Spatial Positioning

(We know it's there, but where is it?)

Point collimation – the rectilinear scanner

Anger camera (Anger Logic): Hal Anger, 1958

 Spatial Positioning

S1 S2 Take Sj = S (1- Dj / D)
where S = S1+S2
PMT
Distance between Tubes
D = D1 + D2
NaI
S1 = S (1- D1 / D)
D1 D2
=(S1 + S2)D2 / D
Colli- Distance from Tube 2
mator D2 = S1 D / (S1 + S2)
Distance from Tube 1
g-Ray
D1 = S2 D / (S1 + S2)

Note: Energy E a (S1 + S2)

 Spatial Positioning
Distance From Center?
S1 S2
Dc = D1 - D / 2
PMT = D1 – 0.5*(D1 + D2)
= 0.5*(D1 – D2)
= 0.5D S2 – S1
NaI
S1 + S2
D1 D2

More generally:
Colli-
mator Dc = 0.5D X+ – X-
X+ + X-
g-Ray
Want to use all PMTs in
array at once …
 Spatial Positioning
S1 S2 S3 S4 S5

W+i 0 1 2 3 4
W-i 4 3 2 1 0

X  = Wi  Si
i
 Spatial Positioning
S1 S2 S3 S4 S5
S1 = 0
S2 = 0
S3 = S
S4 = 0
S5 = 0
W+i 0 1 2 3 4
W-i 4 3 2 1 0

X  = Wi  Si X =k
2S - 2S
=0
i 2S + 2S

X+ -X-
X =k
X+ + X-
 Spatial Positioning
S1 S2 S3 S4 S5
S1 = S
S2 = 0
S3 = 0
S4 = 0
S5 = 0
W+i 0 1 2 3 4
W-i 4 3 2 1 0

X  = Wi  Si 0 - 4S
X =k = -k
i
0S + 4S
X+ - X-
X =k + -
k = radius of array (cm)
X +X
 Spatial Positioning
S1 S2 S3 S4 S5
S1 = 0
S2 = S
S3 = S
S4 = 0
S5 = 0
W+i 0 1 2 3 4
W-i 4 3 2 1 0

X  = Wi  Si X =k
(1S + 2S ) - (3S + 2S ) 1
=- k
i (1S + 2S ) + (3S + 2S ) 4
X+ -X-
X =k + - k = radius of array (cm)
X +X
 Anger
Logic

X = k(X+ - X-)
(X+ + X-)

Y = k(Y+ - Y-)
(Y+ + Y-)

k = radius of
array (cm)

Energy (E)= X+ + X- = Y+ + Y- = Z
 Simple
Anger
Circuit
X = k(X+ - X-)
(X+ + X-)

Y = k(Y+ - Y-)
(Y+ + Y-)
Ohm’s Law
k = radius of V=IR
array (cm)
 Digital centroids

From point of interaction of gamma ion crystal, a cone of

light spreads out with diminishing intensity. The base of
this cone is what is seen by the array of PMTS.
 Intrinsic Spatial Resolution

Multiple Compton scatter within crystal (small effect)

Statistical variation of light photons among PMTs

worse with thicker detector as light spreads more

Resolution depends on energy, less photons produced

means more statistical error. Is Poisson
distributed, so error is proportional to 1/sqrt(N),
or intrinsic resolution proportional to 1/sqrt(E)

Typically about 3mm for a modern clinical camera

 Energy Resolution
• Sources:
•Non-Linear energy response of scint crystal
•Statistical variation in # scint g / keV
•Non-uniform collection efficiency throughout crystal
•Non-uniform sensitivity of PMT photocathode
•Statistical variation in # photoelectrons / scint g
•Statistical variation in gain at each dynode in PMT
•Fluctuations in HV of PMT
•Electrical Noise in PMT / electronics
Typically about 10% FWHM at 140keV, i.e. 14keV
 Energy Resolution

• Given as FWHM
• Typically a % of the photopeak
•10% for NaI at 140keV
•7% at 511 keV
• Goes roughly at 1 / sqrt(E)
• Integration time: better for low-count rate systems
•PET vs SPECT
•Degraded by anything that reduces sensitivity
Ideal Spectrum

Actual
spectrum
(blurred out due to
the 10% energy
resolution of the
camera)
 Energy resolution is measured as the
FWHM of the measured photopeak

Resolution = 46 keV / 662 keV = 7% (at 662keV)

 Photopeak Window

Used to remove majority of scattered photons

Typically set at 20% -- why?

