Lone

R. Pelc,

Gary

H. Gloyer,

PhD #{149} Norbert PhD #{149} Robert

Arterial Noninvasive

J. Pelc, ScD J Herfkens,

Venous

and

Index

terms:

Blood,

961.91,

966.91,

flow dynamics, 984.91 #{149} Blood,

CCURATE

I

From

assessment

904.91,

MR.

the

=

185:809-812

Departments

of Radiology

gery (5CR., L.J.C., D.C.M.), Richard aging, Stanford University School of April 2, 1992; revision requested June part by grants to the Department of Phil N. Allen Trust, and GE Medical ‘ RSNA, 1992

(L.R.P.,

of blood

flow

the major vessels is an essential aspect of cardiovascular evaluation. Current noninvasive techniques have limitations such as the use of ionizing radiation, contrast agents, or both or limited vascular access. Many techniques have been reported for measurement of blood flow by means of magnetic resonance (MR) imaging (1-5). However, the in vivo accuracy of these MR measurements has not been firmly established by comparison with simultaneous measurement of volume flow performed with a technique proved to be accurate. Phase-contrast, gradient-echo MR imaging methods are well suited for measurement of blood flow because of their speed, sensitivity to blood flow across a broad range of flow rates, and inherent quantitative foundation. To measure blood flow in one direction, two complete sets of raw MR data are obtained with pulse sequences that have a different gradient first moment in the direction of interest. For blood flow that is uniform throughout a voxel and constant throughout the image, the change in the phase of the transverse magnetization in a voxel is proportional to the flow velocity in the encoded direction. Phase-contrast data can be used to produce an image in which the pixel intensity depends on the magnitude of the transverse magnetization in each voxel (magnitude images) and a velocity image proportional to the phase shift. The measured phase shift 4 is converted to velocity v by the following equation: V

1992;

Flow: with MR

through

904.1299, 957.1299, 961.1299, 966.1299, 984.1299 #{149} Blood, US, 904.12989, 957.12989, 961.12989, 966.12989, 984.12989 #{149}Magnetic resonance (MR), phase imaging

Radiology

Blood

Quantltatlon

Quantitative measurements of arterial and venous blood flow were obtamed with phase-contrast cine magnetic resonance (MR) imaging and compared with such measurements obtained by means of implanted ultrasound (US) blood flow probes in anesthetized dogs. The US flowmeter was enabled during a portion of each MR imaging sequence to allow virtually simultaneous data acquisition with the two techniques. MR imaging data were gated by means of electrocardiography and divided into 16 phases per cardiac cycle. The rates of portal venous blood flow measured with MR imaging and averaged across the cardiac cycle (710 mL/ mm ± 230 [standard deviation]) correlated well with those measured with the flowmeter and averaged in like fashion (751 mL/min ± 238) (r = .995, slope 1.053). The correspondence in arterial blood flow was almost as good. No statistically significant difference existed between the paired measurements of blood flow obtained with MR imaging and the implanted probe. It is concluded that, as a noninvasive means of accurate quantification of blood flow, phase-contrast MR imaging may be especially useful in deep blood yessels in humans.

957.91,

J.

Stephen C. Rayhill, MD #{149}Luis Castro, MD MD #{149}D. Craig Miller, MD #{149}R. Brooke Jeffrey,

#{149}

N.J.?.,

G.H.G.,

R.J.H.,

R.B.J.)

-yz.M1’

and

Cardiothoracic

(1)

MD

Imaging’ where ment

Mu1 is the change in the two sequences

in first and

moy is

gyromagnetic ratio. The flow-encoding strength is sometimes parameterized by Venc, the velocity that produces a phase shift of 180#{176}. For quantitative flow measurements, flow encoding is generally performed in the section-select direction, a region of interest (ROI) that includes the vessel of interest is defined, and the product of the average velocity within the vessel and the vessel area the

yields

the

flow

rate

in milliliters

per

minute. To measure pulsatile blood flow, this technique is combined with cardiac gating (3) or cine imaging (5). Velocity images have noise that depends on the strength of the flow encoding and on the signal, S. and noise, if, in the magnitude image (5). The standard deviation (SD) in the velocity, (Tv, is calculated as follows: Venc

Thus,

noise

0

(2)

ffv=

ITS

in flow

measurements

is

minimized whenever velocity encoding (Venc) 5 as small as possible. However, because phase shifts can be uniquely measured only over a range of 360#{176}, it is best to ensure that no yelocity of interest exceeds Venc (flow aliasing). Because velocity images are produced by the difference in acquisitions

that

quantitative insensitive geneities

are

almost

identical,

characteristic to magnetic and

many

other

their

is relatively field inhomoimperfec-

tions in the system. However, differential eddy current effects secondary to the gradient changes that produce flow encoding can cause undesired

Sur-

M. Lucas Center for Magnetic Resonance Spectroscopy and ImMedicine, 300 Pasteur Dr. Stanford, CA 94305-5488. Received 3; revision received July 13; accepted July 27. Supported in Radiology from the Society of Computed Body Tomography, Systems. Address reprint requests to L.R.P.

