Dual Energy CT
for Density Measurements
Norbert J. Pelc, Sc.D.
Departments of Radiology and
Bioengineering
Stanford University
What is in a voxel?
Acknowledgements
Souma Sengupta, Uri Shreter and Robert Senzig: GE
Healthcare
Thomas Flohr, Bernhard Schmidt and Karl Krzymyk: Siemens
Healthcare
Gretchen House, Mark Olszewski and Michael Decklever:
Philips Healthcare
Dominik Fleischmann, Carolina Arboleda and Adam Wang:
Stanford University
Cynthia McCollough: Mayo Clinic
Attenuation coefficients
depend on photon energy
CT number depends on:
• inherent tissue properties
(chemical composition,
density)
• x-ray spectrum
• administered contrast media
attenuation coefficient
100
?
10
iodine
1
bone
water
0.1
35
55
75
95
115
135
photon energy
Can we be more specific?
www.uhrad.com/ ctarc/ct153b2.jpg
use CT measurements at
multiple energies for material
specificity and improved
quantitation
Motivation
“Two pictures are taken of the same slice,
one at 100 kV and the other at 140 kV...so
that areas of high atomic numbers can be
enhanced... Tests carried out to date have
shown that iodine (Z=53) can be readily
differentiated from calcium (Z=20)”.
G.N. Hounsfield, BJR 46, 1016-22, 1973.
Genant et al, Inv Radiol, 1977.
Single energy CT:
“water” beam hardening
correction
Beam hardening
14
Accurate for “water-like”
materials of any density
14
12
water
12
10
Methods:
MLE is accurate
polynomials are fast:
p = a1L + a2L2 + a3L3 + …
10
ln(I0/I)
8
6
ln(I0/I)
water
8
6
4
2
4
Provides accurate densities if
Zeff is known and uniform
2
0
0
2
4
6
8
line integral
10
12
14
0
0
2
4
6
8
10
line integral
12
14
Single energy CT:
“Water” beam hardening
correction
Beam hardening
14
Correction fails with
materials of very different
Zeff
water
10
ln(I0/I)
Results in local and distant
errors and artifacts
12
8
calcium
6
4
2
0
0
2
4
6
8
10
12
14
80kVp
line integral
Courtesy of Uri Shreter
GE Healthcare
Single energy CT:
“Water” beam hardening
correction
140 kVp Conventional
CT Hoxworth
Courtesy of Dr. Joseph
Mayo Clinic Arizona
Courtesy of Uri Shreter
GE Healthcare
OUTLINE
14
Correction fails with
materials of very different
Zeff
12
water
ln(I0/I)
10
Results in density errors and
artifacts
8
calcium
6
4
2
Multi-pass corrections can
help if materials and
distribution are known
Can we do better?
0
0
2
4
6
8
10
line integral
12
14
• Physical principles of multi-energy x-ray
measurements
• Signal processing
• Quantitation opportunities and challenges
• Data acquisition
• Summary
I01 I02
energy E1 E2
water
bone
tb
unknown thickness
two known materials
100.0
E1
I01 I02
E2
energy E1 E2
10.0
water
linear attenuation coefficient (cm-1)
linear attenuation coefficient (cm-1)
unknown thickness
two known materials
cortical
bone
water
1.0
bone
tb
0.1
10
20
50
100
200
100.0
E1
water
I1 = I01 e-(w1tw+ b1tb)
I2 = I02 e-(w2tw+ b2tb)
solve for tw and/or tb
dual energy x-ray
absorptiometry
(DEXA)
cortical
bone
1.0
•
•
0.1
10
photon energy (keV)
I1 I2
E2
10.0
20
50
100
200
photon energy (keV)
I1 I2
tb = A{ ln(I01/I1) - (w1/w2)(ln(I02/I2)}
scale for lost
bone signal
subtract
water
makes the water contribution
at E2 match that at E1
material analysis with
absorptiometry
• 2 energies
2 materials
• can we generalize this? N energies for N
materials?
