You are on page 1of 15

PHYSICAL REVIEW E 95, 022415 (2017)

Dephasing and diffusion on the alveolar surface

L. R. Buschle,1,2 F. T. Kurz,1,2 T. Kampf,3 W. L. Wagner,4,5 J. Duerr,5,6 W. Stiller,4,5 P. Konietzke,4,5 F. Wünnemann,4


M. A. Mall,5,6 M. O. Wielpütz,4,5 H. P. Schlemmer,1 and C. H. Ziener1,2
1
German Cancer Research Center - DKFZ, Im Neuenheimer Feld 280, 69120 Heidelberg, Germany
2
Neuroradiology, Heidelberg University Hospital, Im Neuenheimer Feld 400, 69120 Heidelberg, Germany
3
University of Würzburg, Department of Experimental Physics 5, Am Hubland, 97074 Würzburg, Germany
4
University of Heidelberg, Department of Diagnostic and Interventional Radiology, Im Neuenheimer Feld 110,
69120 Heidelberg, Germany
5
University of Heidelberg, Translational Lung Research Center Heidelberg (TLRC), Member of German Center for Lung Research (DZL),
Im Neuenheimer Feld 156, 69120 Heidelberg, Germany
6
University of Heidelberg, Department of Translational Pulmonology, Im Neuenheimer Feld 156, 69120 Heidelberg, Germany
(Received 22 December 2015; revised manuscript received 15 December 2016; published 24 February 2017)
We propose a surface model of spin dephasing in lung tissue that includes both susceptibility and diffusion
effects to provide a closed-form solution of the Bloch-Torrey equation on the alveolar surface. The nonlocal
susceptibility effects of the model are validated against numerical simulations of spin dephasing in a realistic lung
tissue geometry acquired from synchotron-based μCT data sets of mouse lung tissue, and against simulations in
the well-known Wigner-Seitz model geometry. The free induction decay is obtained in dependence on microscopic
tissue parameters and agrees very well with in vivo lung measurements at 1.5 Tesla to allow a quantification of
the local mean alveolar radius. Our results are therefore potentially relevant for the clinical diagnosis and therapy
of pulmonary diseases.

DOI: 10.1103/PhysRevE.95.022415

I. INTRODUCTION tageous to analyze the water signal formation in lung tissue


that is also closely related to the microscopic structure based
Measurements of microstructural lung tissue parameters
on the alveolar geometry. Peripheral lung tissue consist of
improve diagnostics and therapeutic evaluations of pulmonary
very densely packed alveoli [16], comparable to the structure
diseases and, thus, are highly relevant for clinical applications,
of foam [17]. In this geometry, dephasing of signal from
especially for chronic obstructive pulmonary disease [1].
spin-bearing particles is accelerated due to the susceptibility
However, lung tissue structure cannot be resolved with
differences between air-filled alveoli and the surrounding
conventional CT or MR imaging, since typical structural
water [18]. The specific geometry of the pulmonal airway
parameters (such as the mean alveolar radius) are one order
system plays an important role for the magnetization signal
of magnitude smaller than the typical image resolution. An
decay. In this work, we analyze the influence of susceptibility
established method of quantifying lung microstructure is based
on ex vivo lung histology [2]. In addition, high-resolution effects on transverse magnetization using synchrotron-based
CT imaging now reaches the resolution of typical structural μCT data sets of mouse lung tissue and the Wigner-Seitz model
parameters in ex vivo measurements [3,4]. Still, noninvasive as simulation basis.
in vivo measurements without radiation exposure can only Furthermore, diffusion of water molecules influences signal
be performed with magnetic resonance imaging. The most formation in dependence on the alveolar size. Thus, a rigorous
common MRI technique and current gold standard to obtain analytical derivation of the free induction decay caused by
information about the lung’s microstructure consists in lung susceptibility and diffusion effects enables quantifying the
imaging with hyperpolarized gases in external magnetic field local mean alveolar radius.
gradients [5]. To extract microstructural parameters in this The Bloch-Torrey equation [19], a generalization of the
context, some studies made use of analytical techniques [6–8], Bloch equations that includes diffusion terms, is usually
whereas other studies focused on numerical methods [9,10]. applied for the analysis of transverse magnetization in linear
In all proposed techniques, the MR signal of the gas is gradient fields, as in diffusion-weighted imaging [20,21], see
observed and quantified. Such techniques can be used to also Ref. [22] for a detailed review. Stoller et al. gave a
detect deposited aerosol in lung tissue [11], lung airway solution of the Bloch-Torrey equation for a linear gradient
size [6,12] and lung branching structure [13], as well as to in terms of a Green function by using the technique of
quantify the apparent diffusion coefficient [14] and obtain Laplace transform [23]. Barzykin gave an expression of the
apparent diffusion kurtosis maps [8]. Still, MR imaging with Laplace transform of the magnetization [24]. In this case,
hyperpolarized gases is technically challenging and difficult the eigenvalue spectrum of the Bloch-Torrey equation for a
to embed in clinical routine: the production of hyperpolarized linear gradient shows branching points: for large diffusion
gas needs technical expertise and the T1 -relaxation time of the the eigenvalues are purely real, for small diffusion effects,
hyperpolarized gas should be long enough to embed the gas the eigenvalues become complex. In lung tissue, the field
into the patient [15]. inhomogeneities are not created by an external linear gradient
In contrast, measurements of the MR water signal are but by air-filled alveoli that induce local magnetic field
robust, fast, and universally available. Therefore, it is advan- inhomogeneities in the form of dipole fields.

2470-0045/2017/95(2)/022415(15) 022415-1 ©2017 American Physical Society


L. R. BUSCHLE et al. PHYSICAL REVIEW E 95, 022415 (2017)

The general theory of magnetization decay is introduced that are associated with a specific geometry or experimental
in Sec. II. In Sec. III, we analyze the signal evolution condition could be found (e.g., for a linear gradient in
due to microscopic susceptibility effects using numerical Ref. [22] or a two-dimensional dipole field in Refs. [28,29]). In
simulations based on both specific lung geometries as obtained Sec. IV, we derive a closed-form solution of the Bloch-Torrey
from synchrotron-based μCT data sets, and the well-known equation (5) for a specific model geometry that is suitable for
Wigner-Seitz model. Furthermore, we introduce the alveolar lung microstructure.
surface model that provides an analytical approximation of
the magnetization. In Sec. IV, we introduce diffusion effects III. STATIC-DEPHASING SIMULATIONS
into our model and provide a closed-form solution of the
Bloch-Torrey equation on the alveolar surface to describe both A. Synchrotron-based μCT experiments
diffusion and susceptibility effects in lung tissue. We compare All animal studies were approved by the
model results with the Gaussian approximation and the strong Regierungspräsidium Karlsruhe, Germany. Adult 4–6
collision approximation. Using our results, the alveolar radius month old wild type C57BL/6N mice were used in all
is determined from in vivo measurements in human lung tissue synchrotron-radiation-based μCT experiments. Anesthetized
in Sec. V. A summary and conclusion is given in Sec. VI. mice were killed by exsanguination and lungs were perfusion
fixed as previously described [30]. In brief, the trachea
II. GENERAL THEORY was cannulated and lungs were inflated with room air at a
constant pressure of 25 cm H2 O while perfusion fixation was
We analyze the local magnetization of water molecules performed via the pulmonary artery with 30 cm H2 O. After
in lung tissue. The dephasing process in lung tissue is perfusion the trachea was double-tied with a 5-0 surgical silk
mainly influenced by the susceptibility difference between tie to prevent deflation, removed and submerged in fixation
air-filled alveoli and surrounding tissue. The local Larmor solution for 8 h, rinsed with distilled H2 O and exposed to 2%
frequency for a susceptibility distribution χ (r) is given as osmium tetroxide for 8 h, with subsequent freeze drying.
the convolution between χ (r) and dipole kernel D(r) = The samples were scanned at the Tomographic Microscopy
[3 cos2 (θ ) − 1]/r 3 [25,26]: and Coherent Radiology Experiments (TOMCAT) Beamline
ω(r) = γ B0 χ (r) ∗ D(r). (1) at the Swiss Light Source Synchrotron Facility of the Paul
Scherrer Institut (Villigen, Switzerland) [31]. X-rays at energy
In the static-dephasing regime, which does not consider of 20 keV and 410 mA were converted to visible light by
the effects of diffusion on spin-bearing particles, the time a scintillator (20 μm thick LuAG:Ce Crytur Ltd., Turnov,
dependency of the local magnetization m(r,t) is given by [27] Czech Republic). Projection images were further magnified
− Tt by diffraction-limited microscope optics and finally digitalized
m(r,t) = m0 e 2 e−iω(r)t , (2)
by a high-resolution CCD camera with a pixel size of 6.5 μm.
where the relaxation time T2 represents dephasing processes Fourfold magnification optics were used, resulting in a pixel
due to intrinsic spin-spin relaxation. The free induction decay size of 1.625 μm with a field of view of 4.16 mm ×
M(t) is a superposition of the local magnetizations in a voxel 4.16 mm × 1.885 mm. For each measurement, 1001 projec-
with volume V : tions were acquired along with dark and periodic flat field
 images at an integration time of 4 s each without binning. Data
− t
M(t) = m0 e T2 d 3 re−iω(r)t , (3) were postprocessed and rearranged into flat field-corrected
V
sinograms online. Reconstruction of the volume of interest was
and frequency distribution p(ω) and free induction decay M(t) performed on a 16-node Linux PC Farm (Pentium 4, 2.8 GHz,
are connected via a Fourier transform: 512 megabytes RAM) using highly optimized filtered back
 +∞ projection. One exemplary slice of a healthy mouse lung is
− t
M(t) = M0 e T2 dω eiωt p(ω). (4) shown in Fig. 1.
−∞

