CT imaging
using energy-sensitive
photon-counting detectors
Katsuyuki “Ken” Taguchi, Ph.D.
ktaguchi@jhmi.edu
Division of Medical Imaging Physics
Department of Radiology
Johns Hopkins University
Disclosure

No financial interest

Research grants

Past relationships/grants





Siemens Healthcare
Former employee of Toshiba Medical Systems
Former Co-I of a project funded by Philips Healthcare
AHA, NIH R01 and SBIRs (DxRay)
Consultant

Life Saving Imaging Technology (LISIT) (Tokyo, Japan)
K. Taguchi (JHU) – AAPM, July 12-16, 2015
2
Vision 20/20 paper




Detector technology
Imaging technology
System technology
Potential clinical benefits
Med Phys 40(10), 100901 (19 pages) (2013)
K. Taguchi (JHU) – AAPM, July 12-16, 2015
1
Outline
 Current CT detectors to higher dose exams
 Low-dose CT by integrating 3 technologies
Photon counting detectors (PCDs)
Iterative reconstruction for PCDs (PIECE)
Joint estimation with tissue types (JE-MAP)
 Whole-body prototype PCD-CT system
 Other clinical merits of PCD-CT
K. Taguchi (JHU) – AAPM, July 12-16, 2015
4
Major problems of current CT




Relatively high-dose procedure
Insufficient contrast between tissues
CT images are not tissue-type specific
Pixel values, Hounsfield units or linear
atten. coeff., are not quantitative but
qualitative
K. Taguchi (JHU) – AAPM, July 12-16, 2015
5
Number of photons n (E)
Energy weighted integration
1.5E+06
1.0E+06
5.0E+05
0.0E+00
0
20
40
60
80
100 120 140
Energy [keV]
Incident
K. Taguchi (JHU) – AAPM, July 12-16, 2015
Transmitted
Weighted by E
6
2
Energy binning for spectral CT
Linear attenuation coefficient [cm-1] Transmitted x-ray spectrum [A.U.]
Bismuth
Rib
Blood only
Iodine-mixed blood
1.5
1.40E-03
1.20E-03
u(E, blood)
1.01
u(E, Gd)
1.00E-03
u(E, dry_spine)
Gd-mixed blood
u(E, dry_rib)
8.00E-04
u(E, iodine)
BiBi-mixed blood
Spine
Gd
u(E, bismuth)
Gd-blood (0.26%)
6.00E-04
0.5
0.5
I-blood (0.49%)
Bi-blood (0.49%)
4.00E-04
Water
0.00
20
20
40
40
Iodine
2.00E-04
60
60
80
80
100
100
120
120
0
0.00E+00
20
20
Energy [keV]
Four materials would provide an
identical pixel value if scanned
with current CT
40
40
60
60
80
80
100
100
120
120
Energy [keV]
Transmitted spectra with 25 cm
water and 5 cm blood w/ or w/o
contrast agents
K. Taguchi (JHU) – AAPM, July 12-16, 2015
7
Strategy for low-dose CT
Combine the following 3 technologies:
Technologies
Current CT system with FBP*1
Relative amount of
dose reduction
Liver CT
(mSv)
N/A
10
30–40 %
6–7
Iterative reconstruction for PCDs (PIECE)
>35 %
3.9–4.6
Joint estimation with tissue types (JE-MAP)
>35 %
2.6–3.0
Photon counting detectors (PCDs)
Note 1. Dose for current CT with iterative reconstruction may be 6–8 mSv
Note 2. Annual background radiation: 3.1 mSv
K. Taguchi (JHU) – AAPM, July 12-16, 2015
8
Strategy for low-dose CT
Combine the following 3 technologies:
Technologies
FBP*1
Relative amount of
dose reduction
Liver CT
(mSv)
N/A
10
30–40 %
6–7
Iterative reconstruction for PCDs (PIECE)
>35 %
3.9–4.6
Joint estimation with tissue types (JE-MAP)
>35 %
2.6–3.0
Current CT system with
Photon counting detectors (PCDs)
Note 1. Dose for current CT with iterative reconstruction may be 6–8 mSv
Note 2. Annual background radiation: 3.1 mSv
K. Taguchi (JHU) – AAPM, July 12-16, 2015
9
3
Photon counting detector
(PCD)
Incident x-ray
Amplifier
Photon to charge
-
-
-
-
Counting
processing
Charge measured, counted,
and stored digitally
Courtesy of W. Barber, Ph.D. and J. Iwanczyk, Ph.D. (DxRay, Inc)
K. Taguchi (JHU) – AAPM, July 12-16, 2015
10
Detection mechanism
Thresholds to pulse height comparators
Comparators
Input
analog
pulses
Shaper
DAC
Counter 1
DAC
Counter 2
Counter N
DAC
Pulse height
= photon
Threshold #3
energy
N3
N2
Threshold #2
Threshold #1
Digital output
…
…
Preamp
Digital output
N1
Digital output
n3 = N3
n2 = N2 – N3
n1 = N1 – N2
Time
11
“Spectral response (SR)”
Incident x-ray photons
(a)
(b)
(c)
(d)
Sensor
Anode
A 90 keV photon
60 keV
time
30 keV
“Charge sharing”
time
12
4
“Spectral response (SR)”
Incident x-ray photons
(a)
(b)
(c)
(d)
Sensor
Anode
0
20
40
60
80
Probability [a.u.]
Counts
Response to 59.5 keV photons (Am-241 source)
Noise floor
Continuum:
Photo peak
- Charge trapping
- K-escape
- Charge sharing
- Compton scatter
Incident spectrum
Recorded spectrum
100
20
Energy [keV]
40
60
80
100
120
Energy [keV]
13
Pulse pileup
Output count rate [Mcps]
Pulse height (photon energy)
Observed pulses
True pulses
Count rate loss
30
Non-paralyzable detector
( =20 ns)
True
20
10
Paralyzable detector
( =20 ns)
0
10
20
30
40
50
Input count rate [Mcps]
Pulse pileup spectrum