Abbreviations:

RF = radio frequency, ROI of interest, SD = standard deviation, = standard error of the mean, TE = echo time, TR = repetition time, v,. = velocity encoding.

=

region SEE

809

phase

shifts.

These

effects

appear

with

as

additive velocity errors (ie, static structures appear to have nonzero velocity) that vary only slowly across the image. Accurate measurements of blood flow must account for the possible presence of such errors. Phase-contrast measurements of blood flow have been shown to have acceptably

low

interand intraob(less than 5% and

server variability 10% of the measured blood flow, respectively [6]). The accuracy of the method has been repeatedly validated in vitro (4,6,7). Some in vivo validation has also been reported. Generally

good

agreement

has

has

Doppler

ultrasound

(US)

are

complicated

comparisons

also

been

blood flow. We studied vessels: six portal veins, two renal veins, two renal arteries, one carotid artery, and three common iliac ar-

teries.

Two to three centimeters

allowed

to recover

spontaneous

sufficiently

to

respiration.

The

Hoffman-La by Roche

Roche,

Rico) (0.25 mg/kg)

hydrochloride

(2.5-5.0

The

US flowmeter

screened

room,

(approximately

and ketamine

was

inside

2 m)

of cable

flecting surement

brackets within

minimum

were the

production

phase-

of flow rate, whereas Dopis not. Thus, the purpose of study was to validate measure-

Sunnyvale,

MR sequence

the sampling

of the

our

synchronized,

of blood flow obtained in vivo with phase-contrast cine MR imaging by means of comparison with simultaneous measurements obtained with an implanted blood flow probe.

ter samples

dogs

received

an

Because

uniformly

METHODS

was

injection

of sodium

the dogs were was maintained 1%-2%

halothane.

(model

T1OID;

NY)

were

sure

perivascular

of blood 810

One

Transonic

implanted

blood

or two

US

probes

Systems, flow.

invalid

because

of the

of blanking,

throughout

samples

were

missed

the cycle.

dog

The

Ithaca, to mea-

accuracy

obtained

half

the MR

of the last three

car-

periods was generated (9) and stored in the computer (Fig 2). All measurements obtained during flowmeter blanking were

excluded. Flowmeter measurements obtamed during all cardiac cycles throughout the MR acquisition

16 or 32 bins equally

were

spaced

grouped

into

in the cycle.

cardiac

The

trigger

read

(Flowmeter

just before until portion of each

just after sequence.

axis

nulled.

the radiofrequency

(dotted

(Gr,d)

was

The US flowmeBlanking) from the a

=

data flip

acquisition angle of

(RF) pulse.

Rou

__J1_iI_fl_JLJLfl_RJL.fl_.fl_Jl_JLJl.JLJJ_JLJI_I1._J1

Binaking

___

-

“*S’

SR

Figure 2. Schematic representation of cardiac-phase signal generation. Time elapsed since the beginning of the cycle (r) divided by the running average of three preceding RR intervals (running average RR) was en-

tered blood

into a file and flow

used

to partition

measurements

flowmeter

sample

obtained

when

the

was enabled.

The bin to which a sample was assigned was selected with use of the cardiac-phase signal. The average flowmeter measure-

ment

for all acceptable

bin and

the average

ment

during

puted.

Thus,

meter

with

with

each

waves).

first-order-moment ter was blanked

amination.

For comparisons

with

sine

imaging samples

diac of

with different section-select direction gradient waveforms (Gci,ct., dotted and solid lines) were rapidly interleaved throughout the acquisition. Phase encodings (Gptase) were in-

Phase-contrast cine MR produced 16 frames, equally spaced throughout the cardiac cycle, that comprise a composite cardiac cycle averaged throughout the ex-

cycle to the average

and anesthesia of inhalation

in each

flow measurements

Radiology

#{149}

pentothal

intubated, by means

were

(20 mg/kg),

(0.4 mg/kg).

with

blood

of the MR pulse sequence (and flowmeter blanking) were not some of the stored flowme-

intravenous

injec-

maleate

induced

and disabled shortly before acquisition por(Fig 1). Since flow probe and

data, the flowmeter samples were similarly grouped into “frames.” For this purpose, a “cardiac-phase” signal equal to the ratio of the time since the beginning of the

intramuscular

tion of acepromazine Anesthesia

(RF) intercircuit was

external blanking. To enable identification of these samples, the blanking signal was also digitized and stored in the computer.

