implementation and evaluation of the single scatter...
TRANSCRIPT
Implementation and Evaluation of
Implementation and Evaluation of
the Single Scatter Simulation
the Single Scatter Simulation
Algorithm in the
Algorithm in the STIR
STIR
Library
Library
C. Tsoumpas
C. Tsoumpasa, ca, c, P. Aguiar
, P. Aguiar bb, K. Thielemans
, K. Thielemans a, ca, c
aaHammersmith Imanet Limited, Hammersmith Hospital, London, Unite
Hammersmith Imanet Limited, Hammersmith Hospital, London, Unite d Kingdom.
d Kingdom.
bbDept. of Medicine, University of Barcelona, Barcelona, Spain.
Dept. of Medicine, University of Barcelona, Barcelona, Spain.
ccClinical Sciences Center,
Clinical Sciences Center, Dept. of
Dept. of Medicine, Imperial College,
Medicine, Imperial College, United Kingdom.
United Kingdom.
EE-- mail
mail ss: : Charalampos.T
Charalampos.Tsoumpas@
imperial.ac.uk, , [email protected]
[email protected], , [email protected]
STIR Workshop, NSS
STIR Workshop, NSS ––MIC 2004
MIC 2004
Hammersmith
Hammersmith
Imanet
Imanet
2
Overview
Overview
��Scatter in 3D PET
Scatter in 3D PET
��Single Scatter Simulation (
Single Scatter Simulation (SSS
SSS))
��SSS
SSSImplementation in
Implementation in STIR
STIR
��Simulation Characteristics
Simulation Characteristics
••Scanner Sampling
Scanner Sampling
••Simulation Parameters
Simulation Parameters
��Scatter Sinogram Profiles
Scatter Sinogram Profiles
��Evaluation using SimSET
Evaluation using SimSET
��Comparison to Measured Scatter
Comparison to Measured Scatter
��Discussion
Discussion
��Future Work
Future Work
3
SCATTER CASE
Scatter in 3D PET
Scatter in 3D PET
NO SCATTER CASE
Sinogram Radial Profile
Phantom
4
Scatter in 3D PET
Scatter in 3D PET
Scatter Fraction (SF) for a point source at the centre of the
FoV, surrounded by a water filled cylinder (30cm diameter,
20cm length) [Adam et al.]1 :
��SF = (Scattered/Total) Events
SF = (Scattered/Total) Events
��2D PET SF
2D PET SF ≈29%
29%
��3D PET SF
3D PET SF ≈41%
41%
��3D PET Single SF (S
3D PET Single SF (S11F) F)
≈25%
25%
Scatter outside of FoV
Scatter inside of FoV
5
Single Scatter Simulation (SSS)
Single Scatter Simulation (SSS)
��Interpolate Attenuation and Emission Images.
Interpolate Attenuation and Emission Images.
��Select Scatter Points Randomly in the
Select Scatter Points Randomly in the
Attenuation Image, above a Threshold.
Attenuation Image, above a Threshold.
��Sample the Detectors using a Sinogram
Sample the Detectors using a Sinogram
Template.
Template.
��Find the Single Scatter Distribution (Sinogram)
Find the Single Scatter Distribution (Sinogram)
using the algorithm proposed by Watson
using the algorithm proposed by Watson22. .
6
. .
..
. .
..
Single Scatter Simulation (SSS)
Single Scatter Simulation (SSS)
7
SSS
SSSIm
plementation in
Implementation in STIR
STIR (II)
(II)
��Scatter Buildblock
Scatter Buildblock
Create Scatter Viewgram
Create Scatter Viewgram
Sample Scatter Point
Sample Scatter Point
Estimate Scatter from all Scatter Points
Estimate Scatter from all Scatter Points
Estimate Scatter from one Scatter Point
Estimate Scatter from one Scatter Point
Differential and Total Cross Section
Differential and Total Cross Section
Detection Efficiency
Detection Efficiency
Write statistics
Write statistics
Scatter Point to Detector Integrals
Scatter Point to Detector Integrals
Emission and Attenuation
Emission and Attenuation
Cache Factors
Cache Factors
Cache
Cache
Random Points
Random Points
No Cache
No Cache
Reconstructed Image
Reconstructed Image
Attenuation Map
Attenuation Map
Sinogram Template
Sinogram Template
Scatter Sinogram Name
Scatter Sinogram Name
Attenuation Threshold
Attenuation Threshold
Lower & Upper Energy
Lower & Upper Energy
Use of Random Points (?)
Use of Random Points (?)
Use of cache (?)
Use of cache (?)
