Joint CI-JAI advanced accelerator lecture series

Imaging and detectors for medical

physics
Lecture 7: PET

Dr Barbara Camanzi
[email protected]
 Course layout

Day AM 09.30 – 11.00 PM 15.30 – 17.00

Week 1
6th June Lecture 1: Introduction to Lecture 2: Detectors for
medical imaging medical imaging
7th June Lecture 3: X-ray imaging
8th June Tutorial
Week 2
13th June Lecture 4: Radionuclides
14th June Lecture 5: Gamma Lecture 6: SPECT
cameras
16th June Lecture 7: PET
Week 3
22nd June Tutorial

Page 2/52
 Books
1. N Barrie Smith & A Webb
Introduction to Medical Imaging
Cambridge University Press

2. Edited by M A Flower
Webb’s Physics of Medical Imaging
CRC Press

3. A Del Guerra
Ionizing Radiation Detectors for Medical Imaging
World Scientific

4. W R Leo
Techniques for Nuclear and Particle Physics Experiments
Springer-Verlag

Page 3/52
 Positron Emission Tomography
(PET)
Ref. 1 – Chapters 3.13 to 3.21

• Tomographic technique that uses radiotracers

administered to the patient → emission imaging
• Basic principle:
1. Radiotracers used undergo 𝛽+ decay → emit e+
2. e+ travels on average 0.1÷3 mm in tissue depending
on radiotracer → scatters → loses energy → comes to
rest
3. e+ at rest combines with atomic e− to form positronium
4. Positronium decays emitting two back-to-back 511 keV
𝛾-rays

Page 4/52
 +
𝜷 decay
• Proton-rich or neutron Example
deficient radionuclide ejects 𝛽 + decay
15 15
8𝑂 7𝑁 + 𝛽+ + 𝜈 + 𝐸
𝛽+ -particle = e+ = +1 charge
in the process: 𝐸 = shared randomly between 𝑣 and
kinetic energy of 𝛽+
𝑝 → 𝑛 + e+ + 𝜈
Average kinetic energy 𝐸𝛽+ ≅ 𝐸𝛽𝑚𝑎𝑥
+ /3
• Three-body decay → energy
spectrum of e+ = continuum 15
8𝑂
up to a maximum 1.022 MeV
𝛽+ decay
• 𝑍 → 𝑍 − 1, 𝐴 and atomic
weight remain the same 𝐸𝛽𝑚𝑎𝑥
+ = 1.7 MeV
15
7𝑁

Page 5/52
 Common radiotracers for PET
Radionuclide Half-life (min) 𝜷+ fraction Max . kinetic Average 𝜷+ Clinical
energy (Mev) range in water application
Radiotracer (mm)

11
𝐶 20.4 0.99 0.96 1.0 Cardiac
11
𝐶 -palmitate metabolism
13
𝑁 9.96 1.00 1.19 2.0 Cardiac blood
13
𝑁𝐻3 flow
15 2.07 1.00 1.72 2.0
𝑂 Cerebral blood
𝐻2 15𝑂 flow
18
𝐹 109.7 0.97 0.64 0.6 Oncology,
18 inflammation,
𝐹 𝐷𝐺 cardiac viability
82
𝑅𝑏1 1.27 0.95 3.35 2.8 Cardiac
82
𝑅𝑏𝐶𝑙2 perfusion

1Only radioisotope produced at on-site generator and not cyclotron

Page 6/52
 𝟏𝟖
𝑭-fluoro-2-deoxy-D-glucose
(FDG)
• Most common radiotracer used
in 80% of PET studies
• FDG injected into blood stream → 18F
transported to cells across body
• Uptake depends on rate of glucose utilization =
glucose metabolism:
– High glucose metabolic rate characteristic of many
tumours → hot spots in oncological PET scans