Arterial and venous blood flow was measured in eight anesthetized mongrel dogs of either sex that weighed 20-30 kg. All

Calif).

tion of every execution therefore

ments

mea-

fed through the screen-room penetration panel to a personal computer (Macintosh Ilci; Apple Computer, Cupertino, Calif) with data acquisition hardware and software (Workbench; Strawberry Tree,

added to the US flowmeter flowmeter operation from until shortly after the data

imagmea-

to enable

was used to monitor the dog and provide synchronization for the acquisition of MR images. The flowmeter signal was sampled every 20 msec and

surements pler US

AND

the

magnetic field with of artifact. A lead II

To prevent radio-frequency ference, an external blanking

MATERIALS

7 ft

between

used

Blanking

Figure 1. Pulse sequence used for collection of phase-contrast cine MR data for measurement of blood flow. Two sequence types

cremented

the

approximately

many

contrast MR velocity measurements can be expected to be consistently lower than Doppler measurements. Conversely, phase-contrast MR ing is well suited for volumetric

NJ:

Manati,

with

iiii2Hc”:

G Rowmu

flow probe and the meter. Specially designed blood flow probes with brass re-

and While

Thus,

0

mg/kg).

electrocardiogram

periods.

Nutley,

Products,

some MR techniques measure the distribution of velocities within a voxel and can therefore estimate peak yelocity, the most commonly used method yields a value that represents the average velocity within a voxel. Furthermore, all MR measurements represent either averages over the entire imaging time or at least over cardiac

of each

acoustic gel (HR Lubricating Jelly; CarterWallace, New York) was used to fill any air space within the probe bracket. Flowprobe cables were secured and exteriorized, the incision was closed, and the dog

Puerto

the

G

vessel was isolated, a probe of appropriate size was selected and implanted, and

(Valium;

(7). These

peak velocity resolution.

10% of actual

a total of 14 blood

distributed

fact that phase-contrast MR imaging and Doppler US measure different quantities. Doppler US is generally used to measure has high temporal

by the plus or mi-

dog was then transported to the MR imaging facility, and for the duration of the imaging protocol it was sedated with intravenous bolus injections of diazepam

with by

probes is guaranteed to be better than

maintain

been

compared

nus

was

obtained between phase-contrast measurements of ventricular outflow and MR measurements based on planimetry (6,8). Phase-contrast MR imaging

these

manufacturer

the cardiac this

data enabled

in each measure-

cycle were

processing

of the

comparison

comflow-

with

data and also compensated lost during blanking.

All MR imaging a 1.5-T

Systems,

were

readings flowmeter

system

Milwaukee).

used

to select

data

were

(Signa;

obtained GE Medical

Localizer

a plane

MR

for

images

perpendicular

to the direction of blood flow in each yessel of interest, as close as possible to the position of the blood flow probe. In the immediate vicinity of the probe, RF and static field perturbations caused signal loss within the vessel. Typical imaging locations were within 5 mm of the probe. At these locations, the vessel lumen was relatively bright. Any residual phase shifts due to the presence of the probe were

December

1992

b.

a.

Figure

3.

MR images

of canine

256; section thickness, 5 mm; (dotted lines). (b) Magnitude tual data used for calculation tively, and gray as 0 cm/sec.

iliac artery

with

and field of view, 24 cm. image shows ROI (arrow) of blood flow. Gray scale

the following

1000

parameters:

30#{176}; two signals averaged; matrix, 256 x with ROIs used as mask for correction of additive phase errors of blood flow. (c) Magnitude-weighted velocity image contains accm/sec) as black or white in superior or inferior directions, respec-

(a) Magnitude image used for calculation portrays z’ (±100

Because

1200

B

c.

obtained

the

noise

in the

50/12;

measured

flip angle,

veloc-

(n

ity is high in regions with tensity (Eq [2J) and because outside the object do not

low signal inpixels in air exhibit phase-

shift were signal

that excluded weighted by to suppress

air

selected

of

1800

#{149} “noun 600

ani

0

errors, static regions selected. The fit was magnitude squared

noise

from

low

inadvertently

signal

then

intensity.

subtracted

The

from

all

the phase-contrast 200

400

600

Blood

Figure

4.

Mean

flow measured the flowmeter cates the best plus venous); dence intervals. coefficient).

