PowerPoint Presentationamos3.aapm.org/abstracts/pdf/137-41528-446581-142225.pdf · K Taguchi (JHU) 2 Outline AAPM 2018. K Taguchi (JHU) 3 Designtypes Groups PrototypeCT (beta site

8/1/2018

1

Multi-energy CT using photon counting detectors

Katsuyuki “Ken” Taguchi

Russell H. Morgan Department of Radiology and Radiological Science

Johns Hopkins University School of Medicine

(Baltimore, MD)

AAPM 2018. K Taguchi (JHU)

Disclosure

■ No financial interest

■ Research grants and relationships– Siemens Healthineers JHU-2016-CT-1-01-Taguchi_C0022

– NIH R56 HL125680

– Previous: DxRay (NIH SBIRs), Philips (C-arm), Toshiba/Cannon (employee)

■ Former consultantship– Suzhou Bowing Medical Technologies, Life Saving Imaging Technology (LISIT),

JOB Corporation

2AAPM 2018. K Taguchi (JHU)

Outline


Design types Groups Prototype CT (beta site installation)

“Standard” Siemens, Philips, GE, … Siemens (2014), Philips (2015), GE (2007)

“Anti-charge sharing” CERN (and MARS) MARS (–2018)

“Edge-on silicon” KTH Royal Institute of Tech (Sweden) KTH RIT (–2018)

■ Photon counting detectors (PCDs) and properties

■ Algorithms (data correction and image reconstruction)

■ System designs

■ Clinical applications

8/1/2018

2

Outline








■ System designs


Clinical applications


There is no “killer application”

“What is the killer application?”

New generation CT has been developed for


■ improving everything CT does (whole body)

■ adding new dimension to everything CT does

■ enabling new applications

8/1/2018

3



■ Improving everything CT does (whole body)

■ Adding new dimension to everything CT does

■ Enabling new applications

In 1998, MDCT with 4 detector rows

10 mm slice → 2–3 mm slice (→ 0.5 mm)Shorter scan duration, larger coverage, better resolution

2D slice-by-slice view → 3D view (coronal, sagittal, MIP, 3D)

Cardiac CT (coronary artery)




■ Adding new dimension (material info) to everything CT does


In 202x, PCD-CT with 3–6 energy windows

Contrast, contrast-to-noise, radiation dose, contrast dose,spatial reso, accuracy, quantitative values, CT radiomics

Simultaneous image-and-measure (IAM-XX), CT tissue biomarkers

Simultaneous multi-agent, molecular CT, personalized CT

Spatial resolution

9

J. Baek, PMB 2013

Binned system High reso system

Binned detector High reso detector

Low reso detector High reso detector

MTF

NPS

MTF

NPS

8/1/2018

4

Spatial resolution

10A. Pourmorteza, Inv Rad 2018

Eff. (500 µm)2 Eff. (250 µm)2 Eff. (500 µm)2 Eff. (250 µm)2

Molecular CT


Bi

D. Pan, Angew Chem Int Ed 2010

Simultaneous multi-agent imaging


Gd → (10 min) → Iodine → PCD-CT scan

R. Symons, Int J Cardiovasc Imag 2017

8/1/2018

5

Simultaneous multi-agent imaging


Gd → (10 min) → Iodine → PCD-CT scan

R. Symons, Int J Cardiovasc Imag 2017

Material-specific imaging for better diagnosis


DWI

K. Taguchi, Inv Rad 2018

a b c

Normal

Edema

CT “X-map”

Acute ischemic stroke delineated clearly by water density images (for edema)

Dual-energy CT (Force). 68 yo, female, 40 min after symptom onset.



■ improving everything CT does (whole body)

■ adding new dimension (material info) to everything CT does

■ enabling new applications





8/1/2018

6

Outline








■ System designs


PCDs (“Standard,” “ACS,” and “Edge-on Si”)

17AAPM 2018. K Taguchi (JHU) Table I of K Taguchi and J Iwanczyk, Med Phys, Vision 20/20 (2013)

PCD (“Standard” type)

18

Incident x-ray

Convert a photon to

a charge cloud

Charge measured, counted,

and stored digitally

--

--

Amplifier

Counting

processing

Sensor (CdTe, CZT)


Courtesy of W. C. Barber

8/1/2018

7


PCD (“Standard” type)

19

Shaper

Preamp

Counter2Digital output

Input

analog

pulses

Thresholds to pulse height comparators

DAC

Comparators

Counter1Digital outputDAC

CounterNDigital outputDAC

… …

w1 = n1 – n2

w2 = n2 – n3

w3 = n3

Time

n2

n1Threshold #1

Threshold #2

Threshold #3 n3Energy

PCD not flawless: Spectral distortion

■ Independent of x-ray intensities: Charge sharing, K-escape, Re-absorption, Compton scattering

