body and organ fat quantification advances in ct and mri

1

PRECONFERENCE WORKSHOP

Body and Organ Fat QuantificationAdvances in CT

and MRI Techniques

H. Harry Hu (University of Southern California) – organizer

Mark Punyanitya (Columbia University) – organizer

Bret Goodpaster (University of Pittsburg)

Steven Heymsfield (Pennington Biomedical Research Center)

Wei Shen (Columbia University)

Fat Quantification via Imaging

Punyanitya, ShenGoodpaster, Hu

DATA ACQUISITION & IMAGING METHODS

CT, MRI

DATA PROCESSING & IMAGE SEGMENTATION

Theoretical Concepts & Practical Implementation

ROLE OF IMAGING in OBESITY Heymsfield

2

Course Objectives• DESCRIBE

– … principles that underpin CT and MRI fat quantification techniques– … basis of signal contrast between lean and fat tissues / organs

• EMPHASIZE– … tradeoffs between techniques– … the most appropriate methodology for fat quantification of

• Subcutaneous adipose tissue• Visceral adipose tissue• Skeletal muscle fat• Organ fat (liver, pancreas, heart)

• HIGHLIGHT– … post-processing procedures for image segmentation and fat quantification– … advantages and disadvantages of manual and automated segmentation– … the importance of quality control, accuracy, and repeatability

Course Timeline

• 110-140pm (Steven Heymsfield)Importance of 3D Imaging in Obesity Research

IMAGING METHODS• 140-210pm (Bret Goodpaster)

X-ray Computed Tomography

• 210-240pm (H. Harry Hu)Magnetic Resonance Imaging

IMAGE SEGMENTATION• 240pm – 310pm (Mark Punyanitya)

Principles and Methodology

• 310pm – 340pm (Wei Shen)Practical Implementation

• 340pm - 400pm Q/A

3

Magnetic Resonance Imaging of FatPrinciples, Methodology, Utility, and Challenges

Houchun Harry Hu

Radiology & Electrical Engineering

Children’s Hospital of Los Angeles

University of Southern California

Slide / Content Contributors

International Society for Magnetic Resonance in Medicine

http://www.ismrm.org/

• Huanzhou Yu – GE Healthcare

• Shahid Hussain – U of Nebraska, Omaha

• Peter Kellman – NIH

• E. Brian Welch – Vanderbilt University

• Charles McKenzie – University of Western Ontario

• Dimitrios Karampinos – UC San Francisco

• Peter Börnert – Philips Healthcare

• Thomas Perkins – Philips Healthcare

• Lidia Szczepaniak – Cedar Sinai Medical Center

• Gavin Hamilton – UC San Diego

• Mark Bydder – UC San Diego

• Scott Reeder – U of Wisconsin Madison

• Walter Block – U of Wisconsin Madison

• Fritz Schick – University of Tubingen

• Jürgen Machann – University of Tubingen

• Joel Kullberg – Uppsala University

• Russell Low – Sharp’s and Children’s MRI Center

4

• MRI Basics - Principles– Signal origins

– Tissue signal contrasts

– Spectroscopy vs. imaging

– Chemical shift

– Confounding factors

• MRI of Fat (and Water) - Methodology– T1 weighting

– Selective excitation / suppression

– Spectroscopy

– Chemical shift imaging

– Lots of examples

Outline




– Chemical shift


Outline

5

1H

spin population,net magnetization = 0

most common MRI nuclei1H, 13C, 23Na, 31P

“a spin”with magnetic dipole

M in MRI ‐‐‐Magnetization

+ ‐

Mlongitudinal

precession

BO

M in MRI ‐‐‐Magnetization

alignment of spins,net magnetization ≠ 0

BOMlongitudinal

MO

~ Proton Density

z

1.5 Tesla, ~64 MHz

3 Tesla, ~128 MHz

Resonant (Larmor) Frequency

Greater BO, more magnetization BIGGER SIGNAL

6

Mlongitudinal rotated into the transverse plane to induce detectable signal in receivers

R in MRI ‐‐‐ Resonance

Mlongitudinal

transverse plane

Mtransverse

receiver coil wire loops

signal

z

x

y

α

α ‐‐‐ flip angle

{MO, T1, T2} are TISSUE SPECIFIC

and BO DEPENDENT;

SIGNAL EVOLVES WITH TIME.

