1118 ieee transactions on ultrasonics, ferroelectrics, and frequency control, vol. 53, no.
6, june 2006
A New Formalism for the Quantification of
Tissue Perfusion by the
Destruction-Replenishment Method in
Contrast Ultrasound Imaging
Marcel Arditi, Member, IEEE, Peter J. A. Frinking, Member, IEEE, Xiang Zhou,
and Nicolas G. Rognin, Member, IEEE
Abstract—A new formalism is presented for the
destruction-replenishment perfusion quantification ap-
proach at low mechanical index. On the basis of physi-
cal considerations, best-fit methods should be applied us-
ing perfusion functions with S-shape characteristics. These
functions are first described for the case of a geometry with
a single flow velocity, then extended to the case of vascular
beds with blood vessels having multiple flow velocity val-
ues and directions. The principles guiding the analysis are,
on one hand, a linearization of video echo signals to over-
come the log-compression of the imaging instrument, and,
on the other hand, the spatial distribution of the transmit-
receive ultrasound beam in the elevation direction. An in
vitro model also is described; it was used to confirm ex-
perimentally the validity of the approach using a commer-
cial contrast agent. The approach was implemented in the
form of a computer program, taking as input a sequence
of contrast-specific images, as well as parameters related
to the ultrasound imaging equipment used. The generated
output is either flow-parameter values computed in regions-
of-interest, or parametric flow-images (e.g., mean velocity,
mean transit time, mean flow, flow variance, or skewness).
This approach thus establishes a base for extracting infor-
mation about the morphology of vascular beds in vivo, and
could allow absolute quantification provided that appropri-
ate instrument calibration is implemented.
I. Introduction
new formalism is presented for the quantification of
A blood perfusion of organs, based on the analysis of
the replenishment kinetics following the destruction of a
microbubble-based ultrasound contrast agent (UCA) in
the imaging plane. Perfusion parameters are derived from a
best-fit of a parametric function of time with echo-derived
signals during the replenishment phase. This method is
mostly useful in assessing in vivo perfusion in the microvas-
culature, in which velocities are too small to be measured
by Doppler techniques. This approach was first described
by Wei et al. in 1998 [1]. Locally destroying microbubbles
essentially served the purpose of providing what may be
considered as a “negative-bolus” of agent in an organ oth-
erwise under an essentially constant perfusion during the
Manuscript received July 21, 2005; accepted December 12, 2005.
The authors are with Bracco Research SA, Ultrasound
Physics, Plan-les-Ouates, Geneva, Switzerland (e-mail:
marcel.arditi@brg.bracco.com).
0885–3010/$20.00

c 2006 IEEE
time of measurement. These authors observed that moni-
toring the UCA echo levels in intermittent imaging mode,
at moderately high mechanical index (MI) resulting in par-
tial destruction of microbubbles, allowed them to establish
a correspondence between the values of “video intensity”
versus pulsing interval and local flow parameters. With
the advent of sensitive, contrast-specific imaging modes
such as B-mode pulse-inversion, power pulse inversion, or
contrast pulse sequencing [2], other authors noted that a
similar correspondence could be established by using a low
MI in the replenishment phase, and real-time imaging (i.e.,
at rates of typically 8 to 25 frames per second), thus sim-
plifying considerably the method [3].
Published literature has consistently reported using a
best-fit of the replenishment signal with an expression de-
scribing the dilution kinetics of an indicator in a single-
compartment volume, often referred to as the growing mo-
noexponential function, or sometimes as the “single-phase
exponential association equation” [1], [3]–[6]. These heuris-
tic approaches were based on analyses of the video grey
levels of ultrasound imaging instruments, which system-
atically undergo nonlinear compression (also called log-
compression), before they are displayed on a video mon-
itor. Fitting these video data with the monoexponential
function thus allowed producing flow-estimates related to
the actual local organ perfusion, which were then judged
satisfactory.
However, there is no physical basis to model the re-
plenishment signal as a monoexponential function: the de-
tection of the arrival of microbubbles within the imaging
plane following destruction is a process that differs radi-
cally from the process of dilution of an indicator within a
single compartment. Instead, it is governed by the imaging
probe diffraction properties in the elevation direction, in
relation with the steady UCA-concentration replenishment
as a function of time and in the vicinity of the imaging
plane following destruction. Recently, Lucidarme et al. [7],
noting a marked discrepancy between observed S-shape re-
plenishment echo-signals as a function of time and the mo-
noexponential model, proposed a complex, time-delayed,
multicompartment model to account for possible destruc-
tion inducing progressively delayed UCA arrival at indi-
vidual locations within the image plane. Our suggestion is
that, although this effect may be real and playing a signif-
arditi et al.: tissue perfusion by destruction-replenishment method
icant role in the determination of the replenishment func-
tion in some cases, it is not the fundamental reason for
explaining the S-shape replenishment characteristic: our
explanation is actually considerably simpler, as will be ex-
posed hereafter. Other authors [8], [9], noting the sequen-
tial arrival of UCA in B-mode images, caused by the exis-
tence of various flow velocities and entry angles within the
image plane, developed a formalism for the replenishment,
also ignoring the effects of probe transmit (Tx) and receive
(Rx) diffraction properties in the determination of the re-
plenishment function. In their work published in 2004, Pot-
devin et al. [10] correctly recognized the influence of the
acoustical beam point-spread function on the replenish-
ment function, and stressed the importance of performing
the perfusion analysis on acoustical intensity signals. They
also pointed at the possibility of associating the shape of
the replenishment function with perfusion patterns typical
of healthy or diseased tissue. However, these authors did
not develop a complete formalism for deriving physically
meaningful perfusion parameters. The object of this pa-
per is to present such a formalism, based on fundamental
physical principles.
Ignoring the ultrasound equipment properties and set-
tings, such as receiver gain, log-compression, Tx and Rx
focusing, and so on, perfusion estimates remain user-
and instrument-dependent. With the new formalism in-
troduced here, it will be shown how user- and instrument-
independent relative flow estimates may be achieved. With
appropriate probe calibration, this formalism also paves
the way for obtaining absolute perfusion parameters, pro-
vided that proper beam, destruction zone, and attenuation
calibrations are implemented.
Following a brief reminder of basic beam-diffraction
considerations, in the case of linear- or phased-array aper-
tures with rectangular geometries, the new formalism for
the replenishment signal will be presented for the simple
and purely theoretical case of single-flow velocities or tran-
sit times. It will then be extended to the more realistic case
of flow with multiple transit times, known to prevail in
tissue perfusion. Experimental verification of the new for-
malism will be outlined, before exposing its applicability
within perfusion quantification software.
II. Beam Considerations
A. Preamble: Linearity
When addressing the analysis of echo signals arising
from reflections by UCA microbubbles, it is useful to re-
member that, as long as the backscattered ultrasound
energy originates from an ensemble of randomly spaced
and statistically uncorrelated scatterers, the instantaneous
echo power is a quantity proportional to the local concen-
tration of microbubbles [11]. In the context of perfusion
quantification, it is thus important to preserve that pro-
portionality when interpreting parametric values related
to the UCA concentration as a function of time. This pro-
1119
portionality can be obtained by signal linearization. When
dealing with log-compressed video signals, such lineariza-
tion can be obtained by reverting the compression law and
subsequent squaring. When dealing with radio frequency
(rf) or demodulated echo signals, such linearization can be
obtained by squaring these signals. In both cases, propor-
tionality between signal amplitude (i.e., echo power) and
UCA concentration is established. (This proportionality is
verified at constant depths, and in the situation in which
attenuation of the ultrasound beam due to the UCA is
negligible).
B. Linear and Phased-Array Geometries
The acoustic transmit-receive sensitivity distribution in
elevation is discussed because it constitutes the basis of
the present approach. As is well-known from the classical
diffraction theory [12], the acoustic pressure distribution in
the elevation direction y, in the field of a focusing aperture
with rectangular geometry, excited in a continuous-wave
mode, is approximately determined by the function:
p(y) ∼
= Γ · sinc(KT x y), (1)
where the “sinc” function is defined as sinc(q) ≡
(sin(πq)/πq) and KT x is the transmit beam diffraction
parameter in [m−1 ]. All symbols used in this paper are
defined in Table I.
In the case of pulsed wave excitation, as used in echo-
graphic imaging modes, the main lobe of the peak-pressure
distribution is in close agreement with the continuous wave
case at a frequency near the center (or mean) frequency of
the excitation waveform.
The transmit-receive sensitivity distribution, expressed
in terms of echo-signal power, can be determined by the
combined effects of the transmit and receive distributions,
which, in general, may be different. In transmit, the nor-
malized acoustic power distribution PT x (y) in the eleva-
tion direction, in the field of an ultrasound transducer,
is approximately determined by the square of the pressure
distribution given above. For a rectangular aperture, it can
thus be expressed by a function of the form:
PT x (y) ∼
= sinc2 (KT x y). (2)
In practice, and for the purpose of quantifying reperfu-
sion parameters, this power distribution can be approxi-
mated, in the main lobe of a focused aperture with rect-
angular geometry, by a Gaussian-shaped function G(YT x ),
defined as:
G(YT x ) = e−(1.94·YT x ) ,
2
(3)
where YT x ≡ KT x y is a unitless quantity.
The close correspondence of G(YT x ) with sinc2 (YT x ) is
illustrated in Fig. 1(a) for YT x values ranging from −1 to
+1.
In the receive mode, a similar sensitivity distribution is
given by:
G(YRx ) = e−(1.94·YRx ) ,
2
(4)
1120 ieee transactions on ultrasonics, ferroelectrics, and frequency control, vol. 53, no. 6, june 2006