Capture 95% of a Gaussian Function

 Types of Events

Good / valid event

Detector scatter event

Mispositioning due to multiple sites of interaction

Object / Patient scatter event

Scatter is low spatial frequency and lead to loss of
contrast

Septal Penetration
More important for higher-energy photons
Amplifier / Pulse shaper
 Dead-time (Count-rate effects)
Time during which system does not count incoming signals
(radiation)
paralyzable → count rate goes to zero
non-paralyzable → count rate goes to constant

Baseline shift (A)

2nd pulse counted over
tail of 1st (peak is lower E)

Pulse pile-up (B)

signals from 2 events
overlap – coincidence pulse

Baseline Restoration
 Pulse-height Analyzer

Single Channel Analyzer (SCA)

Lower Level Discriminator (LLD)
Upper Level Discriminator (ULD)
Anti-coincidence Circuit
Multi-channel Analyzer (MCA)
Analogue to Digital Convertor (ADC)
Digital Output ( 2^n, typically 8 bit, sometimes 10-12)
Wilkinson (time of capacitor discharge)
Successive Approximation (variable charge / resistance)
Wilkinson ADC converter
 Collimators

We know where it ended up and what

energy it had, but where did it come from?
The Lens of the Gamma Camera

Restrict the direction of gamma-

rays hitting the crystal

Big lead blocks with holes

in them (absorptive
collimation)

Two types, cast and foil

Foil collimators
 Collimators: Septal Thickness

The thickness of the wall of the holes in the collimator

To know where the photons came from, we have to know

that they passed through the holes, that is, that they
did not penetrate the hole wall (septal penetration).

Different lenses for different tasks, close-up vs

panorama

Can 'tune' the lens to be more suitable for different tasks

The first parameter is the septal thickness:
Typical values are
0.1-0.2mm for low-
energy collimators
 Collimators: Resolution

Acceptance Angle: incident angle below which

photons can pass through the collimator

Determined by hole diameter and length

Resolution depends on d & l

l
Collimator Resolution
Collimator Resolution
Collimator Resolution
Collimators: Resolution vs Distance

LEHR
typically has
a resolution
of ∼7mm at
10cm
distance
 Collimators: Geometric Efficiency

Geometric Efficiency: fraction of emitted g reaching crystal

Hole shape
packing efficiency vs rotational symmetry
hexagonal, circular, square

Septal thickness
thicker septa means less holes

Higher resolution
lower acceptance angle so less of solid angle of photon flux

Typical efficiency is 2x10-4 (LEGP)

 Collimators:
Sensitivity with distance

Efficiency is independent of
distance for parallel hole
collimator

Fig 16-14
 Collimators: Geometric Efficiency

q r

Typical efficiency is 2x10-4 (LEGP)

NOTE: Efficiency (g) is proportional to
resolution squared (Rc2 )
 Collimators: Classification
By resolution/efficiency and energy of gamma-ray

Typical collimators are: LEAP (low-energy all purpose)

(LEGP)
LEHR (high resolution)
LEHS (high sensitivity)
MEGP (medium energy – up to 300 keV)
HEGP (high energy – up to 365 keV)

Esoteric ones: LEUHR, UHEGP (511keV)

 System Resolution
Gamma Camera is a Linear system
(propagation of errors, uncertainties add in
quadrature) – convolution of delta input (gamma
ray) by blurring due to collimator and then by
blurring due to intrinsic resolution.

Convolution of two Gaussians with s1 and s2 is a

Gaussian with s = (s12 + s22 ) ½

Rsys2 = Rint2 + Rcoll2

 Intrinsic resolution not usually a factor

Though with ultra-high-resolution imaging, it can be

 Focused Collimators

Another way in which collimators are 'tuned' is

the direction of their holes.

How can we increase the sensitivity of our

collimator?

Capture more of the solid-angle of the emitted

photon flux.

