Application of independent component analysis (ICA) to identify and ...

Application of independent component analysis
(ICA) to identify and separate tumor arterial
input function (AIF) in dynamic contrast
enhanced-MRI
Hatef Mehrabian
Chaitanya Chandrana, Ian Pang, Rajiv Chopra, Anne Martel
Field MITACS conference
June 20, 2011

Illustrations: courtesy of http://www.reishiscience.com/Benefits_2.html
In-vivo images: D. Fukumura et al., MICROVASCULAR RESEARCH, 2007.
Tumor Angiogenesis
100µm

100µm
100µm
Illustrations: courtesy of http://www.reishiscience.com/Benefits_2.html
In-vivo images: D. Fukumura et al., MICROVASCULAR RESEARCH, 2007.
Tumor Vasculature
100µm

100µm
100µm
Illustrations: courtesy of http://www.reishiscience.com/Benefits_2.html
In-vivo images: D. Fukumura et al., MICROVASCULAR RESEARCH, 2007.
Tumor Vasculature

100µm
Pharmacokinetic Modeling
.. ..
..
.
. ...
.. .
. .
.. .
.

100µm
Pharmacokinetic Modeling

100µm
K trans K ep
Pharmacokinetic Modeling

100µm
K trans K ep
Pharmacokinetic Modeling
Renal Cell Carcinoma
DCE – MRI

MR Pixel
0.5mm
100µm
0.02 mm
K trans K ep
Pharmacokinetic Modeling
MR Slice
0.5mm

IV
EV
Dynamic Contrast Enhanced MRI
X

IV
EV
Dynamic Contrast Enhanced MRI
X

x
s IV
s EV
Dynamic Contrast Enhanced MRI
x

x aIV
s IV
s EV
a: contrast Concentration
Dynamic Contrast Enhanced MRI
x

a: contrast Concentration
s: image
s IV
s EV
x aIV sIV
Dynamic Contrast Enhanced MRI
x

a: contrast Concentration
s: image
s IV
s EV
x aIV IV
s a
sE
Dynamic Contrast Enhanced MRI
EV V
x

x aIV IV
a: contrast Concentration
s: image
s a
sE
Dynamic Contrast Enhanced MRI
EV V
x

X
t
Independent Component Analysis
= +
s s
IV
EV
t

X
t
Independent Component Analysis
= +
s s
IV
EV
X AS
1, 2,...,
N
IV , EV
s ,
X x x x
A a a
S s
IV EV
T
t

Estimator:
X
F A
X | s , s
t
IV EV
= +
s s
IV
EV
t
Estimation Model

Estimator: F X | A
s , s
X
t
Maximum Likelihood Estimation
IV EV
Maximum likelihood estimator corresponds to the value
θ ML that makes the obtained measurements more likely.
1 2
ML X | arg max{ p( x , x ,..., x )}
= +
s s
IV
EV
t
N

Estimator:
F A
t
Maximum Likelihood Estimation
X | s , s
IV EV
= +
T
1 2 N
ML X | A
arg max px, x ,..., x | A
X
A
s s
IV
EV
t

Estimator:
F A
t
Maximum Likelihood Estimation
X | s , s
IV EV
= +
T
1 2 N
ML X | A
arg max px, x ,..., x | A
X
A
X AS
p X p AS A p S
1
( ) ( ) det( )
( )
s s
IV
EV
t

Estimator:
F A
t
Maximum Likelihood Estimation
X | s , s
IV EV
= +
T
1 2 N
ML X | A
arg max
p x , x ,..., x
A
1
argmax det A p S | A
| A
X
A
s s
IV
EV
t

Estimator:
F A
t
Maximum Likelihood Estimation
X | s , s
IV EV
= +
T
1 2 N
ML X | A
arg max
p x , x ,..., x
A
| A
1
argmax det A p S | A
A
1
argma x detA p s , |
IV sEV A
X
A
s s
IV
EV
t

Estimator:
F A
t
Maximum Likelihood Estimation
X | s , s
IV EV
= +
T
1 2 N
ML X | A
arg max
p x , x ,..., x
A
| A
1
argmax det A p S | A
If s A 1& s2 are independent
1
argma x detA p s , |
IV sEV A
p( s , s ) p( s )
p( s )
X
A
1 2 1 2
s s
IV
EV
t

Estimator:
F A
t
Maximum Likelihood Estimation
X | s , s
IV EV
= +
T
1 2 N
ML X | A
arg max
p x , x ,..., x
A
| A
1
argmax det A p S | A
A
1
argma x detA p s , |
IV sEV A
A
1
arg mx a det A p s IV | ApsEV | A
X
A
s s
IV
EV
t

Independent Component Analysis
1
ML X | A arg max de t A p s | |
IV A p sEV
A
A

Independent Component Analysis
1
ML X | A arg max de t A p s | |
IV A p sEV
A
A
p s ? i

s , s s ps p p
Probability Density Function (pdf)
i j i j
0
Es 2
p s
? i
i
2
si
E1

Probability Density Function (pdf)
p s
? i
s 2
p s
p
E s
i i
i
s s
2
i i
0

Independent Component Analysis (ICA)
1
ML X | A arg max de t A p s | |
IV A p sEV
A
A
1
2log cosh
exp
p s
i i s

• Dialysis Tubing
2 cm
2.5 cm
Phantom Study

• Dialysis Tubing
• Pore size: 90-970 nm
2 cm
2.5 cm
Phantom Study
300 µm

Phantom Study
1 mm
Raw Data
in-plane resolution=300µm

Intravascular
Extravascular
1 mm 1 mm 1 mm
Separation Results (ICA)
Phantom Study
Raw Data

• High resolution pre-contrast MR image
1 mm 1 mm
Validation
DCE MRI Dataset High Resolution (Pre-contrast)

• High resolution pre-contrast MR image
– Thresholding Binary mask
1 mm 1 mm
DCE MRI Dataset Mask
Validation

Intravascular
Extravascular
1 mm 1 mm 1 mm
Phantom Study
Raw Data
Separation Results (ICA) Separation Results (Mask)

• Tumor in the Thigh muscle
• Imaged with US and MRI
In-vivo Experiment (MR data)
1 cm
Raw Data

• Tumor in the Thigh muscle
• Imaged with US and MRI
In-vivo Experiment (MR data)
1 cm
Raw Data
1 cm 1 cm
Extravascular Intravascular

• Tumor in the Thigh muscle
• Imaged with US and MRI
Ultrasound Contrast Agent (µBubbles)
1 cm
Raw Data

• Tumor in the Thigh muscle
• Imaged with US and MRI
Circulating µbubbles
Ultrasound Contrast Agent (µBubbles)
1 cm
Raw Data

• Ultrasound Imaging
• µBubbles stay intravascular
In-vivo Experiment (MRI vs. US)
1 cm
1 cm
Raw Data- US
Intravascular-MR

Summary & Conclusions
• Tumor vasculature is heterogeneous and leaky
• MRI contrast agent is capable of leaking
• Tumor is characterized based on the tracer dynamics
• ICA is capable of separating the intravascular and
extravascular compartments and identifying arterial
input function
• Results of phantom and in-vivo experiment studies
were promising

Acknowledgements

?
Thank You