1

PET System Design, Acquisition and Image Reconstruction

Frederic H. Fahey DSc

Children’s Hospital BostonHarvard Medical School

frederic.fahey@childrens.harvard.edu

PET: Back in the day…

• Research Device• Single Slice• Run by the research team• Neuroscience

PET III BNL - 1980

Recent Growth in PET

• Oncology• Regional Distribution of 18F FDG• Reimbursement

Migration of PETResearch Lab Clinic

• Simpler to run • More robust• Multi-Slice (15 cm volume)• Multi-Modality (PET-CT)

2

Outline

• Design criteria for State-of-the-Art Scanner• Basics of PET Scanner Design• Review of Current PET Instrumentation

Design Criteria for a PET Scanner

• Image Quality– Sharpness (resolution)– Contrast – Noise (sensitivity)

• Ease of Use (software, patient access)• Fast (throughput, sensitivity & count rate)• Robust• Not TOO Expensive

Positron Emission18F

511 keV

511 keV β+

e-

Detector Ring

3

Detector Blocks (GE Advance NXi)Large Crystal Designs

•Large NaI detectors, PMT array•25x50 cm, 1” thick•Hexagonal array

True, Scatter and Random Coincidence Detections

True

Random

Scatter

Randoms Estimation

• Background Subtraction• Singles Rate Calculation

R = 2 τ N1 N2

• Delay Window Method

4

SCINTILLATOR NaI(Tl) BGO LSO GSO

Rel. Light Yield 100 15-20 75 20-25

Peak Wavelength (nm) 410 480 420 440

Decay Constant (ns) 230 300 12,42 30-60

Density (g/cc) 3.67 7.13 7.40 6.71

Effective Z 51 75 66 59

Index of Refraction 1.85 2.15 1.82 1.85

Hygroscopic ? Yes No No No

New Detector Materials

Sinogram

Image for Each Slice

Ang

le

Image for Each Angle

Projection View

Slic

e

Note: Sinograms and projection views are different ways or showing the same data.

5

PET Sinograms• Point in transverse slice maps to sine wave• Displacement (x) vs Angle (y)• Each row is a projection through the object at

the corresponding angle• Each detector is mapped along a diagonal• Each pixel in the sinogram corresponds to a

particular “line of response” (LOR) i.e. detector pair

18 * *17 * * *16 * * *15 * * *14 * * *13 * * *12 * * *11 * * *10 * * *

9 * * *8 * * *7 * * *6 * * *5 * * *4 * * *3 * * *2 * * *1 * *

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

Span of 3 Michelogram

18 * * * *17 * * * * *16 * * * * * *15 * * * * * * *14 * * * * * * *13 * * * * * * *12 * * * * * * *11 * * * * * * *10 * * * * * * *

9 * * * * * * *8 * * * * * * *7 * * * * * * *6 * * * * * * *5 * * * * * * *4 * * * * * * *3 * * * * * *2 * * * * *1 * * * *

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

Span of 7 Michelogram

6

Acquisition Modes

2D

3Dseptaare

removed

16 * * * * * * * * * * * * * * * *15 * * * * * * * * * * * * * * * *14 * * * * * * * * * * * * * * * *13 * * * * * * * * * * * * * * * *12 * * * * * * * * * * * * * * * *11 * * * * * * * * * * * * * * * *10 * * * * * * * * * * * * * * * *

9 * * * * * * * * * * * * * * * *8 * * * * * * * * * * * * * * * *7 * * * * * * * * * * * * * * * *6 * * * * * * * * * * * * * * * *5 * * * * * * * * * * * * * * * *4 * * * * * * * * * * * * * * * *3 * * * * * * * * * * * * * * * *2 * * * * * * * * * * * * * * * *1 * * * * * * * * * * * * * * * *

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

SliceOrientation

3D Michelogram, RD = 15

16 * * * * * * * * * * * * * * * *15 * * * * * * * * * * * * * * * *14 * * * * * * * * * * * * * * * *13 * * * * * * * * * * * * * * * *12 * * * * * * * * * * * * * * * *11 * * * * * * * * * * * * * * * *10 * * * * * * * * * * * * * * * *

9 * * * * * * * * * * * * * * * *8 * * * * * * * * * * * * * * * *7 * * * * * * * * * * * * * * * *6 * * * * * * * * * * * * * * * *5 * * * * * * * * * * * * * * * *4 * * * * * * * * * * * * * * * *3 * * * * * * * * * * * * * * * *2 * * * * * * * * * * * * * * * *1 * * * * * * * * * * * * * * * *

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Segment 1

Segment 2

Segment 3

18 * * * * * * * * * * * *17 * * * * * * * * * * * * *16 * * * * * * * * * * * * * *15 * * * * * * * * * * * * * * *14 * * * * * * * * * * * * * * * *13 * * * * * * * * * * * * * * * * *12 * * * * * * * * * * * * * * * * * *11 * * * * * * * * * * * * * * * * * *10 * * * * * * * * * * * * * * * * * *

