JACC: CARDIOVASCULAR IMAGING VOL. -, NO.

-, 2022
ª 2022 BY THE AMERICAN COLLEGE OF CARDIOLOGY FOUNDATION

PUBLISHED BY ELSEVIER

ORIGINAL RESEARCH

Coronary Flow Assessment Using

3-Dimensional Ultrafast Ultrasound
Localization Microscopy
Oscar Demeulenaere, MS,a Zulma Sandoval, PHD,a Philippe Mateo, MS,a Alexandre Dizeux, PHD,a
Olivier Villemain, MD, PHD,a Romain Gallet, MD, PHD,b Bijan Ghaleh, PHD,b Thomas Deffieux, PHD,a
Charlie Deméné, PHD,a Mickael Tanter, PHD,a Clément Papadacci, PHD,a Mathieu Pernot, PHDa

ABSTRACT

OBJECTIVES The purpose of this study was to demonstrate 3-dimensional (3D) coronary ultrasound localization
microscopy (CorULM) of the whole heart beyond the acoustic diffraction limit (<20 mm resolution) at ultrafast frame rate
(>1000 images/s).

BACKGROUND Direct assessment of the coronary microcirculation has long been hampered by the limited spatial and
temporal resolutions of cardiac imaging modalities.

METHODS CorULM was performed in isolated beating rat hearts (N ¼ 6) with ultrasound contrast agents (Sonovue,
Bracco), using an ultrasonic matrix transducer connected to a high channel–count ultrafast electronics. We assessed the
3D coronary microvascular anatomy, flow velocity, and flow rate of beating hearts under normal conditions, during
vasodilator adenosine infusion, and during coronary occlusion. The coronary vasculature was compared with
micro-computed tomography performed on the fixed heart. In vivo transthoracic CorULM was eventually assessed on
anaesthetized rats (N ¼ 3).

RESULTS CorULM enables the 3D visualization of the coronary vasculature in beating hearts at a scale down to
microvascular structures (<20 mm resolution). Absolute flow velocity estimates range from 10 mm/s in tiny arterioles
up to more than 300 mm/s in large arteries. Fitting to a power law, the flow rate–radius relationship provides an
exponent of 2.61 (r2 ¼ 0.96; P < 0.001), which is consistent with theoretical predictions and experimental validations
of scaling laws in vascular trees. A 2-fold increase of the microvascular coronary flow rate is found in response to
adenosine, which is in good agreement with the overall perfusion flow rate measured in the aorta (control measure-
ment) that increased from 8.80  1.03 mL/min to 16.54  2.35 mL/min (P < 0.001). The feasibility of CorULM was
demonstrated in vivo for N ¼ 3 rats.

CONCLUSION CorULM provides unprecedented insights into the anatomy and function of coronary arteries at the
microvasculature level in beating hearts. This new technology is highly translational and has the potential to
become a major tool for the clinical investigation of the coronary microcirculation. (J Am Coll Cardiol Img 2022;-:-–-)
© 2022 by the American College of Cardiology Foundation.

From the aPhysics for Medicine, Ecole Supérieure de Physique Chimie Industrielles de Paris, Institut National de la Santé et de la
Recherche Médicale U1273, CNRS UMR 8063, PSL University, Paris, France; and bInserm U955-IMRB, Equipe 03, Université Paris-
Est Créteil, Ecole Nationale Vétérinaire d’Alfort, Maisons-Alfort, France.
The authors attest they are in compliance with human studies committees and animal welfare regulations of the authors’
institutions and Food and Drug Administration guidelines, including patient consent where appropriate. For more information,
visit the Author Center.

Manuscript received August 2, 2021; revised manuscript received January 24, 2022, accepted February 17, 2022.