Both free induction decay and the frequency distribution


depend on the susceptibility distribution χ (r).
The MR signal decay in lung tissue is not only influenced
by microscopic susceptibility effects, but is also subject to
diffusion effects from spin-carrying particles. Although the
size of an alveolus is usually larger than the typical diffusion
length, diffusion effects still contribute to the signal. In general,
the local magnetization m(r,t) is affected by both susceptibility
and diffusion effects as determined by the Bloch-Torrey
equation [19]
 
∂ 1
m(r,t) = D − iω(r) − m(r,t), (5)
∂t T2
FIG. 1. (a) Synchrotron image of healthy lung tissue with an
where  denotes the Laplace operator and D is the diffusion isotropic resolution of 1.625 μm. (b) Frequency of different air
coefficient. This equation is mathematically challenging to volume fractions η in subvolumes with side length 208 μm. The
solve, and, so far, only a few solutions for Larmor frequencies mean volume fraction is η = 0.851 ± 0.045.

022415-2
DEPHASING AND DIFFUSION ON THE ALVEOLAR SURFACE PHYSICAL REVIEW E 95, 022415 (2017)

FIG. 3. Simulation of lung microstructure in the Wigner-Seitz


model [32,33]. Alveoli are represented in white, tissue with spin-
FIG. 2. Frequency distribution p(ω) for different air volume bearing particles in black. The air volume fraction Vair /Vtot ≈ 0.9
fractions η. agrees with values from the literature [34,35]. The tissue model is
calculated using Eq. (8): the thickness of the small layer containing
The data sets were binarized to divide the air-filled alveoli water molecules between alveoli is approximately given as the
and the conducting bronchial system from the surrounding alveolar septum gap
.
tissue. The susceptibility distribution was assumed as χair =
χ and χtissue = 0. Furthermore, without loss of generality, tissue whereas the cell content itself consists of air. To achieve
we assumed the T2 -relaxation time to be infinitely long. The this, we positioned NA alveoli randomly at positions ri within
data sets were divided into cubes with side lengths of 208 μm a cubic voxel of size V = L3 . The soft lung tissue is then
and the local Larmor frequency ω(r) was calculated for each characterized by the following condition:
cube using the Fourier representation of Eq. (1) [26]:
   χ for r2 − r1 >

ω(r) −1 χ̃(k) 3kz2 χ (r) = (8)


=F −1 , (6) 0 else,
δω χ kx2 + ky2 + kz2
where r1 = ranki (|r − ri |,1) represents the shortest distance
where F −1 denotes the inverse Fourier transform with respect to the next alveolus’ position and r2 = ranki (|r − ri |,2)
to k and χ̃ (k) is the Fourier transform of the susceptibility the second-shortest distance. The parameter
adjusts the
distribution χ (r). According to Eq. (3) and Eq. (4), the microstructure and controls the thickness of the thin layer
frequency distribution p(ω) is given as: between the alveoli. A cross section through the simulated
 microstructure is shown in Fig. 3. According to the used
p(ω) = d 3 r δ(ω + ω(r)), (7) dimensionless parameters of
= 2, NA = 40, and L = 200,
V
the air volume fraction in lung tissue follows as η =
where δ denotes Dirac’s delta distribution. Thus, the frequency 0.9060 ± 0.0023, which is in good agreement with literature
distribution p(ω) can be obtained as a histogram of the local values of histological evaluations of lung tissue, and the
Larmor frequency ω(r). Finally, the frequency distribution above synchrotron-based μCT measurements where η ranges
p(ω) for each cube is yielded. between 0.85 and 0.95 (see Fig. 1 and Refs. [34–36]). The
The microstructure in each cube was characterized by the local Larmor frequency can be calculated with the Fourier
volume fraction η = Vair /Vtot and the frequency distributions representation of Eq. (1) given in Eq. (6). The frequency
of cubes with similar volume fractions η were accumulated. distribution was obtained using Eq. (7).
As shown in Fig. 2, the frequency distribution changes for
different air volume fractions η: the peak of the distribution
C. Alveolar surface model
shifts towards negative Larmor frequencies while the width of
the distribution slightly increases. Furthermore, the line shape Considering an alveolar surface model further simplifies
exhibits an asymmetry that is discussed below: see Sec. III C. the description of magnetic field inhomogeneities that occur
through susceptibility differences between tissue and air. In
B. Wigner-Seitz simulations this model, the magnetic perturber is described by a spherical
shell with radius R, neglecting all susceptibility effects from
Since the measurement of synchrotron-based μCT data sets outer tissue. The water molecules that are localized at the
is costly, it is advantageous to a geometrical model of the lung surface of the shell experience a susceptibility difference χ
tissue microstructure such as the Wigner-Seitz model [32,33]. between tissue and air (see Fig. 4). Due to the susceptibility
Briefly, the Wigner-Seitz model (or three-dimensional Voronoi difference χ = χair − χtissue between air and tissue, a local
tesselation) assumes randomly distributed alveoli and con- resonance frequency in form of a dipole field emerges [37]:
structs a Wigner-Seitz cell around each alveolar center. The
surface of each cell is then thickened to describe soft lung ω(θ ) = δω[3 cos2 (θ ) − 1], (9)

022415-3
L. R. BUSCHLE et al. PHYSICAL REVIEW E 95, 022415 (2017)

where θ is the polar angle [see Fig. 4(c)] and the dipole
field strength is given as δω = χ γ B0 /3. Assuming a typical
susceptibility difference of χ = 8 ppm [18] and a magnetic
field strength of B0 = 1.5 T, the dipole field strength is
approximately δω ≈ 150 s−1 . The local Larmor frequency
exhibits the symmetry ω(π/2 − θ ) = ω(π/2 + θ ).
In the alveolar surface model, the alveoli are approximated
by spheres. Microscopic images of lung tissue as shown in
Fig. 4(b) or Ref. [16] may suggest a more complex shape of
the alveoli, such as a honeycomb or a spheroidal geometry,
thus leading to a modified geometry with a different Larmor
frequency ω(r), according to Eq. (1). The local Larmor
frequency for a spheroid was already analyzed in great detail
in Ref. [38]. It is, however, difficult to analytically embed
diffusion effects in these geometrical models.
When modeling alveoli as spherical surfaces, the local
magnetization and the free induction decay can be found as:
− Tt
mSD (θ,t) = m0 e 2 exp(−iδω[3 cos2 (θ ) − 1]t) (10)
 1
MSD (t) − t
= e T2 dx exp(−iδω[3x 2 − 1]t) (11)
M0 0



iδωt− Tt 1+i π 3
=e 2 erfi [1 − i] δωt , (12)
2 6δωt 2

where erfi(x) = −i erf(ix) represents the imaginary error


function and M0 = 4π m0 is the initial magnetization of the
free induction decay. The frequency distribution p(ω) can
be found as the Fourier transform of the magnetization M(t)
according to Eq. (4):