Page 7/52
 PET –vs– SPECT
• Main difference:
– Two 𝛾-rays (PET) instead of one (SPECT)
Advantages
Much higher 𝑆𝑁𝑅𝑃𝐸𝑇 arises from:
1. Collimation not required → no absorption of 𝛾-rays
2. Higher 𝐸𝛾𝑃𝐸𝑇 → less 𝛾-rays attenuation in tissue
3. Use of complete ring of detectors
Significantly better spatial resolution
Two back-to-back 𝛾-rays → two signals in the ring of detectors → two points →
intrinsic line-of-response (LOR) → no collimation needed to select directions
Disadvantages
(On-site) cyclotron needed to produce 𝛽 + emitters
High associated cost
Page 8/52
 PET components
• Detection unit = full ring of scintillating detectors
surrounding the patient
• Scintillation read-out chain:
– PMTs → convert light into electric signal
– Pulse height analyzer
• Annihilation coincidence detection unit

Page 9/52
 Scintillating detector ring
• Large number of small scintillation
crystals placed in circular ring
surrounding patient with diameter:
– 70 or 85 cm for abdominal scanner
– ~45 cm for head scanner
• Up to 48 multiple rings staked
axially with retractable lead
collimation septa in between →
head/foot FOV = ~16 cm
• Ideal geometry = one crystal
coupled to one PMT → better
spatial resolution but too expensive
• Geometry = ‘block detector’ design Courtesy Mike Partridge (Oxford)
Page 10/52
 ‘Block detector’ design
• Large block of scintillating
material ~50 × 50 × 30 cm3
• Partial cuts through filled
with reflective material →
prevent light formed at top
of crystal to produce very
broad LSF while travelling
30 cm to PMT

Courtesy Mike Partridge (Oxford)

Page 11/52
 ‘Block detector’ geometry
• Geometry = 8 × 8 array of
cuts → 64 ‘crystals’ of
~6 × 6 cm2 area and 30 cm
length coupled to 4 PMTs →
multiplexing factor of 16
• The 𝑋-Y position measured
relative to the centre of the
𝑦 block of four PMTs is:
𝑎+𝑏−𝑐−𝑑
𝑋=
𝑎+𝑏+𝑐+𝑑
c a 𝑎−𝑏+𝑐−𝑑
𝑥 𝑌=
d b 𝑎+𝑏+𝑐+𝑑
Courtesy Piero Posocco (Imperial College)
Page 12/52
 Scintillation crystals
• Ideal scintillation material for use in PET has:
1. High detection efficiency for 511 keV 𝛾-rays
2. Short decay time to allow for short coincidence
resolving time
3. High light yield to reduce the complexity and cost of
the system
4. Emission wavelength near 400 nm that corresponds
to maximum sensitivity for standard PMTs
5. Optical transparency at emission wavelength to
minimise reabsorption
6. Index of refraction close to 1.5 to ensure efficient light
transmission between crystal and PMT
Page 13/52
 Common scintillator materials
for PET
Decay time Relative Efficiency Emission Refractive
(ns) light yield1 wavelength index
(nm)
𝐵𝐺𝑂 300 0.15 0.72 480 2.15
𝐿𝑆𝑂(𝐶𝑒) 40 0.75 0.69 420 1.82
𝐵𝑎𝐹2 0.8 prim 0.12 0.34 220, 310 1.49
600 sec
𝐺𝑆𝑂(𝐶𝑒) 60 prim 0.3 0.57 430 1.85
600 sec
𝑁𝑎𝐼(𝑇𝑙) 230 prim 1.0 0.24 410 1.85
104 sec
1Relative to 𝑁𝑎𝐼(𝑇𝑙)

• Scintillator must be 2 cm or more in thickness for high sensitivity

• 𝑁𝑎𝐼(𝑇𝑙) =low efficiency for 511 keV 𝛾-rays → not used in PET
Page 14/52
 Scintillation read-out chain
PMTs Pulse height analyzer
• Standard operation → • Multi-channel analyser
scintillation light = optical receives in input the PMT
photons converted into an signal and converts it into a
amplified electric signal by ‘logic pulse’ typically 6÷10 ns
the PMTs long if PMT signal amplitude
is within pre-set range
• ‘Logic pulse’ sent to
coincidence unit