ISSN 1936-878X/$36.00 https://doi.org/10.1016/j.jcmg.2022.02.008

2 Demeulenaere et al
Coronary Microcirculation Assessment by Ultrasound
C oronary microcirculation (ie, flow in
ABBREVIATIONS
AND ACRONYMS vessels smaller than 300 m m) plays a
key role in the control of cardiac
2D = 2-dimensional
perfusion. 1 Patients with coronary microvas-
3D = 3-dimensional
cular disease (CMD) have poor prognoses
CFR = coronary flow reserve
with significantly higher rates of cardiovas-
CMD = coronary
cular events, including hospitalization for
microcirculation disease
heart failure, sudden cardiac death, and
CorULM = coronary ultrasound
localization microscopy
myocardial infarction (MI).1 Likewise, micro-
vascular obstruction after reperfusion of a
ECG = electrocardiogram
MI, a phenomenon called no-reflow, is the
IQ = in-phase and quadrature
most relevant predictor of adverse outcome.
LAD = left anterior descending
Up to date, although the anatomy of large
LV = left ventricle
vessels can be easily assessed, only func-
LVP = left ventricular pressure
tional assessments of the coronary microcir-
MB = microbubble culation are available either through imaging
MI = myocardial infarction (positron emission tomography,2 cardiac
ULM = ultrasound localization magnetic resonance, 3 and contrast echocar-
microscopy
diography4 ) or through invasive measure-
ments.5 These examinations can provide
hemodynamic information such as myocardial blood
flow and coronary flow reserve (CFR) in response to
the vasodilator adenosine. However, there is
currently no available tool to image directly the entire
coronary microcirculation in clinical or preclinical
settings, which in turn has limited considerably the
knowledge on the coronary microcirculation patho-
physiology and the development of effective thera-
peutic strategies. Indeed, direct visualization of the
full coronary architecture including complex micro-
vascular network remains out of reach of current
cardiac imaging modalities mainly because of the low
spatial and temporal resolutions or sensitivity limi-
tations.6 The association of anatomic and hemody-
namic exploration at the microscopic scale could
allow a better understanding of the adaptation of the
coronary microcirculation to physiological conditions
(aging, exercise) or to pathophysiological stresses
(proximal artery stenosis, O 2 consumption/perfusion
imbalance). Moreover, besides contrast echocardiog-
raphy, none of the examinations for the microcircu-
lation evaluation can be easily repeated because of
their availability, the risk of irradiation, and the risk
of complications for invasive assessment. Conse-
quently, we are currently lacking of a technique that
would allow repeated anatomical and functional
evaluation of the coronary microcirculation to deter-
mine the mechanisms of the functional alterations
but also to address the effect of the different
treatments.
Recently, ultrafast ultrasound broke the conven-
tional resolution barrier via localization of single
circulating microbubbles (MB) in the blood flow.7 Ul-
trafast ultrasound localization microscopy (ULM)
JACC: CARDIOVASCULAR IMAGING, VOL. -, NO. -, 2022
- 2022:-–-
allowed mapping microvascular flows at high spatial
resolution (w10 mm), an order of magnitude
smaller than the ultrasound diffraction limit, and at
depths much greater than the traditionally
frequency-limited imaging depth.8 However, coro-
nary network arrangement is complex and arbitrarily
oriented in all 3 dimensions, which greatly limits 2-
dimensional (2D) ULM assessed in cross-sections.
Furthermore, deformation of the heart as well as
large and rapid out-of-plane motion induced by
physiological events, mostly cardiac and respiratory
motions, create significant errors in localization9 and
cannot be efficiently corrected without 3-dimensional
(3D) imaging.
Here, we overcome these issues and demonstrate
coronary ultrasound localization microscopy
(CorULM) for direct visualization and quantification
of the microvascular flows in beating perfused rat
hearts. Our approach relies on volumetric ultrafast
imaging with a matrix transducer connected to a high
channel–count ultrafast electronics.10 3D-ULM is used
to track the individual MB and quantify the flow ve-
locity in the entire vasculature. 11 A multiscale 3D
motion correction is applied on the localization cen-
ter over a few tens of cardiac cycles, which enabled
reconstructing the microvascular anatomy, flow
velocity, and flow rate of beating hearts (Central
Illustration) under normal conditions, during vasodi-
lator adenosine infusion, and during coronary occlu-
sion. Finally, we demonstrate the in vivo feasibility of
CorULM on closed-chest animals.
METHODS
LANGENDORFF ISOLATED PERFUSED HEART:
EXPERIMENTAL MODEL. Sprague-Dawley male rats
(189  18 g; N ¼ 6) were anesthetized using intra-
peritoneal injection of ketamine (80 mg/kg) and
xylasine (10 mg/kg). The heart was quickly removed
and placed in oxygenated Krebs-Henseleit solution.
The aorta was cannulated and perfused using the
Langendorff method with oxygenated Krebs-
Henseleit solution at constant pressure (60 mm Hg,
temperature 37  0.2  C).12 A latex water-filled balloon
was inserted into the left ventricular chamber and
connected to a pressure transducer for continuous
measurement of isovolumic left ventricular pressure
(LVP), end-diastolic pressure, left ventricle (LV)
developed pressure, and heart rate. All the parame-
ters were recorded using acquisition and analysis
software (Labchart, ADInstruments). The hearts were
electrically stimulated at 350 beats/min via 2 silver
wires; 1 on the right atria, 1 on the LV connected to a
Function Generator (Tektronix).
JACC: CARDIOVASCULAR IMAGING, VOL. -, NO. -, 2022
- 2022:-–-
The perfusion system (isolated heart system, Emka
Technologies) was used to control the perfusion
pressure and flow rate. In constant pressure mode,
the device automatically adapted the coronary flow
by changing the peristaltic perfusion pump velocity.
The perfusion flow rate was computed from the pump
velocity after calibration. Cardiac arrest was obtained
by decreasing temperature of solution containing the
heart and perfusion solution from 37 C to 4  C.
Contrast agent (Sonovue, Bracco) at 0.33% dilution
was added in the perfusion solution. Vasodilation was
-5
obtained by adding adenosine at 10 mol/L under a
constant pressure of 120 mm Hg. Occlusion of the left
anterior descending (LAD) artery was performed by
ligating the artery at the basal-mid level of the LV. In
this experiment, the flow rate was maintained con-
stant and the perfusion pressure was automatically
adapted by the system.
IN VIVO EXPERIMENTS. Transthoracic in vivo imag-
ing of 3 Sprague-Dawley males (208  2.3 g)
was performed under 2% isoflurane anesthesia in
0.8 L/min 70% O 2/30% air. Before positioning the
matrix transducer, a conventional transthoracic
echocardiography was performed with a 12 MHz
linear array (Superlinear 20-6, Aixplorer, SuperSonic
Imaging). MBs were injected by insertion in the
caudal vein of a 25-gauge catheter connected to a
syringe positioned on a syringe withdraw. Electro-
cardiogram (ECG) and breathing were recorded using
subcutaneously positioned electrodes for ECG and a
spirometer connected to an anesthesia face mask for
breathing and the signals were computed in real-time
to synchronize the ultrafast imaging acquisition. All
animals received humane care in compliance with the
2010 European Communities Council Directive (2010/
63/EU), and the study was approved by the institu-
tional and regional committees for animal care
(Comité d’Ethique number 59 “Paris Centre et Sud,”
APAFIS#27733-2020101822261518v2).
3D ULTRAFAST ULTRASOUND IMAGING. Isolated
perfused hearts imaging. The hearts were immersed in
the Krebs-Henseleit solution in a pool covered by a
latex membrane. The transducer was positioned in
the apical view orientation and ultrasound gel was
applied between the membrane and the transducer
(Figure 1A). The placement of the transducer was
controlled using real-time B-mode imaging. A pulse repetition frequency of 6,400 Hz and a frame
customized, programmable, 1,024-channel ultrafast
ultrasound system described previously 10,13
was used
to drive a 32 x 32 matrix-array probe centered at 9
MHz (60% bandwidth@6dB). The system was
composed of 4 research electronics (Vantage, Vera-
sonics), which were assembled and synchronized,
Demeulenaere et al 3
Coronary Microcirculation Assessment by Ultrasound
defining a single device containing 1,024 channels in
transmission and reception. An ultrafast Doppler
imaging sequence consisting of 9 plane waves defined
by pair of angles (i,j) with i and j in (-5.7 , 0 , 5.7  ) as
depicted in Figure 1A was transmitted at a pulse
repetition frequency of 20 kHz during 270 ms. Back-
scattered echoes of cardiac tissues and circulating
MBs were received by each element of the transducer
and the radiofrequency signals were stored in mem-
ories. Receive beamforming and coherent com-
pounding10 were performed on radiofrequency data
to obtain in-phase and quadrature (IQ) demodulated
volumes at an effective imaging rate of 2,220 vol-
umes/s. The high temporal volume rate is paramount
for accurately quantifying the flow velocity. Volumes
of 24  13  13 mm 3 were reconstructed (depth and
lateral directions, respectively) with a voxel size of
0.862  0.197  0.197 mm 3.
Blocks of w100 frames (w50 ms) were then
selected in diastole using both LVP signal and axial
tissue velocity. This selection is described in more
detail in the Supplemental Methods. Each block of IQ
data was filtered using a spatio-temporal clutter-filter
(singular value decomposition) 14 by removing the
first 30 eigenvectors. This sequence was repeated 100
times, each repetition being synchronized with the
LVP signal at the beginning of the systolic phase.
In vivo imaging. The same methods were used for
in vivo acquisitions. In addition to the ECG synchro-
nization, a gating was performed on breathing pres-
sure to trig the acquisition at the R-peaks and when
breathing pressure was low (end of exhalation). The
rat heart rate was about 300 beats/min so that each
block covered at least 1 complete cardiac cycle. The
data were selected in a 37 ms time window in early-
diastole.
2D ULTRAFAST ULTRASOUND IMAGING. Nonbeating
isolated hearts were prepared by cooling the perfused
solution down to 4  C. The probe was placed in long
axis view of the left ventricle. 2D imaging was per-
formed using a linear probe (128 elements, 0.11 mm
pitch, centered at 15.6 MHz) driven by a research
electronics (Vantage 256, Verasonics) using 128 of 256
channels in transmission and in reception. The ul-
trafast imaging sequence consisted in 8 plane waves
transmitted at angles (-7  ; -5  ; -3  ; -1  ; 1  ; 3  ; 5  ; 7  ) at a
rate of 800 Hz during 625 ms. This sequence was
repeated 960 times giving a total accumulation time
of 10 min.
3D-CorULM PROCESSING AND MOTION CORRECTION.
The basic principle of 3D-CorULM along with a lay-out
of the approach used to image the hearts is depicted
4 Demeulenaere et al JACC: CARDIOVASCULAR IMAGING, VOL. -, NO. -, 2022
Coronary Microcirculation Assessment by Ultrasound - 2022:-–-