√1 ω −δω < ω < +2δω
pSD (ω) = 2δω 3[ δω +1] (13)
0 elsewhere.
Naturally, the alveolar surface model is only valid for large
volume fraction η → 1, since the model reduces the tissue
component of alveoli to a thin surface (see Fig. 4). Similar to
the simulations from synchrotron-based μCT images and the
Wigner-Seitz model, the frequency distribution of the alveolar
surface model exhibits a prominent asymmetry (see Fig. 5).
All obtained frequency distributions show a negative peak
position. However, the shape of the curves differ: while the
alveolar surface model exhibits discontinuities at ω = −δω
and ω = +2δω, simulations based on the Wigner-Seitz
model and on μCT images show a continues distribution.
Still, the width of all frequency distributions is similar. The
alveolar surface model therefore has several advantages
and limitations, i.e., it is capable to describe the observed
FIG. 4. Geometrical model of lung tissue. In human lung tissue asymmetry of the frequency distribution in lung tissue and the
(a), alveoli are randomly distributed and the fraction of air volume to frequency distribution width while it requires large air volume
total volume is very large. (b) Synchrotron-based μCT image without fractions to be accurate. Furthermore, as we will show in the
filtering of healthy lung tissue in a mouse with an isotropic resolution next section, the model allows an analytical treatment of the
of 325 nm and a field of view of 325 μm × 325 μm × 162.5 μm. In diffusion process and thus, is a useful tool to understand the
the well-established Wigner-Seitz model, water molecules are located relaxation mechanisms in lung tissue.
at the interfaces of Wigner-Seitz cells [32,33]. Such a cell can be
approximated by a sphere (c). In the alveolar surface model (see main
IV. DIFFUSION EFFECTS
text), local lung tissue is approximated as an ensemble of spherical
surfaces with average radius R. Due to the susceptibility difference In this section, we provide a closed-form solution of the
χ = χair − χtissue , the free induction decay in human lung tissue Bloch-Torrey equation (5) for the geometrical assumptions
shows a large deviation from a monoexponential decay. of the alveolar surface model, together with appropriate

022415-4
DEPHASING AND DIFFUSION ON THE ALVEOLAR SURFACE PHYSICAL REVIEW E 95, 022415 (2017)

With the substitution x = cos(θ ) one finds:


  
1 ∂ ∂2 ∂
τ + m(x,t) = [1 − x 2 ] 2 − 2x (18)
T2 ∂t ∂x ∂x

− iτ δω[3x − 1] m(x,t).
2

Equation (18) is then solved by the eigenfunction expansion


− Tt −2iδωt
 4k + 1 −λ2k,0 (√3iτ δω) t
m(θ,t) = m0 e 2 e ck e τ

k=0
2

× PS2k,0 ( 3iτ δω, cos(θ )), (19)

where the spheroidal eigenvalues λ2k,0 ( 3iτ √δω) and angu-
lar prolate spheroidal eigenfunctions PS2k,0 ( 3iτ δω, cos(θ ))
FIG. 5. Frequency distributions p(ω) as obtained from the alveo- fulfill the spheroidal eigenvalue equation
lar surface model [blue solid line obtained from Eq. (13)], numerical  
∂2 ∂ √
simulations of the Wigner-Seitz geometry (black dotted line) and [1 − x 2 ] 2 − 2x + λ2k,0 ( 3iτ δω)+3iτ δω[1−x 2 ]
from numerical simulations of synchrotron-based μCT images for ∂x ∂x
η = 90–95 % (red dashed line, see Fig. 3). All curves show a clear √
asymmetry of the line shape, the line shape always peaks at a negative
× PS2k,0 ( 3iτ δω,x) = 0 (20)
Larmor frequency. However, there is a clear quantitative difference and are denoted in the notation of Meixner and Schäfke [40].
between the curves. By setting the initial value m(θ,0) = m0 and using the

approximations. This allows us to get new insights in the


interplay between susceptibility and diffusion effects.

A. Exact solution
The local transverse magnetization m(θ,t) = mx (θ,t) +
imy (θ,t) is governed by the Bloch-Torrey equation that treats
the effects of magnetic field inhomogeneities as well as
diffusion effects on the local magnetization:
 
∂ D 1
m(θ,t) =  θ − iω(θ ) − m(θ,t), (14)
∂t R2 T2
where ω(θ ) is the local Larmor frequency at the surface of the
alveoli, see Eq. (9). The polar part of the Laplace operator is
denoted as

1 ∂ ∂
θ = sin(θ ) (15)
sin(θ ) ∂θ ∂θ
and the local magnetization takes the initial values m(θ,t =
0) = m0 = const. The diffusion effects are then characterized
by the diffusion time τ :
τ = R 2 /D. (16)
For typical values of the alveolar size of R ≈ 100 μm [39] and
diffusion coefficient D ≈ 1 μm2 ms−1 , the diffusion time τ is
within the order of several seconds. Since τ is proportional to
the second power of the alveolar radius, it is very sensitive to
changes in alveolar microstructure.
To solve the Bloch-Torrey equation for three-dimensional
dipole fields, we simplify Eq. (14) to
    FIG.
√ 6. (a) Real and (b) imaginary part of the eigenvalues
∂ D 1 ∂ ∂2 λ2k,0 ( 3iτ δω). In the motional narrowing limit τ δω → 0, the
m(θ,t) = cos(θ ) + sin(θ ) 2
∂t R 2 sin(θ ) ∂θ ∂θ eigenvalues become purely real: λ2k,0 (0) = 2k[2k + 1]. The dotted
 lines show the approximations given in Eq. (B7) and Eq. (B8).
1
− iδω[3 cos2 (θ ) − 1] − m(θ,t). (17) Contrary to the case of a two-dimensional dipole field, the eigenvalues
T2 exhibit no symmetries and brunch cuts [45].

022415-5
L. R. BUSCHLE et al. PHYSICAL REVIEW E 95, 022415 (2017)

orthogonality of eigenfunctions and the properties of the band


limited Fourier transform [41], the expansion coefficients ck
can be found as:
 √
ck = 2[4k + 1] [−1]k PS2k,0 ( 3iτ δω,0) (21)

× S(1)
2k,0 ( 3iτ δω,1),


where S(1)
2k,0 ( 3iτ δω,1) denotes the radial prolate spheroidal
function. The spheroidal eigenvalues of order zero are de-
scribed in detail in Ref. [41]. Since the parameter 3iτ δω
is purely imaginary, the numerical evaluation of spheroidal
eigenvalues and eigenvectors should be done with caution [42–
44] (see √ also Appendix C). The odd spheroidal functions
PS2k+1,0 ( 3iτ δω,x) are √antisymmetric in x and, thus, do not
contribute. The prefactor [4k + 1]/2 in Eq. (19) follows from
the Meixner-Schäfke normalization that√is usually used for
spheroidal functions. Eigenvalues λ2k,0 ( 3iτ δω) are shown
in Fig. 6 for arbitrary values of τ δω and discussed in detail
in Appendix B. The expansion coefficient ck are shown in
Fig. 7.
The local magnetization m(θ,t) in dependence on polar
angle θ and time t is shown in Fig. 8. Since the total measured
signal M(t) is a superposition of the local magnetizations, the

FIG. 8. Local magnetization mx (θ,t) = Re(m(θ,t)) and


my (θ,t) = Im(m(θ,t)) in dependence on the normalized time
t/τ and polar angle θ for τ δω = 10. With increasing time, the
x component of the local magnetization decreases while the y
component increases, especially for θ ≈ π/2. Both components
remain symmetric around θ = π/2 for all times.

free induction decay is given as


 π
M(t) = 2π sin(θ )m(θ,t)dθ
0
 +1  1
FIG. 7. (a) Real and (b) imaginary part of the expansion coeffi-
cients ck according to Eq. (21) in dependence on τ δω. With increasing = 2π m(x,t)dx = 4π m(x,t)dx, (22)
−1 0
τ δω, an increasing number of coefficients contribute (see also Table II
in Appendix B).√ In the limit τ δω → 0, the coefficients become: where the symmetry relation m(π/2 − θ,t) = m(π/2 + θ,t)
limτ δω→0 ck = 2δk,0 . shown in Fig. 8 is used. With the local magnetization from

022415-6
DEPHASING AND DIFFUSION ON THE ALVEOLAR SURFACE PHYSICAL REVIEW E 95, 022415 (2017)