α

transverse plane

MO

0

MO

Proton Density

R in MRI ‐‐‐ Relaxometry

decay

recovery

00

7

{TR, TE, flip angle}, {MO, T1, T2} tissue and image contrast

I in MRI ‐‐‐ Imaging

1H1H

RF receptionTissue releases energy via relaxationTissues behave differently ~ (T1, T2)

RF excitationTransmit energy into tissues

RFtime

signal

Pulse Sequence RF excitation RF reception

“an echo”

intrinsicoperator controlled

GMWMGMWM

30° 20° 10° 4°

T1 weightedShort TR, TE

Proton (MO)‐weightedLong TR, short TE

Light T2Long TR, mid TE

Heavy T2Long TR, TE

Tissue Contrast

8

TR Affects T1, TE Affects T2

Squares – white matterCircles – gray matterDiamonds – cerebrospinal fluid

• MRI signal intrinsically depends on (exponentially):

– Tissue MO (proton density), T1 (recovery rate), T2 (decay rate)

• Operator controls signal contrast by adjusting:

– TR, TE, flip angle (pulse sequence)

• MRI detects protons in free (unbound) water and lipids

• Bound water (macromolecules) and solids not detected (very short T2s)

Recap ‐ The MRI Signal

Generally, the nominal value has NO physical meaningDigitized number System dependentARBITRARY, no physical units

Relative tissue‐tissue signal contrast (brighter, darker)

9

Inverse Fourier Transform

image‐space

Complex DataMagnitude and Phase

Digitized DICOM

Magnitude onlyData compressed

PULSE SEQUENCETR, TE, voxel size, partial volume

k‐space trajectory

NOISE!!!!

k‐space

PHYSICS

SYSTEMmagnet field and gradient

imperfections, coil sensitivity, RF penetration … PHYSIOLOGY

tissue perfusion, blood oxygenation, …

OBJECT PARAMETERSproton densityT1, T2, velocity, area, diffusion, …

MODEL

Digitized DICOM

Magnitude onlyData compressed

The Complexity of MRI Signal & Quantification

How is Fat Quantified by MRI?

• Subcutaneous and Visceral Adipose Tissue

– Area, Volume

– Identify the adipose tissue (fat) voxels … sum up

– Does not explicitly depend on nominal value of fat signal

– Visual identification

– Convert to mass (DXA, chemical assay)

• Organs, Muscles

– Fat‐Signal Fraction = {F / (W+F)} x 100%

– Measure separately the fat (F) and water (W) signals (MO, T1, T2)

– Metric explicitly depend on nominal value of F and W signals

– Not volume fraction or mass fraction

– Proton‐density fat fraction

10



– Spectroscopy vs. imaging (very brief)

– Chemical shift (very brief)


Outline

MR Spectroscopy vs. Imaging

RFtime

signal

RFtime

signal

pulse sequence

spatial encoding{x,y,z} space

Fat (CH2)Water

NO spatial encoding{frequency} spacevoxel

11

Δf ~‐ 210 Hz at 1.5T‐ 420 Hz at 3.0T

Methylene (CH2) 1H in fat precesses SLOWER

To the receiver,

it appears to LAG

Chemical Shift H HO

Fat‐Water chemical shift is only ~102 Hz …

resonant frequency is 64 (1.5T) or 128 MHz

frequency

Fat (CH2)Water

“0” Hz

ON‐resonance

Signal receivers tuned to WATER proton

Larmor frequency

Δf

chemically shifted OFF‐resonance

Confounding Factor #1 –Magnet Inhomogeneity• Deviations in the magnetic field BO …

• (BO±δ)

– Imperfect magnet / hardware

– Air / tissue susceptibility

– Placing an object within the magnet

– … mismatch in true vs. assumed (modeled)

resonant frequencies

frequency“0” Hz Δf−δ −δ + Δf

Fat (CH2)Water

Fat (CH2)Water

Fat (CH2)Water

Fat (CH2)Water

Fat (CH2)Water

Fat (CH2)Water

Fat (CH2)Water

Fat (CH2)Water

Fat (CH2)Water

δ + Δfδ

Chemical‐shift is preserved!

