Methods for High Resolution PET

Neal ClinthorneRadiology / Nuclear Medicine

University of MichiganAnn Arbor

2

Ring of PhotonDetectors

• Inject positron emitting radiotracer intopatient

• Tracer localizes according to its metabolicproperties

• Radionuclide decays, emitting β+.

• β+ annihilates with e– from tissue, formingback-to-back 511 keV photon pair

• Photon pairs detected via timecoincidence (<5ns) indicate line alongwhich positron annihilated

• By collecting many (>106) events, activitydistribution can be reconstructed

PET Basics

3

PET Brain Images

PET tracers use “elements of life” (N, C, O, F) and can be designed to follow specific metabolic pathways

Temporal change in distribution is used to estimate parameters in kinetic models

4

• Many tumors have higher than normal uptake.

• Image the whole body to find metastases.

MetastasesShown withRed Arrows

Brain Heart

Bladder

Normal Uptake inOther OrgansShown in Blue

FDG PET for Cancer / Oncology

5

PET

CTFused PET + CT

*Data courtesy of David Townsend, U. Tenn.

PET / CT for Anatomic Correlation

66

479

50 Gy

417

[18F]FLT

[18F]FAZA

Tumor control

Pre-RT 1 week 2 weeks

[18F]FLT

Pre-RT 1 week 2 weeks

35 Gy

[18F]FAZA

No tumor control Small AnimalPET35 Gy total (no tumor control)Mean tumor volume increase (+60%)Decrease in [18F]FLT uptakeIncrease in [18F]FAZA uptake

50 Gy total (tumor control)Mean tumor volume decrease (-20%)Decrease in [18F]FLT uptakeDecrease in [18F]FAZA uptake

Inhomogeneity of intratumoral uptake

7

• Patient port ~60 cm diameter.• 15 cm axial coverage (patient bed moved for larger span).• 4–5 mm fwhm intrinsic spatial resolution.• ~2% solid angle coverage.• ~ $2 million dollars or more with CT

Images courtesy of GE Medical Systems and Siemens / CTI PET Systems

PET Cameras

8

+ “Parallel” Operation– Expensive– Low spatial resolution

Early PET DetectorsThe Cyclotron Corporation PCT 4600

BGO

¾” PMT

9

05001000150020002500300035004000

Counts

X-Ratio

Y-Ratio

Uniformly illuminate block.For each event, compute

X-Ratio and Y-Ratio,then plot 2-D position.

Individual crystals show upas dark regions.

Profile shows overlap (i.e.identification not perfect).

ProfilethroughRow 2

Early BGO Block Detectors

Many individual crystals multiplexed among a few (4) photomultipliersLSO or LYSO scintillators generally used now

10

Position SensitivePhotomultiplier Tube

Fiber OpticBundle

LSO ScintillatorCrystals

(2x2x10 mm)

*Image courtesy of Simon Cherry, UC Davis

17 cmDetector Ring

Diameter

Small Animal PET Cameras

Miniature Version of “Standard” PET CameraMany crystals coupled to PSPMT

Distortions

12

Photon Counting Noise

1M Events 10 M Events

Higher efficiency is better!

13

2-D (w/ Septa)+ Septa Reduce Scatter

+ Simple image reconstruction– Smaller Solid Angle for Trues

3-D (w/o Septa)– No Scatter Suppression

– More difficult reconstruction+ Larger Solid Angle for Trues

Inter-PlaneSepta

NoSepta

Volume (“3D”) Acquisition Increases Efficiency

14

Event detection probability isproduct of individual photondetection probabilities.

Attenuation depends on entirepath length through object

d1

d2

P1 = e!µ"d1

P2 = e!µ"d2

P = e!µ"(d1+d2)

Annihilation Photon Attenuation

15

Distortion Caused by Attenuation

EmissionDistribution

Linear AttenuationDistribution

Measured Sinogram

Reconstruction

Distorted Image

True Emission Sinogram and Image

16

• Simultaneous decays can causeerroneous coincident events(“randoms”)

• For 3-D PET, randoms can be ashigh as 50% of image.

• Random Rate isRate1 x Rate2 x 2 Δt

• Randoms reduced by narrowcoincidence window Δt.

• Time of flight across tomograph ringrequires Δt > 4 ns.

