TEMPLATE DESIGN © 2008

www.PosterPresentations.com

D-CSTAR - Data domain CSTAR

•The objective is to jointly estimate PET activity image

{f1,1}and respiratory motion parameters (βr)

•The relationship between the observation and the

unknowns can be represented as :

•Cost functional:

•PML Cost functional comprises the following terms :

• Data Fidelity term

• Spatial Smoothness Penalty term:

•Respiratory motion penalty term:

Cardiac PET/CT Imaging: Joint reconstruction

and cardiac and respiratory motion correction

Sonal Ambwani, W.C. Karl, Homer Pien

ambwani@bu.edu, wckarl@bu.edu, hpien@partners.org

Abstract

Coronary artery disease (CAD) such as atherosclerosis is the leading

cause of death in industrialized nations.. Accurate assessment,

characterization and localization of this disease through non-invasive

methods is an important step towards the treatment of CAD. It has been

shown that positron emission tomography (PET) is capable of detecting

large vessel inflammation via activated macrophage uptake of FDG.

However, respiratory and cardiac motion during image acquisition leads to

severe blurring of the resulting images thereby rendering the spatial

resolution inadequate for detection of inflammation in coronary arteries.

The objective of this paper is to demonstrate the potential of producing high

resolution PET images to enable imaging of coronary artery inflammation.

In this poster, we demonstrate a novel method for joint super-resolution

reconstruction of coronary PET images and cardiac + respiratory motion

correction Our algorithm features the use of all acquired data for SNR

preservation and enhancement of resolution of PET. Breath-hold CT is

primarily used for cardiac motion estimation. This knowledge of cardiac

motion is incorporated in the framework that iteratively estimates PET

activity images and respiratory motion. We investigated the feasibility of this

technique on simulated cardiac PET/CT data using XCAT and the

preliminary results show a marked qualitative and quantitative improvement

when compared to conventional PET reconstruction.

Motivation

(I) Proposed Algorithm: Image based

Main features

XCAT Simulation Results

(II) Proposed Algorithm: Projection based

Optimization Algorithm

• Alternating minimization (AM).

• Iteratively alternates between estimates of activity

image and respiratory motion.

Simulation results continued ..

References

(1)Ambwani S., Cho, S., Karl, W. C., Tawakol, A. and Pien, H., A Feasibility study of joint

respiratory and cardiac motion correction for coronary PET/CT imaging. ISBI , july 2009, pp.

935-938.

(2)Ambwani S., Karl, W.C., Tawakol, A. and Pien, H, Joint cardiac and respiratory motion

correction and super-resolution reconstruction in coronary PET/CT. To be presented in

ISBI, March 2011.

(3)Martinez-Moller, A., Zikic, D., Botnar, R.M., Bundschuh, R.A., Howe, W., Ziegler, S.I.,

Navab, N., Schwaiger, M. and Nekolla, S.G., Dual cardiac respiratory gated PET:

Implementation and results from a feasibility study. Eur. J. Med. Mol. Imag. 2007

(4)Slomka, P.J., Nishina, H., Berman, D.S., Kang, X., Akincioglu, C., Friedman, J.D.,

Hayes, S.W., Adadl, U.E. and German, G., Motion-frozen display and quantification of

myocardial perfusion, J. Nucl. Med., 2004

• Coronary Artery Disease (CAD) or Atherosclerosis

is the leading cause of mortality in industrialized

nations.

• Positron Emission Tomography (PET) is a non-

invasive imaging modality that provides the

necessary functional information needed to detect

such plaque. Computed Tomography (CT) images

reinforce PET information with high resolution

anatomical information.

(a) (b) (c)

Stages of Atherosclerosis (Ross, R.

N. Engl. J. Med 1999)

(a) Early: Inflammation at the site.

(b) Moderate: Deposition of lipids,

collagen, calcium etc; Plaque formation.

(c) Advanced : Rupture of plaque.

Inflammation vs. FDG

uptake (Tawakol et al, JACC

2006)

• Evidence that there is a

direct correlation between the

radiotracer absorption and

atherosclerotic inflammation

and macrophage concentration.

Problem Statement

PET image CT image PET/CT fusion.

Black arrow indicates the myocardium. Red arrow indicates stenosis in a coronary vessel.

• PET listmode data is binned into cardiac

phases/time-frames:

• Summing across rows yields:

• Inter-frame cardiac motion; Intra-frame respiratory motion.