TABLE I
Symbols Used.

SI
Symbol Quantity unit
x space coordinate within image plane, m
(lateral)
direction orthogonal to beam
y space coordinate across image plane, m
(elevation)
direction orthogonal to beam
z space coordinate within image plane, m
(depth)
direction along the beam
p, q dummy, generic variable —
(A)
p(y) ultrasound pressure distribution Pa
E ultrasound echo power W
P, P E, G unitless power distribution functions —
D special extent of microbubble destruction, m
in elevation (y) direction
f ultrasound frequency Hz
c speed of sound m s−1
λ ultrasound wavelength m
τ flow transit time from edge to center of s
destroyed zone
m mean of the natural logarithm —
s standard deviation of the natural —
logarithm
τmean mean of the lognormal distribution of s
transit times
vmean mean of the lognormal distribution of m s−1
velocities
σ2 variance of the lognormal distribution —
γ skewness of the lognormal distribution —
PD probability density or relative concentra- m−1
tion of microbubbles
A steady-state amplitude —
a half-aperture width in y direction m (B)
v local flow velocity (vx , vy , vz components) m s−1
KT x , KRx parameters on transmit and receive, re- m−1 Fig. 1. Gaussian beam approximation and its normalized integral.
spectively 2a/λz (A) Transmit power-sensitivity profile in elevation, sinc2 (YT x ) and
K transmit-receive parameter determined m−1 Gaussian G(YT x ). (B) Gaussian transmit-receive power sensitiv-
by K 2 = KT2 x + KRx2
ity profile GP E (Y ) in elevation and cumulative normal distribu-
Y unitless variables Y ≡ K y — tion E(Y ).
θ angle between flow direction and normal —
to image plane
β “velocity” term of the monoexponential s−1
function where YRx is defined, in analogy with the transmit case, by
t time s YRx ≡ KRx y and computed considering the receive focus-
Γ arbitrary proportionality constant — ing aperture. In harmonic modes often used for the specific
GL grey level value — detection of UCA echoes, the frequency to be considered
V, Vmax echo amplitude and maximum echo —
amplitude
in KRx is, accordingly, the second-harmonic frequency.
LC dynamic range of log compression dB In the pulse-echo case, the power sensitivity of the ul-
trasound transducer to off-axis targets in the elevation di-
rection may, in a first approximation, be determined by
the product of the transmit and receive distributions. The
transmit-receive power sensitivity-pattern thus can be ap-
proximated by a Gaussian function GP E (Y ) as:

GP E (Y ) = e−(1.94·Y ) ,
2
(5)
arditi et al.: tissue perfusion by destruction-replenishment method 1121

with Y ≡ Ky and K being a transmit-receive parameter to a clinician for assessing local tissue pathologies. The fol-
as defined in Table I. lowing two sections address the cases of single and multiple
flow values, successively.
C. Other Beam Geometries or Imaging Methods