Random Rate ∝ (Activity Density)2

Random Coincidences

17

• Compton scatter in patientproduces erroneous coincidenceevents

• ~15% of events are scattered in2-D PET(i.e. if tungsten septa used)

• ~50% of events are scatteredin 3-D Whole Body PET

• ~30% of events are scatteredin 3-D Brain PET

• Correspondingly small in smallanimal PET

Compton Scatter

18

• Penetration of 511 keVphotons into crystal ringblurs measured position

• Blurring worsens asattenuation lengthincreases

• Can be eliminated bymeasuring depth ofinteraction

RadialProjection

TangentialProjection

Depth-of-Interaction Uncertainty

19

1 cm

Resolution significantly worse at edge of FOVNear Tomograph Center 14 cm from Tomograph Center

Point Source Images in 60 cm Ring Diameter Camera

Resolution Across FOV

20

Positron RangeDistribution

F-18 C-11 N-13 O-15Max energy (MeV) 0.64 0.97 1.19 1.72Mean energy (MeV) 0.25 0.39 0.49 0.74FWHM (mm) 0.10 0.19 0.28 0.50FWTM (mm) 1.03 1.86 2.53 4.14

[by Levin and Hoffman]

Positron Annihilation Point Distribution

Positron Range Before Annihilation

21

Combined Resolution Effects

• Detector resolution (3D)• Acolinearity (0.5° FWHM – 2.2mm / 1000mm)• Positron range (<1mm for F-18, ~5mm for Rb-82)• Subject motion during scan• Additional blurring in reconstruction

22222

det recmotacoltotrrrrrr ++++= !

22

Goals: Submillimeter PET forMice and Rats

1-2 mm PET for humans inspecific regions of interest

Detour

What makes one imaging systembetter than another?

(Especially in light of “resolution recovery”)

24

Example – Coded AperturesWhich aperture is best?

• 3 different apertures – same overall counting efficiency– 1 Pinhole, 10 pixel FWHM resolution– 10 Pinholes, √10 pixel FWHM– 100 Pinholes, 1 pixel FWHM

25

Expected MeasurementsProjection: 1 pinhole

26

Expected MeasurementsProjection: 10 pinholes

27

Expected MeasurementsProjection: 100 pinholes

28

Reconstructions

OK, which is best??

1 10

100 Original

29

It Depends!

• Images were reconstructed at different resolutions

• Need to decide on desired image resolution toaccomplish task as well as tolerable noise

• Need to compare noise at desired resolution (or vice-versa)

30

Noise vs. Resolution Results

0 2 4 6 8 10 12 14 16 18 200

500

1000

1500

2000

2500

3000

3500

4000

Operating Resolution (mm FWHM)

Limiting Standard Deviation

100 Pinholes

10 Pinholes

1 Pinhole

0.5 1 1.5 2 2.5 3 3.5 4 4.50

200

400

600

800

1000

Operating Resolution (mm FWHM)

Limiting Standard Deviation

6 8 10 12 14 16 18 200

5

10

15

20

Operating Resolution (mm FWHM)

Limiting Standard Deviation

Single pinholeperforms better ifPSF width > itsnatural resolutionis desired

10 pinholes performbetter if PSF widith >their naturalresolution is desired

Each system has a different noise-resolution tradeoff!

31

Desirable PET Detector Characteristics

• Circumferential resolution 1mm or less

• Sufficient depth-of-interaction (DOI) resolution

• Timing resolution < 5ns FWHM (even less is better)

• Low deadtime (system singles countrates >107)

• Good energy resolution (<15% FWHM)

• High detection efficiency

• Ability to resolve multiple hits due to Compton scatter

32

Interactions at 511 keV

• The most desirable coincidence interactions are photoelectric-photoelectric

• Photo-Photo coincidence efficiency less than 17%, 10%, and 4% forBGO, LSO, and NaI, respectively

0.20<<1%>99%<1%Si

0.3418%77%5%NaI

0.8833%61%6%LSO

0.9741%53%6%BGO

Attenuation/cmPhotoelectricCompton CoherentMaterial

Compton scatter is most prevalent interaction of 511 keV photons in PET Detectors!

33

Centroid of Energy Distribution

Crystal array

PSPMT16 mm

511 keVphotons

16 mm

BGO LSO NaI

2mm x 2mm x 20mm thick crystals

EGS4 Monte Carlo Simulations

34

Does It Affect Performance?