250 Hz

‐250 Hz

500 Hz

‐500 Hz

L

K

field maps

12

Confounding Factor #2 –Single vs. Multi‐Peak Fat Model

Courtesy of Dr. G. Hamilton (UC San Diego)

For now, assume SINGLE

peak signal model.

Accounts for only 60‐70%

of the total fat signal.

Quantitative accuracy

requires MULTI peaks.




– Chemical shift




– Spectroscopy


Outline

MECHANISMS of

FAT vs. WATER (lean) TISSUE CONTRAST

13

Fat has shortest natural T1 in vivo, fast recovery.

MRI of Fat by T1‐WEIGHTING

On T1w, FAT IS

BRIGHTEST!

signal contrast

Whole Body T1‐weighted MRI (circa 2005)

Courtesy of Dr. J. Machann (Univ. Hospital Tubingen), J Magn Reson Imaging 2005:21:455-462.

LE UET

I II III IV V VI VII VIII

II IV VI VIIIII IV VI VIII

I III VIIVI III V VII

LE UET

I II III IV V VI VII VIII

II IV VI VIIIII IV VI VIII

I III VIIVI III V VII

14

Machann J et al., J Magn Reson Imaging

2005:21:455-462.

segmentation covered by next two speakers

• Principle

– Fat is brighter than other tissue due to shorter T1

• Advantages

– Fast and available on all MRI scanners

– Adipose tissue depots easily visualized

• Drawbacks

– Each voxel is either “all fat” or “no fat”

– Not suitable for organ fat quantification (e.g. fat fraction)

– Other structures can also be bright (e.g. blood, bowel content)

MRI of Fat by T1‐Weighting

THE TRADITIONAL WORKHORSE PULSE SEQUENCE FOR

SUBCUTANEOUS AND VISCERAL ADIPOSE TISSUE QUANTIFICATION;

MANY PUBLICATIONS;

LIKELY THE MOST COMMON MRI PROCEDURE IN TOS COMMUNITY

15

MRI of Fat by T1‐Weighting – Confounding Bright Signals

blood

T1

chemical shift Imaging(coming up)

bowel content

T1

chemical shift imaging

Courtesy of Dr. C. McKenzie

(U Western Ontario)

Alabousi et al., J Magn Reson

Imaging 2011:34:474-479.

Less VAT

Misclassification

SAT and VAT volumes

SINGLE fat peak model

2D and 3D

chemical shift

water or fat signal acquired

Frequency Selective Excitation /

Suppression MRI

SAT and VAT volumes

% fat‐signal fraction(fat:water ratio)

SINGLE and MULTIfat peak models

2D and 3D

chemical shift

water and fat signals are acquired

Chemical Shift MRI


MULTIfat peak model

single voxel

chemical shift


Single‐VoxelMR Spectroscopy

SAT and VAT volumes


2D and 3D

T1 relaxationsignal intensity


T1‐weighted MRI

QuantitativeEndpoints

TypicalDimensionality

Basis of Water‐Fat Differentiation

16

MRI of Fat by T2‐Weighting ?

T1‐weighted T2‐weighted

T2 weighted fat quantification not used.




– Chemical shift




– Spectroscopy


Outline

MECHANISMS of


17

Selective Water‐Fat Excitation / Suppression• Motivation

– Fat is bright on T1 and T2 weighted images.

– In clinical radiology, fat obscures pathology / confounds diagnosis

• Solution

– Develop pulse sequences to …

… suppress fat / excite water “fat sat” … mainly water signal remain

… suppress water / excite fat “water sat” … mainly fat signal remain

Water

Saturation RF pulsecentered on WATERresonant frequency

Fat (CH2) Water Fat (CH2)

Saturation RF pulsecentered on FAT

resonant frequency

Good for clinical workflow. Good for fat quantification.

Fat Sat

Water Sat

Selective Water‐Fat Excitation / Suppression

Machann J et al., Annual Reports on NMR Spectroscopy 2003:50:1-74.

Fat Sat

Water Sat

residual fat Signal

(why?)

no residualwater signal

18


water sat

☺

Looks similar to a T1 weighted image.