In the case of transducer geometries different from rect- III. Single Flow Values
angular apertures—such as apodized rectangular, sparse
array, synthetic array, annular array, or axicon focusing— A flow direction perpendicular to the imaging plane is
the spatial distribution still can be approximated by a first assumed. As UCA microbubbles replenish the slice
Gaussian-shape function in the elevation main lobe. For volume following destruction, the linearized echo signal
any other geometry, actual values of KT x and KRx may be amplitude is determined by the growing proportion of mi-
determined, either from theoretical considerations or ex- crobubbles re-entering that volume, and weighted by the
perimentally, by measuring actual transmit-receive beam transmit-receive beam sensitivity pattern. For a uniform
sensitivity patterns, from the best-fit of Gaussian func- concentration of UCA microbubbles of unity value, having
tions. re-entered the slice until position y ≡ Y /K, the normal-
ized echo power signal E(Y ), resulting from the detection
with the beam sensitivity GP E (Y ) then can be expressed
D. Replenishment Kinetics
as the integral:


As mentioned previously, most prior investigators have
∼ 1.94 Y
based their perfusion quantification approach on the analy- E(Y ) = √ GP E (Y  )dY  . (7)
π −∞
sis of the video signal observed during UCA replenishment
following the destruction of microbubbles in a volume con- The normalized echo power signal E(Y ) can simply be
taining the tomographic image plane. Basing their anal- found with the aid of the error function erf (q), defined
yses on log-compressed, echo-signal data, these authors as [13]:
chose, as a heuristic model of replenishment kinetics, a
q
2
e−p dp.
2
monoexponential function describing the video grey level erf (q) = √ (8)
π 0
values as a function of time, GL(t), according to:
The error function verifies: erf (0) = 0, erf (−q) =
GL(t) = A(1 − e−βt ), (6) −erf (q) and lim erf (q) = 1.
q→∞

where the time origin is taken at the instant immediately
following the last UCA destruction pulses. The values A, β,
and Aβ have commonly been interpreted as quantities pro-
portional to “blood volume,” “blood velocity,” and “blood
flow” within the analyzed region. Considering the nonlin-
ear gain applied to the echo signals, it may be expected
that such analysis of the contrast-replenishment echo sig-
The physical situation of microbubbles replenishing the
destroyed slice is equivalent to assigning −∞ to the lower
integration limit of (8). The normalized echo power signal
is thus determined by the cumulative normal distribution
function, renamed in short perf in the present context of
perfusion estimates. This perf function is defined as:

q
1
e−p dp,
2

nals may be influenced by experimental parameters such perf (q) ≡ √ (9)

as log-compression, receive gain, tissue attenuation, UCA
concentration, and other instrument settings.
In contrast, the formalism exposed in this paper is
founded on the well-established relationship between the
observed echo-power as a function of time and the instan-
taneous UCA concentration. In this way, the actual UCA
replenishment kinetics is estimated by a best-fit approach
based on the spatial transmit-receive sensitivity distribu-
tion in the pulse-echo mode along the elevation direction.
The replenishment signal in the echographic image is thus
determined by two factors. The first one is the re-entry of
fresh UCA microbubbles in the region in which they were
destroyed. The second one is the transmit-receive sensitiv-
ity pattern of the ultrasound probe in the elevation direc-
tion.
Despite the great complexity of organ perfusion (mi-
crovasculature structure, random directions of flow, super-
position of different flow values, etc.), it turns out that the
estimated perfusion parameters yield useful information in
the in vivo case, which may provide valuable information
π −∞
which verifies the following properties: lim perf (q) = 0,
q→−∞
perf (0) = 0.5, and lim perf (q) = 1. These properties
q→∞
are equivalent to noting that half of the steady-state echo
power is achieved when the bubbles reach the beam axis
from either side, and steady state is achieved after re-
plenishment of the whole destroyed section. Furthermore,
perf (q) can be expressed simply in terms of the error func-
tion erf (q) as:
perf (q) = 0.5 · (1 + erf (q)). (10)
With the above definitions, the normalized echo power
signal then may be expressed as:
E(Y ) ∼
= perf (1.94 Y ). (11)
Graphically, this is an S-shape function representing
the cumulated echo power under the Gaussian transmit-
receive beam GP E (Y ) sensitivity profile, integrated from
−∞ to Y , as illustrated in Fig. 1(b).
1122 ieee transactions on ultrasonics, ferroelectrics, and frequency control, vol. 53, no. 6, june 2006

It is now considered that the MI during the destruc-

tion frames is sufficiently high to destroy a majority of
the UCA bubbles in the imaging plane (this is generally
the case in destruction-replenishment UCA-quantification
studies since the highest MI possible is normally chosen).
As a consequence, a region of destroyed microbubbles of
thickness D extends symmetrically on either side of the
image plane in the elevation direction y and is delimited
by sharp edges.
Immediately following the application of the destruc-
tion frames, a flow of fresh microbubbles starts replenish-
ing the destroyed slice in the y direction at a flow velocity
v. Noting that y = vt, the normalized echo power signal
(i.e., the replenishment function) during the nondestruc-
tive monitoring phase as a function of time, E(t), is thus
also represented by a perf function, now expressed as:
  
D
E(t) = A · perf 1.94 · Kv t − , (12) Fig. 2. Replenishment function E(t), in percent, following destruction
2v frames, at four flow velocities. The time origin is at the end of the
destruction frames.
with a time origin at the instant immediately following the
last destruction frame. By introducing the definition of a
transit time, τ , as the time required to replenish half the such that the resulting distribution of transit times follows
bubble destruction slice: a lognormal probability density distribution. This proba-
D bility density function, P D(τ ), is commonly expressed as:
τ≡ , (13)
2v
[ln(τ ) − m]2
−
the replenishment function E(t) becomes: e 2s2
  P D(τ ) = √ , (15)
KD τ s 2π
E(t) = A · perf 1.94 · (t − τ ) . (14)
2τ
where, in this particular case, m and s2 are the mean and
In the general case, when the direction of replenishment variance of the distribution of the natural logarithm of
is at an arbitrary angle θ with respect to the y direction, v transit times, τ , respectively [16]. The normalization of
must be replaced by vθ = v cos θ in (12) and (13). However, the lognormal distribution function is such that:
when the perf function is evaluated in terms of τ (14), the
∞
angle dependence is without object, following the defini- P D(τ )dτ = 1. (16)
tions of τ and D as given above. 0
Thus, the replenishment function for a single flow value
has an initial “heel”, corresponding to the beginning of The lognormal distribution of transit times is character-
the reperfusion process when the microbubbles begin to ized by its mean τmean , variance σ2 and skewness γ, de-
be detected by the edge of the elevation beam. A central, fined by:
fairly linear portion corresponds to fresh bubbles crossing s2
the mid-plane of the imaged volume. A final “shoulder” τmean = em+ 2 ,
portion, approaching the steady-state value of complete s2 +2m s2
reperfusion, corresponds to fresh bubbles finishing to re-
2
σ =e (e − 1), and (17)
  2