This leads to converging collimators where the

holes are focussed on a line or a point.
 Types of focusing collimators

Usually these collimators are designed for low-energy

emitters, though they could be designed for other
energies, or for high resolution, or high sensitivity

Fanbeam – converges to a line

Conebeam (converging) – converges to a point
Pinhole – like a VERY short focal length conebeam
Diverging – converges to a point behind the crystal

Other variations, slant hole, assymetric fanbeam, etc.

Focused Collimators:
Resolution / Efficiency

2
 Focused
Collimators:
Magnification
 What collimators are used for what?

Parallel hole: the workhorse – everything

energy based on isotope
sensitivity based on expected count rate

Fanbeam: heart, brain

Pinhole: spot views of hips or thyroid, cardiac
Diverging: on mobile systems and before they
could grow crystals very large
 QC for SPECT
Good Practice Daily Weekly Quarterly/Ann
ual
Visual check of Extrinsic High count-density Correction map
energy spectrum (intrinsic) low intrinsic flood
count flood

Cine review of Energy Peak Intrinsic/Extrinsic Reconstructed

Projection resolution check SPECT Phantom
Optionally sinogram with bar or hole
data pattern
Background Check Pixel size check
Visual inspection of Center of rotation ‘rest of NEMA’
collimators for (rotating
damage collimators)
 Energy Peaking

Daily on each camera head

Camera’s automatic circuitry working properly
Peak appears at correct energy
Correct shape of spectrum
Point source of appropriate energy (Tc-99m) that give <20 kcps for
uncollimated camera
 Energy Peaking

Correct
peak

Photopeak Window Off-peak

camera
 Off-peak too high Off-peak too low

Co-57 source
measured with
Tc-99m energy
window
 Uniformity

Camera’s ability to produce uniform images of a

uniform source
Method – daily
Extrinsic and qualitative (rarely intrinsic)
Plot data and note trends
Sealed sheet source with radioactive (Co-57, 122 keV, and
T½ 270 days)
Thin Lucite or Plexiglas chamber with a central area filled
with radioactive liquid (Tc-99m)
Point source (Tc-99m 0.5 mL with 7-25 MBq) at distance
>5 UFOV
Flood images visually inspected – 3M to 5M 256x256
 NEMA Uniformity
Intrinsic
Compare against manufacturer specifications
Point source (Tc-99m 0.5 mL with 7-25 MBq) at distance
>5 UFOV
Pixel size is 6.4mm/pixel +/- 30 %
~60-120 M cnts / head (>10 kcnts in central pixel)
Count rate < 20kcps
Window is what is used in clinic and correction map
acquired at.
Edge pixels removed and 9pt smoothing applied
 Uniformity
Global measure
Integral Uniformity, ratio of max. to min. pixel value in
UFOV and CFOV
IU = 100% X (max – min) / (max + min)
Local measure
Differential uniformity, measures contrast over a small
neighborhood
Change in pixel values over 5 adjacent pixels

CFOV UFOV
Typical NEMA Integral 2.5% 3.5%
values for Differential 2.0% 2.5%
Intrinsic Uniformity
 Field of View
UFOV = Useful Field of View
Typically the entire FOV, may be 95% of the FOV
CFOV = Central Field of View
75% of the UFOV

UFOV

CFOV
 Uniformity

Normal flood field Inadequate After

mixing of thorough
Tc in flood mixing
field
phantom
Uniformity

Defective Photomultiplier
tube
 Uniformity
Collimator
defect

Crack in NaI Without With computer

crystal computer correction
correction

Essential of Nuclear Medicine Imaging, fifth edition

 Non-Uniformity Artifacts

Pin-cushion Barrel

Pin-cushion and Barrel distortions are

caused by the non-linear response of the
PMT
 Uniformity

Detector 1, good uniformity Detector 2, need tuning

Uncorrected Corrected
Point source too close to the detector
 Uniformity

10% Energy window 15% Energy window

30M floods with

symmetric energy
windows suggest
problems with PMT
tuning
20% Energy window
Ring Artifact
 Ring Artifact
Fig 31. Transaxial slices reconstructed in an identical manner
from two studies of a uniform elliptical phantom. Total counts
were (A) 5 x 106 and (B) 100 x 106.
Ring Artifact