9 * * * * * * * * * * * * * * * * * *8 * * * * * * * * * * * * * * * * * *7 * * * * * * * * * * * * * * * * * *6 * * * * * * * * * * * * * * * * *5 * * * * * * * * * * * * * * * *4 * * * * * * * * * * * * * * *3 * * * * * * * * * * * * * *2 * * * * * * * * * * * * *1 * * * * * * * * * * * *

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

1

2

3

4

5

6

7

8

9

10

11

GE NXi 3D Projection view and Michelogram

7

3D PET• Sensitivity drops off towards edges • 4-5X increased sensitivity overall• Increased scatter (15% to 40%)• Increased randoms from out-of-field activity• Rebinning algorithms to apply 2D

reconstruction• Some devices can acquire in 2D or 3D whereas

some can only acquire in 3D• 3D in Brain, 2D (or 3D) in Whole Body

PET Attenuation Correction Methods• Calculated

– No noise but possibly inaccurate• Measured

– Accurate but noisy• Segmented, Measured

– Less noise => less time• Singles• CT

Calculated Attenuation Correction Measured Attenuation Correction

8

Measured Attenuation CorrectionSegmented, Measured Attenuation Correction

•Noise added from measured attenuation correction•Rel err in unif phantom (10 min EM)

•9% with calc atten•16% with 10 min TR•18% with 5 min TR

•Segmentation classifies by tissue type•Smoothes lung areas•Substantial reduction in noise added

Measured

Segmented

Singles-Based Attenuation Correction

•Single-photon source (Cs-137)•Source shielded from “near”detectors•Energy resolution (NaI) allows separation of 2 peaks•High count rate reduces noise•Susceptible to scatter

Courtesy of GE Medical SystemsPET-CT

9

PET-CT Attenuation Correction PET-CT Attenuation Correction

Iterative Reconstruction Feedback Loop

Backprojection

Simulated Projections

ActualProjections

Compare

Use to improve current estimate

Current Estimate

Error

Courtesy of Jerold W. Wallis, M.D.

Maximum Likelihood Reconstruction (ML-EM)

• Maximize the likelihood that the estimatedactivity distribution in the body (the reconstructed transaxial slices) would lead to the measured projections

• Use the expectation maximization (EM) algorithm to iteratively estimate the activity distribution

10

pixelj

projection bini

aij is the probability that a photon emitted from pixelj is detected at projection bini.

EM-ML Reconstruction

• aij can contain physical information (effects of spatial resolution, scatter, attenuation…)

• EM-ML algorithm takes into account the nature of the noise (quantum mottle) in the projection data

• Can yield a more accurate reconstruction

λ’(k)j = λ’(k -1)j Σ [aij {d/(d’(k-1)}]Σ [aij]

λ’(k)j is the newest estimate of the object pixel value

λ’(k -1)j is estimate of the object at last iteration

d is the projection data

d’ is the calculated projection data

OSOS--EM AlgorithmEM Algorithm• Ordered-subset expectation maximization• At each step, project and backproject at only some

angles (i.e. a subset)• Perform the steps in an ordered way to include all

angles• Data start to converge even before the 1st iteration

is complete• Convergence achieved in 3 - 10 iterations• Computation time of a few minutes• For GE, we use 28 subsets (12 projections per

subset) and 2 iterations.

11

OSEM Iterative Reconstruction

• Filtered Back-Projection– Fast– Robust– Subject to noise &

streaks• OSEM

– Almost as fast– Handles noise &

streaks

3D Reconstruction

• 2D reconstructions are much easier than 3D• Rebinning 3D data to be reconstructed as

2D data– Single-slice– Fourier

• Fully 3D Reconstruction

Rebinning3D Data Acquisition

N2 ObliqueSinograms

2N-12D Sinograms

2D Reconstruction

3D Object

2DSlices

Single Slice Rebinning (SSRB)

12

Fourier Rebinning

• More accurate approach to rebinning• Better estimate of determining into which

parallel plane oblique data should be placed• Based on the frequency-distance

relationship (Value of Fourier transform of a sinogram receives contributions mainly from sources at a fixed distance t=-k/ω)

Fourier Rebinning (FORE)

• Initialize a stack of Fourier transforms of 2D sinograms

• For each oblique sinogram– Take 2D FT– For each pixel in Fourier space, calculate the

interpolated plane location {z’ = z – k/ω tan(θ)}– Add values from oblique sinogram to 2D sinogram at z’

• Normalize to take into account over-sampled areas• Take inverse Fourier transform• Reconstruct interpolated sinograms as 2D

Note that this only requires a 1D interpolation along z.

