Transcript otws4 7875

Diffusion, reaction, and spin-echo signal
attenuation in branched structures
Denis S. Grebenkov
Laboratoire de Physique de la Matière Condensée
CNRS – Ecole Polytechnique, Palaiseau, France
Workshop IV “Optimal Transport in the Human Body: Lungs and Blood” , 22 May 2008, Los Angeles, USA
Outline of the talk

Branched structure of the lung acinus
E. Weibel
H. Kitaoka and co-workers

Oxygen diffusion and lung efficiency
B. Sapoval and M. Filoche
M. Felici (PhD thesis)

Toward lung imaging and understanding
G. Guillot
Pulmonary system
O2
O2
O2
1 acinus ~ 8 generations
10 000 alveoli
Gas exchange units
Dichotomic branching
Pulmonary system


Densely filling the volume
With a large surface area for
oxygen transfer to blood
O2
in the bulk
c = 0
O
O
D nc = Wc on the boundary
1 acinus ~ 8 generations
10 000 alveoli
c = c0 at the entrance
Gas exchange units
2
2
Geometrical model of the acinus





Dichotomic branching
Filling of a given volume
Controlled surface area and
other physiological scales
Random realizations
Simplicity for numerical use
Kitaoka et al. J. Appl. Physiol. 88, 2260 (2000)
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
• when chosen, suppress 1,
shift other indexes by -1
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
• when chosen, suppress 1,
shift other indexes by -1
1
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
• when chosen, suppress 1,
shift other indexes by -1
1
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
• when chosen, suppress 1,
shift other indexes by -1
1
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
• when chosen, suppress 1,
shift other indexes by -1
1
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
• when chosen, suppress 1,
shift other indexes by -1
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
• when chosen, suppress 1,
shift other indexes by -1
1
2
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
• when chosen, suppress 1,
shift other indexes by -1
1
2
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
• when chosen, suppress 1,
shift other indexes by -1
1
21
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
• when chosen, suppress 1,
shift other indexes by -1
1
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
• when chosen, suppress 1,
shift other indexes by -1
3
2
1
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
• when chosen, suppress 1,
shift other indexes by -1
2
1
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
• when chosen, suppress 1,
shift other indexes by -1
2
1
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
• when chosen, suppress 1,
shift other indexes by -1
2
1
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
• when chosen, suppress 1,
shift other indexes by -1
2
1
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
• when chosen, suppress 1,
shift other indexes by -1
21
1
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
1
• when chosen, suppress 1,
shift other indexes by -1
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
• when chosen, suppress 1,
shift other indexes by -1
1
2
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
1
3
2
• when chosen, suppress 1,
shift other indexes by -1
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
1
3
2
• when chosen, suppress 1,
shift other indexes by -1
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
1
3
2
• when chosen, suppress 1,
shift other indexes by -1
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
3
2
1
• when chosen, suppress 1,
shift other indexes by -1
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
3
2
1
• when chosen, suppress 1,
shift other indexes by -1
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
2
• when chosen, suppress 1,
shift other indexes by -1
1
3
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
Idea: to fill densely a
given volume with a
branching structure
• the current box is always 1
• when chosen, suppress 1,
shift other indexes by -1
• the previously current box
takes the largest index + 1
• the new box takes the
largest index + 2
Kitaoka model
2D labyrinth
its skeleton
Felici et al., PRL 92, 068101 (2004); Grebenkov et al., PRL 94, 050602 (2005)
Kitaoka model
2D labyrinth
its skeleton
Felici et al., PRL 92, 068101 (2004); Grebenkov et al., PRL 94, 050602 (2005)
Kitaoka model
Since there is no memory, diffusion
averages out the effect of the specific
geometrical features of the domain
DIFFUSION IS NOT SENSITIVE TO
LOCAL GEOMETRICAL DETAILS
Is this model geometry accurate enough?
Outline of the talk

Branched structure of the lung acinus

Oxygen diffusion and lung efficiency

Toward lung imaging and understanding
Finite element resolution
c = 0 in the bulk
D nc = Wc on the boundary
c = c0 at the entrance
solving the
discretized
equations…
Felici et al. J. Appl. Physiol. 94, 2010 (2003)
Felici et al. Phys. Rev. Lett. 92, 068101 (2004)
Felici et al. Resp. Physiol. Neurob. 145, 279 (2005)
Diffusion on a skeleton tree
2D labyrinth
Felici et al., Phys. Rev. Lett. 92, 068101 (2004)
its skeleton
A step further: analytical theory
Diffusion-reaction on a tree can
be solved analytically using a
“branch-by-branch” computation
Grebenkov et al., PRL 94, 050602 (2005).
One branch analysis
Continuous problem
=D/W
Dnc = Wc
c0 a c = 0
al
Discrete problem
=(1+a/)-1
½(ck-1+ck+1)-ck = ck
c0
Grebenkov et al., PRL 94, 050602 (2005)
0
1
2
…
l-1 l
l+1
ext = Wcl+1
One branch analysis
Continuous problem
=D/W
Dnc = Wc
c0 a c = 0
al
Discrete problem
=(1+a/)-1
½(ck-1+ck+1)-ck = ck
c0
0
1
2
…
l-1 l
ext = Wcl+1
ent = D c0/’
ul+a
’ = fl() = a
(u2l- v2l) + aul
ext = D cl+1/
Grebenkov et al., PRL 94, 050602 (2005)
l+1
Branch-by-branch computation
…
0
l-1 l
1
2
l+1
0
1
2
Grebenkov et al., PRL 94, 050602 (2005)
…
…
l1-1 l1
l1+1
l2-1 l2
l2+1
Branch-by-branch computation
1ent = D c0/fl 1()
…
l-1 l
0
l+1
1
1
2
2
2ent = D c0/fl 2()
At branching point:
ext = 1ent+2ent
cl+1= c10 = c20
1ext = D cl +1/
…
…
1
l1-1 l1
l1+1
l2-1 l2
l2+1
2ext = D cl +1/
2
ext
=
D[1/f
()+1/f
()]
l
l
cl+1
Grebenkov et al., PRL 94, 050602 (2005)
1
2
= D/’
Branch-by-branch computation
1ent = D c0/fl 1()
…
l-1 l
l+1
ext = 1ent+2ent
cl+1= c10 = c20
1
ext = D cl+1/’
2ent = D c0/fl 2()
At branching point:
1ext = D cl +1/
2ext = D cl +1/
2
ext
=
D[1/f
()+1/f
()]
l
l
cl+1
Grebenkov et al., PRL 94, 050602 (2005)
1
2
= D/’
Symmetric trees
m
1
m
1
1
=
=
=

