methods are used on CT and MR in order to evaluate the methods and the imaging modalities in a better way.


18

References

[1] Orta, K. and Kilickesmez, O. Clear Depiction of Inflammatory Abdominal Aortic Aneurysm with Diffusion-Weighted Magnetic Resonance Imaging. Cardiovascular and Interventional Radiology. (2010) 33:379-382.

[2] Prince, M. et al. Gadolinium-enhanced Magnetic Resonance Angiography of Abdominal Aortic Aneurysms. Journal of Vascular Surgery. (1995) Volume 21. Number 4.

[3] Jodoin, P. et al. Markovian Method for 2D, 3D and 4D segmentation of MRI. (2008) 15th IEEE International Conference on Image Processing. [4] Sonka, M. et al. Early Detection of Aortic Aneurysm Risk from 4D MR Image Data. (2006) Computers in Cardiology.[5] Kang Li et al. Optimal Surface Segmentation in Volumetric Images - A Graph-Theoretic Approach (2006) IEEE Transactions on Pattern Analysis and Machine

Intelligence. Volume 28. Number 1.[6] Crawford, C. et al. Abdominal Aortic Aneurysm: An Illustrated Narrative Review. (2002) Journal of Manipulative and Physiological Therapeutics. Volume 26.

Number 3. [7] Marleen de Bruijne and Bram van Ginneken and Max A. Viergever and Wiro J. Niessen. Interactive Segmentation of Abdominal Aortic Aneurysms in CTA

Images. 2004.[8] TF. Cootes and A. Hill and C.J.Taylor and J.Haslam and Manchester M Pt. The Use of Active Shape Models For Locating Structures in Medical Images. 1994.[9]T.F. Cootes and C.J. Taylor and Manchester M Pt. Statistical Models of Appearance for Computer Vision. 2000.[10] Cootes, T. F. and Taylor, C. J. and Cooper, D. H. and Graham, J. Active shape models - their training and application. Comput. Vis. Image Underst. 1995. Volume

61. Issue 1. Pages 38-59. [11] A. Hill and A. Thornham and C. J. Taylor. Model-Based Interpretation of 3D Medical Images. In British Machine Vision Conference. 1993. BMVA Press. [12] Michael Kass and Andrew Witkin and Demetri Terzopoulos. Snakes: Active contour models. International Journal of Computer Vision. 1988. Volume 1.

Number 4. [13] Derek Magee and Andrew Bulpitt and Elizabeth Berry. Level Set Methods for the 3D Segmentation of CT Images of Abdominal.[14] Steven C. Mitchell and Boudewijn P. F. Lelieveldt and Rob J. van der Geest and Hans G. Bosch and Johan H. C. Reiber and Milan and Milan Sonka. Multistage

Hybrid Active Appearance Model Matching: Segmentation of Left and Right Ventricles in Cardiac MR Images. IEEE Transactions on Medical Imaging. 2001. Volume 20.

[15] Wikipedia. Level Set Method. December 2010. $http://en.wikipedia.org/wiki/Level\_set\_method$[16] Wikipedia. Isosurface. December 2010. $http://en.wikipedia.org/wiki/Isosurface$[17] Wikipedia. Marching Cubes. December 2010. $http://en.wikipedia.org/wiki/Marching\_cubes$[18] George G. Hartnell. FRCR, FACC. Imaging of Aortic Aneurysm and Dissection: CT and MRI. Journal of Theoretical Imaging. 2001.


19

Appendix - AAA Segmentation through level set method


Defining static and evenly spaced mesh in the image using Mesh values are updated using the speed functionSpeed function F depends on :

Advection term (constant value) Curve term based on zero level set Image term (Based on the edges)

Updating the curve and Image terms based on nearest neighbor in the zero level set !! Computationally expensive

Reducing the computational cost by updating the mesh restricted to the zero level set area ; Narrow band with the level set method

01 tFt

20


Elastin Responsible for the elastic recoil of

the arteries due to the pulsatile blood flow

Degradation of elastin fibres will shift load to collagen fibres Elasticity decrease Diameter increase Aortic rupture

Genetic and environmental factors contribute to AAA development

Great clinical importance to determine aortic diameter

EnvironmentAtherosclerosisSmokingInflamation

Genetic Factors

Elastase activity increase vs. Inhibition decrease

Elastin Destruction

Failure of Elastin

Increased load on collagen

Failure of Collagen

Aneurysmal Dilation

Rupture

Ageing

HypertensionCollagen

remodelling

Alteration in vessel geometry

IntroductionPathogenesis of an Abdominal Aortic Aneurysm

21

Traditional MRI Widely used to retrieve anatomical and physiological information of

patients

Vascular imaging limitations Flow artifacts generated in different pathologies

Aneurysms – slow, swirling flow Stenotic vessels – turbulent flow


MR Imaging of AAATraditional MRI limitations on Vascular Imaging

Documents

CT and MR Imaging of Abdominal Aortic Aneurysm