Study of 3D Reconstruction
Method for Electron-Tracking
Compton Camera
Annual Report Meeting by GCOE Postdocs
March 7th, 2011
Shigeto Kabuki
Contents
Medical imaging
Electron-Tracking Compton Camera (ETCC)
3D reconstruction method Simple Back Projection
List mode Maximum Likelihood Expectation Maximization
Modified Source Space Tree Algorithm
Imaging results
Summary
Radioisotope drug
Accumulation to affected part
g rayinjection
detector
imaging
Diagnosticimaging
Flow of Medical imaging
X ray CT
Xray
absorption
SPECT(single photon
computed
tomography)
E < 300 keV
PET(Positron Emission
Tomography)
E=511keV
RI distribution
Conventional Imaging Detector
Motivation
PET & SPECT
High quality algorithm for imaging.
It is very important that imaging algorithm decide the image
quality because the detector performance is almost same
as any company product.
ETCC ??
Compton camera has no good imaging algorithm because
reconstruction method is very complex compared with
other gamma camera.
ETCC can perform the one directional 3D reconstruction
Scintillator
μPICGEM
e-
Determination of direction and energy of each incident
gamma ray
Electron-Tracking Compton Camera μTPC(Time Projection Chamber)
μPIC(μ Pixel Chamber)+GEM+Drift cage
10 x 10 x 10 cm^3 TPC
uPIC:10x10cm^2, 400um pitch, Gain x3000
GEM: 10x10cm^2, Gain x 10
TPC spatial resolution ~300um
3D track and energy of Compton-recoil
electron
PSA (Pixel Scintillation Array)
GSO or LaBr3 + PMT(H8500:HPKK)
Pixel size: 6 x 6 x 13 mm,
Energy resolution:11%
Position and energy of scattered gamma
ray
TPC
PSA
180cm
70cm120cm
RI reagent
10cm15cm
10cm
10cm
Developed ETCC
Reconstruction method
Simple Back Projection
(SBP)
Source Space Tree
Algorithm (SSTA)
Reconstruction Filter
(RF)
List mode Maximum
Likelihood Expectation
Maximization (List mode
MLEM)
Parametric Wiener filter
(PWF)
RFSBP
SSTA
ListmodeMLEM
PWF
Reconstruction flow
Back Projection Region
Defined the reconstruction region
Using voxel 32 x 32 x 32
Voxel size is 200mm / 32 = 6.25mm20
20
20
10
(unit: cm)
TPC
PSA
1010
reconstruction
region
10
source
voxel
Simple Back Projection (SBP)
Compton point
absorption point
back projection
region
Defined the reconstruction region
Using voxel 32 x 32 x 32
Voxel size is 200mm / 32 = 6.25mm
Defined the reconstruction
function
Cauchy distribution
Source Space Tree Algorithm (SSTA)
Defined the reconstruction region
Using voxel 32 x 32 x 32
Voxel size is 200mm / 32 = 6.25mm
Defined the reconstruction
function
Cauchy distribution
Defined the weight of voxel
Defined the line
Located the reference points
(sampling length is Δd)
Calculate the weight at the reference
point which is included the voxel. R.C. Rohe, M.M. Sharfi, K.A. Kecevar et al., 1997.
SSTA 3D Images
Reconstruction image
1 event
ETCC
Camera
SSTA 3D Images
Reconstruction image
5 events
SSTA 3D Images
Reconstruction image
6000 events
SSTA problem
Point source (365keV)
XY slice
Spatial Resolution 50.1mm
XZ slice
Spatial Resolution 187.9mm??
Depth peak is corresponded
with the location of source
Depth direction has the tail of
reconstruction distribution.
These spatial resolutions are too
worth to image an affected part.