True
Time
Probability
Dead
Live
Recorded
Time
Events on detector
Energy [keV]
14
Spectral distortion
 Spectral response (SR): Charge sharing,
K-escape, Compton, etc.
Incident x-ray photons
 Larger, slower PCD
(a)
(b)
(c)
(d)
Sensor
Anode
 Pulse pileups
Pulse height (photon energy)
Observed pulses
True pulses
 Smaller, faster PCD
K. Taguchi (JHU) – AAPM, July 12-16, 2015
15
5
Strategy for low-dose CT
Combine the following 3 technologies:
Relative amount of
dose reduction
Technologies
Current CT system with FBP*1
Liver CT
(mSv)
N/A
10
30–40 %
6–7
Iterative reconstruction for PCDs (PIECE)
>35 %
3.9–4.6
Joint estimation with tissue types (JE-MAP)
>35 %
2.6–3.0
Photon counting detectors (PCDs)
Note 1. Dose for current CT with iterative reconstruction may be 6–8 mSv
Note 2. Annual background radiation: 3.1 mSv
K. Taguchi (JHU) – AAPM, July 12-16, 2015
16
Image artifacts due to SR and PP
Observed
spectrum
PCD
pixels
PCD
Bowtie
filters Object, x(E)
nR(E)
Ray
X-ray
focus
nt(E)
Energy, E
Recorded
counts
n0(E)
y={y1, y2, y3, y4}
nt(E)
Energy, E
Reconstruct
images
Energy, E
x={x1, x2, x3, x4}
K. Taguchi (JHU) – AAPM, July 12-16, 2015
17
Image artifacts due to SR and PP
x2
Observed
spectrum
PCD
pixels
PCD
Bowtie
filters Object, x(E)
nR(E)
Ray  [cm-1]
X-ray
focus
nt(E)
Photoelectric
Energy, E
Compton Scattering
Recorded
K-edge(Iodine)
counts
𝑥 𝐸n (E)=
𝑤𝑚 Ψ𝑚 𝐸n (E)
0
t
𝑚
Energy, E
K. Taguchi (JHU) – AAPM, July 12-16, 2015
y={y1, y2, y3, y4}
Reconstruct
images
Energy, E
x={x1, x2, x3, x4}
Photon energy [keV]
18
6
PCD compensation
PIECE = Physics-modeled Iterative reconstruction for
Energy-sensitive photon Counting dEtector
Line integrals,
Bowtie
𝑣Wj
filters
Observed
spectrum
PCD
pixels
PCD
nR(E)
Ray
nt(E)
X-ray
focus
Energy, E
Recorded
counts
n0(E)
y={y1, y2, y3, y4}
nt(E)
Energy, E
Energy, E
𝑝 𝑦𝑣
Estimate object v by maximizing the likelihood
19
PCD model
Transmitted spectrum
Spectral response1
𝑛𝑆𝑅𝐸 𝐸 =
Energy
𝑎𝑆𝑅𝐸 𝐸, 𝐸𝑘 𝑛𝑡 𝐸𝑘
𝑘
or
𝒏𝑺𝑹𝑬 = 𝐴𝑆𝑅𝐸 𝒏𝒕
Spectral response
Pulse pileups2,3
Pulse pileups
𝑛𝑅 𝐸 = 𝑛𝑆𝑅𝐸 𝐸 𝑃𝑟 "𝑟𝑒𝑐" = 1
Recorded spectrum
×
𝑃𝑟 𝑙 "𝑟𝑒𝑐" = 1 𝑃𝑟 𝐸 𝑙
𝑙
l = # of photons piled up for one recorded count
Energy
1: J. Cammin, Med. Phys. 41: 041905 (2014)
2: K. Taguchi, Med. Phys., 37: 3957 (2010)
3: K. Taguchi, Med. Phys., 38: 1089 (2011)
20
Monte Carlo simulation study
for Siemens PCD
25 mA
800 mA
16,000
600
SRE+DE
SRE+pileups
PCD data
DLR = 0.9%
COV = 2.4%
W
200
100
measured
DE only
SRE+DE
SRE+pileups
SRE+PPE((E))
SRE, DL
transmitted, DL
12000
P, unipolar
It = 25 mA
300
H2O, Al
P, unipolar
It = 800 mA
10000
DLR = 25.3%
COV = 1.5%
8000
W
PCD data
6000
4000
2000
20
20
40
60
60
100
80
100
Energy (keV)
120
140
140
Energy (keV)
SRE+PPE, P, unipolar, no Gd
SRE x DL, P, unipolar, no Gd
SRE, P, unipolar, no Gd
trans x DL, P, unipolar, no Gd
trans, P, unipolar, no Gd
160
180
0
0
20
20
40
60
60
100
80
100
Energy (keV)
120
140
140
160
180
Energy (keV)
800 mA
140.0
120.0
100 mm water, 3.5 mm Al, 0.9 mm Ti; 120 kVp;
pulse height analyzer; unipolar pulse
200 mA
100.0
COVW (%)
14000
DE only
400
0
16000
SRE+PPE((E))