163
 Ring Artifact
Fig 34. Transaxial image from the study shown in Fig 33. The ring
artifact (arrow) is now easily identified and is concentric with
the center of the image (cross-hair).
 Spatial Resolution / Linearity
Camera’s ability to produce image detail and sharpness
Method – weekly
Similar to uniformity test with a resolution bar phantom
between camera detector and flood source

4 quadrants bar phantom with 4 sets of bars of different

width and spacing in each quadrant, arranged at 90
degree angles to each other

Spatial distortion (non-linearity) – straight bars depicted

wavy lines
Cause of linearity problems is linked to uniformity
 Spatial Resolution

Example Bar Phantom

7mm 4mm

5mm
6mm
 Non-Uniformity Artifacts

Pin-cushion Barrel

Pin-cushion and Barrel distortions are

caused by the non-linear response of the
PMT
 Non-Uniformity/Linearity Artifacts

Flood – no corrections Linearity – no corrections

Linearity phantom,
1mm slits at 30mm spacing
 Non-Uniformity/Linearity Artifacts

Corrections applied

Corrections: PMT Gain – spatially varying photopeak position

Count-skimming
Linearity map corrections
More accurate positioning algorithms
 Spatial Resolution and Linearity

Normal
flood-field
and bar
phantom

Rotate to assess
different directions

Essential of Nuclear Medicine Imaging, fifth edition

 Spatial Resolution and Linearity

Essential of Nuclear Medicine Imaging, fifth edition

 NEMA Intr Spatial Linearity
Annual
Uses same acquisition as spatial resolution
Differential linearity – average std deviation of peak
Absolute linearity – max. deviation from the true value

NEMA values for intrinsic

spatial linearity

CFOV UFOV
Absolute 0.4 mm 0.6 mm
Differential 0.15 mm 0.2 mm
 System Spatial Resolution

Annual
Collimator dependent
Quoted typically at 10cm: FWHM + FWTM
With and without scattering media (10cm in
front and 5cm behind)
Collect min 10k cnts in peak of each line
source
Use capillary tubes with <1mm int diameter
 Energy Resolution
Point source suspended 5 x UFOV
Intrinsic
Convert spectrum for channels to energy
Scan Tc-99m and Co 57 to calibrate system
Resulting pulse-height spectrum is analyzed to
measure FWHM

NEMA value of energy resol (9.5mm crystal): 9.8%

 Count Rate Performance
Dead Time – required to detect, amplify, shape, mix
and integrate pulses
Method: Intrinsic (quarterly)
Several small sources (minimal scatter)
Different known activity
Each measured over time
Plot:
20% loss (B)
Max. count rate (C)

Typical NEMA value: max. count rate at 200 kcps

20% loss at 160 kcps
 Count Rate Performance
Paralyzable
- R t
R 0 = R te
Non-Paralyzable
Rt
R0 =
(1 + R t )
Ro
Rt =
(1 - R o ) True Count Rate (cps)
True Count Rate (cps)

R0 = observed rate; Rt = true rate;  = deadtime

 SPECT

Single
Photon Type of radioactive decay
Emission
Computed CT: 3D imaging
Tomography
Clinical Cameras Detector Heads

ECam

Infinia Hawkeye-4

NM 530c (CZT)
 SPECT : Basic Process

Patient Camera

Radioactive
Sources

Camera
Collimator
Crystal (NaI)
Photomultiplier Tubes
Computer
 Projections

Patient Camera

Sources
 Sum all Data in Direction of Camera

Patient Camera

Sources

S
 Planar (Static) Heart Imaging
camera

Ant

LAO
Potential Problems:
Cannot differentiate overlapping structures
Image contrast reduced
L.Lat
(LAO-70)
Modes of Acquisition

Static
Dynamic
ECG Gated
SPECT
 MAG3 Renal Study

Dynamic
Planar
Images
 ECG Gating – What is it?

Split-up or rebin the data based

on the phase of the cardiac cycle.
Usually achieved by using the
ECG signal and separating the
data based on the length of time
since the R-wave.
Data is typically binned into 8 or
16 gates.
 Gating – How?

ECG Signal
 Gating – How?

ECG Signal
 Gating – Issues

Beat rejection, +/- 10% - 15%, bad beat (+ 1)

Buffering of each beat
Loss of counts (actual period longer than expected)
Dynamic adjust. of the period vs fixed definition
Prospective vs Retrospective gating
Count fall off in last frame

Counts / gate issue

Sum for Ungated
 Multiple Gated Acquisition (MUGA)
camera

Ant

LAO

L.Lat
(LAO-70)

Gated Blood-pool imaging

 Gating – Why?