Niiy ,,2,1}{

,,,,,,,, 2121 NN CCCCCC

Cycle R1 Cycle R2

• Image based analysis: Post-reconstruction

processing.

• Use of all acquired data, resulting in SNR

preservation.

• Sequential correction of cardiac and respiratory

motion.

• The recovery of the underlying HR image can be

stated as the following inverse problem :

Where S = Subsampling matrix, Bpsf = Blurring matrix due to the imaging system, Wcard

= Cardiac Motion matrix, Bresp = Matrix representing motion blur due to patient’s

respiration. Y = vector of stacked LR images, x = unknown HR image and η =

Gaussian noise vector.

• PML Cost functional :

Explicit

correction for

cardiac motion

in a super-

resolution

framework

Implicit

correction for

residual

respiratory

motion blur.

Estimation of cardiac motion via

optical flow using Breath-hold CT-AC images

Blind deconvolution to solve

for LTI respiratory blur.

Niiy ,2,1}{ x̂

High level notional diagram .

(Estimated HR image)

xMBSBxBWSBY RpsfrespcardPSF 0

2

22

2

201

2

200 ||||||||||||),( resprespcardpsfresp DbDxYxBWSBbxL

Extended Cardiac Torso (XCAT) simulation.

Photon count rate : 1 million/second

Experiment 1: FBP CSTAR

Reference Slice Conventional PET Cardiac Motion removed

Both motions removed

Reference Slice Conventional PET Cardiac Motion removed

Both motions removed

Experiment 2: ML-EM CSTAR

5D Binning of PET

Listmode data

1,1, )()(][ fTHTgE krrk

ikr

k r i

irkikrrrkdata fTHTgfTHTfgJ ])()(log[][])()([}){,},({ 1,1,1,11,1,

1 step toGo : 3 Step

}}){,(minarg}ˆ{ :2 Step

}){,(minargˆ :1 Step

}ˆ{ and ˆ Initialize :0 Step

)1(

1,1}{

)1(

)(

1,1

)1(

1,1

)0()0(

1,1

1,1

r

nn

r

n

rf

n

r

fJ

fJf

f

r

XCAT Simulation Results

Experiment 1: 3D plaque studies and comparison

with peer methods.

• Simulated plaque of size 4 x 4 x 3 mm.

• Plaque activity to left ventricle activity ratio = 4:1

• Inserted respiratory and cardiac cycle irregularities for

realistic simulation.

Experiment 2: Studying spatial regularization with

background removed.

This work was supported in part by CenSSIS, the Center for Subsurface

Sensing and Imaging Systems, under the Engineering Research Centers

Program of the National Science Foundation

(a)

Reference Cardiac Sequence.

(b)

JRMC: L2 regularization.

Plaque recovery: 36%

(c))

JRMC: TV spatial regularizer.

Plaque Recovery: 59%.

(d)

JRMC: TV + L1

Plaque Recovery: 67%

100CR

CR1Recovery Plaque

1 original

tedreconstruc xK

K

k

CSTAR - Cardiac Shape Tracking and Adjustment for Respiration.

})({)(}){,},({[minarg})ˆ{,ˆ( 1,11,1,}{,

1,11,1

rmotionsmoothrrkdataf

r JfJfgJfr

1

11,11

1

11,101,1 ||||||||)( DfffJ smooth

r

rrrmotionJ 2

12 ||||})({

Reference Slice

(axial view)

Conventional ML-EM

(no motion correction)

Motion-frozen method

(Slomka et al).

Dual-gated approach

(Martinez-Moller et al)

D-CSTAR method.

Reference Slice

(sagittal view)

D-CSTAR method.

Comparison of line-profiles.

Table 1. Comparison based on Plaque

Recovery (PR) percentage and

Sum of Squared Differences (SSD).

Clinical PET/CT data

PET/CT overlay with stent.

•Definite presence of plaque.

•Blurring of PET signal.

Objective Proposed algorithm flow.

• Siemens Biograph-64 PET/CT machine.

• Understanding the forward model.

• Deciphering the raw PET data (listmode) and

generating dual-gated projection data.

• Data correction methods and choice between pre-

correction and model-based correction.

Conclusions

• Two novel methods for increasing the conventional

resolution in cardiac PET images to facilitate detection

of arterial inflammation.

• Image based – CSTAR.

• Data/Projection based – D-CSTAR.

• Successful demonstration on simulations.

• Applications in clinical PET/CT.