• limitation: two strong interaction
mechanisms
2 energies
materials
2
Compton scattering and photoelectric
absorption
Barring a K-edge in the spectrum, the energy
dependence of each is the same for all
elements!!
basis material decomposition:
• barring a K-edge:
Basis material decomposition
1000
any material acts like a combination of pure photoelectric and
Compton
any material can be modeled as a weighted sum of two other
materials
100
Cu
O
Ca
Cu
Ca'
10
Ca
1
.61*O + .04*Cu
O
0.1
0
Basis material decomposition
20
40
60
80
100
120
140
basis material decomposition:
• barring a K-edge:
I0
.04 M grams of
Cu
.61 M grams of
O
I0
M grams of Ca
=
I
I
Indistinguishable at any x-ray energy above their Kedge
Common “basis materials”:
iodine and water, aluminum and plastic
any material acts like a combination of pure photoelectric and
Compton
any material can be modeled as a weighted sum of two other
materials
“basis material” decomposition
in any projection measurement, we can only isolate two
materials
• Material parameters:
effective atomic number and electron density
amounts of two basis materials
can be measured using 2 x-ray energies
linear attenuation coefficient (cm-1)
K-edge subtraction
100.0
• reconstruct images in the normal
manner, and combine HU images
E1 E2
10.0
water
cortical
bone
easy to implement
iodine
• combine projection data prior to
reconstruction
1.0
0.1
10
20
50
100
200
photon energy (keV)
Very specific material information
2 narrow spectra, or 3 spectra
Dual energy processing
~80 kVp
iodine contrast
“water” contrast
Dual-energy processing
somewhat more difficult
requires aligned projections
enables “exact” beam hardening correction
Dual energy processing
~80 kVp
monochromatic 55 keV simulation
comparable to ~ 80 kVp
~140 kVp
iodine contrast
“water” contrast
monochromatic simulation
comparable to 80/150 kVp
Dual energy processing
~140 kVp
100
attenuation coefficient
~80 kVp
E2
~80 kVp
~140 kVp
iodine image
(water cancelled)
water image
(iodine cancelled)
10
iodine
1
water
0.1
35
iodine image
(water cancelled)
E1
Dual energy processing
55
75
photon energy
Iodine image = a • (Image80 - b •
Image140)
restore reduced
iodine signal
95
Water image =
c • (Image80 - d •
Image140)
amplify 140 kVp
water signal
Virtual non-contrast (VNC) image
Dual energy processing
~80 kVp
~140 kVp
optimal combination
(“mixed” image)
iodine CNR=7.9
iodine CNR=3.8
iodine CNR=10
water SNR=67
water SNR=71
iodine image
water image (VNC)
combined image
has high SNR
material cancelled images
have increased noise
SNR=3.4
SNR=37
Noise depends on dose
allocation
Noise
~80 kVp
~140 kVp
iodine image
Iodine image =
a • (Image80 - b • Image140)
iodine image
SNR=3.4
water image
SNR=37

=
a2
•
(80
+
b2
•
140)
depends on dose
allocation to the 80 kVp
and 140 kVp images
water image
dose allocation that
maximizes iodine SNR
SNR=3.4
SNR=37
iodine image
water image
80 kVp dose, 140 kVp
dose
same total dose
SNR=1.3
SNR=34
Dual energy processing
Dual energy Basis material decomposition
ethanol
LL = ln(I0L/IL), LH = ln(I0H/IH)
mA, mB = amounts of basis materials
acrylic
With monochromatic beams, L’s are linear functions of m’s, so
mA = a0LL + a1LH, mB = b0LL + b1LH
With polychromatic beams, functions are nonlinear. Approaches:
1) iterative solutions (e.g., MLE)
accurate but slow
2) polynomial approximation
mA = a0LL + a1LH + a2LL2 + a3LH2 + a4LLLH+…
accuracy depends on polynomial order, dynamic range, etc.