Fat is bright, muscle is dark.

T2‐weighted T2‐weighted fat sat

Water Fat (CH2)

Selective Water SuppressionWhy Does it Fail?Magnet Inhomogeneity (Revisited 1)

water sat

☺

BO inhomogeneity worsens with

higher field strengths.

SPATIALLY VARYING!

spectrum is shifted …

19

water sat

Peng Q et al., J Magn Reson Imaging

2005:21:263-271.

T1‐weighted

T1‐weighted water sat spectroscopy

W

F(CH2)

Machann J et al., Magn Reson Med2006:55:913-917.

• Principle

– Chemical shift; resonance frequency difference

– RF pulse is tuned to selectively excite/suppress one or the other

• Advantage

– Available on all MRI scanners

– Adipose tissue / lean tissue / organs easily separated

– Can detect fat presence in organs

• Drawbacks

– Can fail if there is large BO inhomogeneity

– Each voxel is still either “all fat” or “no fat” (SAT and VAT volumes)

– Not suitable for organ fat quantification


20

Fat Sat

Water Sat



Fat Sat

Water Sat

residual fat signal

no residualwater signal

‐HC=CH‐

SAT and VAT volumes


2D and 3D

chemical shift



Suppression MRI

SAT and VAT volumes



2D and 3D

chemical shift


Chemical Shift MRI


MULTIfat peak model

single voxel

chemical shift



SAT and VAT volumes


2D and 3D



T1‐weighted MRI




21




– Chemical shift




– Spectroscopy


Outline

MECHANISMS of


Single‐Voxel MR Spectroscopy (MRS)


Qayyum A. Radiographics 2009:29:1653-1664.

water peak

fat (CH2)

fatpeaks

22

• Principle

– Chemical shift

• Advantages

– Very high spectral frequency resolution

– Can detect very low amounts of fat

– Can detect ‐HC=CH‐ (unsaturation)

• Drawbacks

– Not always available on all MRI systems

– Usually single voxel

– Highly operator dependent; motion sensitive

– Post‐processing burden

MRS HAS BEEN THE GOLD‐STANDARD FOR FAT QUANTIFICATION IN

ORGANS (liver, pancreas, heart) and SKELETAL MUSCLES.

Single‐Voxel MR Spectroscopy (MRS)

area under fat peaksfat fraction (%) 100%

(area under fat + water peaks)= ×∑

∑

Magnet Inhomogeneity (Revisited 2)Line Broadening

water

As peaks get broader, area quantification becomes challenging.

Courtesy of Dr. F. Schick (Univ. Hospital Tubingen)

23

MRS at higher BO field strengths benefit from greater chemical shift and sharper peaks.

Ren, et al. J Lipid Research 2008:49:2055-2062.

Courtesy of Dr. G. Hamilton (UC San Diego)

T2 decay of water signal

T2 decay of fat signal

fat‐signal fraction =

Fat (signal, area)

Fat (signal, area) + Water (signal, area)

Fat {MO, T1, T2}

Fat {MO, T1, T2} Water {MO, T1, T2}

Fat (MO)

Fat (MO) + Water (MO)

Long TR, Short TE ‐> Proton Density Weighted

= proton‐density fat fraction

24

Courtesy of Dr. J. Machann (Univ. Hospital Tubingen), Diabetes, Obesity and Metabolism 2004:6:239-248.

IMCL

droplets

EMCL

striations

MRS of Skeletal Muscle Fat

Lingvay I, et al. J Clin Endocrinol Metab 2009:94:4070-4076.

MRS of the Pancreas

25

Lingvay I, et al. Circulation 2007:116:1170-1175.

Hammer, et al. J Clin Endocrinol Metab, 2008:93:497-503.

Van der Meer, et al. Radiology, 2007:245:251-257.

note the scale

MRS of the Heart

LV

4 chamber view

2 chamber view

Interventricular septum

myocardium

MRS Challenge: Motion

• MRS of abdominal organs (liver, pancreas, heart) typically requires

respiratory gating (bellows) and cardiac gating (ECG).

• Synchronize pulse sequence to natural physiological motion.