plenish the imaged volume in the elevation direction. This γ = es2 − 1 2 + es .
behavior is illustrated in Fig. 2, for an example with a du-
ration of destruction frames of 1 s, A = 100, D = 8 mm, The probability density function in (15) can also can
K = 0.276 mm−1 , θ = 0◦ and different values of v at 0.5, be expressed, as an alternative, in terms of flow velocity v,
1, 2, and 4 mm s−1 , respectively. instead of transit time τ . In the Appendix, a law of trans-
formation is derived for converting (15) into a lognormal
distribution expressed in flow velocity v.
IV. Multiple Flow Values
In the presence of multiple flow transit times, the
A. Lognormal Distribution of Transit Times replenishment function E(t) may, in this case, be ex-
pressed as:
As mentioned by, for example, Karshafian et al. [4], [14]
∞  
KD
or Qian and Bassingthwaighte [15], the vascular arrange- E(t) = A P D(τ )perf 1.94 · (t − τ ) dτ.
ment of blood vessels in most regularly perfused organs is 0 2τ
(18)
arditi et al.: tissue perfusion by destruction-replenishment method
Fig. 3. Lognormal distribution of transmit times τ with m = 0.835
and s = 0.726, corresponding to τmean = 3 s and σ = 2.5, decom-
posed into four distribution segments τ1 to τ4 , separated at τ = 2,
4, and 6 s, respectively.
In this context, E(t) is named the lognormalperf replen-
ishment function, and it is expressed in terms of elemen-
tary perf functions, weighted by the probability of density
distribution P D(τ ).
In Fig. 3, an example of a lognormal probability dis-
tribution of transit times with a mean at τmean = 3 s is
shown, considering four arbitrary transit time intervals,
between 0 and 2 s, 2 and 4 s, 4 and 6 s, and above 6 s. The
four corresponding contributions of these subranges to the
overall replenishment function E(t) are illustrated in Fig. 4
as the numbered curves Eτ1 (t) to Eτ4 (t), as obtained by in-
tegration of (18) in the corresponding τ intervals. It may
be observed that the individual replenishment functions
of the four relatively narrow distributions have an appear-
ance close to the S-shape perf function of a single-flow
value. In contrast, the overall replenishment function dis-
plays a distorted or skewed appearance, with a more rapid
“heel” at the onset of replenishment and slower “shoulder”
near the end of the replenishment. It reflects the contri-
butions of the various transit times, each with their own
weights as determined by the relatively wide probability
distribution function.
B. Other Distributions of Transit Times
If flow quantification is to be applied to tissues that
are believed not to obey lognormal distributions of tran-
sit times, but are better described by another probabil-
ity distribution (e.g., first passage, random walk, gamma
variate, or arbitrary distributions), (15) then may be ex-
pressed by using this other distribution for P D(τ ). There
is, however, strong evidence that the perfusion of most
physiological tissues, including abnormal structures such
as neoplasms or neovasculature in tumors, are well rep-
1123
Fig. 4. Replenishment functions corresponding to the four distribu-
tions τ1 to τ4 of Fig. 3, as well as the overall replenishment function
determined by the sum of the four contributions.
resented by lognormal probability distributions of transit
times. In these cases, and as discussed by Karshafian et
al. [4], [14], the analyses of the replenishment functions by
best-fit approaches outline differences in skewness of the
lognormal probability distribution, which then may bear
diagnostic significance in the assessment of their state of
perfusion.
V. Experimental Verification
A. Objective
A controlled-flow system was assembled to be imaged
in a contrast-specific ultrasound imaging mode. By ap-
plying the destruction-replenishment technique, the im-
age database collected was analyzed in two ways: one us-
ing the monoexponential replenishment model function on
the video signals, and the other using the new approach
as described above. The objective was to assess how re-
liable each approach was while outlining their respective
advantages and limitations. In particular, the same mea-
surements were repeated at two different receive-gains and
log-compression settings in the imaging instrument, to as-
sess the instrument-dependence of the techniques.
B. Materials and Method
A variation on the model described by Veltmann et al.
[17] was used to establish stable and reproducible flow
conditions to be imaged by a commercially available ul-
trasound scanner. The general setup is shown in Fig. 5. A
bundle of 170 microfibers with 240-µm lumen diameter and
18-cm length (AN69 type, Hospal SA, Lyon, France) was
modified by removing its clear polymethyl methacrylate
(PMMA) tubing over half its circumference over a couple
of centimeters in length. In this way, the microfibers were
1124 ieee transactions on ultrasonics, ferroelectrics, and frequency control, vol. 53, no. 6, june 2006
Fig. 5. Double-flow circuit (HI and LO), modified from Veltmann
et al. [17], allowing a stable agent concentration in the microfiber-
bundle at all flow values imposed by the downstream peristaltic
pump.
exposed directly, in water immersion, to the ultrasound
field of a linear-array probe (Esaote LA532E, 2.6/5.2 MHz
Tx/Rx-frequencies), coupled to an EsaTune instrument
(Esaote, Florence, Italy). The ultrasound scanner was op-
erated at a frame rate of 43 Hz, in harmonic B-mode, at
a low acoustic power (MI 0.07), with the probe placed at
45-mm distance from the center of the fiber bundle. With
an aperture width in elevation of 5.5 mm, these imaging
conditions (i.e., using a transmit frequency of 2.6 MHz
and a receive frequency of 5.2 MHz for KT x and KRx ,
respectively) determined a value K = 0.480 mm−1 . The
bundle was characterized as having a linear relationship
between the fluid volume flow ϕ in milliliters per min-
utes and the mean flow velocity v in millimeters per sec-
ond by the relation ϕ = 0.46 v. While gravity feeding a
HI-flow circuit with deionized water at a constant rate of
270 mL/minute, SonoVueTM microbubbles (Bracco SpA,
Milan, Italy) were injected in this circuit at a constant
rate of 0.5 mL/minute by using a VueJect
infusion pump
(Bracco Research SA, Geneva, Switzerland), determining
a dilution factor of 1:1620, equivalent to approximately 2.5
105 bubbles per milliliter. A small fraction of that flow was
sucked into a LO-flow circuit from a small mixing chamber
(approximately 2 mL), using a MiniPuls 3 peristaltic pump
(Gilson, Villiers-le-Bel, France) placed downstream from
the bundle. In this way, a UCA suspension with stable con-
centration was forced through the bundle, with nine differ-
ent flow values between 0.46 and 13.8 mL/minute, corre-
sponding to mean flow velocities between 1 and 30 mm s−1 .
With the imaging probe positioned to scan a cross sec-
tion of the fiber bundle, at an angle θ of 60◦ , sequences of
real-time, echo rf-data were captured by a Femmina digital
interface (University of Florence, Italy [18]) coupled to the
EsaTune instrument by a fiberoptic link, at a sample rate
of 40 MS/s. The UCA destruction sequences consisted of
43 frames of destructive imaging (maximum acoustic out-
put, for a duration of 1 s). Three recordings were made at
each of nine flow values. The rf-data sequences then were
demodulated using the Hilbert transform, scan-converted,
log-compressed and stored as 8-bit grey scale TIF im-
ages with 256 possible grey levels (GL), using an in-house
program written in Matlab (The Mathworks Inc., Natick,
MA). Three sets of TIF images were generated, using two
values of log-compression and two relative receiver-gain
values (0 and −10 dB), in order to simulate the effects
of different user choices of instrument settings. The log-
compression function, expressed as a function of the input
echo amplitude variable V , was:
 