Dual energy Basis material decomposition
water image
Iso-intense
@ 140 kVp
acquire
80 kVp
calcium image
ethanol + NaCl
Iso-intense
@ 80 kVp
calcium image
140 kVp
basis
material
decomposition
water image
fully
characterizes
object
Image Based Methods
 Calculation of Energy-Selective Images: Noise @ different energies
fully
characterizes
object
Monoenergetic: 55
keV
Monochromatic energy is
a display parameter.
Monoenergetic: 68
keV
Similar to “mixed” image
Contrast and SNR vary
1.25*Ca + .22*Water with selected energy.
.78*Ca + .19*Water
B: Heismann, B. Schmidt, T. Flohr
Short course 987
SPIE Medical Imaging Conference 2010
Application of monoenergetic
images
Beam hardening correction
If basis material assumption holds (e.g., no K-edge materials),
• nonlinear decomposition exactly handles polychromaticity
• exact beam hardening correction
• projection domain processing is preferred
• image interpretation
high SNR, tissue characterization
• extrapolation to higher energies
MV, SPECT, PET
Discovery CT750 HD
Monochromatic CT from HDCT projectionbased recon
Posterior Fossa Artifact Reduction
Potential for beam hardening streak-free images
80kVp
Clinical Value
140kVp
•Spectral or Monochromatic
images have a reduced beam
hardening effect vs polychromatic
reconstructed images
Projection based
Image based
Water
Aluminum
Water
Aluminum
•Beam hardening artifacts can
obstruct the the clinician’s
interpretation of important brain
anatomy
Monochromatic
•Notice the visualization
improvement of the brain anatomy
between the petrous bones
140 kVp Conventional CT
75 keV
Projection based MD reduces beam hardening
Courtesy of Uri Shreter, GE
Healthcare
Images Courtesy of Dr. Joseph Hoxworth
Mayo Clinic Arizona
Posterior Fossa Streak Artifacts are Removed /
Lesion Verification
Monochromatic CT from HDCT projectionbased recon
Potential for beam hardening streak-free images
80kVp
140kVp
Projection based
Image based
Water
80
kVp
100 keV
Aluminum
Water
Aluminum
Monochromatic
MD Iodine
Projection based MD reduces beam hardening
Images Courtesy of Dr. Amy Hara, Mayo Clinic, Scottsdale, AZ
Beam hardening correction
If basis material assumption holds (e.g., no K-edge materials),
• nonlinear decomposition exactly handles polychromaticity
• exact beam hardening correction
• projection domain processing is preferred
• starting with image domain data is possible
Courtesy of Uri Shreter, GE
Healthcare
Dual energy processing
accurate beam
hardening built-in
material cancelled
images
monoenergetic
images
Maab et al, Med Phys,
2011
Lehmann et al: Med Phys 8, 659-67, 1981.
Quantitation
• Decomposition yeilds local density of basis materials
• Can convert basis materials – i.e., calculate density of any two
materials
• Quantifies actual materials only if they match basis materials
• Suppose bases are a and b, but voxel has a and c
Truth: (ma, mc)
Appearance: (ma,0)+ (ma’, mb’) = ([ma+ma’] , mb’)
• (Local) errors results if the materials are wrong (e.g., tissue vs.
water vs. fat)
• Virtual non contrast ≠ precontrast image
Scatter
ideal:
mA = a{LL – (BL/BH) LH}
w/ scatter:
mA’ = mA – a{ln(1+SPRL) - (BL/BH) ln(1+SPRH)}
20 cm
water
absorber
Ghafarian et al, IEEE Nuclear Science Symposium, 2007.