1 k=1 fl () fl () fl ()
m=2
k
1= fl ()
 
Grebenkov et al., PRL 94, 050602 (2005)
Symmetric trees
m
1
m
1
1
=
=
=

1 k=1 fl () fl () fl ()
m=2
1= fl ()
2= fl (1)
1 1 1 1
Grebenkov et al., PRL 94, 050602 (2005)
Symmetric trees
m
1
m
1
1
=
=
=

1 k=1 fl () fl () fl ()
m=2
structures
1= Branching
fl ()
against
2= flare
(1robust
)
2
2
…permeability change
Total flux at the root: () = D c0/n
a(l+1)
n= fl (fl (fl (…fl ()…)))
n
m-1
Grebenkov et al., PRL 94, 050602 (2005)
Application to human acini
Haefeli-Bleuer and Weibel, Anat. Rec. 220, 401 (1988)
Human acinus
Approximation by a symmetric tree
 of
the same total area
 of the same average length of branches
 of the same branching order (m=2)
Grebenkov et al., PRL 94, 050602 (2005)
Human acinus
1
10
symmetric acinus
( )
real acinus
0
10
-1
10
-2
10
/L
-3
10 -2
10
-1
10
0
10
Grebenkov et al., PRL 94, 050602 (2005)
1
10
2
10
p
Outline of the talk

Branched structure of the lung acinus

Oxygen diffusion and lung efficiency

Toward lung imaging and understanding
Schematic principle of NMR
Static magnetic field B0
z
y
x
local magnetization
90° rf pulse
Grebenkov, Rev. Mod. Phys. 79, 1077 (2007)
Schematic principle of NMR
Static magnetic field B0
z
y
x
Phase at time T
Grebenkov, Rev. Mod. Phys. 79, 1077 (2007)
Schematic principle of NMR
Static magnetic field B0
z
Inhomogeneous magnetic field
z
y
x
Phase at time T
Grebenkov, Rev. Mod. Phys. 79, 1077 (2007)
y
x
Monte Carlo simulations

Spin trajectory Xt is modeled as a sequence
of normally distributed random jumps, with
reflections on the boundary of the acinus
Grebenkov et al., JMR 184, 143 (2007)
Healthy acinus
1
S
0.9
Fixed
gradient
direction
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0
g
2
4
6
Grebenkov et al., JMR 184, 143 (2007)
8
10
mT/m
Healthy acinus
1
S
0.9
Fixed
gradient
direction
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0
g
2
4
6
Grebenkov et al., JMR 184, 143 (2007)
8
10
mT/m
Healthy acinus
1
S
0.9
Averaged
gradient
direction
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
g
2
4
6
Grebenkov et al., JMR 184, 143 (2007)
8
10
mT/m
Emphysematous acini
How can one model
emphysematous acini?
Emphysema may lead to
enlargement of the alveolar ducts
partial destruction of the alveolar tissue
Emphysematous acini
Emphysema may lead
to partial destruction of
the internal alveolar tissue
Emphysematous acini
1
S
=0.6
=0.5
=0.4
=0.3
=0.2
=0.1
=0.0
0.8
0.6
0.4
0.2
0
0
ν = 0.6
gmT/m
2
ν = 0.5
4
ν = 0.4
6
8
ν = 0.3
10
ν=0
Conclusions
Branching
structures present peculiar
properties which have to be taken into
account to understand the lungs
The
Kitaoka model of the acinus is
geometrically realistic and particularly
suitable for numerical simulations
Conclusions
Oxygen
diffusion can be studied on the
skeleton tree of the model or realistic
human acinus… a crucial simplification!
Diffusion-reaction
on a tree can be
solved using a “branch-by-branch” trick
which allows for very fast computation
and derivation of analytical results
Conclusions
Tree
structures are robust against the
change of the permeability (mild edema)
Partial
destruction of
branched
structure by emphysema can potentially be
detected in diffusion-weighted magnetic
resonance imaging experiments with HP
helium-3
Perspectives
Further theoretical, numerical and
experimental study of restricted
diffusion in branched or porous
structures are important
Thank you for your attention!!!
[email protected]
http://pmc.polytechnique.fr/pagesperso/dg
If you see this slide, the talk is about to end… sorry
What shall we do during
“discussion” 4:00 – 4:50?
Please do not leave!!!
Lung imaging with helium-3
Normal volunteer
Can one make a reliable
Healthy smoker
diagnosis at earlier stage?
Patient with
severe emphysema
van Beek et al. JMRI 20, 540 (2004)
Human acinus
Haefeli-Bleuer and Weibel, Anat. Rec. 220, 401 (1988)