Page 15/52
 Data acquisition
• Two different modes:
1. 2D multi-slice mode
2. Full 3D mode → becoming more common due to far
superior 𝑆𝑁𝑅

2D 3D
• Collimation septa extended • Collimation septa retracted
• Requires fully 3D
reconstruction algorithm

Page 16/52
 2D PET –vs– 3D PET
2D 3D
• Advantages: • Advantages:
1. Reduced amount of scattered 1. Factor 10 higher sensitivity
𝛾-rays than 2D PET → higher 𝑆𝑁𝑅
2. Uniform sensitivity profile 2. For same 𝑆𝑁𝑅 as in 2D PET
along axial direction → two order magnitude
reduction in scan time
• Disadvantages:
1. More random coincidences
and scattered 𝛾-rays
2. Sensitivity profile in axial
direction higher at the centre
than at ends

Page 17/52
 2D PET and 3D PET
in whole-body PET
• Whole-body PET = bed is moved several times
along head/foot directions to cover entire body
length
2D 3D
• Uniform sensitivity profile • Non-uniform sensitivity profile
along axial direction → along axial direction →
overlap between successive overlap between successive
bed positions = only ~1÷2 cm bed positions = as high as
50% → more bed positions
required

Page 18/52
 2D multi-slice acquisition
• Image planes formed:
1. Between two crystals in
same ring = direct planes
2. From crystals in adjacent
rings = cross planes
• For system with n rings:
– n direct planes
– n-1 cross planes
– Total 2n-1 planes

Courtesy Piero Posocco (Imperial College)

Page 19/52
 Full 3D acquisition
• Image planes in a much
more complicated
configuration

Courtesy Piero Posocco (Imperial College)

Page 20/52
 Annihilation coincidence
detection unit
• Time
– Fixed ‘coincidence resolving
time’ = time window for each
PET system
– Each signal recorded in a
crystal given a time-stamp
with precision 1÷2 ns to
Line-of-response (LOR)
account for different arrival
Courtesy Piero Posocco (Imperial College)
time of two 𝛾-rays at detector
ring
• Position
– Geometric ‘coincidence arcs’
= two arcs at 180° formed by
set number of crystals

Page 21/52
 Annihilation coincidence
detection
• Time
– 1st 𝛾-ray detected at time 𝑡1 = 0
– 2nd 𝛾-ray detected at time 𝑡2
• Position
– 1st 𝛾-ray detected assigned to crystal 1
– 2nd 𝛾-ray detected assigned to crystal 2
• Coincidence
– If 𝑡2 falls into time window → 2nd 𝛾-ray assigned to same
annihilation
– If crystals 1 and 2 are operated in coincidence → two 𝛾-rays
accepted as ‘true event’ and LOR drawn
Page 22/52
Types of coincidence events
in PET

True
Both 𝛾-rays escape without
scatter and interact in detectors

Random (accidental)
Scatter Two 𝛾-rays from
One or both 𝛾-rays separate emissions
scatter in tissue strike the detectors at
the same time

Page 23/52
 Scatter events
• Two main sources of scatter:
1. Within the body
2. Within the scintillator
• Leads to reduction in image contrast

Page 24/52
 Scatter contribution
• High contribution due to:
1. Only one of two annihilation 𝛾-rays has to scatter
2. BGO and LSO poor intrinsic energy resolution
compared to 𝑁𝐴𝐼(𝑇𝑙) = cannot discriminate scattered
from unscattered 𝛾-rays → scatter events significant
fraction
1. Ex: detection window for BGO 450-650 keV while 𝛾-ray
scattered at 45° loses only ~115 keV
3. Higher in 3D PET due to elimination of collimator
2D 3D
10÷15% Up to 50%

