What do PET clinicians/researchers want? What can the PET physicist deliver?

1
Radiologic Physics Nuclear Medicine
PET Imaging and Quantification
Suleman Surti surti_at_mail.med.upenn.edu (215)
662-7214
2
Positron decay
11C t1/2 20 minutes 13N t1/2 10 minutes
15O t1/2 2 minutes 18F t1/2 110 minutes
3
Positron Emission Tomography
A primary goal and usefulness of a tomographic
imaging modality such as PET is to achieve images
where the intensity of each voxel in the image is
proportional to the activity concentration
present in the corresponding location in the
patient
4
True, Scatter, Random coincidences in PET

• Trues ?2 . A
• A Activity
• stopping power
• Scatters k . Trues
• k energy threshold
• (depends on energy resolution)
• Randoms 2? . (? . A) 2
• 2? coincidence timing window
• (depends on decay time/light)

5
Count-rate Performance
70-cm long phantom (20-cm diameter) NEMA NU2-2001
Noise Equivalent Count-rate NEC
T/(1S/TR/T)
Philips Gemini TF Univ. of Pennsylvania
6
Limits on spatial resolution

• Positron range, R

• Photon non-collinearity
• FWHMNC0.0022 X scanner diameter
• (2-mm for a 90-cm diameter)

18F 11C 82Rb
Rmax (mm) 2.6 3.8 16.5
FWHMp (mm) 0.22 0.28 2.6

• Detector resolution (FWHMd )

7
PET Instrumentation Design

• Scintillators stopping power, speed,
  light output
• Detector configuration scintillator -
  photo-sensor coupling
• Scanner geometry field-of-view (axial)
• 2-dimensional vs. 3-dimensional
• Time-of-flight PET
• Data processing / image reconstruction scatter,
  randoms and attenuation correction
• iterative reconstruction algorithms

8
Comparison of Scintillators
9
Scintillation Detector
Photo-Multiplier Tube (PMT)
Scintillator
10
Small crystals require position encoding
Block Detector
CTI HR (1995)
BGO 8 x 8 array 4 x 4 x 30 mm3 19 mm PMTs (4)
18,432 crystal elements (32 rings) 1,152 PMTs
11
Block vs. Quadrant Sharing
Quadrant Sharing Block (W.-H. Wong)
Standard Block (Casey-Nutt)
Similar spatial resolution with larger
PMTs or Better spatial resolution with similar
size PMTs
12
Continuous optical coupling
More uniform light output -gt better energy
resolution Similar spatial resolution with larger
PMTs
Example Philips Allegro (2001)
17,864 crystal elements (GSO) 420 PMTs
13
2D (septa) vs. 3D (no septa)
2D Imaging
3D Imaging
14
Energy threshold reduces scatter random
coincidences- particularly in 3D
15
NEC Count-rates - 2D vs. 3D
GE Advance
70-cm long phantom NEMA 2001
S.Kohlmeyer and T. Lewellen University of
Washington
2D 2001
3D 2001 (380 keV) (300 keV)
0.14 ?Ci/cc
16
High count-rate capability in 3D PET requires
fast, dense scintillator with good energy
resolution
LSO ? 40 ns 0.81/cm BGO ? 300 ns 0.91/cm
Compare 3 CTI scanners LSO Accel, BGO EXACT, BGO
HR (2D)
3D Accel
2D HR
NEC (cps)

• Both measurements assume randoms smoothing
• Courtesy of CTI, inc

3D EXACT
17
Time-of-flight PET

• Can localize source along line of flight -
  depends on timing resolution of detectors
• Time of flight information reduces noise in
  images - weighted back-projection along LOR

?t uncertainty in measurement of t1-t2
?x uncertainty in position along LOR
c . ?t/2
D/?x reduction in variance or gain in
sensitivity
18
Does Noise-Equivalent Count-rate (NEC) infer
Image Quality?
NEC Trues / (1 Scatter/Trues
Randoms/Trues) NEC1/2 Signal / Noise

• NEC includes global effects
• Trues
• Noise from scatter and randoms

• NEC does not include local effects
• Spatial resolution - variations within FOV
• Image reconstruction
• Accuracy of scatter and randoms correction
• Attenuation correction
• Deadtime corrections and normalization

19
Fully 3D Iterative Reconstruction improves image
quality
Philips Allegro
Filtered Backprojection
3D Ramla
20
Positron Emission Tomography
What is needed to achieve quantitative PET images?