SRE, DL
transmitted, DL
H2O, Al
500
0
measured
Counts/keV
Counts/keV
Counts/keV
700
Counts/keV
700
80.0
60.0
DE = Detection efficiency (%)
COVw = Weighted coefficient of variation (%)
40.0
20.0
0.0
0.0
20.0
40.0
60.0
deadtime loss ratio: 1-nrecorded/nt (%)
100-DE (%)
80.0
J. Cammin, presented at IEEE NSS/MIC 2014
21
7
Evaluation of PIECE-1
CNR 6.0
CNR 6.0
CNR 4.1
CNR 4.9
22
Strategy for low-dose CT
Combine the following 3 technologies:
Relative amount of
dose reduction
Technologies
Current CT system with FBP*1
Liver CT
(mSv)
N/A
10
30–40 %
6–7
Iterative reconstruction for PCDs (PIECE)
>35 %
3.9–4.6
Joint estimation with tissue types (JE-MAP)
>35 %
2.6–3.0
Photon counting detectors (PCDs)
Note 1. Dose for current CT with iterative reconstruction may be 6–8 mSv
Note 2. Annual background radiation: 3.1 mSv
K. Taguchi (JHU) – AAPM, July 12-16, 2015
23
Joint estimation with tissue types
Sinogram
Characteristic coefficients
𝒚
𝒘
Tissue type
𝒛
Projection
noise
𝐸
K. Nakada, et al., IEEE MIC (2013); Med Phys (2015)
24
8
Joint estimation with tissue types
Sinogram
Characteristic coefficients
𝒚
Tissue type
𝒘
𝒛
Projection
noise
𝐸
Geometry
Statistical relationship
𝑤pe
0.20
𝒛𝑖 = 𝒛𝑗 = blood
0.10
Iodine-mixed
blood
rib
Smooth
𝒛𝑖 ≠ 𝒛𝑗
spine
lung fat
0.00
blood
2
→ 𝛼 𝒘𝑖 − 𝒘𝑗
→𝛽
muscle
muscle
Discontinuous
liver
location
0.00
0.10
0.20
0.30
0.40 𝑤
cs
K. Nakada, et al., IEEE MIC (2013); Med Phys (2015)
25
Joint estimation with tissue types
Sinogram
Characteristic coefficients
𝒚
Tissue type
𝒘
𝒛
Projection
noise
𝐸
Geometry
Statistical relationship
𝒛𝑖 = 𝒛𝑗 = blood
𝒘
→ 𝛼 𝒘𝑖 − 𝒘𝑗
2
Smooth
𝒛
blood
𝒛𝑖 ≠ 𝒛𝑗
→𝛽
Discontinuous
muscle
location
K. Nakada, et al., IEEE MIC (2013); Med Phys (2015)
Truth
FBP
PML
JE-MAP
K. Nakada, CERN wrokshop (2015)
@70keV, WW 400HU, WL -50HU
26
27
9
Truth
FBP
PML
JE-MAP
K. Nakada, CERN wrokshop (2015)
@70keV, WW 400HU, WL -50HU
28
Joint estimation with tissue types
Noise reduction at FWHM=3 pixels
PML
FBP
by 30%: FBP  PML
by 33%: PML  JE-MAP
by 53%: FBP  JE-MAP
JE-MAP
K. Taguchi (JHU) – AAPM, July 12-16, 2015
29
K. Nakada, et al., IEEE MIC (2013)
Photon counting low-dose CT
 Current CT detectors lead to higher dose exams
 Low-dose CT by integrating 3 technologies
 Photon counting detectors (PCDs)
 Iterative reconstruction for PCDs (PIECE)
 Joint estimation with tissue types (JE-MAP)
Technologies
Current CT system with FBP*1
Relative amount of
dose reduction
Liver CT
(mSv)
N/A
10
30–40 %
6–7
Iterative reconstruction for PCDs (PIECE)
>35 %
3.9–4.6
Joint estimation with tissue types (JE-MAP)
>35 %
2.6–3.0
Photon counting detectors (PCDs)
K. Taguchi (JHU) – AAPM, July 12-16, 2015
30
10
Outline
 Current CT detectors to higher dose exams
 Low-dose CT by integrating 3 technologies
Photon counting detectors (PCDs)
Iterative reconstruction for PCDs (PIECE)
Joint estimation with tissue types (JE-MAP)
 Whole-body prototype PCD-CT system
 Other clinical merits of PCD-CT
K. Taguchi (JHU) – AAPM, July 12-16, 2015
31
Whole-body prototype PCD-CT system
Developed by Siemens, installed at Mayo Clinic