Scatter
three known materials
water
bone
1 = w1fw + b1fb + I1fI
iodine
2 = w2fw + b2fb + I2fI
fw+ fb + fI = 1
D. Tran and D.
Fleischmann
Vetter et al, Med Phys 1998.
solve for fw , fb , fI
three materials
water
fat
bone
• Bone mass from single
energy CT is inaccurate if
fat fraction is unknown
Goodsitt et al: Inv Radiol 22, 799, 1987.
three materials
water
fat
bone
• Bone mass from single
energy CT is inaccurate if
fat fraction is unknown
• DE can be more accurate
but is less precise
Three known materials
dual energy CT
Reliable separation requires
large
(R1 - R2)2
calcium
iodine
where R = high/low, depends
on material and energies
Works best for one high Z and
one lower Z material, and very
different x-ray energies
Kelcz et al: Med Phys 6, 418-25,
1979.
Choice of photon energies
Photon counting detectors
• DE, EL, EH
• very critical for SNR efficiency,
separation robustness, etc.
• implementations
• main challenges:
– different kVp and/or filtration
– layered detector
– photon counting with energy analysis
count rate capability, translates to scan speed
imperfect and count-rate dependent energy response
• very promising in the long term but not yet
ready for clinical use
Spectral separation
• very critical for SNR efficiency, separation
robustness, etc.
• implementations
photon counting with K-edge filter*
photon counting with energy analysis*
different kVp and filtration
different kVp
layered detector
better
spectral
separation
and
dose
efficiency
Dual energy implementations
• Sequential scans at different kVp
motion sensitivity > 50% Trot
• Two sources at 90º on the same gantry
some motion sensitivity (~ 25% Trot ?)
• Switching kVp within a single scan
• Energy discriminating detectors
better
immunity
to
motion
layered detector, photon counting
* assumes good energy response
contrast material sensitivity very low concentration CNR
• optimized single energy CT
advantage: highest CNR
limitation:
inhomogeneous background
• temporal subtraction (post - pre)
advantage: perfect background suppression (w/o
motion)
limitation:
motion misregistration
lower CNR at the same dose
• rapid dual energy CT
advantage: motion immunity
limitations: nonuniform Z background?
lowest CNR at the same total dose
Summary
• more material specificity than single energy
CT(e.g., average material properties,
material cancelled images)
• perfect beam hardening correction (prerecon)
effective monoenergetic images, more accurate
RTP and PET attenuation correction
• significant challenges but also many
opportunities
Summary
• virtual pre-contrast image
Thank You
perfectly registered and simultaneously acquired
beware of noise propagation. Separate optimized scans
probably have lower total dose
• isolate contrast media from calcified plaque
difficult, especially for small amounts of either
• lower dose?
not likely, compared to optimized protocols
• molecular imaging?
I don’t think so
Dose
• Two scans. Do we have to double the
dose?
Depends on the goal
Start by splitting the dose to both energies
Dose
higher Z task
thicker object
image quality
at given dose
beam energy
both spectra of DE system can’t be optimal!
Summary of commercial
systems
Dose
• Two scans. Do we have to double the
dose?
Depends on the goal
Start by splitting the dose to both energies
• DE not likely to provide the same noise
performance as optimized single energy
protocols
• penalty, if any, is low
Dose comparison
water image
optimal combination
(virtual non-contrast image)(post contrast image)
SNR=37
pre-contrast
dose ~
0.15D
iodine CNR=10
postcontrast
dose ~
0.79D
DUAL ENERGY
55/80 keV acquisition
ideal dose allocation
(~1:1)
total dose = D
CONVENTIONAL
55 keV pre/post
contrast scans
total dose 0.94 D
Under ideal conditions, DE scan has slightly higher dose
• Siemens: two sources, different kVp and
filtration, image-based processing
• GE: single source with rapid switching,
same filter for both kVps, projectionbased processing
• Philips: sequential scans at different kVp,
same filtration, image-based processing
• Lots of R&D work
Dose comparison
water image
optimal combination
(virtual non-contrast image)(post contrast image)
SNR=34
pre-contrast
dose ~
0.13D
iodine CNR=5.5
postcontrast
dose ~
0.23D
DUAL ENERGY
55/80 keV acquisition
poor dose allocation
(~1:8)
total dose = D
CONVENTIONAL
55 keV pre/post
contrast scans
total dose 0.36 D
Under non-ideal conditions, DE scan has much higher dose
Applications of Dual Energy CT
Another image based application : characterization of kidney stones
Principle of Dual Energy CT – Image Based Evaluation
Each material is characterized by its „Dual Energy Index“
x80 and x140 are the Hounsfield numbers at 80 kV and 140 kV, resp.