C E NT R AL IL L U STR AT IO N CorULM

Microbubbles 3D Ultrasound Localization Microscopy

corULM of the rat
contrast agent
3D Ultrasound IQ volume After SVD clutter filter coronary vasculature
Rat beating
heart t1
t2
t3

t4
2 mm Motion
500 μm
correction

z
y

Matrix array
ultrasound probe

250 μm

2 cm

Microbubble center position localization and tracking

Paves the way for new

Visualizes directly Provides quantitative flow
microcirculation functional
microcirculation (ø 20 - 100 μm) measurements
assessment with ultrasound imaging

Demeulenaere O, et al. J Am Coll Cardiol Img. 2022;-(-):-–-.

CorULM relies on the injection of MB contrast agents into the coronary vasculature and 3D ultrafast ultrasound imaging with a matrix array pobe. Thanks to its high-
volume rate, individual MBs are localized and tracked over time. After correction of cardiac contraction and breath motion, CorULM provides 3D super-resolved images
of the anatomy and flow velocity of the coronary vasculature in beating hearts at a scale down to microvascular structures (<20 mm resolution). CorULM also provides
insights into the function of coronary arteries at the microvasculature level. This new technology is highly translational and has the potential to become a major tool
for the clinical investigation of the coronary microcirculation. 3D ¼ 3-dimensional; CorULM ¼ coronary ultrasound localization microscopy; IQ ¼ in-phase and
quadrature; MB ¼ microbubble; SVD ¼ singular value decomposition.

in Figure 1A, B, and C. 3D-CorULM was based on
several successive processing steps. All the steps are
fully described in the Supplemental Methods.
VISUALIZATION AND QUANTIFICATION. 3D repre-
sentations of CorULM results were computed using
the volren function of Amira (Amira v.6.0.1 software,
FEI). Automatic quantification of vessel diameters
was performed on density maps reconstructed with
isotropic voxels of 40 mm. A Gaussian filter was
applied (s ¼ 40 m m) and the skeletonization of the
vasculature was performed using the “auto-skeleton”
function in Amira. As a result of the skeletonization,
the diameters were quantified at every centerline
point detected.
For the flow quantification, we assumed the
maximum velocity Vmax of the velocity profile to be at
the skeleton centerline. The flow rate was eventually
computed assuming a Poiseuille flow: Q ¼ pr 2  Vmax/2
(Q: flow rate, r: radius) for every vessel.
To quantify the effect of vasodilation on micro-
vascular vessels, we computed the variation of vessel
radius, flow velocity, and flow rate in response to
adenosine. Skeletons were extracted from acquisi-
tions in baseline and under adenosine-induced
vasodilation and spatially registered using Coherent
Point Drift registration.15 After registration, vessels
whose centerline points were closer than 150 m m were
considered identical in the 2 acquisitions. We then
performed several analyses of the flow and diameter
variation for different classes of vessels.
In a first analysis, we quantified the radius, veloc-
ity, and flow rate in a class of microvascular vessels.
We extracted all the vessels between 30 and 40 m m in
radius at baseline, and computed the median value of
radius, velocity, and flow rate in baseline and during
adenosine infusion for each heart (N ¼ 6). In a second
analysis, we investigated the vasodilation as a func-
tion of vessel size. We extracted vessels of different
JACC: CARDIOVASCULAR IMAGING, VOL. -, NO. -, 2022 Demeulenaere et al 5
- 2022:-–- Coronary Microcirculation Assessment by Ultrasound