Eq. (19), the following expression for the free induction decay B. Gaussian approximation
can be found: As mentioned above, the exact solution of the Bloch-Torrey

 √ equation (14) contains an infinite series of spheroidal eigen-
M(t) 1 − Tt
= e 2 e−2iδωt ck2 e−λ2k,0 ( 3iτ δω) τt
. (23) functions [see Eq. (19)], which makes numerical calculations
M0 2 k=0 cumbersome. It can therefore be useful to use analytical
approximations for the local magnetization that allow a less
In the static-dephasing limit D → 0, this expression coin- rigorous mathematical treatment with similar results. One
cides with the expression given in Eq. (12). In the opposite simple approach is the Gaussian approximation. The Gaussian
limit D → ∞ (motional narrowing limit), the differential approximation considers the probability that a particle moves
equation (18) simplifies to a Legendre differential equation. from θ0 to θ during the time interval t. Using a cumulant
Thus, the eigenvalues are given as λ2k,0 (0) = 2k[2k + 1] and expansion up to the second order, an analytical expression for
only the first addend with k = 0 contributes to the infinite sum the local magnetization can be found as [48]
in Eqs. (19) and (23) as shown in Fig. 7. Therefore, the free
induction decay is dominated by the T2 relaxation mechanism: − Tt 1 2
−β(θ,t)
m(θ,t) = m0 e 2 eiα(θ,t)+ 2 [α(θ,t)] , (24)
− t
M(t) = M0 e T2 .
For the numerical evaluation it is helpful to utilize Eq. (23) where exponents α and β are associated with the transition
at initial time t = 0 to estimate the number of contributing probabilities of first and second order. As detailed shown in
coefficients as shown in Appendix B. With increasing values Appendix D, they can be found as
of the parameter τ δω, the number of contributing addends α(θ,t) 1  
= P2 ( cos(θ )) e−6 τ − 1
t
in Eq. (23) increases. Thus, for small values of τ δω  3, (25)
τ δω 3
the free induction decay is well approximated by a mono-  
β(θ,t) 1 t
exponential decay with a relaxation√rate that depends on 1 − e−6 τ − 6
t
=− (26)
the first spheroidal eigenvalue λ0,0 ( 3iτ δω) ≈ −2iτ δω + [τ δω] 2 45 τ
2
[τ δω]2 as given in Eq. (B7) in Appendix B, resulting in    
15 √ 2 t
+ P2 ( cos(θ )) 1 − 6 + 1 e−6 τ
t

M(t) = M0 exp(−t/T2 − 2iδωt − λ0,0 ( 3iτ δω)t/τ ) = 63 τ


− t
M0 e T2 e− 15 τ δω t , which also agrees with Eq. (18) in Ref. [46]
2 2

3  t 
P4 ( cos(θ )) 7 + 3e−20 τ − 10e−6 τ .
t
and Eq. (12) in Ref. [47]. For τ δω  11, the free induction +
decay can be described by a nearly biexponential decay with 1225
complex decay rates. Naturally, with increasing τ δω, the shape Here, Pi denote the ith Legendre polynomials with P2 (x) =
of the free induction decay becomes more complicated. [3x 2 − 1]/2 and P4 (x) = [35x 4 − 30x 2 + 3]/8. As shown in

FIG. 9. Transverse components of the local magnetization m(θ,t) for large diffusion coefficients [(a) and (b)], and small diffusion coefficients
(large value of τ δω) [(c) and (d)] in the exact solution [black lines, obtained from Eq. (19)] and in the Gaussian approximation [red lines,
obtained from Eq. (24)]. The solid lines in (a) and (b) show the local magnetization at t/τ = 0.1, the dashed lines at t/τ = 0.2. The blue lines
in (c) and (d) show the local magnetization in the static-dephasing limit obtained from Eq. (10).

022415-7
L. R. BUSCHLE et al. PHYSICAL REVIEW E 95, 022415 (2017)

TABLE I. Estimation of the susceptibility difference χ and the alveolar radius R. The free induction decays measured in two healthy
volunteers are fitted with Eq. (23). According to Eq. (16), the alveolar radius was calculated with a typical value of D = 1μm2 ms−1 [53].

voxel 1 2 3 4 5 6 7 8 mean ± standard deviation


χ [ppm] 8.88 8.03 8.33 6.50 7.78 8.40 8.23 7.30 7.93 ± 0.75
R [μm] 95 140 147 78 116 192 79 138 123 ± 39

Fig. 9, the local magnetization as predicted by the Gaussian where the complex valued parameter v is defined as the zero
approximation agrees very well with the exact solutions for of the equation
short times as well as for small and large diffusion effects. √ √
In the static-dephasing limit (D → 0), the local magneti- iτ δω v + 2 arccoth( v) = 0. (28)
zation in the Gaussian approximation agrees with Eq. (12), as A comparison of the strong collision approximation with
shown in Eqs. (D5) and (D6) in Appendix D. the exact free induction decay is shown in Fig. 11 in the
Appendix.

C. Strong collision approximation V. EXPERIMENT


In the strong collision approximation the diffusion operator
The theoretical results allows determining the local mean
D is replaced by a simpler Markov operator. Thus, an
alveolar radius in human lung tissue. Thus, the free induction
analytical expression for the free induction decay can be
decay in human lung tissue of two healthy volunteers was mea-
obtained that is valid for large diffusion effects as well as
sured with a point resolved spectroscopy (PRESS) sequence in
small diffusion effects [49–51]. Details of the derivation are
15 mm × 15 mm × 15 mm large voxels in expiration [52]. The
provided in Appendix E:
measurements were performed at a Magnetom Aera scanner
 (Siemens Healthcare, Erlangen, Germany) with an external
MSC (t) − t [τ δω]2 v[1 − v] −6 t −i[3v−1]δωt field strength of 1.5 T, a repetition time of 1.5 s, an echo
= e T2 e τ (27)
M0 2 + iτ δω[1 − v] time of 30 ms and twenty averages (NA = 20) during a
 1  single breathhold. The voxels are put in the peripheral tissue,
dz z2 e−i[3z −1]δωt e−6 τ
2 t

+  2  2 , trying to minimize macroscopic susceptibility artifacts. One


0 z − τ2iδω arctanh(z) − τ πδω can estimate the alveolar radius R with the expression for

FIG. 10. Comparison of measured free induction decay with theoretical fits from Eq. (23) for eight voxels, see also Table I.

022415-8
DEPHASING AND DIFFUSION ON THE ALVEOLAR SURFACE PHYSICAL REVIEW E 95, 022415 (2017)