HU at 80 kV
high Z
low Z
HU at 140 kV
Material
DEI
Bone
0.1148
Liver
0.0011
Lung
-0.0021
Soft Tissue
-0.0052
Skin
-0.0064
Proteins
-0.0087
Fat
-0.0194
Gall fluid
-0.0200
based on
subtraction, noisy.
Dual energy CT can measure chemical composition!
 Uric acid stones can be differentiated from other renal calculi
Courtesy of University Hospital of Munich - Grosshadern / Munich, Germany
Image Based Methods
Image Based Methods
 Modified 2-material decomposition: Separation of two materials
 Assume mixture of blood + Iodine (unknown density)
and bone marrow + bone (unknown density)
Separation line
600
Iodine pixels
500
400
Bone pixels
Blood+Iodine
300
Marrow+bone
200
Soft
100 tissue
Blood
Marrow
0
-100
-100 0 100 200 300 400 50 600
HU at 140 kV
0 at low HU numbers
Short course 987
 Additional postprocessing to improve classification
B: Heismann, B. Schmidt, T. Flohr
SPIE Medical Imaging Conference 2010
HU at 80 kV
Courtesy of B. Krauss, B. Schmidt, and Th. Flohr, Siemens Medical
Solutions
 Modified 2-material decomposition: Separation of bone and Iodine
 Automatic bone removal without user interaction
 Clinical benefits in complicated anatomical situations:
 Base of the skull
 Carotid arteries
 Vertebral arteries
 Peripheral runoffs
Courtesy of Prof. Pasovic,
University Hospital of Krakow,
Poland
B: Heismann, B. Schmidt, T. Flohr
Short course 987
SPIE Medical Imaging Conference 2010
noise in processed images
low energy
high energy
material 1
high, correlated noise material 2
usually high noise
material cancelled images
Kalender, Computed Tomography, Publicis Corporate Publishing, 2005.
equivalent monoenergetic images
Dual kVp, dual filtration
Principle of Dual Energy CT
Data acquisition with different X-ray spectra: 80 kV / 140 kV
85 kVp
0.1 mm erbium
Mean Energy:
56 kV
can be low noise
76 kV
135 kVp
1.5 mm bronze
Tube 1
Tube 2
Different mean energies of the X-ray quanta
• switched filtration
improves separation
• different mA helps
apportion dose
Courtesy of B. Krauss, B. Schmidt,
and Th. Flohr, Siemens Medical
Solutions
Lehmann et al: Med Phys 8, 659-67, 1981.
Dual Source Challenge: Inconsistent scans
syngo Dual Energy
- Principle of Dual Energy
Moving Objects
SOMATOM Definition Flash
140kV
kV + SPS
SS11::80
80kV
kV SS22::140
Moving Phantom
Simulation
Ideal
4
x 10
x
104
quanta
of quanta
number
number of
15
15
80
80kV
kV
140
140kV
kV
140 kV + SPS
10
10
Does not see movement
Dual Source system
55
00
SPS= Selective Photon Shield
50
50
100
100
150
150
photon
photonenergy
energy(keV)
(keV)
Dual energy implementations
• Sequential scans at different kVp
motion sensitivity > 50% Trot
Courtesy of R. Senzig, GE
Healthcare
Rapid kVp switching
Dual energy CT
• Requires fast
generator and
detectors
• Two sources at ~90º on the same gantry
some motion sensitivity (~ 25% Trot)