��Scatter
Scatter (EXECUTABLE)
(EXECUTABLE)
��Scatter Inline
Scatter Inline
8
Scanner Characteristics
82
5R
ing
dia
met
er (
mm
)
15
5.2
Sca
nn
er l
eng
th (
mm
)
9.2
9.2
19
.41
9.4
4.8
5R
ing
Dis
tan
ce (
mm
)
16
16
88
32
Nu
mb
er o
f R
ing
s
14
47
21
44
72
57
6D
etec
tors
per
rin
g
IVII
III
IE
XA
CT
HR
+C
ha
ract
eris
tics
Scanner Sampling (I, ..., IV)
BGO crystals resolution: 25% at 511keV
Energy window: 350-650 keV
Wider Energy window (W-SSS): 320-650 keV
9
Phantoms for Evaluation
Phantoms for Evaluation
Simulation
AB
C
A) Point source:Sphere of Emission Voxel Size diameter.
B) Line Source:Emission Voxel Size diameter, Scanner length.
C) Cylinder Source:20cm diameter, Scanner length.
Attenuation Cylinder:20cm diameter, Scanner length.
10
Simulation Parameters
2.7
61
9.5
7/4
32
91
0×
10
×1
07
2×
72
×(1
6+
2)
IV–
(in
)dir
ect
2.8
31
7.5
4/4
32
91
0×
10
×1
07
2×
72
×1
6IV
–d
irec
t
3.3
21
.15
/11
57
10
×2
0×
20
36
×36
×16
III
–d
irec
t
2.7
74
.64
/23
31
20
×1
0×
10
72
×7
2 ×
8II
–d
irec
t
3.2
00
.30
/71
22
0×
20
×2
03
6 ×
36
×8
I –
dir
ect
Rel
ati
ve
Sta
nd
ard
Dev
iati
on
**
(%)
for
B
CP
U T
ime/S
catt
er
Po
ints
*fo
r C
, ~
min
Att
enu
ati
on
/Em
issi
on
Vo
xel
(m
m3)
Sin
og
ram
siz
eS
am
pli
ng
* CPU time approximately needed for W-SSSby a processor AMD Athlon MP 1900+, for the
simulation C.
**Relative Standard Deviation over all views through the centreof a sinogram, for simulation B.
Attenuation / Emission Image Interpolation
and Relative Standard Deviation
11
Comparing Profiles for different schemes of the SSS for simulation A,
along radial direction, through a single view of a direct sinogram, and
along axial direction through the center of a viewgram.
Single Scatter Sinogram Profiles
Single Scatter Sinogram Profiles
�Shape of the distribution has significant importance.
�Random scatter points reduce discretisation artefacts.
�Essentially, all radial profiles have the same shape.
Scatter Points
RandomlyIn the center
12
Single Scatter Sinogram Profiles
Single Scatter Sinogram Profiles
Comparing Profiles for different schemes of the SSS for
simulation C,along radial and axial directions.
Essentially, all profiles axialand radialhave the same shape.
13
��Data acquired on an ECAT EXACT HR
Data acquired on an ECAT EXACT HR++scanner:
scanner:
••NEMA
NEMA ’’94
94 ––Scatter Acceptance Tests.
Scatter Acceptance Tests.
••Uniform cylinder, 20cm diameter, filled with water.
Uniform cylinder, 20cm diameter, filled with water.
••GeGeline source, 2mm diameter.
line source, 2mm diameter.
••Line Source, radial, at ~0mm and ~80mm.
Line Source, radial, at ~0mm and ~80mm.
��Perform
Perform SSS
SSSusing:
using:
••Energy Window (350
Energy Window (350 ––650) keV.
650) keV.
••Wider Energy Window (320
Wider Energy Window (320 ––650) keV (
650) keV (W
W ––SSS
SSS).).
••Scheme I.
Scheme I.
��Perform SimSET simulation using:
Perform SimSET simulation using:
••Energy Window (350
Energy Window (350 ––650) keV
650) keV
••Detector sampling characteristics based on Scheme I.
Detector sampling characteristics based on Scheme I.
Comparison to Measured Data
Comparison to Measured Data
14
Scatter Sinogram Profiles
Scatter Sinogram Profiles
�Mean values of the profiles are the same after global scaling.
�The total scatter distribution is wider than the SSS, as expected.
�SSShas excellent agreement in both positions.
�W-SSS has quite good agreement with SimSET, but not excellent.
Comparing Profiles along radial direction for vertical view of t
Comparing Profiles along radial direction for vertical view of t he he
simulation
simulation BBand SimSET (Scheme I)
and SimSET (Scheme I)using Experimental Data.
using Experimental Data.
