iterative reconstruction for metal artifact reduction in ct
DESCRIPTION
Iterative reconstruction for metal artifact reduction in CT. the problem projection completion polychromatic ML model for CT local models, bowtie,… examples. Katrien Van Slambrouck, Johan Nuyts Nuclear Medicine, KU Leuven. the problem. CT. iron. y. ln(b/y). the problem. - PowerPoint PPT PresentationTRANSCRIPT
Iterative reconstruction for metal artifact reduction in CT
1
• the problem
• projection completion
• polychromatic ML model for CT
• local models, bowtie,…
• examples
Katrien Van Slambrouck, Johan Nuyts
Nuclear Medicine, KU Leuven
the problem
2
y
CTCT
jj ijlii eby
L
xd)x(e),s(b),s(y
ln(b/y)
iron
the problem
Double hip prosthesisDouble knee prosthesis Dental fillings
Cause of metal artifacts:•Beam hardening•Nonlinear partial volume effects•Noise•Scatter•resolution (crosstalk, afterglow)•(Motion)
Mouse bone and titanium screw (microCT)
3
I. Beam hardeningPolychromatic spectrum, beam hardens when going through the objectLow energy photons are more likely absorbed
Artifacts in CT
Energy (keV)
10 cm water
10 cm water
Energy (keV) Energy (keV)
Nor
mal
ized
inte
nsity
(%)
Nor
mal
ized
inte
nsity
(%)
Nor
mal
ized
inte
nsity
(%)
Typical artifact appearance: dark streaks in between metals, dark shades around metals (and cupping)
Iron in water Amalgam in PMMA
II. (Non)-linear partial volume effects• Linear: voxels only partly filled with particular substance• Non-linear: averaging over beam width, focal spot, …
I0
I
µ1µ2
Artifacts in CT
Typical artifact appearance: dark and white streaks connecting edges
Iron in water Amalgam in PMMA
III. Scatter• Compton scatter: deviation form original trajectory • Scatter grids?
Artifacts in CT
I0
Iron in water Amalgam in PMMA
Typical artifact appearance: dark streaks in between metals, dark shades around metals (and cupping)
IV. Noise• Quantum nature: ± Poisson distribution
Artifacts in CT
Iron in water Amalgam in PMMA
Typical artifact appearance: streaks around and in between metals
projection completion
Initial FBP reconstruction Segment the metals and project Remove metal projections for sinogram Interpolate (e.g. linear, polynomial, …) Reconstruct (FBP) and paste metal parts
8
• Kalender W. et aI. "Reduction of CT artifacts caused by metallic impants." Radiology, 1987• Glover G. and Pelc N. "An algorithm for the reduction of metal clip artifacts in CT reconstructions." Med. Phys., 1981• Mahnken A. et al, "A new algoritbm for metal artifact reduction in computed tomogrpaby, In vitro and in vivo evaluation after
total hip replacement." Investigative Radiology, 2003
projection completion
9
window 600 HU
Fe
PMMAH2O
projection completion
10
true object FBP projection completion
window 600 HU
1
projection completion
11
2
• Muller I., Buzug T.M., "Spurious structures created by interpolation-based Ct metal artifact reduction." Proc. of SPIE, 2009• Meyer E. et al, "Normalized metal artifact reduction (NMAR) in computed tomography." Med. Phys., 2010
zeroed metal trace
linear interpolation
NMAR
12• Muller I., Buzug T.M., "Spurious structures created by interpolation-based Ct metal artifact reduction." Proc. of SPIE, 2009• Meyer E. et al, "Normalized metal artifact reduction (NMAR) in computed tomography." Med. Phys., 2010
sinogram interpolated sinogram ofsegmentation
normalizedsinogram
window 600 HU
NMAR
13
1
2
sinogram,metals erased
sinogram ofthe segmentedreconstruction
NMAR
14
1
2
normalizedsinogram,metals erased
interpolatedsinogram
NMAR
15
unnormalizedinterpolatedsinogram
proj.completion and NMAR
16
true object FBP projectioncompletion
window 300 HU
NMAR
17
CTCT
jj ijlii eby
Maximum Likelihood for CT
L
xd)x(e),s(b),s(y
Maximum Likelihood for CT
18
CTCT
jj ijlii eby
data recon
computing p(recon | data) difficult inverse problem
computing p(data | recon) “easy” forward problem
one wishes to find recon that maximizes p(recon | data)
Bayes:
p(recon | data) = p(data | recon) p(recon)
p(data)
data recon
~
Maximum Likelihood for CT
19
MAP
ML
Maximum Likelihood for CT
p(recon | data) ~
p(data | recon)
projection Poisson
j
j ijjii lexpby
j = 1..Ji = 1..I
i i
yiy!y
yei
i
i
ii )y|y(p
i
iiii )!yln(yylnyln(p(data | recon)) = L(data | recon) = ~
p(data | recon)recon data
20
Maximum Likelihood for CT
i
iii yylnyL(data | recon) j ijjl
ii eby
21
i k kikiij
i iiijjj lyl
yyl
iterative maximisation of L:
0j
22
MLTR
convex algorithm [1]
[1] Lange, Fessler, “Globally convergent algorithms for maximum a posteriori transmission tomography”, IEEE Trans Image Proc, 1995
[2] JA Fessler et al, "Grouped-coordinate ascent algorithm for penalized likelihood transmission image reconstruction." IEEE Trans Med Imaging 1997.