Page 25/52
 Random coincidences
• Random coincidences are due to two separate
disintegrations occurring very close in time
• Uniformly distributed across FOV → significant
errors in areas of very low activity
• Rate 𝑅𝑟𝑎𝑛𝑑𝑜𝑚𝑠 for a time window 𝜏 is given by:
𝑅𝑟𝑎𝑛𝑑𝑜𝑚𝑠 = 𝑅𝑆𝑖 × 𝑅𝑆𝑗 × 2𝜏
𝑅𝑠𝑛 = detection rate at scintillation detector 𝑛
• Contribution:
– For head scanners ~20%
– For body scanners close to 50%
Page 26/52
 Image formation
• Two steps:
1. Correction of data for:
a. Attenuation effects
b. Scatter
c. Accidental and multiple coincidences
d. Dead-time
2. Tomographic image reconstruction = SPECT → CT
reconstruction methods:
a. Filtered backprojection method
b. Iterative method

Page 27/52
 Attenuation correction
• Old method (pre PET/CT)
Identical to SPECT = transmission-based
calibration with 68𝐺𝑒 used to estimate attenuation
• Current method (PET/CT)
CT images segmented according tissue type1 →
standard value of 𝜇 at 511 keV assigned →
attenuation map smoothed to match PET
resolution and applied to PET image
1Muscle, lipid, bone, etc.

Page 28/52
 Scatter correction
• Three different approaches:
1. Background subtraction approach → simplest method
a. Measure signal intensity outside the patient → fit values to a Gaussian to estimate
amount of scatter in patient = function
b. Function subtracted from raw data → corrected image
c. Works well for homogeneous organs (brain) but not abdomen
2. Dual-energy window approach
a. Two options:
i. Lower window 190÷350 keV + upper window 350÷650 keV
ii. Lower window 450÷650 keV + upper window 550÷650 keV → overlap =
Estimation of trues method (ETM)
b. Scatter data scaled appropriately and subtracted from data in photopeak window
→ corrected image
3. Iterative reconstruction approach
a. Uses simulations and CT-derived attenuation maps
b. Most sophisticated, computationally intensive and time-consuming
Page 29/52
 Random coincidences
corrections
• Two methods:
1. Use of additional parallel timing circuitry = most common
a. Second time window starts typically 60 ns after event is recorded
b. Standard window measures total number of coincidences +
delayed window records only random events → subtracted from
total
2. Use relation between 𝑅𝑟𝑎𝑛𝑑𝑜𝑚𝑠 and 𝑅𝑆𝑖
a. Measured values of 𝑅𝑆𝑖 used to determine 𝑅𝑟𝑎𝑛𝑑𝑜𝑚𝑠 → subtracted
from acquired data

Page 30/52
 Multiple coincidences
corrections
• Multiple = more than two events recorded during one
time window
• One method:
1. Total number of multiple coincidences 𝑅𝑚𝑢𝑙𝑡𝑖𝑝𝑙𝑒 estimated
as:
𝑅𝑚𝑢𝑙𝑡𝑖𝑝𝑙𝑒 ≈ 𝑅𝑟𝑎𝑛𝑑𝑜𝑚𝑠 × 𝑅𝑆𝑖 × 𝑅𝑆𝑗 × 𝜏
2. Multiple event coincidences discarded before image
reconstruction

Page 31/52
 Dead-time
• Maximum count rate the system can record due to
components’ finite response and recovery time
• Fractional dead time = ratio of measured count rate
to theoretical count rate with zero dead-time
• Major sources of dead-time in PET:
1. Time taken to integrate the charge from the PMTs
2. Processing time of a coincidence event
3. Multiple coincidences → data are discarded
• Corrections performed by:
1. Characterising the dead-time of each component
2. Estimating the number of multiple coincidences expected
Page 32/52
 Sensitivity
• True coincidence count rate 𝑅𝑡𝑟𝑢𝑒 for a 𝛽 + -emitter in
air near midpoint between pair of detectors is:
𝑅𝑡𝑟𝑢𝑒 = 𝑅e+ 𝜀 2 2𝐺
𝑅e+ s−1 = rate of e+
𝛾
𝑁𝑟𝑒𝑐𝑜𝑟𝑑𝑒𝑑
𝜀= 𝛾 = intrinsic detector efficiency
𝑁ℎ𝑖𝑡𝑡𝑖𝑛𝑔
𝐴
𝐺= = geometric efficiency of individual detector
4𝜋𝑟 2
of effective area 𝐴 and radius 𝑟
• Determined primarily by detector efficiency and solid
angle coverage
Page 33/52
 Signal-to-noise ratio
• Factors affecting 𝑆𝑁𝑅:
1. Dose administered to patient
2. Targeting efficiency
3. 𝛾-ray attenuation in body → lower than in SPECT due
to higher 𝐸𝛾
4. System sensitivity → greater than in SPECT due to lack
of collimator and higher 𝐸𝛾
SPECT 2D PET 3D PET
𝛾-rays detected 0.01÷0.03% 0.2÷0.5% 2÷10%