DOI uncertainty included in both images

True first interaction position Centroid of scattered E distribution

- EGS4 Monte Carlo simulations- BGO PET ( 17.6 cm I.D. 16 cm length segmented with 3 mm x 3mm x 20 mm crystals)- Point sources at 0, 3, 6, 9, 12, 15, and 18 mm from center of FOV- Filtered back projection reconstruction

High Resolution Small Animal PETArtist’s Conception

2nd detector(non position sensitive)

1st detector(high resolution)

First Concept

36

“Compton” PET

BGOdetector

Sidetector

Si-Si

BGO-BGO

Si-BGO

Very high resolution achievable in small field-of-view

37

High Resolution Imaging Probes

• Do not need complete inner detector – partial high resolutiondetector sufficient in many cases

• Can potentially be used in conjunction with existing PETinstruments

• Probes for head & neck cancer and prostate imaging currentlyunder development

511 keV photon Scattered photon

LOR (position uncertainty)

BGO - BGO

FOV

Si - BGO

PET Probe

detector

Position

tracker

38

Effects of Detector Resolution

• Already large uncertainty along path of annihilation photons(undone by tomographic reconstruction)

• Resolution determined primarily by uncertainty transverse to thephoton paths

θ1 θ2

D1 D2

Detectors

( ) ( ) ( )( )2

22

22

22

222

11

22

11

22cossincossin135.2

CDCDDR !"!"#!"!"# +++$%

α (1-α)

RDAnnihilationPhoton Path

39

Spatial Resolution• Detectors of 1mm, 2mm, and 3mm FWHM in coincidence with

6mm FWHM detector

0 5 10 15 20 25 30 35 400

1

2

3

4

5

6

Distance from probe face (cm)

Resolution (mm FWHM)

Probe resolution = 1mm FWHM

2 mm

3 mm

Spatial resolution improves close to detector with good resolution

40

Can It Have Enough Efficiency?

0

5

10

15

20

25

0 10 20 30 40 50 60

BGO diameter (cm)

Ab

so

lute

eff

icie

ncy (

%) BGO-BGO

Si-BGO

Si-Si

0

2

4

6

8

10

0 0.8 1.6 2.4 3.2 4 4.8 5.6 6.4 7.2

Silicon thickness (cm)

Ab

so

lute

eff

icie

ncy (

%) Si-BGO

Si-Si

Further evaluate a system having the following characteristics

• Silicon 4cm ID, 7.2cm OD (16 layers of 0.3 mm × 0.3 mm × 1 mm elements)

• BGO detector 17.6 cm diameter, 16 cm length, and 2 cm thicknesssegmented into 3 mm × 3 mm × 20 mm crystals

Simulated System and Results

3 x 3 x 20 mm3 BGO Crystals

0.3 x 0.3 x 1 mm3 Silicon Pixels, 16 layers

16 cm

4 cm

4 cm

1.6 cm

2 cm

17.6 cm

Lower E Threshold: 350 keV

Interaction Selection Method:BGO crystal with Maximum E

Si-Si Sensitivity: 1.0 % *FWHM = 230 µm

Si-BGO Sensitivity : 9.0% *FWHM = 790 µm

BGO-BGO Sensitivity: 21.0 % *FWHM = 1.45 mm

Image reconstruction: FBP

* Does not include acolinearity and positron range

Overall Spatial Resolution

Spatial Resolution (mm FWHM) Event Geometric Geometric Overall + Acolinearity F-18 C-11 N-13 O-15 Si-Si 0.234 0.241 0.393 0.443 0.492 0.553 Si-BGO 0.788 0.816 1.062 1.261 1.419 1.742 BGO-BGO 1.452 1.458 1.977 2.270 2.490 3.069

Si-Si Si-BGO BGO-BGO

Simulated Multiple Disk Sources Object

Diameters of Disks : 1, 2, 4, 6, 8, and 10 mm

Center of disks : 1 cm from center of FOV

-2 -1 0 1 2

x (cm)

-2

-1

y (cm) 0

1

2

2D Images of Simulated Multiple Disk Sources

Si-Si (100k) Si-BGO (800k) BGO-BGO(1.9M)

- Filtered back projection (FBP)- BGO crystal with Maximum Energy was used- Images were reconstructed with different system efficiencies

45

Combined Reconstruction Using ML

- Maximum likelihood Expectation Maximization (ML-EM)- Iteration number = 200

Si-Si (160k) + Si-BGO (1.4M) + BGO-BGO (3.1M)

4 cm x 4 cm

46

Can a Small Amount of Hi Res Data Havean Effect on Performance?