the total magnetization in Eq. (23). Fitting the theoretical functions and eigenvalues. The local magnetization exhibits
result to the measured free induction decay leads to values the same symmetry relations as the local Larmor frequency
for δω and τ , and, thus, to the susceptibility difference and agrees very well with that of the Gaussian approximation.
χ and, with Eq. (16), alveolar radius R, respectively. The Furthermore, the exact free induction decay is very close
results are shown in Table I and the corresponding fits in to the free induction decay obtained by the strong collision
Fig. 10. approximation.
The obtained parameters for susceptibility differences χ The comparison of the theoretical results with in vivo
and alveolar radii R agree very well with expectations of χ = measurements gives an approximation of the alveolar radius
8–10 ppm [18,54] and R = 100 μm [39]. R. As shown in Table I the estimated mean alveolar radii are
very close to the expected value.
Since the alveolar surface model shows deviation from
VI. SUMMARY AND CONCLUSION
simulations on synchrotron-based μCT images, systematic
In this work, we analyze the influence of both susceptibility uncertainties in the estimation of the alveolar radius may
and diffusion effects on the dephasing process in lung occur. Deviations of the model-estimated alveolar radius
tissue. Numerical simulations based on synchrotron-based from alveolar radii as determined from histological lung
μCT images and the Wigner-Seitz model were performed in tissue slices could be analyzed within an animal model.
the regime of static dephasing. The simulations predict an However, such a procedure would go beyond the scope of this
increasing width of the frequency distribution with increasing work.
air volume fractions η, and an asymmetric line shape. The The difference of estimated radii in different voxels can be
proposed alveolar surface model assumes spherically shaped explained by a low signal to noise ratio and macroscopic field
alveoli with small soft tissue compartments that are located inhomogeneities, which cause further dephasing processes.
at the surface of the sphere. This simplified model allows Thus, it will be interesting to add a linear term to the local
an analytical determination of the line shape. Moreover, the Larmor frequency, that occurs for macroscopic field inhomo-
model provides a closed-form solution of the Bloch-Torrey geneities. The free induction decay is influenced by diffusion
equation to allow an analytical treatment of the diffusion effects, especially at long times according to Eqs. (19), (12),
process. and (24). Therefore, the low signal to noise ratio at long times
The assumption that spin-bearing particles are localized causes an increase in the deviation of measured alveolar radii.
on the surface of spherical alveoli may seem restricting; However, the influence of both effects could be limited by
however, this assumption represents a logical simplification considering additional field inhomogeneities and improved
of the Wigner-Seitz foam model applied to lung tissue with sequence designs [55]. For typical parameters of human lung
high relative air volume and foamlike tissue boundaries [35]. tissue at 1.5 T, the susceptibility effects may predominate the
In addition, microscopic measurements of alveolar structure diffusion effects. Thus, the sensitivity of the free induction
support this geometrical model [16]. decay towards lung tissue microstructure can be increased by
The line shape in the alveolar surface model shows a using lower magnetic field strength. This will also decrease
pronounced asymmetry and discontinuities at ω = −δω and macroscopic susceptibility artifacts. Furthermore, the suscep-
ω = +2δω. Apart from these deviations from numerical tibility χ between air and lung tissue can be determined as
simulations, it qualitatively is able to explain the formation shown in Table I and, therefore, it is in principle possible to
of the line shape in lung tissue: due to the asymmetry of estimate the oxygen saturation. The presented methods are a
the three-dimensional dipole field, whose values are limited complementary ansatz to imaging with hyperpolarized gas, the
to the interval −δω  ω  +2δω, the frequency distribution latter producing correct results but being difficult to embed in
in lung tissue is mainly limited to the same interval (see clinical routine. Finally, this noninvasive method is a promising
Fig. 2). Deviations from this restriction only occur due to the routine for the quantification of alveolar susceptibility and
superposition of different dipole fields from different alveoli. alveolar size.
The shifted peak position of the frequency distribution towards
negative frequencies can also be qualitatively understood in
the alveolar surface model: for an arbitrary position of the
ACKNOWLEDGMENTS
alveolar surface, the probability distribution to find the angle
θ is proportional to sin(θ ). Specifically, it is more likely to We want to thank Dr. Pablo Villanueva, Paul Scherrer
find angles close to θ = π/2 than θ = 0 or θ = π . Thus, Institute, for expert technical support at the TOMCAT beam-
it is more likely that the local Larmor frequency for an line. Kerstin Bahr and Dr. Maximilian Ackermann, Johannes
arbitrary position on the surface of the sphere is negative Gutenberg University Mainz, for osmation and freeze drying
since ω(θ ) ∝ 3 cos2 (θ ) − 1 and, consequently, the peak of of synchrotron specimens. This work was supported by grants
the frequency distribution is shifted towards negative Larmor from the Deutsche Forschungsgemeinschaft (Contract Grants
frequencies. In Sec. IV, diffusion as well as susceptibil- No. DFG ZI 1295/2-1 and No. DFG KU 3555/1-1). Further,
ity effects are analyzed in the alveolar surface model, by this work was supported in part by grants from the German
solving the Bloch-Torrey equation for this specific model Federal Ministry of Education and Research to the DZL
geometry. (82DZL00401 and 82DZL004A1). F.T.K. was supported by a
The provided closed-form solution of the Bloch-Torrey postdoctoral fellowship from the medical faculty of Heidelberg
equation allows an analytical expression of the local mag- University and the Hoffmann-Klaus foundation of Heidelberg
netization and the free induction decay in terms of spheroidal University.

022415-9
L. R. BUSCHLE et al. PHYSICAL REVIEW E 95, 022415 (2017)

APPENDIX A: MODEL LIMITATIONS TABLE II. The maximum value of τ δω, for which the free
induction decay can be numerically calculated, in dependence of the
When obtaining the expectation value of R 2 , denoted number of considered summands N . The values are obtained with the
as R 2 , by  fitting the diffusion time τ = R 2 /D, the  −3
condition |2 − N k=0 k | < 10
c 2
given in Eq. (B5).
difference of R 2 and the actual expectation value of the
alveolar radius, R , is not equal to zero when the standard N 1 2 3 4 5 10
deviation σ of radius values around R is different from zero.
In fact, assuming a normal distribution of radii inside the voxel, τ δω 3.01 11.16 25.11 44.83 70.35 282.47
it is easy to show that [36]
⎡  ⎤
for small values of the parameter τ δω the free induction
  σ 2
δR = R 2 − R = R 2 ⎣1 − 1 − 2 ⎦. decay is approximately monoexponential and exhibits more
R oscillating components with decreasing diffusion effects (see
Fig. 11).
(A1) For small values
 √ of the parameter τ δω the spheroidal
For a coefficient of variation σ/ R < 0.15, the term
2 eigenvalues λ2k,0 ( 3iτ δω) can be approximated as (see
δR is negligible, but it continually grows to reach a Eq. (10) in Chapter 3.2 in Ref. [40] or Eq. (3.44) in Chapter
 3.1 in [41]):
value of δR/ R 2 ≈ 0.13 for coefficient of variation of
0.5, implyinga 13% correction of the originally deter- √
λ2k,0 ( 3iτ δω) ≈ 2k[2k + 1] (B7)
mined radius R 2 . However, the coefficients of variation
 
for healthy human subjects were found to be small ≈ 3 1
0.1 [39]. Thus, this effect can be safely neglected in this + iτ δω −1
2 [4k − 1][4k + 3]
analysis.

9 [τ δω]2 [2k + 1]2 [2k + 2]2
+
APPENDIX B: SPHEROIDAL EIGENFUNCTIONS AND 2 4k + 1 [4k + 3]3 [4k + 5]
EIGENVALUES 
[2k − 1]2 4k 2
The Bloch-Torrey equation (18) is solved by an eigen- − .
function expansion with [4k − 3][4k − 1]3
√ spheroidal eigenfunctions. These
eigenfunctions PS2k,0 ( 3iτ δω, cos(θ )) exhibit the symmetry √
For large values τ δω the eigenvalues λ2k,0 ( 3iτ δω) can be
relations [56]
approximated as [56,57]
√ π 
PS2k,0 3iτ δω, cos −θ √
2 λ2k,0 ( 3iτ δω)
√ π  √
= PS2k,0 3iτ δω, cos +θ (B1) −3iτ δω + [2k+1] 3iτ δω − 14 [2k 2 +2k + 3] k even
2 ≈ √
−2ik 3iτ δω − 12 [k 2 + 1] k odd.
∂ √  (B8)
PS2k,0 3iτ δω, cos(θ ) 
∂θ θ=0
∂ √ 
=0= PS2k,0 3iτ δω, cos(θ )  . (B2)
∂θ θ=π

For the numerical evaluation it is useful to utilize the free


induction decay at the initial time point t = 0:

M(t = 0) = M0 (B3)
M0
M
(t = 0) = − . (B4)
T2
Inserting the obtained analytical result for the free induction
decay given in Eq. (23), the following relations hold:

 
N
2= ck2 ≈ ck2 (B5)
k=0 k=0
∞
√ FIG. 11. Absolute value of the free induction decay M(t) given
−4iτ δω = ck2 λ2k,0 ( 3iτ δω). (B6)
in Eq. (23) compared with the strong collision approximation given
k=0
in Eq. (27). With increasing value of the parameter τ δω, the decay
Therefore, the number N of contributing addends in the infinite becomes faster and shows a larger deviation from a monoexponential
sum in Eq. (23) can be estimated as shown in Table II. Thus, decay.