5. Image acquisition time

6. Image post-processing
Page 34/52
 Contrast-to-noise ratio
• Factors affecting 𝐶𝑁𝑅:
1. Same factors affecting 𝑆𝑁𝑅
2. Non-specific uptake of radiotracer in healthy tissues
surrounding pathology being studied
3. Corrections for Compton scattered 𝛾-rays

Page 35/52
 Spatial resolution
• Factors affecting the spatial resolution:
1. Effective e+ range in tissue → 𝛿𝑅𝑟𝑎𝑛𝑔𝑒
2. Non-colinearity of two 𝛾-rays → 𝛿𝑅180°
3. Dimension of the scintillating crystals → 𝛿𝑅𝑑𝑒𝑡𝑒𝑐𝑡𝑜𝑟
• Overall spatial resolution 𝛿𝑅𝑠𝑦𝑠 given by:
2 2 2
𝛿𝑅𝑠𝑦𝑠 = 𝛿𝑅𝑟𝑎𝑛𝑔𝑒 + 𝛿𝑅180° + 𝛿𝑅𝑑𝑒𝑡𝑒𝑐𝑡𝑜𝑟

• ‘Double detection’ of two 𝛾-rays reduces depth

dependence of 𝑃𝑆𝐹 → spatial resolution less
dependent from depth in body than in SPECT
Page 36/52
 +
Effective e range
• The e+ travels a certain
distance before coming to
rest → effective e+ range in
tissue → error in the
determination of the point
where e+ was produced
• This contribution is intrinsic
and cannot be eliminated

Error due to effective e+ range

Page 37/52
 Effective e+ range distribution
• Example of an effective e+ range distribution:

• The distribution is better described by the root mean

square 𝑟𝑚𝑠 or 𝜎 than 𝐹𝑊𝐻𝑀
Page 38/52
 Effect of e+ range on
spatial resolution
• Effective e+ range increases → spatial resolution
decreases with:
1. Maximum kinetic energy of the emitted e+ = the higher the
energy the longer distance e+ needs to travel to lose all its
energy
2. The tissue density = the lower the density of the tissue the
less interactions e+ undergoes per unit length → the less
energy e+ loses per unit length → the longer distance e+
needs to travel to lose all its energy

Page 39/52
 Non-colinearity
• Non-colinearity of two 𝛾-
rays = small random
deviation from 180° angle
between two trajectories
due to residual motion of e+
at point of annihilation
• Distribution in angles
between two trajectories
Error due to non-colinearity centred around 180° and
has 𝐹𝑊𝐻𝑀 ≅ 0.5°

Page 40/52
Effect of non-colinearity on
spatial resolution
• The larger the diameter of
the detector ring the greater
the effect on the spatial
resolution
𝐷 0.25𝜋
𝛿𝑟180° = ×
2 180
= 0.0022 × 𝐷
𝐷 = diameter of the PET
scanner