PET Instrumentation

• High-Resolution• Medium-Resolution• PET-CT

PET Instrumentation

• High-Resolution– Siemens HR+– GE Advance– Philips Allegro

• Medium-Resolution– Siemens EXACT/ACCEL– Philips CPET+

13

PET InstrumentationSiemens HR+ GE Advance Philips Allegro

Detector Dimension (mm) 4.1 x 4.4 x 30 3.9 x 8.2 x 30 4 x 6 x 20# of Detectors 18,432 12,096 17,864Detector Material BGO BGO GSOSpatial Resolution (mm) 4.6 4.8 4.8Sensitivity (kcps/uCi/mL) 200/900 200/1060 /800

Siemens EXACT Siemens Accel Philips CPET+Detector Dimension (mm) 6.8 x 6.8 x 20 6.8 x 6.8 x 20 500x300x25# of Detectors 9,216 9,216 6Detector Material BGO LSO NaISpatial Resolution (mm) 6 6.2 5Sensitivity (kcps/uCi/mL) 180/780 180/780 /450

GE Advance NXi

PET-CT Scanners

• Siemens Biograph (BGO or LSO)• Siemens Hi-Rez• GE Discovery LS• GE Discovery ST• Philips Gemini

GE Discovery LS

• LightSpeed Plus CT (4-16 slice, 0.5sec gantry).• Full-featured Advance NXi PET.

– Retractable septa for 2D and 3D imaging.

LightSpeed PlusAdvance NXi

Courtesy of GE Medical Systems

14

Courtesy of GE Medical Systems

GE Discovery ST

GE Discovery ST

CT PET

GE Discovery ST (vs LS)

• Larger Detectors (6 vs 4 mm)• Shorter septa (5 vs 10 cm)• Higher Sensitivity• Larger patient bore (70 vs 50 cm)• No rod sources• 2D and 3D imaging

15

• ECAT EXACT HR+:• Siemens Somatom

Emotion: High performance, spiral CT

• 70 cm patient port • Optimized bed design

Siemens Biograph

Courtesy ofSiemens Medical Systems

Siemens Hi-Rez Detectors

Courtesy of Siemens

CURRENT HI-REZDetector material LSO LSOBlock matrix 8 x 8 13 x 13 Creates smaller voxels!Crystal size 6.4 mm x 6.4 mm 4.0 mm x 4.0 mm Enables finer resolution!Crystal thickness 25 mm 20 mmNumber of blocks 144 144Total number of crystals 9,216 24,336 Allows greater sensitivity!Axial field of view 16.2 cm 16.2 cmNumber of crystal rings 24 39 Permits more slices!Number of image planes 47 81 More slices!Plane spacing 3.4 mm 2.0 mm Excellent sampling!Ring diameter 83 cm 83 cmSpatial resolution 6.3 mm 4.6 mm Finer spatial resolution!Volumetric resolution 250 mm3 98 mm3 Finer volumetric resolution!

Siemens Biograph with LSO HI-REZ detectors

*510(k) pending, not for sale in the US

HI-REZ*

HR+

CONVENTIONAL

Note definition of sulcus and gyrus!

Courtesy of Siemens

Philips Gemini

Allegro PET Scanner andMx8000 16 slice CT scanner

Courtesy of Philips Medical Systems

16

PET-CT Scanners

GE Discovery LS GE Discovery ST Philips GeminiDetector Dimension (mm) 4 x 8 x 30 6.2 x 6.2 x 30 4 x 6 x 20# of PET Detectors 12,096 10,080 17,864PET Detector Material BGO BGO GSOSpatial Resolution 4.8 6.2 4.92D/3D 2D/3D 2D/3D 3DAtten Corr CT&Ga-68 CT CT&Cs-137

Siemens Biograph BGO Siemens Biograph LSO Siemens Hi-Rez LSODetector Dimension (mm) 4.1 x 4.4 x 30 6.5 x 6.5 x 25 4 x 4 x 20# of PET Detectors 18,432 9,216 23,336PET Detector Material BGO LSO LSOSpatial Resolution 4.5 6.3 4.62D/3D 3D 3D 3DAtten Corr CT CT CT

Special PET Devices

• Brain • Breast • Small Animal

CTI ECAT HRRT• Max Planck Institute in

Cologne • Dual LSO Phoswich for

DOI Determination• Octagonal Design• 936 blocks with 128 2.1x2.1

dual detectors each• 120,000 crystals• Reconstructed resolution of

less than 2.5 mm

Courtesy of CTI

Positron Emission Mammography (PEM)

Courtesy of Wake Forest Universityand PEM Technologies

17

Positron Emission Mammography (PEM)

Courtesy of Wake Forest University

Concorde MicroPET Scanner

650g Rat

[18F] Fluoride

4 bed positions30 min each

Courtesy of Concorde MicroSystems

GE eXplore Vista Scanner•Rodent system•11.8 cm diameter•4.6 cm axial FOV•1.6 mm resolution in CFOV

Philips Mosaic

•16,680 GSO crystals•2.1 mm resolution•11.8 cm axial FOV•137Cs Transmission

18

Summary

• Scanners specifically optimized for clinical (oncologic) imaging

• Sensitivity and count rate capability• New detector materials (faster!)• PET-CT• Small animal imaging