■ Dependent on x-ray intensities: Pulse pileups


Anode

Sensor

Charge sharing

21

A 90 keV photon

time

60 keV

time

30 keV“Charge sharing”


Incident x-ray photons

8/1/2018

8

Anode

Sensor


Fluorescence x-ray emission (K-escape)


A 90 keV photon

time

67 keV

time

23 keV

“K-escape”

Anode

Sensor

Re-absorption of fluorescence x-rays


A 90 keV photon

time

67 keV23 keV

67 keV

23 keV

time

70 keV

Re-absorption of K-escape x-ray

Slower PCD

Faster PCD


Anode

Sensor

Compton scattering


A 90 keV photon

time

50 keV30 keV

time

60 keV

“Compton scattering”

Slower PCD

Faster PCD50 keV

30 keV

10 keV

10 keV


8/1/2018

9


Spectral distortion

25

Continuum:- Charge trapping- K-escape- Charge sharing- Compton scatter

0 20 40 60 80 100

Energy [keV]

Co

un

ts

Response to 59.5 keV photons (Am-241 source)

Photo peak

Noise floor

20 40 60 80 100

Energy [keV]120

Pro

bab

ility

[a.

u.]

Incident spectrum

Recorded spectrum

Anode

Sensor


Pulse pileups (count rate dependent)

26

Time

Events on detector

Live

Dead

True pulses

Observed pulses

True

Recorded

Input count rate [Mcps]

Outp

ut count

rate

[M

cps]

10 20 30 40 500

10

Paralyzable detector

( =20 ns)

Non-paralyzable detector

( =20 ns)

True20

30

Count rate loss

Time

Pulse height (photon energy)

Pulse pileup spectrum

Energy [keV]

Pro

babili

ty

𝜏

PCD not flawless: Spectral distortion

■ Charge sharing, K-escape, etc.AlwaysLower energy, more countsSolution: Larger, slower PCD

■ Pulse pileupsOnly near skinsHigher energy, fewer countsSolution: Smaller, faster PCD


200 400 600 800 1000 1200 1400 1600 1800

500

1000

1500

2000

25000

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2x 10

7

Φ16 cm

Detector channels (fan angles)

Pro

ject

ion

s (v

iew

an

gles

) 20

10

0

Count rates [Mcps]Heart

Synthesized from PCD

Measured by GOS-25 -20 -15 -10 -5 0 5 10 15 20 250

1

2

3

4

5

6

7

8

Lin

e in

tegr

als

Radial distance [cm]0 10-10

PCDs

Synthesized

Measured

Difference

8/1/2018

10

Spectral distortion: How to mitigate it?

■ Balanced design + PCD model + compensation algorithm


X-ray focus

PCD

Ray

PCD pixels

Energy, E

nt(E)

nR(E)

Energy, E

Energy, E

n0(E)

Object, x(E)nt(E)

Bowtie filters

y={y1, y2, y3, y4}

Observed spectrum

Recorded counts

x={x1, x2, x3, x4}

Reconstruct images




X-ray focus

PCD

Ray

PCD pixels

Energy, E

nt(E)

nR(E)

Energy, E

Energy, E

n0(E)

Object, x(E)nt(E)

Bowtie filters

y={y1, y2, y3, y4}

Observed spectrum

Recorded counts

x={x1, x2, x3, x4}

Reconstruct images

𝑥 𝐸 =

𝑚

𝑤𝑚Ψ𝑚 𝐸

x2

[cm-1]

Photon energy [keV]

Photoelectric

K-edge(Iodine)

Compton Scattering




X-ray focus

PCD

Ray

PCD pixels

Energy, E

nt(E)

nR(E)

Energy, E

Energy, E

n0(E)

nt(E)Bowtie filters

y={y1, y2, y3, y4}

Observed spectrum

Recorded counts

Line integrals,Wj

𝑝 ො𝑦 𝑊

Estimate object W by maximizing the likelihood

8/1/2018

11




(a) Truth (b) Without pulse pileup model (c) With pulse pileup model

WW 400 HU, WL 50 HU

Taguchi K, et al, IEEE MIC 2013

Outline








■ System designs



■ Anti-charge sharing circuit (“CSM” in Medipix 3RX)

33AAPM 2018. K Taguchi (JHU) Courtesy of M. Campbell

55

The winner takes all CSMNon CSM

Spectrum

Th. Koenig, et al., IEEE TNS (2013).