F I G U R E 1 3D-CorULM Principles

A Set-Up and Acquisition

Diluted microbubbles Perfusion of Krebs−

solution (0.3%) Henseleit solution

Plane Wave
Compounding Acquisition Synchronized with
Left Ventricular Pressure (LVP)

Acquisition

LVP [mm Hg]

120

90
matrix array
(32 × 32 elements) 100 200
ultrasound probe Time [ms]

B 3D Ultrasound Localization Microscopy

3D ultrasound IQ volume After SVD clutter filter

t1

t2

t3

t4
2 mm

500 μm

250 μm

Continued on the next page

Continued on the next page
6 Demeulenaere et al JACC: CARDIOVASCULAR IMAGING, VOL. -, NO. -, 2022
Coronary Microcirculation Assessment by Ultrasound - 2022:-–-

F I G U R E 1 Continued

C Multi Scale Motion Correction

LVP [mm Hg]
Motion correction 120
between diastole
90 Time [ms]
100 200 300

STEP 1: Coarse 3D
registration from
Power Doppler volumes

STEP 2: Sub-wavelength correction STEP 3: Accumulation of

from micro bubbles coordinates corrected coordinates

(A) Experimental set-up: plane waves are transmitted at high repetition rate at different angles during diastole in a perfused isolated beating
rat heart. (B) 3D ULM processing: after SVD filtering, MBs are localized with a 3D paraboloid bn and are tracked over acquisition times ti to
determine tracks. (C) 3D intensity-based rigid registration is computed to correct motion between diastolic phases. Then, a fine ICP correction
is applied at a subwavelength scale. Corrected MB coordinates are eventually aggregated to create a 3D vascular volume.
3D ¼ 3-dimensional; CorULM ¼ coronary ultrasound localization microscopy; ICD ¼ iterative closest point; IQ ¼ in-phase and quadrature;
LVP ¼ left ventricular pressure; MB ¼ microbubble; SVD ¼ singular value decomposition; ULM ¼ ultrasound localization microscopy.

classes of radius (20-30; 30-40; 40-50;
>60 mm). For each vessel we computed its radius
variation as the ratio of radius (radenosine/r baseline) and
quantified the median value per class and per heart.
In a third analysis, we computed the flow rate varia-
tion as the ratio (qadenosine/q baseline ) for each vessel
and compared the median value with the ratio of
perfusion flow measured using the perfusion system
during the same experiment.
For 2D acquisitions, density and velocity maps
were reconstructed with a pixel size of 5 x 5 m m 2. We
analyzed the flow velocity profile across the section of
a large vessel. The vessel cross-section was divided
into 5 segments of the same length (26 m m). In each
segment, we evaluated the velocity of individual
flowing MBs assuming that each MB was an inde-
pendent event.
50-60; determine the exponent of the flow rate-radius rela-
tion. Median values of radius, velocity, and flow rate
in baseline and under vasodilation were compared for
each heart (N ¼ 6) using a paired, 1-sided Student’s t-
test. Relative variations of radius were compared for
several classes of vessels with a paired, 1-sided Stu-
dent’s t-test. Eventually, variations of flow rate were
compared with the control measurements with a
paired, 2-sided Student’s t-test. For all box plots,
center line (median), box (first and third quartiles),
and whiskers, 1.5  IQR with all data points,
were individually plotted. Statistical significance
was inferred for P < 0.05 with no significance (NS):
*P < 0.05, **P < 1e-2, ***P < 1e-3, ****P < 1e-4. Values
are presented as mean  SD unless specified.
RESULTS

STATISTICS. Spatial variations of velocity profiles
were analyzed using unpaired 2-sided Student’s t-test
to evaluate the significance of difference at several
locations of the flow profile assuming each MB event
to be independent as explained in the Methods sec-
tion. Linear regression was used on log-values to
CORONARY ANATOMY. CorULM provides high-
resolution images of the whole heart coronary
vasculature as shown in Figure 2Ai. When zooming at
a smaller sale, the coronary microcirculation is shown
in Figure 2Aiv. 3D-CorULM images can be best visu-
alized in Video 1. Note that the spatial resolution in
JACC: CARDIOVASCULAR IMAGING, VOL. -, NO. -, 2022 Demeulenaere et al 7
- 2022:-–- Coronary Microcirculation Assessment by Ultrasound

F I G U R E 2 CorULM of Perfused Rat Hearts

3D CorULM of the Isolated Beating Rat Heart

A B C

D E F

G H I J 10−1

100
log(y) = 2,61log(x) − 13,35
Flow Rate (mL/min)

R2 = 0.96
10−1

10−2

10−3

10−4
101 102
Radius (μm)

(A to J) Example of 3D-CorULM of a perfused beating rat heart. (A) 3D-Density map, (B) directional flow velocity map that identifies the arterial (red) and venous (blue)
flow, and (C) 3D-flow velocity distribution. Microvessels are better visualized at smaller scale. (D) 3D-density map, (E) quantification of the diameter, and (F) velocity
after skeletonization. For comparison, the low-resolution Power Doppler map is shown in (G) as well as micro computed tomography imaging (H) and gross photography
(I) of the same fixed heart. (J) Analysis of flow rate as a function of radius in a broad range of vessels reveals a power law with an 2.61 exponent. (K to M) 2D-CorULM
density maps (K) and velocity (L) obtained in a nonbeating rat heart (LV-free wall long-axis view). (M) A cross-section of several vessels (1, 2, and 3 annotated in (K))
is performed on the density map to quantify the diameters. (N) Analysis of the velocity profile in a large vessel (vessel 4 in (L)) shows a Poiseuille-like flow profile.
2D ¼ 2-dimensional; LV ¼ left ventricle; other abbreviations as in Figure 1.