022415-10
DEPHASING AND DIFFUSION ON THE ALVEOLAR SURFACE PHYSICAL REVIEW E 95, 022415 (2017)

In the computer algebra program Mathematica (Wolfram APPENDIX C: DISCRETIZATION OF THE


Research, Inc., Champaign, IL, USA [58]) the spheroidal BLOCH-TORREY EQUATION
eigenvalues can be obtained by the command
√ √ To examine the correctness of the numerical evaluation of
λ2k,0 ( 3iτ δω) = SpheroidalEigenvalue[2k,0, 3iτ δω] the spheroidal eigenvalues, the Bloch-Torrey-equation can be
solved in a discretized version [59]. Therefore, the substi-
(B9)
tution x = cos(θ ) is used and the following grid points are
and the spheroidal functions are given by the commands defined:
√ √ 2
S(1)
2k,0 ( 3iτ δω,x) = SpheroidalS1[2k,0, 3iτ δω,x] (B10) xi = −1+ [i − 1]h for i = 1, . . . ,n and h= .
√ n−1

PS2k,0 ( 3iτ δω,x) = SpheroidalPS[2k,0, 3iτ δω,x]. (C1)
(B11)
The polar part θ of the three-dimensional Laplace operator
given in Eq. (15) can be represented as a n × n matrix:

⎛  2    ⎞
2 x3 − 1 2 1 − x 23
⎜ 2 2

⎜1 − x 2 x 21 + x 23 − 2 1 − x 23 ⎟
⎜ ⎟
⎜ ⎟
1
2 2 2 2
⎜ ⎟
⎜ .. .. .. ⎟
⎜ . . . ⎟
⎜ ⎟
ˆx = ⎜
1 1 − xi− 1 + xi+ 1 − 2 1 − xi+ ⎟
2 2 2 2
 ⎜ 1 xi− 1 ⎟.
h ⎜
2 2 2 2 2

⎜ .. .. .. ⎟
⎜ . . . ⎟
⎜ ⎟
⎜ ⎟
⎜ 1 − xn− ⎟
⎜ 3 + xn− 1 − 2 1 − xn− ⎟
2 2 2 2
3 xn− 1
⎝ 2
 2 2 2  2 2
 ⎠
2 1 − xn− 1 2 xn− 1 − 1
2 2
(C2)
Introducing the explicit expressions for the grid points given in Eq. (C1), the discretized Laplace operator is given as:
⎛1 − 2 2
−1

2 h h 2
⎜12    ⎟
⎜2 − − − ⎟
1 5 4 3 2 3

⎜ ⎟
h 2 2 h 2 h 2

⎜ ⎟
⎜ .. .. .. ⎟
⎜ .

.
 
.
   ⎟
x = ⎜
ˆ
⎜ i − 32 h2 − i + 32 4 1− h
i
+ 2i 2 − 4i + 5
2
i − 12 h2 − i + 12 ⎟.

⎜ .. .. .. ⎟
⎜ . . . ⎟
⎜ ⎟
⎜      ⎟
⎝ n − 52 h2 − n + 52 2[n − 1]2 − 4 n −h 2 − 4[n − 1] + 5
2
n − 32 h2 − n + 32 ⎠
2
h
− 1
2
1
2
− 2
h
(C3)

Due to particle number conservation, the sum of each row value equation shows a good agreement of the numerical
vanishes (see also Eq. (45) in Ref. [59]). The discretized form eigenvalues with the eigenvalues given in Eq. (B9).
of the Bloch-Torrey equation (18) can be written in the form
of the eigenvalue equation: APPENDIX D: GAUSSIAN APPROXIMATION

[iτ δω[3[ − 1 + [i − 1]h]2 − 1]1̂ −  ˆ 2k,0 ( 3iτ δω)
ˆ x ]PS In the Gaussian approximation an analytical expression for
√ √ both components of the local magnetization m(θ,t) can be
= λ2k,0 ( 3iτ δω)PS
ˆ 2k,0 ( 3iτ δω), (C4) found. The probability p(θ0 ,φ0 ,θ,φ,t) that a particle moves
from (θ0 ,φ0 ) to (θ,φ) during time t is given as [60–63]
where 1 represents the n-dimensional unity matrix and the
∞ 
 +l
eigenvector
e−l[l+1] τ Ylm (θ,φ)Y∗lm (θ0 ,φ0 ),
t
p(θ0 ,φ0 ,θ,φ,t) =

ˆ 2k,0 ( 3iτ δω)
PS l=0 m=−l
√ √ (D1)
= (PS2k,0 ( 3iτ δω,x1 ), . . . ,PS2k,0 ( 3iτ δω,xn ))T (C5) where

contains [2l + 1][l − m]! m
√ the values of the discretized eigenfunction Ylm (θ,φ) = Pl ( cos(θ ))eimφ (D2)
PS2k,0 ( 3iτ δω,xi ). The numerical examination of this eigen- 4π [l + m]!

022415-11
L. R. BUSCHLE et al. PHYSICAL REVIEW E 95, 022415 (2017)

denotes the spherical harmonics with indices l and m. The


azimuthal angles are denoted as φ and φ0 . Rewriting the
three-dimensional magnetic dipole field in terms of spheroidal
harmonics leads to ω(θ ) = 4δω π5 Y2,0 (θ,φ). Thus, the phase
exponent α and attenuation exponent β can be obtained with
the transition probability p(θ,φ,θ0 ,φ0 ,t) given in Eq. (D1) and
the local Larmor field given in Eq. (9):
 t  2π  π
α(θ,φ,t) = − dξ dφ0 dθ0 sin(θ0 )ω(θ0 ,φ0 )
0 0 0
× p(θ,φ,θ0 ,φ0 ,ξ ) (D3)
 t  t−η  2π  π
β(θ,φ,t) = + dη dξ dφ0 dθ0 sin(θ0 )
0 0 0 0
 2π  π FIG. 12. Dependency of the zero v on the parameter τ δω
× dφ1 dθ1 sin(θ1 )ω(θ0 ,φ0 ) according to Eq. (28). For small values of τ δω 1 the real part
0 0 of v takes the constant value Re(v) = 1/3 and the imaginary part
× p(θ0 ,φ0 ,θ1 ,φ1 ,η)ω(θ1 ,φ1 )p(θ1 ,φ1 ,θ,φ,ξ ). diverges like Im(v) = 2/[τ δω] [see Eq. (E7)]. In the opposite limit
τ δω 1, the real part of v decreases like Re(v) = π 2 /[τ δω]2 and
(D4) the imaginary part decreases like Im(v) = 4π 2 /[τ δω]3 as shown in
The Larmor frequency is independent on azimuthal angle Eq. (E6).
ω(θ,φ) = ω(θ ). Inserting the dipole field ω(θ ) and the prob-
ability function p(θ,φ,θ0 ,φ0 ,t) from Eq. (D1) leads to the
connected to the static-dephasing free induction decay
exponents α and β as expressed in Eq. (25) and Eq. (26). Due
[65]:
to the symmetry of the dipole field, also the exponents α and β
! −1
"
do not depend on the azimuthal angle φ. In the static-dephasing M̂SC (s) M̂SD s + τSC
limit D → 0, the exponents α and β become = −1
! −1
". (E1)
M0 M0 − τSC M̂SD s + τSC
αSD (θ,t) = lim α(θ,t) = −δωt[3 cos2 (θ ) − 1] (D5) The correlation time τSC is defined as
τ →∞

βSD (θ,t) = lim β(θ,t) = 12 [αSD (θ,t)]2 K(t) − t


= e τSC = e−6 τ ,
t
(D6) (E2)
τ →∞
K0
and, thus, the local magnetization obtained from the Gaus-
sian approximation coincides with the static-dephasing local where the correlation function K(t) is given in Table 2 in
magnetization given in Eq. (12). The Gaussian approximation Ref. [66] and in Eq. (46) in Ref. [67]. The Laplace transform
is a result of a cumulant expansion method [64] up to the of the static-dephasing
#∞ free induction decay is denoted as
second order in δωt as shown in [48]. Thus, the Gaussian M̂SD (s) = 0 MSD (t)e−st dt. Using the expression for the
approximation is valid in all diffusion regimes for small times. static-dephasing free induction decay given in Eq. (12) it
This agrees with numerical results as presented in Fig. 9. reads [51]:

M̂SD (s) i 3δω
APPENDIX E: STRONG COLLISION APPROXIMATION =
M0 3δω i[s + 1/T2 ] + δω
In the strong collision approximation the diffusion operator  
D is replaced by a simpler Markov operator. In a Markov pro- 3δω
× arctanh . (E3)
cess, the diffusion transition rate between different positions i[s + 1/T2 ] + δω
is only dependent on the equilibrium probability of the final
state [49,50]. Therefore, the diffusion process is approximated According to Eq. (E1) the Laplace transform of the free
by a much simpler process and the free induction decay can be induction decay in the strong collision approximation can be
found as for all diffusion regimes as