8/1/2018

12

Outline








■ System designs


Prototype PCD-CT systems






■ Siemens (Mayo Clinic and NIH)

– Dual-source CT, PCD for 27.5 cm x 8–24 mm, EID for 50 cm

– (225 µm)2, 2 thresholds (staggered 4 thresholds), (450–900 µm)2 output

– 128x106 counts/s/mm2 with 13.5% loss, 256x106 with 25.2% loss

■ Philips (Louis Pradel University Hospital, Bron, France)

– Brilliance 64 CT, 16.8 cm x 2.5 mm

– (500 µm)2, 5 thresholds (*1)

– >150x106 counts/s/mm2 (*1)

1. R. Steadman, et al., Nucl Inst Meth Phys Res A (2017)







■ University of Canterbury, Christchurch, NZ (on site)

– Their own open gantry CT, PCD for ? cm x ? mm

– (110 µm)2, 8 thresholds, anti-charge sharing mode (Medipix3RX by CERN)

– 10–20x106 counts/s/mm2 with ?% loss

■ KTH Royal Inst of Tech, Stockholm, Sweden (on site)

– Philips CT(?), ? cm x ? mm

– (400 µm)x(500 µm), 8 thresholds/layer x 16 layers/pixel (KTH silicon strip det)

– 200x106 counts/s/mm2 with 2% loss, 600x106 with 25% loss

New York Times, July 17, 2018.https://www.nytimes.com/2018/07/17/health/3d-color-xrays-cern.html

8/1/2018

13







■ University of Canterbury, Christchurch, NZ (on site)

– Their own open gantry CT, PCD for ? cm x ? mm

– (110 µm)2, 8 thresholds, anti-charge sharing mode (Medipix3RX by CERN)

– 10–20x106 counts/s/mm2 with ?% loss

■ KTH Royal Inst of Tech, Stockholm, Sweden (on site)

– Philips CT(?), ? cm x ? mm

– (400 µm)x(500 µm), 8 thresholds/layer x 16 layers/pixel (KTH silicon strip det)

– 200x106 counts/s/mm2 with 2% loss, 600x106 with 25% loss

H. Bornefalk, Nucl Inst Meth Phys Res A (2010)

Photon Counting Toolkit (PcTK, pctk.jhu.edu)

38

• “Spectral response” model of photon counting detectors

• Medical Physics 2018;45(5):1985–1998

• Matlab programs

• Available for academic research, free of charge

• Download and send the application

Probabilities

0.00

0.10

0.05

(a)Ver.3.2 E1=60keV

E1=90keV

E1=120keV

Recordedenergy,E3 (keV)0 6040 10080 14012020

(b)aBN

(d)vST (e)vBN

(f)m1 (g)m2 (h)m3 (i)m4

(j)fR,1 (k)fR,2 (l)fR,3 (m)fR,4

(n)ft,1 (o)ft,2 (p)ft,3 (q)ft,4

(a)aST aST=0.97 aST=1.00(c)

FPJ FPJ

Randomnumbergenerator

FBP FBP FBP FBP

Otherdata(Sec.2.C)(A)Init. x-rayspectrum(B)Initial#ofphotons(D)µk(E)of materialk(E)nCov3x3w byPcTK

(C)Material-specificsinograms

Material-specificimages

(j)fR,1

(n)f t ,1

Synthesized from the input images (a–b)

Reconstructed from noisy PCD data

Summary






■ PCD-CT with 3–6 energy windows in Year 202x that

– Improves contrast, noise, spatial reso, quant. accuracy, etc.

– Add material-axis to 3-D images → How to use it effectively?

– Enables multi-agent CT, molecular CT, personalized CT

– No killer application!

■ Algorithms needed for data correction and image recon

http://pctk.jhu.edu/

8/1/2018

14

Acknowledgemnet


Mayo Clinic■ Cynthia H. McCollough, Ph.D.■ Shuai Leng, Ph.D.■ Kishore Rajendran, Ph.D.

TITech■ Kenji Amaya, Ph.D.■ Kento Nakada, M.Sc.

JHU (my group)

■ Okkyun Lee, Ph.D.■ Somesh Srivastava, Ph.D.■ Jochen Cammin, Ph.D.■ Qiulin Tang, Ph.D.■ Zhihui Sun, M.Sc.■ Mengxi Zhang, M.Sc.

JHU ■ Eric C. Frey, Ph.D. ■ Benjamin M.W. Tsui, Ph.D.

Siemens Healthineers■ Steffen Kappler, Ph.D.■ Karl Stierstorfer, Ph.D.■ Christoph Polster, Ph.D.■ Thomas G. Flohr, Ph.D.■ Matthew K. Fuld, Ph.D.■ George S. K. Fung, Ph.D.

DxRay■ Jan S. Iwanczyk, Ph.D. ■ William C. Barber, Ph.D. ■ Neal E. Harsough, Ph.D.




■ Adding new dimension (material info) to everything CT does






Documents

PowerPoint Presentationamos3.aapm.org/abstracts/pdf/137-41528-446581-142225.pdf · K Taguchi (JHU) 2 Outline AAPM 2018. K Taguchi (JHU) 3 Designtypes Groups PrototypeCT (beta site