Dual-source CT, EID for 50 cm, PCD for 27.5 cm
(225 m)2 pixel, 2 thresholds (staggered 4 thresholds)
128x106 counts/s/mm2 with 13.5% loss, 256x106 with 25.2% loss
Shading artifacts due to energetic cross-talks (not shown)
In-vivo swines
Cadavers
Courtesy of Cynthia H. McCollough, Ph.D. (Mayo Clinic)
32
Clinical benefits of spectral CT
Improvement of current CT images
1) Contrast-to-noise ratio (CNR) and
contrast of CT images, or doses of
radiation & contrast agents
2) Quantitative CT imaging
3) Material- or tissue type-specific imaging
4) Accurate K-edge CT imaging
5) Simultaneous multi-agent imaging
6) Molecular CT with nanoparticles
7) Personalized medicine
New class of CT imaging
K Taguchi, Med Phys 40(10), 100901 (19 pages) (2013)
33
11
Molecular CT with nanoparticle
contrast agents or probes
Bi
D. Pan, Angew Chem Int Ed 49: 9635-9639 (2010)
34
Imaging macrophages in
atherosclerotic plaques
Nano agent
Nano agent
Conventional
Injection
Nano agent
F. Hyafil, Nat Med 13(5): 636-641 (2007)
Conventional
agent
35
The use of PCDs and coded-aperture for
simultaneous absorption, phase, and
differential phase contrast imaging
Low-dose single scan followed by
a novel retrieval algorithm
5cm average breast tissue
1.5 mGy dose
6 energy bins, CdTe detector
M. Das, AAPM 2015, TH-AB-204-07 (8:30 am)
Absorption
Phase
Diff. phase
36
12
Summary
 Current CT detectors cause the major limitations of
CT images
 PCDs address them and enable new applications
 Low-dose CT by integrating 3 technologies
Photon counting detectors (PCDs)
Iterative reconstruction for PCDs (PIECE)
Joint estimation with tissue types (JE-MAP)
 A few whole-body prototype PCD-CT systems
being evaluated at hospitals
K. Taguchi (JHU) – AAPM, July 12-16, 2015
37
Acknowledgment