Continued on the next page

8 Demeulenaere et al JACC: CARDIOVASCULAR IMAGING, VOL. -, NO. -, 2022
Coronary Microcirculation Assessment by Ultrasound - 2022:-–-

F I G U R E 2 Continued

Spatial Resolution and Quantification of 2D CorULM

K M
Density Profiles

12 3
10
2

Density
8 1
6 61 μm
4 23
16
2 μm
μm 10 μm
0
Distance [μm]

L N Velocity Profile
50 n = 184
n = 92 n = 120
40

Velocity [mm/s]
30 n = 36 n = 48
+
+
20
10
0
****
**** **** ****

14 40 66 92 118
Distance [μm]

CorULM depends on the number of MBs passing
through each reconstructed voxel and thus could
potentially be improved by using longer acquisition
time. Approximately 10 MBs are localized in each
reconstructed voxel (20 x 20 x 20 m m3 ) in the present
result. 3D spatial registration was performed to cor-
rect the motion (Supplemental Figure 1). Quantifica-
tion of vessel diameters is performed automatically
by the skeletonization algorithm. Diameters range
from 20 m m for tiny arterioles to w300 m m for larger
branches of the right and left coronary arteries
(Figure 2Av). The coronary vasculature is compared
with micro computed tomography performed on the
fixed heart (Figure 2Aviii). The vessel radius is quan-
tified on 15 large vessels (134  35 m m) visualized in
both modalities (Supplemental Figure 3) with a mean
absolute error of 12 m m between the 2 modalities.
Gross photography of the fixed heart shows the
overall anatomy of the rat heart (Figure 2Aix).
For comparison, CorULM is performed in 2D on
arrested hearts using high-frequency imaging. Note
that in 2D, CorULM cannot be performed on the
beating hearts because of strong out of plane motion
artefacts and requires the hearts to be arrested using
low-temperature liquid perfusion. The vasculature
and flow velocity images of the left ventricular free
wall are shown in Figure 2Bi and Figure 2Bii. Vessels as
small as 16 m m are visible (Figure 2Biii). It was further
possible to accurately assess the velocity profile
across a relatively large vessel, revealing a Poiseuille-
like flow profile (Figure 2Biv).
CORONARY FLOW ASSESSMENT. The high temporal
resolution of the 3D ultrafast imaging system further
enables estimating the flow velocity via particle
tracking (see Methods section for a detailed descrip-
tion). Figure 2Aii shows the base to apex flow direc-
tion: downward (red) corresponding mainly to
arterial flow and upward (blue) to veinous flow. Ab-
solute flow velocity estimates range from 10 mm/s in
tiny arterioles up to more than 300 mm/s in large
arteries (Figure 2Aiii and 2Avi). Flow velocity can be
better visualized in Video 2. CorULM is compared
with low-resolution power Doppler imaging
(Figure 2Avii) obtained with the same IQ data. 16 Flow
velocity were estimated as a function of perfusion
pressure (Supplemental Figure 3).
The flow rate–radius relation is shown in
Figure 2Ax for a large range of arterial radius (10-
JACC: CARDIOVASCULAR IMAGING, VOL. -, NO. -, 2022 Demeulenaere et al 9
- 2022:-–- Coronary Microcirculation Assessment by Ultrasound

F I G U R E 3 Flow Quantification

A
3D Mapping of Radius and Flow Velocity During Adenosine Infusion
Baseline Adenosine

2 mm 2 mm

Right coronary arteries [μ m]

160

140

120

100
Radius

80

60

40

20
Left coronary arteries

2 mm

[mm.s−1]
300

250

200
Flow Velocity

150

100

50

2 mm

(A) Quantification of vessel radius and flow velocity in baseline (left) and during adenosine-induced vasodilation (right). (Top) Flow velocity
overlaid onto the density map (grayscale). (Middle) Quantification of radius in a ROI. (Low) Quantification of flow velocity in a ROI. (B) The
velocity distribution is shifted toward higher velocities during adenosine-induced vasodilation. (C) Radius, flow velocity, and flow rate
computed on an ensemble of microvascular vessels (Ø 60-80 mm) increase significantly with adenosine. (D) Relative variations of radius are
analyzed for different classes of vessels and are significantly higher in small vessels. (E) The relative increase of flow rate (N ¼ 6), which is in
good agreement with the control measurement of the perfusion system. CFR ¼ coronary flow reserve; ROI ¼ region of interest.

Continued on the next page

10 Demeulenaere et al JACC: CARDIOVASCULAR IMAGING, VOL. -, NO. -, 2022
Coronary Microcirculation Assessment by Ultrasound - 2022:-–-

F I G U R E 3 Continued

B ×10−3 Flow Velocity Distribution

Probability Density
4

0
0 50 100 150 200 250 300 350 400 450
Velocity [mm.s−1]
Baseline Adenosine

C Median Radius, Flow Velocity and Flow Rate in [Ø 60-80 μm] Vessels
*** ***
**
N=6 N=6
60 N=6 150 40

50
30
Flow Velocity [mm.s−1]

40 100 Flow Rate [μL.min−1]

Radius [μm]

30 20

20 50
10
10

0 0 0
Baseline Adenosine Baseline Adenosine Baseline Adenosine

D E Microcirculation Flow
Variation of Radius for Different Classes of Vessels Variation Compared to CFR
* ×3 n.s
N=6 N=6
***
2.5
n.s
×2
* 2

1.5
1.5

1
1

0.5
0.5

0 0
20-30 30-40 40-50 50-60 >60μm Microcirculation CFR
Classes of Vessels Per Radius Flow Variation
JACC: CARDIOVASCULAR IMAGING, VOL. -, NO. -, 2022 Demeulenaere et al 11
- 2022:-–- Coronary Microcirculation Assessment by Ultrasound

F I G U R E 4 LAD Coronary Occlusion

A
3D Density Maps

Before After ligation of LAD artery

LCA LCA
RCA RCA

LAD

2 mm

B
Quantification of Flow Velocity on AHA Standardized Nomenclature

Before 600 After ligation of LAD artery

LAD LCA
LCA
500

400

300

200

100

RCA 0 RCA
Flow velocity (mm.s−1)

(A) 3D density maps before and after LAD coronary occlusion. RCA and LCA are identified. (B) The flow velocity is projected onto the bull’s-
eye American Heart Association standardized representation of the LV. LAD ¼ left anterior descending; LCA ¼ left circumflex artery;
RCA ¼ right coronary artery; other abbreviations as in Figure 1 and Figure 2.