20 dB V LC
GL(V ) = 255 log10 10 20 dB ,
LC Vmax
(19)
where Vmax was chosen as the maximum echo amplitude,
and kept constant for the whole data set. LC was set at
values of 40 and 70 dB (using the 0 dB gain only), and
any negative values of GL were truncated to zero.
Image sequences of 8 s of UCA replenishment were an-
alyzed in two different ways. In one data path, the com-
pressed GL images were analyzed using an in-house video-
densitometric program in a region-of-interest (ROI), hand-
drawn to include the whole fiber bundle region, to gener-
ate what was called video replenishment functions. As an
alternative data path, the compressed GL images were lin-
earized by applying, in sequence, the inverse of (19) and
squaring, to obtain replenishment functions proportional
to the instantaneous echo-power from the UCA, E(t), with
the aid of the function E(GL):
 2
E(GL) = Vmax 10( 255 −1) 20 dB ,
GL LC
(20)
with E(GL) truncated at (Vmax )2 for any values GL > 255.
The echo-power signals so computed then were averaged
in the same ROI, effectively representing the mean square
(rms2 ) amplitude in that region, resulting in what was
called linearized replenishment functions.
The recorded data sets were analyzed using the curve
fitting toolbox (v.1.1.2) of Matlab. The monoexponential
function GL(t) of (6) and the perf function, E(t) of (14)
were used as fitting functions. The GL(t) function had A
and β as fitting parameters, and the E(t) function had A
and τ . Note that the perf function model was used be-
cause the simple geometry of the bundle of parallel fibers
determined single flow velocity values in one direction. The
experimental conditions were such that there were no satu-
rated values in both the video and linearized data sets, and
negligible offset levels. Each measurement was repeated
three times, and sets of parameters were computed and
averaged for both functions. Estimates of the flow veloci-
ties then were considered as:
D D
vest ≡ = , (21)
2τ cos 60◦ τ
for the perf function, and as β for the monoexponential
function. The value of D was found to be D = 7.2±0.1 mm.
It was measured in a separate experiment by optical ob-
servation of the extent of destroyed bubbles embedded in
an agarose gel, exposed at the same distance as the fibers,
but perpendicular to the ultrasound imaging beam.
arditi et al.: tissue perfusion by destruction-replenishment method
Fig. 6. Results of curve fits with the perf function on linearized data,
at flow velocities of 4 and 2 mm s−1 .
C. Results
In Fig. 6, two linearized replenishment curves are
shown, for individual measurements performed at 4 and
2 mm s−1 mean flow velocities. Parametric perf functions
were fitted to the data, determining sets of correspond-
ing A and τ values. The S-shape characteristics of the
replenishment may be clearly recognized. The compara-
tively large noise amplitude at the plateau level is caused
by fluctuations in the instantaneous agent concentrations
and speckle patterns in the ROI, magnified by the lin-
earization process. In Fig. 7, the values of vest are plotted
against the nine flow velocity values v. The linear regres-
sion slope coefficient a1 was found to be 0.924, a result
in close agreement with the value of 1.0 expected with an
absolute quantification method.
In Fig. 8, replenishment functions at 8 mm s−1 are
shown for GL amplitude data, obtained with 40 dB log-
compression, at two relative front-end gain settings of 0 dB
and −10 dB (applied before log-compression). The cor-
responding best-fit parametric monoexponential functions
are able to adjust well to the experimental data near the
plateau region, but the fits are poor in the early phases
of replenishment. Fig. 9 illustrates the system gain depen-
dence as β is plotted against the nine values of v; although
the linearity in each case may be considered as adequate,
an overestimate of 50% on the slope a1 is made at 0 dB
gain, compared to the slope value a2 obtained at −10 dB
gain. Similarly, Fig. 10 shows the β estimates using a fixed
gain of 0 dB, but at two different LC values of 40 and
70 dB. Here, also, estimates of the linear regression slopes
are significantly affected by the log-compression settings.
D. Discussion
In the previous section, experimental results obtained
in a simple flow model were presented. The simplified ge-
ometry allowed comparative assessments of both the perf
1125
Fig. 7. Flow-velocity estimates vest as a function of flow velocity
between 1 and 30 mm s−1 using the perf function. The actual esti-
mated values of D (7.2 mm) and θ (60◦ ) were used in the fits. The
slope a1 of the linear regression on vest is 0.924.
and monoexponential models in terms of transit time or
flow-velocity estimates, in the case of a single flow direc-
tion. In these conditions, the perf model is clearly able to
provide more consistent and robust estimates of flow ve-
locity, because it is based on a physical description of the
UCA replenishment process as measured by a given imag-
ing beam. Different gain values for the estimation of flow
velocity with the perf model were not considered, as this
is irrelevant: the perf analysis is intrinsically a linear pro-
cess. Changing the gain value, or equivalently, the concen-
tration of UCA bubbles, results in identical flow velocity
estimates with the perf model, in contrast with the mo-
noexponential model applied on video grey levels, which
produces estimates that are instrument-setting dependent.
The monoexponential model also was applied on the lin-
earized replenishment data set (results not shown). In this
case, the best-fit analysis was performed with and with-
out a manual adjustment of the time origin to the instant
of detectability of the UCA following destruction. This op-
tional adjustment was considered in an attempt to improve
the quality of the fits, especially at the low-flow velocities.
In both cases, the quality of the fits and the linearity of
the β estimates with flow velocity v were still poorer than
when applying the monoexponential model on compressed
video grey levels.
In reality, the shape of the UCA destruction zone might
be more complex than was considered here, e.g., with a
Gaussian or truncated Gaussian profile as described in
[10]. In the present formalism, the value of D represents a
destruction zone delimited by sharp edges. In our experi-
ence, this assumption proved to be justified, as supported
1126 ieee transactions on ultrasonics, ferroelectrics, and frequency control, vol. 53, no. 6, june 2006
Fig. 8. Results of curve fits with the monoexponential function at
the flow velocity of 8 mm s−1 . Two different front-end gains of 0 dB
and −10 dB were used. The log-compression was set to LC = 40 dB.
Fig. 9. Estimates of β with the monoexponential function, as a func-
tion of flow velocity between 1 and 30 mm s−1 , at different front-end
gains of 0 dB and −10 dB, at a log-compression with LC = 40 dB.
The slopes a1 and a2 of the linear regression functions are strongly
influenced by the choices of gain values.
by the good agreement between the perf function and the
experimental replenishment data.
Relative estimates of blood-volume flow have commonly
been associated with the product Aβ of the monoexpo-
nential model. Clearly, such products cannot physically
represent blood flow when applied to compressed video
grey level data, because the plateau value A is not pro-
portional to local UCA concentration (or regional blood
volume, RBV), and the estimate of β is subject to the de-
pendence on settings discussed above. Using the perf or
lognormalperf analyses, the expression of A/τ or A/τmean
are improved estimates of the local blood-volume flow,