• Bulk motion: subject moves between voxel placement and scan start.

fatpeaks

water peak

Are these fat peaks real?

26

SAT and VAT volumes


2D and 3D

chemical shift



Suppression MRI

SAT and VAT volumes



2D and 3D

chemical shift


Chemical Shift MRI


MULTIfat peak model

single voxel

chemical shift



SAT and VAT volumes


2D and 3D



T1‐weighted MRI







– Chemical shift




– Spectroscopy (1D chemical shift)


(“Dixon”method)

Outline

fat‐separated water‐separated fat fraction (%)

27

“Dixon” Method (2‐point) Fat‐Water MRI

z

x

y

t = 0 - t = 0 + … … …oppose‐phase (OP)= W ‐ F

Fat is precessing slower, accrues phase periodically with respect to water.

{W,F}θ =2π(Δf)(TE)

θ{W,F}

… … …

in‐phase (IP) = W + F

W F

2 ( )( )[ ]n

n

i TE fTES F eW π Δ= + ⋅

IPθ=0°= W + FOP θ=180°= W- F

fat only water only

F W

no difference

difference‐> fatty liver

28

OP = W‐F IP = W+F

Opposed‐Phase In‐Phase

η =F

W + F=

IP − OP( )2 IP

0%

50%

Courtesy of Dr. S. Reeder, U. Wisconsin Madison.

“Dixon” Method (2‐point) Fat‐Water MRI

34 %fat‐signal fraction

e.g. 34% of the liver signal originates from fat

1984 2‐pt. methods

1990 ‐ 19913 and 4‐pt. methods

Mid‐1990s, early‐2000Extensive clinical useSubtle variations

2003 – present6‐pt. methods

“Dixon” Method Evolution

• General Electric

– Lava‐Flex (2 pt.) – commercial, IDEAL (3‐6 pt.) – commercial + research

• Philips

– mDixon (2 pt.) – commercial, (3‐6 pt.) – research

• Siemens

– Dixon (2‐3 pt.) – research

towards method robustness and

fat fraction accuracy

Fat signal modeled with single (CH2) peakIP, OP, magnitude‐base

multi‐peak fat signal modelcomplex‐base

use in liver, heart

29

*2/2 2(( ) )ki t i t Ttf

ka es t W eF e π ψπ −+= ∑

Generalized 6+ pt. Dixon Method

fat chemical shift term

magnet non‐uniformity term

signal relaxation

water fat

measured MRI signal

Σ multiple fat peaks‐CH2‐

‐HC=CH‐‐CH3

complex unknown (2)

complex unknown (2)

real unknown (2)

2 ( )( )[ ]n

n

i TE fTES F eW π Δ= + ⋅

Generalized 6+ pt. Dixon MethodFat

Water *2/2 2(( ) )ki t i t Ttf

ka es t W eF e π ψπ −+= ∑

. . .

Water

Fat

IP = W+F

OP = W‐F

Fat Fraction (%) T2* map (ms)BO Inhomogeneity

Field Map(Hz)

iterativealgorithm

QUANTITATIVE !!!

30

Generalized 6+ pt. Dixon Method ‐ Advantages

Courtesy of Dr. S. Reeder, U. Wisconsin Madison, J Magn Reson Imaging 2011:34:729-749.

Full 0‐100% fat fraction scale

Whole organ assessment

Courtesy of Dr. S. Reeder, U. Wisconsin Madison, J Magn Reson Imaging 2009:29:1332-1339.

Single vs. Multi Peak Fat Model

Generalized 6+ pt. Dixon Method ‐ Advantages

31

Magnet Inhomogeneity (Revisited 3)Water‐Fat Signal Swap

Courtesy of Dr. R. Low Courtesy of S. Sharma

(June 2010‐LEFT, 24.9% in liver), (April 2011‐RIGHT, 3.8% in liver)

Color bar is set from 0 (0%) to 1000 (100%) fat fraction

32

3D “Dixon” Imaging

Courtesy of Dr. P. Kellman, National Institutes of Health, Curr Cardiovasc Imaging Rep 2010:3:83-91.