 ∞
M̂SC (s) MSC (t) −st
= e dt (E4)
M0 0 M0
 
1 + τ6 s + T12 − 6i τ δω τ
= $     !$    " − 6 . (E5)
6
τ
+s+ 1
T2
− iδω − τ6 13 1 + δω
i
s + T12 + τ6iδω arccoth 13 1 + i
δω
s+ 1
T2
+ 6i
τ δω

The expression for the free induction decay MSC (t) given in Eq. (27) is obtained by an inverse Laplace transform of M̂SC (s) given
in Eq. (E5) in the complex plane. The dependency of the complex zero v defined in Eq. (28) on τ δω is visualized in Fig. 12. For

022415-12
DEPHASING AND DIFFUSION ON THE ALVEOLAR SURFACE PHYSICAL REVIEW E 95, 022415 (2017)

large values of τ δω a Taylor expansion of Eq. (28) around v ≈ 0 lead to an approximation of the zero in the complex plane:
π2 4π 2
v≈ 2
+i . (E6)
[τ δω] [τ δω]3
In the opposite limit τ δω → 0 the position of the zero v in the complex plane can be estimated as:
1 2
+i v≈ . (E7)
3 τ δω
As shown in Fig. 11 the exact free induction decay given in Eq. (23) agrees very well with the free induction decay obtained in
the strong collision approximation given in Eq. (27).

[1] R. A. Pauwels and K. F. Rabe, Burden and clinical features [12] M. S. Conradi, B. T. Saam, D. A. Yablonskiy, and J. C.
of chronic obstructive pulmonary disease (COPD), Lancet 364, Woods, Hyperpolarized 3He and perfluorocarbon gas diffusion
613 (2004). MRI of lungs, Prog. Nucl. Magn. Reson. Spectrosc. 48, 63–83
[2] C. C. W. Hsia, D. M. Hyde, M. Ochs, and E. R. Weibel, (2006).
An official research policy statement of the american thoracic [13] J. Parra-Robles and J. M. Wild, The influence of lung airways
society/european respiratory society: Standards for quantitative branching structure and diffusion time on measurements and
assessment of lung structure, Am. J. Respir. Crit. Care Med. models of short-range 3He gas MR diffusion, J. Magn. Reson.
181, 394 (2010). 225, 102 (2012).
[3] K. Murata, H. Itoh, G. Todo, M. Kanaoka, S. Noma, T. Itoh, M. [14] C. Wang, G. W. Miller, T. A. Altes, E. E. de Lange, G. D.
Furuta, H. Asamoto, and K. Torizuka, Centrilobular lesions of Cates Jr, and J. P. Mugler, Time dependence of 3He diffusion
the lung: demonstration by high-resolution CT and pathologic in the human lung: Measurement in the long-time regime using
correlation, Radiology 161, 641 (1986). stimulated echoes, Magn. Reson. Med. 56, 296 (2006).
[4] R. Yuan, T. Nagao, P. D. Pare, J. C. Hogg, D. D. Sin, M. [15] H. E. Möller, X. J. Chen, M. S. Chawla, B. Driehuys, L. W.
W. Elliott, L. Loy, L. Xing, S. E. Kalloger, J. C. English, Hedlund, and G. A. Johnson, Signal dynamics in magnetic
J. R. Mayo, and H. O. Coxson, Quantification of lung sur- resonance imaging of the lung with hyperpolarized noble gases,
face area using computed tomography, Respir. Res. 11, 153 J. Magn. Reson. 135, 133 (1998).
(2010). [16] E. Namati, J. Thiesse, J. de Ryk, and G. McLennan, Alveolar
[5] E. J. R. van Beek, J. M. Wild, H.-U. Kauczor, W. Schreiber, J. P. dynamics during respiration: are the pores of Kohn a pathway
Mugler, and E. E. de Lange, Functional MRI of the lung using to recruitment?, Am. J. Respir. Cell Mol. Biol. 38, 572 (2008).
hyperpolarized 3-helium gas, J. Magn. Reson. Imaging 20, 540 [17] E. M. Scarpelli, The alveolar surface network: A new anatomy
(2004). and its physiological significance, Anat. Rec. 251, 491 (1998).
[6] D. A. Yablonskiy, A. L. Sukstanskii, J. C. Leawoods, D. S. [18] R. Mulkern, S. Haker, H. Mamata, E. Lee, D. Mitsouras, K.
Gierada, G. L. Bretthorst, S. S. Lefrak, J. D. Cooper, and M. S. Oshio, M. Balasubramanian, and H. Hatabu, Lung parenchymal
Conradi, Quantitative in vivo assessment of lung microstructure signal intensity in MRI: A technical review with educational
at the alveolar level with hyperpolarized 3He diffusion MRI, aspirations regarding reversible versus irreversible transverse
Proc. Natl. Acad. Sci. USA 99, 3111 (2002). relaxation effects in common pulse sequences, Concepts Magn.
[7] D. A. Yablonskiy, A. L. Sukstanskii, J. C. Woods, D. S. Gierada, Reson. A 43, 29 (2014).
J. D. Quirk, J. C. Hogg, J. D. Cooper, and M. S. Conradi, [19] H. C. Torrey, Bloch equations with diffusion terms, Phys. Rev.
Quantification of lung microstructure with hyperpolarized 3He 104, 563 (1956).
diffusion MRI, J. Appl. Physiol. 107, 1258 (2009). [20] T. M. de Swiet and P. N. Sen, Decay of nuclear magnetization
[8] R. Trampel, J. H. Jensen, R. F. Lee, I. Kamenetskiy, G. by bounded diffusion in a constant field gradient, J. Chem. Phys.
McGuinness, and G. Johnson, Diffusional kurtosis imaging in 100, 5597 (1994).
the lung using hyperpolarized 3He, Magn. Reson. Med. 56, 733 [21] M. D. Hürlimann, Effective gradients in porous media due to
(2006). susceptibility differences, J. Magn. Reson. 131, 232 (1998).
[9] S. Fichele, M. N. J. Paley, N. Woodhouse, P. D. Griffiths, E. J. R. [22] D. S. Grebenkov, NMR survey of the reflected Brownian motion,
van Beek, and J. M. Wild, Investigating 3He diffusion NMR in Rev. Mod. Phys. 79, 1077 (2007).
the lungs using finite difference simulations and in vivo PGSE [23] S. D. Stoller, W. Happer, and F. J. Dyson, Transverse spin
experiments, J. Magn. Reson. 167, 1 (2004). relaxation in inhomogeneous magnetic fields, Phys. Rev. A 44,
[10] D. S. Grebenkov, G. Guillot, and B. Sapoval, Restricted 7459 (1991).
diffusion in a model acinar labyrinth by NMR: Theoretical and [24] A. V. Barzykin, Theory of spin echo in restricted geometries
numerical results, J. Magn. Reson. 184, 143 (2007). under a step-wise gradient pulse sequence, J. Magn. Reson.
[11] M. Sarracanie, D. S. Grebenkov, J. Sandeau, S. Coulibaly, A. 139, 342 (1999).
R. Martin, K. Hill, J. M. Pérez Sánchez, R. Fodil, L. Martin, E. [25] J. D. Jackson, Classical electrodynamics (John Wiley and Sons
Durand, G. Caillibotte, D. Isabey, L. Darrasse, J. Bittoun, and Ltd., New York, 1999).
X. Maı̂tre, Phase-contrast helium-3 MRI of aerosol deposition [26] J. P. Marques and R. Bowtell, Application of a Fourier-based
in human airways, NMR Biomed. 28, 180 (2015). method for rapid calculation of field inhomogeneity due to