Fig. 10. Estimates of β with the monoexponential function, as a
function of flow velocity between 1 and 30 mm s−1 , at two different
log-compressions (LC) of 40 and 70 dB, with a front-end gain of 0 dB.
The estimated slopes a1 and a2 of the linear regression functions are
influenced by the choices of log compression settings.
compared to the monoexponential model. However, these
flow estimates remain relative, and not absolute, consider-
ing the fact that the A estimates remain dependent on the
actual UCA concentration, system gain and the round-trip
ultrasound attenuation between the probe and the depth
of analysis.
The improvements of the perf analysis on echo-power
data may thus be summarized as allowing possibly abso-
lute estimates of flow velocity or transit time, and rela-
tive estimates of blood-volume flow, independently of log-
compression settings of the imaging instrument. Further-
more, the lognormalperf analysis allows estimates of the
morphology of tissue vascularization in an ROI, possibly
opening new avenues of tissue characterization for tumor
staging, antiangiogenic drug efficacy assessment, or differ-
ential lesion diagnosis.
VI. Application in Echo-Contrast Imaging
A. Linearizing Video Data
Although the experimental examples described above
were collected using a dedicated rf-grabber, the new for-
malism may be implemented on video data, collected on a
personal computer using any video acquisition hardware.
In some cases, equipment manufacturers provide a tool
to generate echo data without log-compression (e.g., Q-
lab by Philips Medical Systems, Bothell, WA). Otherwise,
linearizing the video data to generate echo power data
may be implemented in one of several different ways. One
way is to obtain the log-compression law from the equip-
ment manufacturer, and to apply the corresponding anti-
log function. If that information is not available, another
arditi et al.: tissue perfusion by destruction-replenishment method
option is to record images of a fixed tissue-mimicking ma-
terial at increasing gain values, often displayed as decibels
on the monitor of the echo instruments to estimate the
log-compression law by analyzing the resulting grey lev-
els within individual pixels or small groups of pixels. Yet
another method consists of recording video sequences of
echo-contrast images produced at known and increasing
concentrations of UCA, in order to derive the grey level
map as a function of UCA concentration, thus allowing lin-
earization of the data. In any case, the preferred approach
is for the equipment manufacturer to provide direct access
to the raw echo data (rf-signals, quadrature-demodulated
signals, linearized image data, or similar), or to incorpo-
rate the quantification analysis within the instrument.
B. Practical Considerations
In a clinical application of the UCA destruction-
replenishment method, it can happen that the echo level
immediately after the last destruction frame is not negli-
gible. This can be caused by two factors: a residual tissue
echo signal, due to incomplete cancellation by the contrast-
specific imaging mode, or a nonzero UCA echo signal, due
to incomplete destruction. The first situation can be han-
dled by subtracting baseline echo power data, recorded
before UCA administration, from the replenishment data.
The second situation can be handled by including D as
a fitting parameter. In this case, the estimate of abso-
lute flow velocity remains possible, provided constraints
are considered on the acceptable values of D as a function
of depth.
C. Parametric Imaging
The new formalism, described above in the case of com-
puting best-fit functions of replenishment echo data in
ROIs, also is applicable at the scale of individual pixels or
small groups of pixels, as required in the case of display-
ing parametric images depicting the spatial distribution
of flow parameters (e.g., local RBV, volume flow, veloc-
ity, transit time, etc.). In this case, the signal-to-noise ra-
tio of echo-power data naturally is degraded, due to large
fluctuations in signal amplitude during the replenishment
phase. However, alternative approaches are possible, and
a detailed description of the use of the present formalism
for the purpose of constructing parametric flow images is
in preparation and will be the object of a separate article.
D. Clinical Areas of Interest
As mentioned by many investigators, the destruction-
replenishment, perfusion-quantification technique finds
applications in many areas of medical contrast ultrasound.
The formalism described above has been implemented
within an in-house developed software program, which has
been tested in a wide variety of pathological conditions
(myocardial perfusion deficits, renal artery stenosis, liver
disease, prostate and breast cancer, etc.). In most cases,
1127
analyses carried out with the lognormalperf model func-
tion provided successful curve fitting and useful quantifi-
cation information when applied to assess perfusion within
user-drawn ROI in both preclinical and clinical settings.
VII. Conclusions
A new formalism has been presented for the
destruction-replenishment, perfusion-quantification ap-
proach. Based on a physical analysis of the transmit-
receive imaging conditions during the UCA replenishment,
the model function to be used for a best-fit analysis
on echo-power data should be with S-shape characteris-
tics. More precisely, this function is a linear combination
of elementary cumulative normal distribution functions
weighted by the probability density distribution of flow
transit times within the region being analyzed. The char-
acteristics of such an elementary function are determined
by the spatial transmit-receive sensitivity distribution in
the elevation direction, and the extent of the zone in which
UCA microbubbles have been destroyed, also in the eleva-
tion direction. Using a simple in vitro flow model, it was
shown how absolute flow velocity estimates may be made
with knowledge of the extent of the bubble destruction
zone.
Compared to previously described approaches, the new
formalism is at least as simple to implement in a software
program, and it has the advantage of providing perfusion
parameters that are independent of user settings or instru-
ment characteristics. As a counterpart, information about
the destruction zone and beam profile is required, as well
as log-compression characteristics if raw-echo data are not
available.
The formalism is applicable to the computation of
perfusion-parameter values either in ROIs, or within in-
dividual pixels or groups of pixels for parametric imaging.
Furthermore, the approach presented allows the extraction
of information about the morphological distribution of vas-
cular beds in vivo, and thus could become a useful tool of
tissue characterization for tumor staging, antiangiogenic
drug efficacy assessment, or differentiating pathologies.
Appendix A
Correspondence Between Lognormal
Probability Density Distributions
Expressed in τ or in v.
A law of transformation is derived for converting a log-
normal probability density distribution expressed in terms
of a transit-time variable τ into a lognormal distribution
expressed in flow velocity v. The reverse transformation
is found by reciprocity in similar terms. In the following,
the angle θ formed between the flow direction and the el-
evation direction is considered to be zero. The case of an
arbitrary angle θ may be treated, as indicated in the dis-
cussion about (13) and (14), by replacing v hereafter by
vθ = v cos θ.
1128 ieee transactions on ultrasonics, ferroelectrics, and frequency control, vol. 53, no. 6, june 2006