• Switching kVp within a single scan1, 2
• Dose allocation
controlled by dwell
time
• Difficult to switch
filters
1. Lehmann et al: Med Phys 8, 659-67, 1981.
2. Kalender et al, Med Phys 13, 334, 1986.
Courtesy of Uri Shreter, GE
Healthcare
Layered detector
Multilayer Detector
• simultaneous dual energy sensing
• relatively poor spectral separation
X-Rays
Photons
100%
~50%
SCINT1
SCINT2
Low Energy Raw data
~50%
E1 image
+
High Energy Raw data
E2 image
---------------------------------------
=
Weighted combined Raw data
CT image
Carmi R, Naveh G, and Altman A: IEEE
NSS M03-367, 1876-78, 2005
81
Investigational device
Dose efficiency
Dual Layer Detectors - Challenges
 Complex technical realization
 Reduced dual energy performance compared to dual kV
3 component decomposition, each technique optimized
Relative
std dev
24.1
36.2
 Performance depends
on scintillator types
and thicknesses
 Dual energy performance
less than 70% compared
to dual kV technology
(same radiation dose!)
4X overall difference in dose efficiency
Kelcz et al: Med Phys 6, 418-25,
1979.
Courtesy S. Kappler,
Siemens
Healthca
Short course
987
Page 84
B: Heismann, B. Schmidt, T. Flohr
SPIE Medical Imaging Conference 2010
Photon Counting Spectral CT – Detector Principle
Absorbed single X-ray photon
(CZT crystal: Cadmium Zinc Telluride)
High
Voltage
Discriminating
thresholds
Counter 4
Counter 3
Counter 2
Counter 1
Direct conversion
material
Pixelated electrodes
Photon Counting Prototype*
Early results
Electronics
• Two systems in Philips Research labs
• Research system installed at Washington
University-Saint Louis in Nov. 2008
Charge pulse
Pulse height proportional
to x-ray photon energy
Stored counts of all energy windows,
for each reading time period
Counts
Photon energy
Direct- Conversion Detector efficiently translates X-ray photons into large electronic signals
These signals are binned according to their corresponding X-ray energies
‘K-Edge Imaging’ w/ Univ Ulm (Radiology Oct 08)
Novel contrast agents targeting fibrin (clots) w/ Wash U
*Works-in-Progress: Pending commercial availability and regulatory clearance
Confidential
Investigational device
Dual energy implementations
• Sequential scans at different kVp
86
8
syngo Dual Energy
- Principle of Dual Energy
SOMATOM Definition
SOMATOM Definition Flash
motion sensitivity > 50% Trot. Helical?
• Two sources at ~90º on the same gantry
Two X-ray tubes, one at 80 kVp, second at 140
kVp
attenuation and
material identification
Image Based Methods
 Calculate material specific images: 2 materials of unknown density
I0
 Noise amplification in the material specific images!
But what material is it?
2000
Bone
T
1500
HU at 80 kV
Image noise 
1000
Bone image
500
Soft tissue
0
I = I0 e-T
-500
Soft tissue image
-1000
-1000
-500
0
HU at 140 kV
500
1000
Short course 987
SPIE Medical Imaging Conference 2010
B: Heismann, B. Schmidt, T. Flohr
Three known materials
dual energy CT
water
bone
HUlow
iodine
bone and iodine
in water
water +
iodine
water +
bone
identity
Can calculate both iodine and
bone.
HUhigh
Requires known mixing
properties
mb = A{ HUlow - ( 1,I/  2,I) HUhigh}
and consistent water density
Kelcz et al: Med Phys 6, 418-25,
1979.
T = ln(I0/ I)
Even more confusing if T
is unknown