1. Deadtime correction
2. Data Normalization
3. Scatter correction
4. Randoms correction
5. Attenuation correction

21
Deadtime correction

• Deadtime High count-rate effect present in
  radiation detectors
• Two manifestations
• Pulse pileup Events are collected but
  measurements such as energy and spatial position
  are affected (reduced image quality)
• Loss of counts Due to electronics deadtime and
  determined mainly by scintillator decay time
• Loss of counts corrected by measuring collected
  counts vs activity in a uniform cylinder

22
Data normalization

• Normalization non-uniformities in event
  detection over the full scanner
• Two sources
• Variation in amount of scintillation light
  collection due to crystal non-uniformities and
  detector design (detector effect)
• Difference in detection sensitivity due to angle
  of incidence

23
Data normalization techniques
24
Scatter Correction (SSS)

• Contribution to LOR AB
• from each scatter point
• Activity distribution and
• Klein-Nishina equation
• Repeat for all LORs to get
• scatter sinogram

P188
25
Randoms Correction Delayed window technique
26
Why do we need attenuation correction?

• More accurate activity distribution uniform
  liver, cold lungs
• Improved lesion detectability deep lesions
• Reduce image artifacts and streaking reconstru
  ct using consistent data
• Improved image quality with iterative
  reconstruction include attenuation into model

Butattenuation correction must be FAST -
compared to emission scan ACCURATE - e.g. near
lung boundary LOW NOISE - minimize noise
propagation
27
Attenuation correction can be calculated
directly in PET
patient
Total path length, Dd1d2 D can be
independently measured and allows an accurate
correction
PET High energy photons with small ?, but pair
of photons must traverse entire body width.
d1
d2
I/I0 e -?d1 e-?d2 e-?(d1d2)
?(511kev) 0.095/cm
I/I0 e-?d1 e-?d2 0.06 for D30cm
28
Transmission sources for attenuation measurements

1. PET transmission source (68Ge/68Ga) - source of
   coincident annihilation photons (mono energetic _at_
   511 keV), 265 day half life
2. Single photon source (137Cs) - source of single
   ?-rays (mono energetic _at_ 662 keV), 20 yr
   half-life
3. X-ray CT scan - source of X-rays with a
   distribution of energies from 30 to 120 keV. We
   can assume an effective energy of 75 keV

(Recall that the PET emission data is attenuated
at 511 keV)
29
Transmission Scan
137Cs point source 662 keV, t1/2 30 yr
d2
d1
d1 d2 D
Emission
I / I0 e-?d1 . e-?d2 e-?D
Transmission
I / I0 e-?D
30
Post-injection transmission scan
Philips Allegro

• 20 mCi 137Cs pt src
• 40 sec Tx acquisition
• Energy scaling
• EC subtraction
• Segmentation
• Interleaved Em-Tx
• 7 Em frames 9 Tx frames

University of Pennsylvania PET Center
31
CT-based attenuation correction threshold method
STEP 1 Separate bone and soft tissue using
threshold of 300 H.U.
STEP 2 Scale to PET energy 511 keV.
Scale factors (51170 keV) bone 0.41, soft
tissue 0.50
STEP 3 Forward project to obtain attenuation
correction factors.
Kinahan PE, Townsend DW, Beyer T, et al. Med
Phys. 1998 25(10) 2046-2053.
32
Potential problems for CT-based attenuation
correction

• Difference in CT and PET respiratory patterns
• Can lead to artifacts near the dome of the liver
• Use of contrast agent
• Can cause incorrect values in PET image
• Truncation of CT image due to keeping arms down
  in the field of view to match the PET scan
• Can cause artifacts in corresponding regions in
  PET image
• Bias in the CT image due to beam-hardening and
  scatter from the arms in the field of view