FWHM
50.1 mm
Depth peak
96.9 mm
Y
X
Z
X
List mode MLEM
n
k
l
knk
nnm
m
l
ml
mc
yc
S
1
m,k :Image bin
n :event
l : iteration
λml+1 : Image bin m at iteration level l
Sm : Probability that an event emitted in image bin m is detected
Cn m : Position of the event that intersects the image
yn : Probability that an event emitted in image bin m is detected as event bin n
List mode Maximum Likelihood Expectation Maximization
Imaging results & Iteration level
Level 0 (mSSTA) Level 10 Level 100Level 5
Y
X
Z
X
Point source 365keV
Bin 32
Event 1000
Quantities of MLEM
Quantities of
MLEM depends
on number of
events.
Quantities is less
than 5%
(>200events).
5% is same level
as PET and
SPECT
Point source 365keV
Bin 32
Iteration 30
Multi heads 3D study
Multi heads ETCC
has the 2 big merits
Effective area
Depth
information
3heads and 30
iterations condition
is enough to image
the point source
Level
5
Level
10
Level
30
2 heads 3 heads 5 heads1 head
Z
X
List mode MLEM problem
MLEM problem
Big memory size
Number of Events x Num of X bin x Num of Y bin x Num of Z bin.
Long reconstruction time
Computer power
Parametric Wiener filter
Parametric Wiener filter (PWF)
M(u,v,w) : filter function
H(u,v,w) : Fourier transform of a PSF function
H*(u,v,w): complex conjugate of H (u,v,w)
Γ : parameter to reduce the effect of noise
Imaging Result (Point source)
Number of Bin is 128.
PWF does not need
the big memory
size.
The resolution of XY
slice is 16 mm and
XZ is 26 mm
Depth resolution is
better than MLEM
Y
X
Z
X
SSTA PWF
Point source 365keV
Bin 128
Event 6000
Quantitiesof PWF
Quantities of
MLEM depends
on number of
events.
Quantities is less
than 5%
(>1000events).
5% is same level
as PET and
SPECT
Point source 365keV
Bin 32
PWF problem
200 events 1000 events 2000 events
PWF
MLEM
Y
X
Imaging power for
low statistics is
very important
part for medical
gamma camera.
PWF has many
artifact at low
statistics image
PWF is used a
deconvolution
algorithm and need
precise data.
Comparison of algorithm
Advantage & disadvantage
Comparison of List mode MLEM and PWF
Spatial
Resolution
At 365 keV
Statistics Reconstru
ction time
Memory
size
3D
reconstruc
tion
List mode
MLEM
XY : 11mm
XZ ; 21mm
5%(>200events )
X X 〇(Multi head)
PWF XY: 16mm
XZ: 26mm
5%(>1000events )
〇 〇 〇
One directional imaging thyroid phantom
3D imaging of
Thyroid grand
phantom
6M events
Human phantom
I-131 (365keV)
Other area was
filled with neck
bone and water
Amount of material
same as human
PWF
45mm
I-131 Cross linked micelle imaging
Cross linked micelle
imaging
Nagasaki lab. Tsukuba Univ
365keV
4hours data
Accumulation of the micelle in
the tumor.
PWF
Comparison of medical gamma
cameras
Spatial
Resolution
Placement
(FOV)
Energy
Dynamic
Range
Effective
area
3D
reconstructi
on of one
direction.
SPECT 10mm
@365keV
Ring or rotate
(diameter)
<365keV 〇 X
PET 5mm
@511keV
Ring
(diameter)
511keV 〇 X
ETCC 11mm
@365keV
8mm
@511keV
free
(1str)
300~1500keV
X
? (Multi
heads)
〇
Imaging detector
分子プローブ
Molecular
Probe
Visualize
Imaging
detector
Drug design・Clinicapplication
Clinic
29
Medical Imaging
Summary
Development of reconstruction algorithm SSTA
List mode MLEM
Parametric wiener filter
Spatial resolution 11 mm @ 365keV and quantities < 5%
Phantom & Animal 3D reconstruction
Future work Multi heads
Data acquisition algorithm of TPC
New RI reagent