022415-13
L. R. BUSCHLE et al. PHYSICAL REVIEW E 95, 022415 (2017)

spatial variation of magnetic susceptibility, Concepts Magn. [44] J. A. Stratton, P. M. Morse, L. J. Chu, J. D. C. Little, and
Reson. B 25, 65 (2005). F. J. Corbató, Spheroidal Wave Functions (Chapman and Hall,
[27] F. T. Kurz, L. R. Buschle, T. Kampf, K. Zhang, H.-P. Schlemmer, London, 1956).
S. Heiland, M. Bendszus, and C. H. Ziener, Spin dephasing in [45] C. H. Ziener, M. Rückl, T. Kampf, W. R. Bauer, and H. P.
a magnetic dipole field around large capillaries: Approximative Schlemmer, Mathieu functions for purely imaginary parameters,
and exact results, J. Magn. Reson. 273, 83 (2016). J. Comput. Appl. Math. 236, 4513 (2012).
[28] C. H. Ziener, T. Kampf, G. Reents, H. P. Schlemmer, and W. R. [46] J. H. Jensen and R. Chandra, NMR relaxation in tissues with
Bauer, Spin dephasing in a magnetic dipole field, Phys. Rev. E weak magnetic inhomogeneities, Magn. Reson. Med. 44, 144
85, 051908 (2012). (2000).
[29] C. H. Ziener, F. T. Kurz, and T. Kampf, Free induction decay [47] F. T. Kurz, T. Kampf, S. Heiland, M. Bendszus, H.-P. Schlemmer,
caused by a dipole field, Phys. Rev. E 91, 032707 (2015). and C. H. Ziener, Theoretical model of the single spin-echo
[30] D. M. Vasilescu, L. Knudsen, M. Ochs, E. R. Weibel, and E. relaxation time for spherical magnetic perturbers, Magn. Reson.
A. Hoffman, Optimized murine lung preparation for detailed Med. 71, 1888 (2014).
structural evaluation via micro-computed tomography, J. Appl. [48] J. Stepišnik, A new view of the spin echo diffusive diffraction
Physiol. 112, 159 (2012). in porous structures, Europhys. Lett. 60, 453 (2002).
[31] M. Stampanoni, A. Groso, A. Isenegger, G. Mikuljan, Q. [49] W. R. Bauer, W. Nadler, M. Bock, L. R. Schad, C. Wacker,
Chen, A. Bertrand, S. Henein, R. Betemps, U. Frommherz, A. Hartlep, and G. Ertl, The relationship between the BOLD-
P. Böhler et al., Trends in synchrotron-based tomographic induced T2 and T2*: A theoretical approach for the vasculature
imaging: the SLS experience, In SPIE Optics+ Photonics, of myocardium, Magn. Reson. Med. 42, 1004 (1999).
p. 63180M (International Society for Optics and Photonics, [50] W. R. Bauer, W. Nadler, M. Bock, L. R. Schad, C. Wacker,
Bellingham, 2006). A. Hartlep, and G. Ertl, Theory of Coherent and Incoherent
[32] A. G. Cutillo, Application of Magnetic Resonance to the Study Nuclear Spin Dephasing in the Heart, Phys. Rev. Lett. 83, 4215
of Lung (Futura, Armonk, 1996). (1999).
[33] R. A. Christman, D. C. Ailion, T. A. Case, C. H. Durney, [51] L. R. Buschle, F. T. Kurz, T. Kampf, S. M. F. Triphan, H.-P.
A. G. Cutillo, S. Shioya, K. C. Goodrich, and A. H. Morris, Schlemmer, and C. H. Ziener, Diffusion-mediated dephasing in
Comparison of calculated and experimental NMR spectral the dipole field around a single spherical magnetic object, Magn.
broadening for lung tissue, Magn. Reson. Med. 35, 6 (1996). Reson. Imaging 33, 1126 (2015).
[34] K. C. Stone, R. R. Mercer, B. A. Freeman, L. Y. Chang, and J. D. [52] P. A. Bottomley, Spatial localization in NMR spectroscopy in
Crapo, Distribution of lung cell numbers and volumes between vivo, Ann. N.Y. Acad. Sci. 508, 333 (1987).
alveolar and nonalveolar tissue, Am. Rev. Respir. Dis. 146, 454 [53] M. Matoba, H. Tonami, T. Kondou, H. Yokota, K. Higashi,
(1992). H. Toga, and T. Sakuma, Lung carcinoma: Diffusion-weighted
[35] S. H. Baete, Y. De Deene, B. Masschaele, and W. De Neve, MR imaging–preliminary evaluation with apparent diffusion
Microstructural analysis of foam by use of NMR R2 dispersion, coefficient, Radiology 243, 570 (2007).
J. Magn. Reson. 193, 286 (2008). [54] E. D. Pracht, J. F. Arnold, T. Wang, and P. M. Jakob, Oxygen-
[36] F. T. Kurz, T. Kampf, L. R. Buschle, H.-P. Schlemmer, S. enhanced proton imaging of the human lung using T2, Magn.
Heiland, M. Bendszus, and C. H. Ziener, Microstructural Reson. Med. 53, 1193 (2005).
analysis of peripheral lung tissue through CPMG inter-echo [55] F. Carinci, C. Meyer, F. A. Breuer, and P. M. Jakob, In vivo
time R2 dispersion, PloS One 10, e0141894 (2015). imaging of the spectral line broadening of the human lung
[37] C. H. Ziener, T. Kampf, P. M. Jakob, and W. R. Bauer, Diffusion in a single breathhold, J. Magn. Reson. Imaging, 44, 745
effects on the CPMG relaxation rate in a dipolar field, J. Magn. (2016).
Reson. 202, 38 (2010). [56] C. Flammer, Spheroidal Wave Functions (Stanford University
[38] M. Kraiger and B. Schnizer, Potential and field of a homoge- Press, Stanford, 1957).
neous magnetic spheroid of arbitrary direction in a homogeneous [57] B. E. Barrowes, K. O’Neill, T. M. Grzegorczyk, and J. A. Kong,
magnetic field in cartesian coordinates, COMPEL 32, 936 On the asymptotic expansion of the spheroidal wave function
(2013). and its eigenvalues for complex size parameter, Stud. Appl.
[39] M. Ochs, J. R. Nyengaard, A. Jung, L. Knudsen, M. Voigt, T. Math. 113, 271 (2004).
Wahlers, J. Richter, and H. J. Gundersen, The number of alveoli [58] S. Wolfram, The Mathematica Book (Cambridge University
in the human lung, Am. J. Respir. Crit. Care Med. 169, 120 Press, New York, 1999).
(2004). [59] C. H. Ziener, S. Glutsch, P. M. Jakob, and W. R. Bauer,
[40] J. Meixner and F. W. Schäfke, Mathieusche Funktionen und Spin dephasing in the dipole field around capillaries and cells:
Sphäroidfunktionen (Springer-Verlag, Berlin, 1954). numerical solution, Phys. Rev. E 80, 046701 (2009).
[41] A. Osipov, V. Rokhlin, and H. Xiao, Prolate Spheroidal Wave [60] B. Halle, Theory of spin relaxation by diffusion on curved
Functions of Order Zero (Springer Science+Business Media, surfaces, J. Chem. Phys. 94, 3150 (1991).
New York, 2013). [61] K. Yosida, Brownian motion on the surface of the 3-sphere, Ann.
[42] J. Meixner, F. W. Schäfke, and G. Wolf, Mathieu Functions Math. Statist. 20, 292 (1949).
and Spheroidal Functions and Their Mathematical Foundations [62] D. R. Brillinger, A particle migrating randomly on a sphere, J.
(Springer-Verlag, Berlin, 1980). Theor. Prob. 10, 429 (1997).
[43] L.-W. Li, X.-K. Kang, and M.-S. Leong, Spheroidal Wave [63] V. Tulovsky and L. Papiez, Formula for the fundamental solution
Functions in Electromagnetic Theory (John Wiley and Sons, of the heat equation on the sphere, Appl. Math. Lett. 14, 881
New York, 2002). (2001).

022415-14
DEPHASING AND DIFFUSION ON THE ALVEOLAR SURFACE PHYSICAL REVIEW E 95, 022415 (2017)

[64] R. Kubo, Generalized cumulant expansion method, J. Phys. Soc. [66] C. H. Ziener, T. Kampf, and F. T. Kurz, Diffusion propagators for
Jpn. 17, 1100 (1962). hindered diffusion in open geometries, Concepts Magn. Reson.
[65] C. H. Ziener, T. Kampf, G. Melkus, V. Herold, T. Weber, A 44, 150 (2015).
G. Reents, P. M. Jakob, and W. R. Bauer, Local frequency [67] C. H. Ziener, T. Kampf, V. Herold, P. M. Jakob, W. R.
density of states around field inhomogeneities in magnetic Bauer, and W. Nadler, Frequency autocorrelation function of
resonance imaging: effects of diffusion, Phys. Rev. E 76, 031915 stochastically fluctuating fields caused by specific magnetic field
(2007). inhomogeneities, J. Chem. Phys. 129, 014507 (2008).

022415-15

You might also like