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

10

20

30

40

50

60

norm(f-g)/norm(f)

sqrt(variance)

Desired resolution = 1mm FWHM Gaussian

Low Res data only

Low & Med Res

All data / Med & Hi Res data

0 0.2 0.4 0.6 0.8 10

0.05

0.1

0.15

0.2

0.25

norm(f - g) / norm(f)

sqrt(variance)

Low Res

Low + Med

Med + High

All

4mm FWHM Gaussian

47

Experiments

48

Silicon Pad Detectors for Compton Camera

• Double-metal Si paddetectors

• 1.4mm x 1.4mm pads, 16x 32 array

• 0.5mm and 1mmthicknesses

• Full depletion ~180V for1mm

• Readout via 4 x IdeasVATA GP-3 ASICs

49

VATA Readout ASICs

GP-3: 2.5-3 µs shaping in slow channel200 ns peaking time in fast channel

Serial, sparse, sparse + adjacent channel readout

Energy Resolution

Tc-99m (140.5 keV)Am-241 (59.5 keV)

FWHM = 1.39 keV (0.99%)Pb Kα1 = 74.969 keV, Kα2 = 72.804 keV,Kβ1 = 84.936 keV, and Compton edge = 49.8 keV

FWHM = 1.49 keV (2.5 %)

Experimental Setup

Silicon detector

Source turntable

Lead shielding 1mm Tungsten Slit

Laser

Silicon detector

52

Experimental System with BGO

BGO detectors flanksilicon for energy resolution

53

Lines of Response for Point Source

Silicon detector 1LOR

Silicon detector 2

F-18 Source

Random Coincidences

Lots of randoms – coincidence window was ~2.5 µs

Point Source ComparisonMicroPET R4Compton PET

ML-EM Image reconstruction (no detectormodeling)Si-Si coincidence events only

0 1 2 3 4 5 cm

5

4

3

2

1

0

0 1 2 3 4 5 cm

5

4

3

2

1

0

Source

1.1~1.2 mm

0.4 mm

F-18

Glass wall0.2 mm

F-18 in glass capillary tubes

MAP Reconstruction

Intrinisic Resolution Measurment

Needle 25G (ID = 0.254 mm, OD = 0.5mm, SS_steel wall = 0.127 mm)

5

4

3

2

1

0

0 1 2 3 4 5 cm

0.254 mm

0.127 mm

Image Resolution= 700 µm FWHM

SS_steel wall

F-18

Resolution Uniformity

0 1 2 3 4 5 cm

5

4

3

2

1

0

Source pairs at 5, 10, 15, & 20mm off-axis

Sinogram

The sources in each pair are clearly separated at appropriate sinogram angles

57

New Pad Detectors

1040 (26 x 40) 1mm x 1mm pads, 1mm thick Co-57 Spectrum

Should allow 0.5 – 0.6mm FWHM spatial resolution

58

Challenges

• Number of electronics channels (1.25M for simulated system)

• Packaging

• Triggering threshold

• Time resolution

• Event classification

• Comparing performance with more conventional PET

59

Packaging

Double-sided 14-layer hybrid (FR4) allows 4mm active detector in 5.6mm (~70%)Won’t work for 1mm x 1mm 1040 pad detectors!

Stacked hybrids, side viewNew stackable hybrid

60

Packaging -- TAB Experiements

Al traces 8 µm thick x17 µm wide

40 µm dielectric inserts

Packing fraction >90%

61

Triggering Threshold

• New hybrid routes analog anddigital signals to separatecables

• Trigger threshold can be set aslow as 3 keV

• 5.8 keV emission of Fe-55clearly seen (source <10 cps)

• But threshold spread too broadfor trimDAC alignment on VATAGP-3

62

Timing• Desired time resolution <10ns

FWHM

• Marginal timing is evident

• Slower signal generation fromevents near backplane

• Large range of pulse-heightcoupled with leading-edge triggeris biggest issue

• Large time-walk is the resultBGO-Silicon timing spectrum for 511 keV source

63

Simple Signal Generation Model

0 50 100 150 200 250 3000

0.5

1

1.5

2

2.5

x 10-9

Time (ns)