150 m m). Fitting to a power law provides an exponent
of 2.61 (r2 ¼ 0.96; P < 0.001), which is consistent with
theoretical predictions and experimental validations
of scaling laws in vascular trees, also known as the
generalized extension of Murray’s law.17
We then investigated the flow variation
response to adenosine-induced vasodilation
(Figure 3). Analysis of the velocity histograms reveals
a shift of the velocity distribution in response to
vasodilation (Figure 3B). The median radius, flow
velocity, and flow rate (Figure 3C) analyzed in an
ensemble of microvascular vessels (w100 segments of
Ø 60-80 mm per heart; N ¼ 6 hearts) increase signifi-
cantly with adenosine. We analyze the vasodilation as
a function of the vessel size (Figure 3D) and the
relative radius variation decreased with vessel size:
small vessels (20-30 m m) are dilated by 1.57  0.26,
whereas larger vessels (>60 mm) changed only by 1.07
 0.2 (P < 0.01). The median flow rate is analyzed
over the entire vasculature and a 2-fold increase of
in the microvascular coronary flow rate is found in
response to adenosine, which is in good agreement
with the overall perfusion flow rate (control mea-
surement) that increases from 8.80  1.03 mL/min to
16.54  2.35 mL/min (P < 0.001). This ratio known as
the CFR provides an indirect functional assessment of
the microcirculation, which is used clinically. We
show here that CorULM provides a local and direct
quantification of the microvascular coronary flow
variation (Figure 3E).
12 Demeulenaere et al JACC: CARDIOVASCULAR IMAGING, VOL. -, NO. -, 2022
Coronary Microcirculation Assessment by Ultrasound - 2022:-–-

F I G U R E 5 In Vivo Imaging

A
Animal preparation Acquisition

Anesthetized rat Synchronization with

breath and ECG
Temperature, heart and
breath rate monitored

Microbubbles solution
perfusion
(Sonovue®, 6 mL/h) Matrix array
(32 × 32 elements)
ultrasound probe

B C
Trig at ECG R peak when breath motion is low
Inspiration Expiration Rest

Acquisition 1 Acquisition 2

Spirometer
signal (a.u)

ECG (a.u)

Time [s]
0 0,5 1 1,5
Selection of early diastole
Time Motion Mode
0 100 200 300 Time
[ms]
7
Depth [mm]

10

13

16
Systole DIastole

(A) Set up and acquisition: in vivo acquisitions are performed using the matrix transducer positioned on the rat chest. Panel created with BioRender.com.
(B) 2D B-mode image of in vivo beating rat heart, showing the heart structures. (C) Diastole selection during acquisition: a short time window (37 ms) in
early diastole is selected from the ultrasound data. (D-G) 3D CorULM of in vivo beating rat heart. The Power Doppler map shows the coronary and left
ventricular flow at low resolution (D), whereas coronary vessels are revealed at high resolution using 3D-CorULM (E). Quantification of radius (F) and
velocity (G) is shown at a smaller scale. ECG ¼ electrocardiogram; other abbreviations as in Figure 1 and 2.

Continued on the next page

Finally, the main LAD artery is ligated to induce a occlusion. Projection of the flow velocity on the 17-
complete occlusion of the LAD. Figure 4A shows the segment model of the American Heart Association
3D reconstruction of the coronary vasculature and the enables a clear delineation of nonperfused territory
disappearance of a large part of the LAD after (Figure 4B).
JACC: CARDIOVASCULAR IMAGING, VOL. -, NO. -, 2022 Demeulenaere et al 13
- 2022:-–- Coronary Microcirculation Assessment by Ultrasound