Myocardial Lipodystrophy

Pericardial Fat

3D “Dixon” Imaging of Heart

33

fattyinfiltration

MI

Courtesy of Dr. P. Kellman, National Institutes of Health

Lipoma

11yr male

Courtesy of Dr. J. Kullberg, Uppsala University, J Magn Reson Imaging 2009:30:185-193.

34

Comprehensive Fat Quantification with 3D “Dixon” Imaging

Water‐separated Fat‐separated In‐Phase Opposed‐Phase % fat fraction

SAT and VAT volume

Looks similar to heavy T1w or water‐saturated image

organ fat fraction content

numerically meaningful

anatomical guidance

Chemical Shift Imaging

• Principle

– Chemical shift

• Advantages

– Data yields a comprehensive set of images

• Fat‐only, Water‐only

• In‐phase (W+F), Opposed‐phase (W‐F)

• Fat‐Fraction, Water‐Fraction (full 0‐100% range)

– Ability to generate adipose tissue volumes and organ % fat content

• Drawbacks

– Longer scan time

– Not widely available (commercially) … yet

35

SAT and VAT volumes


2D and 3D

chemical shift



Suppression MRI

SAT and VAT volumes



2D and 3D

chemical shift


Chemical Shift MRI


MULTIfat peak model

single voxel

chemical shift



SAT and VAT volumes


2D and 3D



T1‐weighted MRI




36

Emerging Application – Double Bond Mapping with “Dixon” CSI

Potential to assess different

‘ndb’ in adipose tissue

depots and organs

Courtesy of Dr. M. Bydder, UC San Diego.

Magn Reson Imaging 2011, in press.

‐CH=CH‐

• MRI is very flexible in highlighting fat

– T1 weighting


– Spectroscopy


– … new contrast mechanisms are in development

• There are tradeoffs to each method

• Commercial availability

• Scan time (speed)

• Quantitative endpoint (volume, fat fraction)

• Required level of operator expertise / post‐processing burden

Take Home Message

SPATIAL RESOLUTIONpartial volume effects

2D multi-slice or 3D volume

RESPIRATIONbreath-hold

motion artifacts

SCAN TIMEsignal-to-noise ratio

subject comfort

Method?2D or 3D?

TR? TE? Flip angle?Spatial resolution?Slice thickness?

37

Chemical Shift Imaging (is here to stay)

Vol 34:729-749.October 2011

January 2010

http://ismrm.org/workshops/FatWater12/

FACULTY

W. Tom Dixon ‐ GE Healthcare

Gary Glover ‐ Stanford

Shahid Hussain ‐ U Nebraska Omaha

Jeffrey Schwimmer ‐ UC San Diego

Qing‐San Xiang ‐ U British Columbia

Holger Eggers ‐ Philips Healthcare

Walter Block ‐ UW Madison

Catherine Hines ‐Merck

Mark Bydder ‐ UC San Diego

Angel Pineda ‐ CSU Fullerton

Russell Low ‐ Sharp & Children’s MRI

John Wood ‐ Children’s Hospital LA

Richard Bergman ‐ Cedar Sinai

Gavin Hamilton ‐ UC San Diego

Lidia Szczepaniak ‐ Cedar Sinai

Chris Boesch ‐ U of Bern, Switzerland

Fritz Schick ‐ U of Tubingen, Germany

Michael Goran ‐ USC

Jürgen Machann ‐ U of Tubingen, Germany

Rosa Branca ‐ UNC Chapel Hill

38

[email protected]@chla.usc.edu

http://mrel.usc.edu/

Courtesy of Dr. P. Bornert, Philips Healthcare

Magnet strength measured in TESLA. 1 Tesla = 10,000 GAUSS. Earth’s pole fields are only 0.3‐0.6 GAUSS !!!

Tesla Gauss

~90021

~60014

~50011.7

~4009.4

~3007

~2004.7

~1283

~641.5

~431

Frequency (MHz)Tesla

Most common clinical for humanFDA approved

Non FDA approved for clinical humanResearch only

Small Animal

Greater BO, more magnetization BIGGER SIGNAL

39

fat water

Dixon: 22.9%MRS: 23.4%

Dixon: 4.4%MRS: 3.1%

water peak

fat (CH2)

fatpeaks

body and organ fat quantification advances in ct and mri

Documents