The m and s parameters of the lognormal distribution Acknowledgment

in transit times of (15), P D(τ ), are renamed mτ and sτ ,
to be distinguished from the corresponding parameters mv The authors wish to thank E. Gaud, A. Broillet, F. Yan,
and sv of the following lognormal distribution expressed in T. Messager, and M. Schneider of Bracco Research, as well
velocity P D(v): as F. Bertora and M. Cerofolini of Esaote Biomedica, for
their precious support, inputs, and collaboration.
[ln v−mv ]2
− 2s2v
e
P D(v) = √ . (A1)
v sv 2π References
Considering that: [1] K. Wei, A. R. Jayaweera, S. Firoozan, A. Linka, D. M. Skyba,
and S. Kaul, “Quantification of myocardial blood flow with
 
D D −D ultrasound-induced destruction of microbubbles administered as
v= , ln v = ln − ln τ and dv = dτ, a constant venous infusion,” Circulation, vol. 97, pp. 473–483,
2τ 2 2τ 2 1998.
(A2)
[2] P. Rafter, P. Phillips, and M. A. Vannan, “Imaging technologies
and techniques,” Cardiol. Clin., vol. 22, pp. 181–197, 2004.
the probability distribution in v, for positive dτ incre- [3] T. Schlosser, C. Pohl, C. Veltmann, S. Lohmaier, J. Goenechea,
ments, yields: A. Ehlgen, J. Koster, D. Bimmel, S. Kuntz-Hehner, H. Becher,
and K. Tiemann, “Feasibility of the flash-replenishment concept

 2 in renal tissue: Which parameters affect the assessment of the
− ln τ + ln D 2 − mv contrast replenishment?,” Ultrasound Med. Biol., vol. 27, pp.
− 937–944, 2001.
e 2s2v [4] R. Kharchakdjian, P. N. Burns, and M. Henkelman, “Fractal
P D(v)dv = √ dτ modeling of microbubble destruction-reperfusion in unresolved
τ sv 2π vessels,” in Proc. IEEE Ultrason. Symp., 2001, pp. 1669–1673.