[3] Fessler, Donghwan, "Axial block coordinate descent (ABCD) algorithm for X-ray CT image reconstruction.“ Fully 3D 2011
patchwork: local update [2,3]
i k kikiij
i iiijjj lyl
yyl
MLTR
MEASUREMENT
REPROJECTION
COMPAREUPDATE RECON
23
MLTR
24
validationSiemens Sensation 16
Siemens MLTR
models for iterative reconstruction
25
I
iiii yylnyLPoisson Likelihood:
measured data
data computed from current reconstruction image
iy
iy
J
jjijii lexpby
Projection model:
• monochromatic:
iy
bi
models for iterative reconstruction
I
iiii yylnyL
J
jjijii lexpby
waterref
waterk
kk
J
jjijkiki PlPexpby
Poisson Likelihood:
Projection model:
• monochromatic:
• 1 material polychromatic:
26
energy k
intensity bik
measured data
data computed from current reconstruction image
iy
iy
energy“water correction”
MLTR_C
models for iterative reconstruction
27
J
jjijii lexpby
J
jjkij
K
kiki lexpby
• Full Polychromatic Model – IMPACT
I
iiii yylnyLPoisson Likelihood:
energy k
intensity bik
Projection model:
jk = j ∙ photok + j ∙ Comptonk
models for iterative reconstruction
28
J
jjijii lexpby
J
jjkij
K
kiki lexpby
• Full Polychromatic Model – IMPACT
water
Comptonphoto-electric
attenuation
al
jk = photo-electric + Compton at energy k
Comptonk = Klein-Nishina (energy)Photok ≈ 1 / energy3
models for iterative reconstruction
29
J
jjijii lexpby
J
jjkij
K
kiki lexpby
• Full Polychromatic Model – IMPACT
and (1/cm)
mono (1/cm)
jk = j ∙ photok + j ∙ Comptonk
jk = j∙ photok + j ∙ Comptonk
and (1/cm)
mono (1/cm)
models for iterative reconstruction
30
patches, local models
31
MLTR
convex algorithm [1]
[1] Lange, Fessler, “Globally convergent algorithms for maximum a posteriori transmission tomography”, IEEE Trans Image Proc, 1995
[2] JA Fessler et al, "Grouped-coordinate ascent algorithm for penalized likelihood transmission image reconstruction." IEEE Trans Med Imaging 1997.
[3] Fessler, Donghwan, "Axial block coordinate descent (ABCD) algorithm for X-ray CT image reconstruction.“ Fully 3D 2011
patchwork: local update [2,3]
i k kikiij
i iiijjj lyl
yyl
bowtie, BHC
32
e-
energy k
intensity bik
• raw CT data not corrected for beam hardening• send spectrum through filter and bowtie
bik = spectrum(k) x bowtie(i)
patches, local models
IMPACT is complex and slow, MLTR and MLTR_C are simpler and faster
Find the metals
PATCH 3
PATCH 2
PATCH 1
Define patches
IMPACT in metalsMLTR_C elsewhere
33
PATCH 4
clinical CT (Siemens Sensation 16)Body shaped phantom
34
sequential CT (Siemens Sensation 16)Body shaped phantom
35
FBP Regular PC PC NMAR
IMPACT PATCH MLTR_C + IMPACT
IMPACT
20 iter x 116 subsets
sequential CT (Siemens Sensation 16)Body shaped phantom
36
Black = FBPBlue = PC-NMARRed = IMPACT PATCH
water aluminumCoCr..Ti Al V PMMA water
helical CT
37
sequential 2 x 1mm helical 16 x 0.75mm
helical CT
38
MIP
IMPACT
FBP
NMAR
metal patches,uniform init.
no patches,NMAR init.
metal patches,NMAR init.
5 iter x 116 subsets
helical CT
39
IMPACT
FBP
NMAR
metal patches,uniform init.
no patches,NMAR init.
metal patches,NMAR init.
MIP
helical CT
40
FBP NMAR5 it10 it
IMPACT
helical CT
41
We give patches same x-y sampling but increased z-sampling:
z-sampling x 3impact, regular z
to do
42
• after 5..10 x 100 iterations with patches still incomplete convergence• persistent artifacts near flat edges of metal implants
• we currently think it is noto scattero non-linear partial volume effecto crosstalk, afterglowo detector dead space
43
thanks
better physical model
better reconstruction
Katrien Van SlambrouckBruno De Man
Karl Stierstorfer,David Faul, Siemens