Signa;

200 V = Vdep

Threshold

0 50 100 150 200 250 3000

0.5

1

1.5

2

2.5

x 10-9

Threshold

500V = Vdep + 300V

Time (ns)

Signal

holes

electrons

p+ implant

n+ backplane + Alelectrode

n-bulk

1.4 mm

1 mm InteractionLocations

Simplified Pad Detector

• Depth dependence of signal evident

• Non-linear effect with depth

• Higher bias helps

64

But…It gets worse!

• Pad size is nearly same asdetector thickness

• Weighting potential dependson x & y in addition to z

• Result is additional jitter due tounknown 3D interactionlocation

65

T-CAD Simulation / Measurements

Energy deposited in Si detector vs. triggering time

66

Solutions

• VATA GP-7 with 50 ns peaking time in fast channel• Low trigger threshold + secondary cut on energy

• Use 2 pad-to-pad 0.5mm thick detectors• Detector redesign – readout from both sides• Sum pad + 8 neighbors for more uniform triggering• Complete redesign of readout ASICs

Immediate

Longer Term

Reduction of scatter,

random,

and misclassified events

Event Classification -- Compton KinematicsCompton Kinematics

Angular Uncertainty Factors

Si

BGO

ESi

Doppler Broadening Detector Element Size Energy Resolution Photon Acolinearity

Si pad:

0.3 mm x .3 mm x 1 mm

BGO crystal:

3 mm x 3 mm x 20 mm

BGO

Si

EBGO

68

Reducing Positron Range

Distance (mm)

0 Tesla

-4 -2 0 2

-4

-3

-2

-1

0

1

2

3

Distance (mm)

9 Tesla XZ-Plane

-4 -2 0 2

-4

-3

-2

-1

0

1

2

3

9 Tesla XY-Plane

-4 -2 0 2

-4

-3

-2

-1

0

1

2

3

• Embed PET FOV in strong magnetic field (Raylman, Hammer, etc.)• Positrons spiral and range is reduced transverse to B-field vector• Not very effective for F-18 positrons• Potentially useful for emitters with higher endpoint energies (I-124, Tc-94m,

etc.) increasingly being used in small animal imaging

Simulated PSF for I-124 at 0T and 9T

69

MRI-Safe Silicon PET Imager

System rebuilt using no ferromagnetic materials…

…and inserted into boreof 7T MRI magnet at OSU

70

Ga-68 Resolution Improvement at 7T

71

Next Steps for Compton PET?

• Complete BGO / silicon PET test bench

• Test 1mm detectors with VATA GP-7s

• Attempt comparison in terms of noise-resolutiontradeoff with more conventional PET techniques

72

Prostate Cancer

73

C-11 Choline PETMay be correlated with prostate cancer aggressiveness

PET/CT Fusion Image PET / Histoloogy Fusion

Promising, however, difficult to detect small tumors even with high uptake

74

Internal High Resolution Probe

Geant 4 Monte Carlo simulation of internal prostate probe

External PET ringInternal probe

75

Noise Advantage vs. Resolution

1 2 3 4 5 6 7 8 9 100

2

4

6

8

10

12

14

Reconstructed Resolution (mm FWHM)

Hi-Res Noise Advantage

76

Probe Construction and Performance

Courtesy of Stan Majewski and James Proffitt -- JLab

77

Acknowledgments

UMichLes RogersScott WildermanLi HanSam HuhSang-June ParkBob KoeppeDave RaffelMorand Piert

OSUHarris KaganKlaus HonscheidDon BurdetteEric CochranShane Smith

CERNPeter WeilhammerEnrico ChesiAlan Rudge

IFIC/CSIC (Valencia)Carlos LacastaJuan FusterGabriela Llosa

IJS (Ljubljana)Marko MikuzGregor KrambergerAndrej StudenDejan Zontar

Gamma-Medica IdeasEinar NygardDirk MeierBjørn SundahlSindre Mikkelsenetc.

LEPSI (Strasbourg)Wojtek Dulinski

LBNLBill Moses

JLABStan MajewskiJames Proffitt