33
Attenuation correction for PET
Types of transmission images
Single photon Cs-137 (662 keV) lower noise 5-10
min scan time some bias lower contrast
X-ray (30-130 keV) no noise 1 min scan
time potential for bias high contrast
Coincident photon Ge-68/Ga-68 (511 keV) high
noise 15-30 min scan time low bias low contrast
Alessio AM, Kinahan PE, Cheng PM, et al. Radiol.
Clin. N. America 2004 42(6) 1017-1032.
34
Attenuation correction - increased confidence of
liver lesion
No AC
AC
Philips Allegro
University of Pennsylvania PET Center
35
Attenuation correction - better comparison of
relative activity of deep (mediastinum) vs.
superficial (axilla) lesions
No AC
AC
Philips Allegro
University of Pennsylvania PET Center
36
Image quality degrades with heavy patients
Slim 58 kg Normal 89 kg
Heavy 127 kg
Increasing attenuation (less counts)
Increasing scatter (more noise)
Increasing volume (lower count density)
37
How can we improve image quality?
2D - counts limited by septa and maximum allowed
dose 3D - counts limited by dead-time and randoms
Scintillator High stopping power - higher
coincidence fraction Fast decay - lower
dead-time and randoms Energy resolution - lower
scatter and randoms Geometry Sensitivity
(Axial FOV)2 (increased scintillator and
PMT cost) Time-of-flight Requires very fast
scintillator with excellent timing resolution
38
Time-of-flight PET

• Can localize source along line of flight -
  depends on timing resolution of detectors
• Time of flight information reduces noise in
  images - weighted back-projection along LOR

?t uncertainty in measurement of t1-t2
?x uncertainty in position along LOR
c . ?t/2
D/?x reduction in variance or gain in
sensitivity
39

Time-of-flight PET
PET scanner 70-cm bore 18-cm axial FOV
CT scanner Brilliance 16-slice
Philips Gemini TF Univ. of Pennsylvania
PET shows increased FDG uptake in region of porta
hepatis CT demonstrates that this uptake
corresponds to the gallbladder representing acute
cholecystitis, not bowel activity
40
Phantom measurements
41
Gemini TF
Heavy-weight patient study
13 mCi 2 hr post-inj 3 min/bed
Colon cancer
119 kg BMI 46.5
MIP
non-TOF
LDCT
TOF
42
Clinical 18F-FDG imaging

• Clinical 18F-FDG imaging essentially involves two
  tasks
• Identifying regions with abnormal uptake (lesion
  detection)
• Deriving a measure of glucose metabolism in these
  regions (lesion estimation task)

43
Factors affecting lesion detection and activity
estimation

• Accuracy of scanner normalization and corrections
  for deadtime, scatter, randoms, attenuation
• Remove biases with minimal noise propagation
• Spatial resolution
• Lesion size and partial volume effects
• Lesion activity uptake relative to background
• Scan time
• Reduced noise
• Patient habitus
• Determines amount of Sc, R, and attenuation
• Reconstruction
• Determines amount of noise in image and for
  iterative algorithms plays off contrast recovery
  with noise

44
Summary

• PET scanner design is still an evolving area of
  research with new scintillators and
  photo-detectors being developed
• Current generation of clinical scanners achieve
  spatial resolution of 4-5 mm
• Fully-3D imaging is imaging mode of choice
• PET is still count limited
• TOF PET can help improve the statistical quality
  of PET images
• PET/CT as a multi-modality imaging device has
  increased the confidence in interpreting PET
  images
• Future direction - PET/MRI scanners

45
18F-Fluoro-Deoxy-Glucose (FDG)
OH
Ido et al. 1978
O
H
Glucose Blood -gt tissue -gt cell phosphorylation -
glycogen
OH
OH
OH
FDG Blood -gt tissue phosphorylation
18F
Patient injected activity 10 mCi 3.7 x 108
dps Tracer kinetics 6 pico-mole 1
nano-gram
46
Lesion detectability
Non-TOF
TOF

• Improved lesion detectability with TOF achieved
  with short scan time
• and reduced reconstruction time ( of
  iterations)
• Spheres are just barely visible with a 5 minute
  scan in non-TOF
• After a 2-3 minute scan in TOF the spheres
  become visible

6-to-1 contrast 35-cm diam. cyl. 10-mm diam.
spheres
6.4mCi in all phantoms
47
Time-of-flight scanners need investigation of new
data processing and image reconstruction methods

• Scatter correction - can incorporate
  timing information - energy based methods -
  statistical weighting
• Image reconstruction - list-mode ML-EM -
  optimize use of TOF - include data
  corrections in system model - spatial recovery
• Data quantification - SUV estimation -
  convergence of lesion contrast improves with TOF
• Image evaluation - lesion detectability
  measures - how does TOF improve SNR in image?