Ejection Fraction (EF)

ROI counts where heart

is fully expanded
(end diastole) and fully
contracted (end systole)

EF = (ED counts – ES counts )

(ED counts – background)
Gating – Why?

Phase Analysis
 ERNA: Phase TAC Curve
Activity Near Heart Wall Boundary

Counts in voxel
voxel Time
 ERNA: Phase TAC Curve
Activity Near Heart Wall Boundary

Counts in voxel
voxel Time
 ERNA: Phase TAC Curve
Activity Near Heart Wall Boundary

Counts in voxel
voxel Time
 ERNA: Phase TAC Curve
Activity Near Heart Wall Boundary

Counts in voxel
voxel Time
 ERNA: Phase TAC Curve
Activity Near Heart Wall Boundary

Counts in voxel
voxel Time
 ERNA: Phase TAC Curve
Activity Near Heart Wall Boundary

Counts in voxel
voxel Time
 ERNA: Phase TAC Curve
Activity Near Heart Wall Boundary

Counts in voxel
voxel Time
 ERNA: Phase TAC Curve
Activity Near Heart Wall Boundary

Counts in voxel
voxel Time
 ERNA: Phase TAC Curve
Pixel
Activity Near Heart Wall Boundary Time-Activity Curve

Counts in voxel
voxel Time

Phase Analysis

f (t ) = A0 + A1 cos(q (t ) +  )
 Gating – Why?

Phase Analysis
f (t ) = A0 + A1 cos(q (t ) +  ) ROI
TAC
Pixel
Time-Activity Curve

92

Measurements of
46 Dissynchrony

SD, Bandwidth,
0 etc
0 90 180 270 36
0
 Camera Rotates Around Patient

SPECT acquisition times typically ~15min

 Sum all Data in Direction of Camera

Patient Camera

Sources

S
Projections
Projections
Projections
Patient Motion
Projections = Set of 2D planar images

Take pictures from many directions, then

transform x,y,q into x,y,z

But how do we do it, really?

 Sinograms

A way of representing the raw acquired data from a

rotating imaging modality (SPECT, or PET, or CT, etc.)

A plot of transaxial position versus rotation angle

Sinograms

r
Sinograms

q
r
 Sinograms Equivalent
SPECT
projection

q
r

All LORs
at same q
contribute
to the
same
projection
Small cylinder
Small cylinder projection
Small cylinder sinogram
 Small cylinder sinogram

0 degrees rotation
Angular direction

Transaxial direction
 Small cylinder sinogram
Angular direction

45 degrees rotation

Transaxial direction
 Small cylinder sinogram
Angular direction

90 degrees rotation

Transaxial direction
 Small cylinder sinogram
Angular direction

Transaxial direction
 Basic Reconstruction: FBP

Ie 
- m ( x ) dx

-log = -log
I
=  m ( x)dx

SPECT is like CT
“sum of signal”
∴use FBP for reconstruction
 Central Section Theorem

Fourier transform of a projection at angle q,

P(k,q), is equal to a sample of the 2D Fourier
transform of the image taken on a line
through the origin at angle q.

Also called projection-slice theorem.

Backproject
BP 1
BP 2
BP 3
BP 32
BP 64
BP 128
Star artifact – 1/r blurring

Due to sampling issues

How fix?

p. 396
 Central Section Theorem

Fourier transform of a projection at angle q,

P(k,q), is equal to a sample of the 2D Fourier
transform of the image taken on a line
through the origin at angle q.

Also called projection-slice theorem.

 Star artifact –
Fourier sampling is like a wheel

# spokes / (2p R )

Sampling density falls off as 1/r in frequency domain

Problem is Overlapping Beams
Ramp filter: spatial domain

p. 398
Applying the ramp filter
FBP 1
FBP 16
FBP 32
FBP 64
FBP 128
 Compare BP and FBP

Truth Backprojection FBP

Reconstruct with Filtered
Backprojection (FBP)

MCAT model
FBP 1
FBP 2
FBP 3
FBP 8
FBP 64
FBP 128