Page 41/52
 Effect of scintillating crystals
dimension on spatial resolution
• The smaller the scintillating crystals the better the
spatial resolution due to:
1. Area = the spatial resolution decreases with area =
given approximately by half the crystal diameter
The precision in locating the position = spatial resolution
at which the 𝛾-ray reaches the scanner increases for
crystals with smaller area
2. Length = the spatial resolution decreases with length
The uncertainty in the depth-of-interaction (DOI) = point
within the crystal where the 𝛾-ray is absorbed and
creates scintillation increases for longer crystals → the
spatial resolution decreases for longer crystals
Page 42/52
 Factors affecting the
spatial resolution
1. Effective e+ range in tissue = distance increases
with:
2. Non-colinearity of two 𝛾-rays = small random
deviation from 180° angle between two trajectories
→ distribution around 180° with 𝐹𝑊𝐻𝑀 ≅ 0.5° →
the larger the diameter of the detector ring the
greater the effect
3. Dimension of the scintillating crystals → spatial
resolution given by approximately half the crystal
diameter + depth-of-interaction uncertainty
increases with length of crystal
Page 43/52
 Time-Of-Flight (TOF) PET
• Very accurate measurement of
exact time at which each 𝛾-ray
arrives at the detector →
localisation within LOR:
– Time difference between signals
from two crystals measured
– Annihilation point along LOR
directly calculated
• Technology for TOF PET:
– Fast scintillating materials:
D1
time-of-flight
envelope
• LSO material of choice:
line of
coincidence time = ~450 ps
response instead of 3 ns for BGO
D2
– Fast photon detectors:
• SiPMs being investigated

Page 44/52
 TOF in context
• For a system with timing resolution ∆𝑡, the position
resolution ∆𝑥 along LOR is given by:
𝑐∆𝑡
∆𝑥 =
2
𝑐 = speed of light
• Timing resolution of commercial scanners ~500 ps
• R&D goals:
∆𝒕 (ps) ∆𝒙 (cm)
100 3
30 <1

Page 45/52
 Advantages of TOF PET
• ~500 ps timing resolution → ~7.5 cm spatial resolution
> spatial resolution of conventional scanners
• Length of LOR constrained down to ~7.5 cm from that
of conventional scanners (≫ 7.5 cm) → statistical
noise in the measurement reduced → 𝑆𝑁𝑅 improved
• Noise variance1 reduction factor 𝑓 for patient of size 𝐷:
𝐷 2𝐷
𝑓= =
∆𝑥 𝑐∆𝑡
• Better 𝑆𝑁𝑅 → higher sensitivity and specificity

1Noise variance proportional to length of LOR

Page 46/52
 Hybrid PET/CT
• Stand-alone PET scanner
almost entirely replaced by
hybrid PET/CT scanners:
1. Two separate systems one
next to the other
2. Bed that slides between two
systems
• Rationale:
– Improved attenuation
correction
– Ability to fuse anatomical
(morphological) and functional
information

Page 47/52
 Image fusion in PET/CT

CT PET PET + CT

Courtesy Mike Partridge (Oxford)

Page 48/52
 Clinical applications of PET/CT
• PET/CT scans currently represent ~5÷10% of all
nuclear medicine imaging and increase each year
• Clinical investigations:
1. Oncology (~90% of all investigations)
− Whole body PET imaging used to identify both primary and
secondary metastatic disease away from primary tumour
2. Cardiology
3. Neurology

Page 49/52
 Whole-body PET/CT
• Used in oncology for staging
of cancer = determined
based on number of
secondary lesions =
metastases spread away
from primary lesion
• Radiotracer used: 𝐹𝐷𝐺 →
metabolised more by
malignant cells
• Typical scan time 30÷60 min

Page 50/52
 PET/CT in brain imaging
• Used to investigate:
1. Neurodegenerative dementias such
as Alzheimer's (AD) and Parkinson's
diseases and distinguish them from
from other dementias → AD
characterised by low metabolism
2. Brain tumours: also to distinguish recurrences from radiation
induced necrosis
3. Trauma
4. Developmental abnormalities
5. Epilepsy

Page 51/52
 Cardiac PET/CT studies
• Used for the study of coronal artery diseases →
myocardial viability and perfusion

• PET/CT use in cardiac studies is increasing →

where available is used instead of SPECT
Page 52/52