F I G U R E 5 Continued

D Power Doppler Volume dB E Density Map

0

−10

−20

−30

−40

−50

−60

F Radius
[μ m]
100

80

60

40

20

G Flow Velocity

[mm.s−1]
150

100

50

IN VIVO EXPERIMENTS. The feasibility of 3D-CorULM
was demonstrated in 3 closed-chest rats. The matrix
transducer was positioned on the rat chest (Figure 5A)
in parasternal view instead of the apical view because
of ergonomic limitations caused by the transducer
dimensions. Figure 5B shows that a large portion of
the LV can be reached with this orientation. One can
notice, however, that several regions are shadowed
by the ribs. In the post-processing step, a short time
window was determined in early diastole based on
the analysis of the motion of the LV wall (Figure 5C).
The 3D Power Doppler map shows the coronary and
14 Demeulenaere et al
Coronary Microcirculation Assessment by Ultrasound
left ventricular flow at low resolution (Figure 5D).
After 3D-CorULM processing, the coronary vessels are
revealed at high resolution (Figure 5E). It should be
noted that the MBs in the cavity are automatically
removed by the tracking algorithm because the bub-
ble concentration is too high. After vessel skeletoni-
zation, quantification of radius (Figure 5F) and flow
velocity (Figure 5G) are performed and visualized. The
radius of the vessels ranged from w30 m m for small
arterioles to w100 m m for larger branches of the cor-
onary arteries. Corresponding flow velocity estimates
ranged from 10 mm/s up to 300 mm/s with a negative
gradient toward small branches.
DISCUSSION
We hereby demonstrate that CorULM can visualize
and quantify coronary microvascular flows in the
beating heart. The ability to image in 3D and subse-
quently to apply motion correction strategies em-
powers the recently developed ULM approach and
enables imaging of the microvascular flows in beating
hearts. 3D-ultrafast imaging is an important prereq-
uisite for successful application of CorULM as high-
volume imaging rate (2220 volumes/s) is key to
correct 3D tissue motion and track successfully the
MBs. Imaging of tiny coronary vessels <20 mm in
diameter was achieved across the whole heart
(w20 mm depth). Ultrafast imaging was previously
used to image coronary vessels of about 100 m m.18 We
demonstrate here that CorULM can drastically
improve the spatial resolution by 1 order of magni-
tude. Apart from its superb spatial resolution and
mapping of the whole coronary vasculature with ac-
curate diameter quantification, CorULM enables
quantification of flow rate and velocity.
The flow rate increase in microvascular vessels is
demonstrated and quantified after adenosine-
induced vasodilation. It allows the direct functional
assessment of the microvascular coronary arterioles
and is correlated to the CFR. To our knowledge, it is
the first imaging modality that can provide direct
anatomical and functional visualization of coronary
microvascular flow of the whole heart. The micro-
vessels imaged using CorULM correspond to coronary
arteries and arterioles (20-300 mm), which have a
fundamental role in the metabolic regulation of cor-
onary blood flow and are the site of regulation of flow
resistance and thus cardiac perfusion.
In a broader perspective, CorULM can be used to
facilitate early diagnosis based on biomarkers asso-
ciated with microcirculatory alterations in cardio-
vascular disorders, while additionally providing new
insights into disease progression, efficacy of drugs,
JACC: CARDIOVASCULAR IMAGING, VOL. -, NO. -, 2022
- 2022:-–-
and other therapeutic interventions. CorULM should,
therefore, strongly benefit fundamental research on
coronary microvascular dysfunction and the devel-
opment of effective treatments. The setup of isolated
retrograde perfused heart is commonly used as a
preclinical model in cardiovascular research in phys-
iology and drug developments and is particularly
useful for the ex vivo investigation of ischemia-
reperfusion injury. This ex vivo model has proved to
be an irreplaceable and highly adaptable tool
for research into coronary vascular function in
physiological, pathologic, or pharmacological in-
vestigations, and into myocardium biochemistry and
metabolism.19 CorULM is particularly well adapted to
isolated perfused heart of rats or mice because it
provides the unique capability to visualize and
quantify the microvascular flows of an entire heart
with little modification of the Langendorff setup.
In vivo feasibility of transthoracic CorULM was
demonstrated on 3 rats by slightly modifying the
acquisition system to take into account the respira-
tory motion. The potential of CorULM for the diag-
nostic of CMD will be further investigated on
pathologic animal models. In the long term, trans-
lation to the human heart could be achieved with
lower frequency matrix transducers. 3D-ultrafast im-
aging of the human heart with a large field of view has
been recently demonstrated using matrix transducers
with diverging wave emissions 13,20,21 and could be
applied to CorULM.
Like any other cardiac imaging modalities, CorULM
has limitations. First, it requires acquisitions dura-
tions of a few tens of cardiac cycles (about 10s). Sec-
ond, the current spatial resolution, estimated at 20 x
20 x 20 m m 3 could be improved by using matrix arrays
with a refined one-wavelength spatial pitch leading to
large angular aperture and smaller focal spots or by
increasing frequency up to 15 MHz. The concentration
and volume of Sonovue used in this study should not
be extrapolated directly to human applications.
Further studies will be required to determine the
optimal concentration and volume for coronary im-
aging application in large animals and human pa-
tients. Finally, we analyzed the coronary flow only
during the diastolic phase instead of the full cycle. In
fact, there is also flow during systole that could be
identified during isovolumic systole and during late
systole but the phasic blood flow pattern differs be-
tween large epicardial and penetrating coronary ar-
teries. Indeed, mid-systole coronary blood flow
velocity is retrograde in the intramyocardial arteries
but antegrade in large epicardial arteries. Neverthe-
less, CorULM imaged small vessels where it is known
that the majority of coronary blood velocity occurs
JACC: CARDIOVASCULAR IMAGING, VOL. -, NO. -, 2022 Demeulenaere et al 15
- 2022:-–- Coronary Microcirculation Assessment by Ultrasound

during diastole. 22 Further technological and imaging
development will allow achievement of full cycle
analysis, taking into account the overall pattern of
blood flow during the whole cardiac cycle.
3D-ULM was applied to coronary mapping in this
study but could be translated to any vascularized
organs such as the brain or the liver. For instance, it
could have a crucial impact on brain tumor applica-
tions, detecting early the angiogenesis responsible for
recurrence after glioblastoma tumor resection.
European Research Council Grant Agreement number 311025 and by
the Fondation Bettencourt-Schueller under the program “Physics for
Medicine.” The authors have reported that they have no relationships
relevant to the contents of this paper to disclose.
ADDRESS FOR CORRESPONDENCE: Dr Mathieu Per-
not, Physics for Medicine, Ecole Supérieure de Physique
Chimie Industrielles de Paris, 17 rue Moreau, 75012 Paris,
France. E-mail: mathieu.pernot@inserm.fr.
PERSPECTIVES

CONCLUSION
We developed CorULM to image and quantify the
microvascular coronary flow of beating hearts, which
provided unprecedented insights into the anatomy
and function of coronary arteries. We anticipate that
the advantages provided by CorULM will massively
impact a large number of studies on clinically rele-
vant coronary diseases and further foster the growing
use of ultrafast ultrasound as a biomedical imaging
tool.
COMPETENCY IN MEDICAL KNOWLEDGE: Patients with
CMD have poor prognoses with significantly higher rates of
cardiovascular events, including hospitalization for heart failure,
sudden cardiac death, and MI. Direct visualization of the full
coronary architecture including complex microvascular network
remains out of reach of current cardiac imaging modalities mainly
caused by inadequate temporal resolutions or sensitivity limita-
tions. The results of this study show that direct and noninvasive
imaging of the coronary microcirculation at the microscopic scale
could be performed with ultrasound combined with MBs. It may

ACKNOWLEDGMENTS We acknowledge the ART offer a nonionizing and portable technique for the visualization

(Technological Research Accelerator) biomedical ul- and quantification of microvascular flows.

