Post on 16-Feb-2021
Image Reconstruction and Motion Correction using
STIR
MIC 2009 STIR Workshop
Nikolaos DikaiosSupervisor: Tim Fryer
CAMBRIDGE
UNIVERSITY OF
Why STIR?
ν Well evaluated reconstruction algorithms
ν User friendly
ν Useful utilities
ν Easy to develop
Image ReconstructionNCAT (GE Advance)
OSEM (8 iterations, 12 subsets)Sinogram
Motion Correction using STIR
ν Calculate motion parameters from registration (ITK library)
ν Motion correction methods
– post-reconstruction registration
– motion compensated 4D image
reconstruction
Target Image
Source Image
RegistrationDeformation
Field
Motion Parameters
Source Image Target Image Deformation Field
Evaluate Registration
Target - RegisteredTarget - Source
SSD CC NMI
Before registration 64.4 0.962 1.439
After registration 34.9 0.989 1.645
Motion Correction (I)
Post-Reconstruction Registration (PRR)
N
fWf
f
N
n
nn∑=
→
+
=2
11 ~~
~
Reconstruct gated
projection data
separately
Register
reconstructed
images – produce
DOFs
Deform reconstructed
images & calculate the
average
Motion Correction (II)
Motion Compensated 4D Image Reconstruction (MCIR)
1
1PWP
n
n
→
=
∑∑ =
=
=
N
n n
nT
nN
n
T
n
fP
gP
P
ff
1
1
~
where
Features
• Unlike PRR all acquired data are taken into account when updating the estimate
• Computationally expensive
LOR
n
bvp
pbv1
Evaluation Using NCAT
Target Frame
No Motion Correction
PRR
MCIR
Lesion Quantification
−⋅=
etT
etTframesAllbiasPercentage
arg
arg100
-35
-30
-25
-20
-15
-10
-5
0
Liver Lesion Left Lung Lesion Right Lung Lesion
No motion correction
PRR
MCI
Add Poisson Noise
Target Frame
No Motion Correction
PRR
MCIR
Lesion Quantification
−⋅=
etT
etTframesAllbiasPercentage
arg
arg100
-35
-30
-25
-20
-15
-10
-5
0
Liver Lesion Left Lung Lesion Right Lung Lesion
No motion correction
PRR
MCI
PET+MR: Cambridge Scanner
Image ReconstructionMOBY (Focus 120)
OSEM (8 iterations, 12 subsets)
Summary
ν We have implemented motion correction in STIR– PRR
– MCIR
ν Motion correction improved lesion quantification in NCAT
ν STIR-based motion correction has also been applied to real MR/pseudo-PET (poster M05-253)