Kenji Amaya, Ph.D.
Kento Nakada, B.Sc.
Cynthia H. McCollough, Ph.D.
Jan S. Iwanczyk, Ph.D.
William Barber, Ph.D.
Neal E. Harsough, Ph.D.
Steffen Kappler, Ph.D.
Thomas G. Flohr, Ph.D.
Karl Stierstorfer, Ph.D.
Eric C. Frey, Ph.D.
Jochen Cammin, Ph.D.
Somesh Srivastava, Ph.D.
Benjamin M.W. Tsui, Ph.D.
Current and former students
K. Taguchi (JHU) – AAPM, July 12-16, 2015
(TITech)
(TITech)
(Mayo Clinic)
(DxRay, Inc)
(DxRay, Inc)
(DxRay, Inc)
(Siemens Healthcare)
(Siemens Healthcare)
(Siemens Healthcare)
(JHU)
(JHU)
(JHU)
(JHU)
and colleagues in DMIP and JHU
38
13
Thresholds to pulse height comparators
Comparators
Input
analog
pulses
Shaper
DAC
Counter 1
DAC
Counter 2
Digital output
Digital output
…
…
Preamp
Counter N
DAC
Digital output
Clinical benefits of spectral CT
1) Contrast-to-noise ratio (CNR) and contrast of CT images,
or doses of radiation & contrast agents
2) Quantitative CT imaging
3) Material- or tissue type-specific imaging
4) Accurate K-edge CT imaging by >2 energy windows
5) Simultaneous multi-agent imaging
K. Taguchi (JHU) – AAPM, July 12-16, 2015
41
Joint estimation with tissue types
Sinogram
Characteristic coefficients
𝒚
𝒘
Tissue type
𝒛
Projection
noise
𝐸
𝒘
𝒛
blood
muscle
location
K. Nakada, et al., IEEE MIC (2013); Med Phys (2015)
42
14