15
Scatter Sinogram Profiles
Scatter Sinogram Profiles
Comparing
Comparing SSS
SSSand
and WW-- SSS
SSSProfiles along radial direction for vertical view of the
Profiles along radial direction for vertical view of the
simulation
simulation BB--Scheme I and normalised measured sinogram, at ~0mm and ~80 mm.
Scheme I and normalised measured sinogram, at ~0mm and ~80 mm.
�Mean values of the profiles are the same, ignoring the peak.
�The y-axis is scaled such that the maximum in the measured data corresponds to 1.
�Estimated single scatter distribution too narrow, as expected.
�The W-SSSovercomes, somehow, this problem.
�Reasonable agreement with the measured data.
16
How scatter is corrected with
How scatter is corrected with
STIR
STIR ??(1st release)
(1st release)
1) Sample the detectors of the scanner using a Sinogram Template,
constructed by STIR.
2) Reconstruct Emission and Attenuation image (Em, Att) doing the
appropriate pre-corrections to the projection data.
3) Rescale the reconstructed images and perform the SSSor W-SSS.
4) Interpolate the Scatter Sinogram (SS) to have the same dimensions
with the Emission Sinogram (ES).
5) Find the sum of the tails, at the SSand ESand define a global scaling
factor (s).
6) Remove the scatter by subtraction and multiply with the Attenuation
Sinogram (AS) to obtain a scatter Corrected Sinogram (CS):
CS = AS �(ES -s�SS)
7) Reconstruct the Corrected Sinogram (CS).
17
Why to use
Why to use STIR
STIR
Scatter Correction ?
Scatter Correction ?
�Easy to use. Package will include Visual C++
scatter project.
�All needed functions/utilities will be included into
the first release.
�Flexible choice of the voxel sizes for the Emission
and Attenuation images and of the detector
sampling characteristics.
�Easy to change between W-SSSto SSS.
�Open Source under Lesser GNU License.
18
Future Work
Future Work ––MIC 2005 (?)
MIC 2005 (?)
��Find a robust scale factor of the scatter
Find a robust scale factor of the scatter
distribution to the emission distribution.
distribution to the emission distribution.
��Evaluate using Brain and whole Body phantoms.
Evaluate using Brain and whole Body phantoms.
��Test if the indirect scatter sinogram has
Test if the indirect scatter sinogram has
significant information.
significant information.
��Integrate the Scatter Estimation into Analytic
Integrate the Scatter Estimation into Analytic
and Iterative Reconstruction Algorithms.
and Iterative Reconstruction Algorithms.
��Include it into
Include it into STIR
STIR55and make it available to
and make it available to
PET community.
PET community.
19
References
References
[1] [1] Adam L E, Karp J S and
Adam L E, Karp J S and Brix
BrixG, "G, "Investigation of scattered radiation in 3D whole
Investigation of scattered radiation in 3D whole-- body positron emission
body positron emission
tomography using Monte Carlo simulations
tomography using Monte Carlo simulations ." ." Phys. Med. Biol.
Phys. Med. Biol. 44 2879
44 2879––95, 1999.
95, 1999.
[2] C. C. Watson, New, faster, image
[2] C. C. Watson, New, faster, image --based scatter correction for 3D PET,
based scatter correction for 3D PET, IEEE Trans. Nuclear Sci.
IEEE Trans. Nuclear Sci.47 (4)
47 (4)
1587
1587-- 1594, 2000.
1594, 2000.
[3] NEMA Standards Publication NU 2, Performance Measurements of
[3] NEMA Standards Publication NU 2, Performance Measurements ofPositron Emission Tomography:
Positron Emission Tomography:
National Electrical Manufacturers Association, 1994.
National Electrical Manufacturers Association, 1994.
[4] SimSET web site:
[4] SimSET web site: http://
http://depts.washington.edu/simset/html/simset_home.html
depts.washington.edu/simset/html/simset_home.html , June 2004.
, June 2004.
[5] [5] STIR
STIR1.3 software library and documentation
1.3 software library and documentation http://stir.
http://stir. HammersmithImanet
HammersmithImanet.com
.com, , September
September2004.
2004.
The authors would like to thankDr Terry Spinks, who provided the experimental data,
Dr Robert Harrison from Imaging Research Lab, at University of Washington Medical
Center and Dr Albert Cot for the assistance during the SimSET simulations,Dr George
Kontaxakis, Dr Oliver Nix,Dr Falk Poenisch and Dr Alexander Werling for the
assistance during the SSS implementation.C. T. also thanks Dr George Loudos and Dr
Sofia Tzimopoulou for many valuable discussions.
Acknowledgments
Acknowledgments