trasound program of INSERM. The authors thank Lotfi

TRANSLATIONAL OUTLOOK: The major impact of this study
Slimani for the support with micro computed to-
is the development of 3D-CorULM to visualize and quantify the
mography acquisitions and fruitful discussions.
coronary microcirculation on a beating heart. This technology is
FUNDING SUPPORT AND AUTHOR DISCLOSURES highly translational and has the potential to become a major tool
for the clinical investigation of the coronary microcirculation.
This study was supported by the European Research Council under
the European Union’s Seventh Framework Program (FP/2007-2013)/

REFERENCES

1. Camici PG, d’Amati G, Rimoldi O. Coronary
microvascular dysfunction: mechanisms and func-
tional assessment. Nat Rev Cardiol. 2015;12:48–
62.
2. Gould KL, Johnson NP, Bateman TM, et al.
Anatomic versus physiologic assessment of cor-
onary artery disease. Role of coronary flow
reserve, fractional flow reserve, and positron
emission tomography imaging in revasculariza-
tion decision-making. J Am Coll Cardiol. 2013;62:
1639–1653.
3. Henningsson M, Shome J, Bratis K, et al. Diag-
nostic performance of image navigated coronary
CMR angiography in patients with coronary artery
disease. J Cardiovasc Magn Reson. 2017;19:68.
4. Watanabe N, Akasaka T, Yamaura Y, et al.
Noninvasive detection of total occlusion of the left
anterior descending coronary artery with trans-
thoracic Doppler echocardiography. J Am Coll
Cardiol. 2001;38:1328–1332.
5. Fihn SD, Blankenship JC, Alexander KP, et al.
2014 ACC/AHA/AATS/PCNA/SCAI/STS focused
update of the guideline for the diagnosis and
management of patients with stable ischemic
heart disease: a report of the American College
of Cardiology/American Heart Association Task
Force on Practice Guidelines, and the American
Association for Thoracic Surgery, Preventive Car-
diovascular Nurses Association, Society for Car-
diovascular Angiography and Interventions, and
Society of Thoracic Surgeons. J Thorac Cardiovasc
Surg. 2015;149:e5–e23.
6. Villemain O, Baranger J, Jalal Z, et al. Non-
invasive imaging techniques to assess myocardial
perfusion. Expert Rev Med Devices. 2020;17:1133–
1144.
7. Errico C, Pierre J, Pezet S, et al. Ultrafast ul-
trasound localization microscopy for deep super-
resolution vascular imaging. Nature. 2015;527:
499–502.
8. Couture O, Hingot V, Heiles B, Muleki-Seya P,
Tanter M. Ultrasound localization microscopy and
super-resolution: a state of the art. IEEE Trans-
actions on Ultrasonics, Ferroelectrics, and Fre-
quency Control. 2018;65:1304–1320.
9. Cormier P, Porée J, Bourquin C, Provost J. Dy-
namic myocardial ultrasound localization angiog-
raphy. IEEE Transactions on Medical Imaging.
2021;40:3379–3388. https://doi.org/10.1109/TMI.
2021.3086115
10. Provost J, Papadacci C, Esteban Arango J,
et al. 3D ultrafast ultrasound imaging in vivo. Phys
Med Biol. 2014;59:L1–L13.
11. Heiles B, Correia M, Hingot V, et al. Ultrafast
3D ultrasound localization microscopy using a
3232 matrix array. IEEE Trans Med Imaging.
2019;38:2005–2015. https://doi.org/10.1109/
TMI.2018.2890358
12. Abi-Gerges A, Richter W, Lefebvre F, et al.
Decreased expression and activity of cAMP phos-
phodiesterases in cardiac hypertrophy and its
impact on beta-adrenergic cAMP signals. Circ Res.
2009;105:784–792.
13. Papadacci C, Finel V, Villemain O, Tanter M,
Pernot M. 4D ultrafast ultrasound imaging of
naturally occurring shear waves in the human heart.
IEEE Trans Med Imaging. 2020;39:4436–4444.
14. Demené C, Deffieux T, Pernot M, et al.
Spatiotemporal clutter filtering of ultrafast ultra-
sound data highly increases Doppler and ful-
trasound sensitivity. IEEE Trans Med Imaging.
2015;34:2271–2285.
16 Demeulenaere et al
Coronary Microcirculation Assessment by Ultrasound
15. Myronenko A, Song X. Point set registration:
coherent point drift. IEEE Transactions on Pattern
Analysis and Machine Intelligence. 2010;32:2262–
2275.
16. Correia M, Maresca D, Goudot G, et al. Quan-
titative imaging of coronary flows using 3D ultra-
fast Doppler coronary angiography. Phys Med Biol.
2020;65:105013.
17. Kassab GS. Scaling laws of vascular trees: of
form and function. Am J Physiol Heart Circulatory
Physiol. 2006;290:H894–H903.
18. Maresca D, Correia M, Villemain O, et al.
Noninvasive imaging of the coronary vasculature
JACC: CARDIOVASCULAR IMAGING, VOL.
using ultrafast ultrasound. J AmColl Cardiol Img.
2018;11:798–808.
19. Skrzypiec-Spring M, Grotthus B, Szela˛g A,
Schulz R. Isolated heart perfusion according to
Langendorff—still viable in the new millen-
nium. J Pharmacol Toxicol Methods. 2007;55:
113–126.
20. Papadacci C, Finel V, Villemain O, et al. 4D
simultaneous tissue and blood flow Doppler im-
aging: revisiting cardiac Doppler index with single
heart beat 4D ultrafast echocardiography. Phys
Med Biol. 2019;64:085013.
21. Salles S, Espeland T, Molares A, et al. 3D
myocardial mechanical wave measurements:
-, NO. -, 2022
- 2022:-–-
toward in vivo 3D myocardial elasticity mapping.
J Am Coll Cardiol Img. 2021;14:1495–1505.
22. Chilian WM, Marcus ML. Phasic coronary blood
flow velocity in intramural and epicardial coronary
arteries. Circ Res. 1982;50:775–781.
KEY WORDS blood flow, coronary,
imaging, microcirculation, microscopy,
ultrasound, volumetric imaging
AP PEN DIX For an expanded Methods
section, as well as supplemental figures and
videos, please see the online version of this
paper.