2
ln τ − −mv + ln D 2
[5] S.-J. Rim, H. Leong-Poi, J. R. Lindner, D. Couture, D. Elle-
− gala, H. Masson, M. Durieux, N. F. Kasse, and S. Kaul, “Quan-
2s2v (A3)
e tification of cerebral perfusion with real-time contrast-enhanced
= √ dτ ultrasound,” Circulation, vol. 104, pp. 2582–2587, 2001.
τ sv 2π [6] T. H. Murthy, P. Li, E. Locvicchio, C. Baisch, I. Dairywala,
= P D(τ )dτ, W. F. Armstrong, and M. Vannan, “Real-time myocardial blood
flow imaging in normal human beings with the use of myocardial
contrast echocardiography,” J. Amer. Soc. Echocardiogr., vol.
provided that: 14, pp. 698–705, 2001.
[7] O. Lucidarme, S. Franchi-Abella, J. M. Correas, S. L. Bridal,
[ln(τ )−mτ ]2 E. Kurtisovski, and G. Berger, “Blood flow quantification with
− 2s2τ contrast-enhanced US: Entrance in the section phenomenon—
e
P D(τ ) = √ , (A4) phantom and rabbit study,” Radiology, vol. 228, no. 2, pp. 473–
τ sτ 2π 479, 2003.
[8] M. Krix, F. Kiessling, N. Farhan, K. Schmidt, J. Hoffend, and
with: S. Delorme, “A multivessel model describing replenishment ki-
netics of ultrasound contrast agent for quantification of tissue
  perfusion,” Ultrasound Med. Biol., vol. 29, no. 10, pp. 1421–
D 1430, 2003.
mτ = −mv + ln and sτ = sv ≡ s.
2 [9] M. Krix, C. Plathow, F. Kiessling, F. Herth, A. Karcher, M. Es-
(A5) sig, H. Schmitteckert, H. U. Kauczor, and S. Delorme, “Quan-
tification of perfusion of liver tissue and metastases using a mul-
The correspondences between the mean of the lognor- tivessel model for replenishment kinetics of ultrasound contrast
agents,” Ultrasound Med. Biol., vol. 30, no. 10, pp. 1355–1363,
mal distributions are found with the definitions: 2004.
[10] T. C. Potdevin, J. B. Fowlkes, A. P. Moskalik, and P. L. Car-
s2 s2 son, “Analysis of refill curve shape in ultrasound contrast agent
vmean = emv + 2 and τmean = emτ + 2 , studies,” Med. Phys., vol. 31, no. 3, pp. 623–632, 2004.
(A6) [11] N. De Jong and L. Hoff, “Ultrasound scattering of Albunex mi-
crospheres,” Ultrasonics, vol. 31, no. 3, pp. 175–181, 1993.
[12] L. E. Kinsler, A. R. Frey, A. B. Coppens, and J. V. Sanders,
which yield: Fundamentals of Acoustics. New York: Wiley, 1982.
    [13] W. Gautschi, “Error function and fresnel integrals,” in Handbook
D s2 s2 of Mathematical Functions. M. Abramowitz and I. A. Stegun,
−mv +ln 2 + 2 D − mv + 2 +s2
τmean = e = e . Eds. New York: Dover, 1972, pp. 297–329.
2 [14] R. Karshafian, P. N. Burns, and M. R. Henkelman, “Transit time
(A7) kinetics in ordered and disordered vascular trees,” Phys. Med.
Biol., vol. 48, no. 19, pp. 3225–3237, 2003.
and: [15] H. Qian and A. Bassingthwaighte, “A class of flow bifurcation
models with lognormal distribution and fractal dispersion,” J.
D 2
Theoretical Biol., vol. 205, pp. 261–268, 2000.
τmean = es , (A8) [16] D. Zwillinger and S. Kokoska, “Continuous Probability Distri-
2vmean butions,” in CRC Standard Probability and Statistics Tables and
Formulae. Boca Raton, FL: Chapman and Hall/CRC, 2000, pp.
which, together with the relations in (A5), provides the 143–144.
[17] C. Veltmann, S. Lohmaier, T. Schlosser, S. Shai, A. Ehlgen, C.
rules for converting the probability density distributions Pohl, H. Becher, and K. Tiemann, “On the design of a capillary
from τ to v and conversely. flow phantom for the evaluation of ultrasound contrast agents
arditi et al.: tissue perfusion by destruction-replenishment method
at very low flow velocities,” Ultrasound Med. Biol., vol. 28, no.
5, pp. 625–634, 2002.
[18] M. Scabia, E. Biagi, and L. Masotti, “Hardware and software
platform for real-time processing and visualization of echo-
graphic radiofrequency signals,” IEEE Trans. Ultrason., Fer-
roelect., Freq. Contr., vol. 49, no. 10, pp. 1444–1452, 2002.
Marcel Arditi (M’82) was born in Alexan-
dria, Egypt, in 1951. He graduated in 1975
with a M.Sc. degree in physics at the Univer-
sity of Geneva, Switzerland. He then obtained
his Ph.D. degree in 1982 at the Department
of Medical Biophysics, University of Toronto,
Toronto, ON, Canada, specializing in several
aspects of medical ultrasound imaging (annu-
lar array transducers, beam forming, and sig-
nal processing).
From 1982 to 1985, he was with SRI In-
ternational in Menlo Park, CA, where he con-
tributed to research projects for various industrial and governmental
organizations, still in the ultrasound imaging field. In 1985, he joined
Battelle-Europe in Geneva, Switzerland, and managed contract re-
search projects for corporations in the optical and automotive indus-
tries, in fields related to image and signal processing. Since 1989, he
has been associated with Bracco Research SA, also in Geneva, where
he contributed to the acoustical specifications and characterization
of novel microbubble-based contrast agents for medical ultrasound.
His current interests focus on the quest for new contrast-specific
detection methods and for making medical ultrasound a reliable
quantitative modality.
Peter J. A. Frinking (M’02) was born in
Bussum, The Netherlands, in 1966. After sec-
ondary school, he received his M.Sc. degree
in applied physics from the Delft Univer-
sity of Technology, Delft, The Netherlands, in
1995. He received his Ph.D. degree in medical
biophysics at the Thoraxcentre of the Eras-
mus University Rotterdam, Rotterdam, The
Netherlands, in 2000. During his stay in Rot-
terdam, he worked on the acoustic character-
ization of new ultrasound contrast agents, de-
veloping novel detection methods for contrast
ultrasound imaging, and he did preliminary work on ultrasound me-
diated drug delivery.
1129
Currently, he is working for Bracco Research SA, Geneva, Switzer-
land, where he is involved in microbubble physics and contrast ul-
trasound imaging. He is working on advancing the understanding
of the interaction of ultrasound beams and microbubbles, promot-
ing novel detection schemes and exploring new applications of ul-
trasound contrast agents like molecular imaging, sonoporation, and
ultrasound-directed drug delivery.
Xiang Zhou was born in Wuhan, P.R. China,
in 1972. He obtained his M.D. degree in the
Tongji Medical College, Huazhong Science
and Technology University, Wuhan, Hubei,
P.R. China, in 2000. His M.D. thesis fo-
cused on perfusion quantification and func-
tional imaging studies on kidney tumors, us-
ing ultrasound contrast agents. In 2002, he
worked as a postdoctoral fellow at Bracco
Research SA, Geneva, Switzerland, where he
contributed to studies on perfusion quantifi-
cation.
In 2003 and 2004, he was a clinical fellow at the Friedrich-
Alexander-Universität Erlangen-Nürnberg, Germany, where he stud-
ied the hepatic transit time of ultrasound contrast agents for the
evaluation of liver tumor radiofrequency ablation.
Dr. Zhou is now a clinician at the Union Hospital, Tongji Medical
College, Huazhong University of Science and Technology (HUST),
Wuhan, Hubei, P.R. China.
Nicolas G. Rognin (M’05) was born in Vo-
iron, France, in 1974. He received his M.Sc.
degree in signal, image, and speech process-
ing from the National Institute of Applied
Sciences, INSA-Lyon, France, in 1998, with a
thesis on brain perfusion using magnetic res-
onance imaging.
He completed his Ph.D. degree in 2002 at
Creatis (Centre de Recherche et d’Applica-
tions en Traitement de l’Image et du Signal),
Lyon, France, a unit of the National Centre
for Scientific Research (CNRS), focusing on
simulation and quantification tools in ultrasound contrast imaging.
From 2000 to 2002, he had a teaching position in computer sciences at
Claude Bernard University, Lyon, France. Since 2003, Dr. Rognin has
been with Bracco Research SA, Geneva, Switzerland. His research in-
terests include ultrasound contrast imaging, parametric imaging, and
image registration.