800

flow(rnl/min)

arterial

iou

MIII

and

venous

blood

with MR imaging (x axis) and (y axis). The solid line mdilinear fit to all data (arterial the dashed lines, 95% confir2 = r (sample correlation

can

be sensitive

Care ders

was taken consistently

studies. ROIs tude images to the

fore

ROI

and

window

to the

two

phase-contrast

surements

and

were

subtracted.

contrast a 256

cine MR x 256 matrix

data were and two

ages. Respiratory compensation

compensation in the readout

The

parameters

following

of view,

mm;

18-24

cm;

flip angle,

msec/echo and

collected signal were

time

(TE) per

used:

=

field

velocity

24-50/9.5-12;

cycle.

out

causing

flow

gions tures

(5). A correction of these effects that were

contained identified

error image, the measured regions and function

Volume

of space,

185

in the

flow

means

both

3c).

Be-

level

set

signal Whenever

and

regression. blood

flow

with

both

arterial

modalities

was

images errors magnetic for was

are suscaused field

cycle) plot-

a cor-

RESULTS The

the possible used. Re-

sured

Number

#{149}

C11 + C2y

3

(5).

(3)

mean

venous

blood

flow

for each

is

flow

vessel.

Results

flows

combined.

The

mean

dif-

ference between measurements obtamed with MR imaging and those obtained with the flowmeter in all vessels was -11 mL/min ± 11 (standard error of the mean). Average blood flow measured with both methods was calculated in each vessel and compared by means of lin-

regression

.995

(Fig 4). The

of correlation,

(standard

[SEE]

=

mea-

with MR imaging was 710 mL/ mm ± 230 (SD) versus mean venous blood flow of 751 mL/min ± 238 measured with the flowmeter (ii = 8). The corresponding mean arterial blood flow was 137 mL/min ± 6 (measured with MR imaging) and 131 mL/min ± 6 (measured with the flowmeter)

of the

error

3.59

technique

following

r, equal to estimate

mL/min). The intercept zero and the slope was 1.0, indicating that

is accurate

without

lyzed tions

fit to the

data:

flowmeter (where

x

MR

imaging),

=

mL/min;

flowmeter r = .980,

SEE In addition, cardiac cycle

flow

blood r

for =

+

arterial 0.88x

flow

+ 0.95x 5.71

blood mL/

measured

.963, SEE = blood flow,

=

venous

73.6 =

for 5.87

=

with

blood

venous ana-

separately, the following equafor the lines described the best

mm 1.66

the

need

for a calibration curve. When and arterial blood flows were

flow,

no vascular struc(Fig 3a). A velocity

t’() +

blood

linear

as follows: =

measurements

in blood

of the paired Student t test were not significant for differences between portal or arterial blood flows or all

was nearly approximately

blood

over

representative cardiac cycle to assess relation of pulsatile blood flow.

difference

coefficient MR

data

measurement

paired

equation for the line described the best linear fit to all data (venous plus arterial): flowmeter = -7.54 + 1.053x mL/min (where x = blood flow measured with MR imaging), with the

from

(ie,

of

ear

average

cardiac

range

intensity sub-

In addition,

a representative

the

of the large

flows measured, a more relecomparison of the techniques

to half

calculated.

flow calculated with flowmeter

resolved

during

cycle

were

of linear

ted for each

e(x, y), was calculated with velocity data in these static a least squares fit to a linear

e(x, y)

(10). bor-

aliasing.

Phase-contrast velocity ceptible to additive velocity by eddy current-induced changes presence

3b,

window

were

blood compared

measured

Velocity

encoding (v) was varied from 20 cm/sec for venous flow to 100 cm/sec for arterial flow. The was chosen in each case to be near the minimum value possible with-

(Fig

intraluminal image.

16 frames

Average data was

flow

data

display

width

temporally

(TR)

vascular and within

drawn on the magnicine set and then ap-

selection,

of blood

by

3-5

time

msec

cardiac

rate

of ROI

vessel motion during the cycle was separate ROIs were selected for each The rate of blood flow during each

of the

flow used.

thickness,

30#{176}; repetition

16 frames

with aver-

and

were

section

stantial seen, frame.

meaPhase-

of

of blood

to estimate between

the maximum in the magnitude common

images

to choice

were of the

was

set.

measurements

flow

plied

image

velocity

cine

Phase-contrast

1200

areas

error

6). Because

=

blood vant

mL/min,

mL/min.

plots of the combined were made for arterial

measured

with

both

meth-

ods. flow

A representative plot of blood in the iliac artery is shown in Figure 5. Excellent agreement is seen.

Radiology

811

#{149}

Blood

DISCUSSION This

study confirms of phase-contrast

the

in vivo cine MR

curacy aging for the measurement of blood flow. A large range of mean blood flows were included in these measurements

(Fig

4). This

was

acim-

purposely

accomplished by inclusion of different blood vessels to represent the physiologic range of blood flows that might be seen clinically and to avoid artificially inflated correlation coefficients that can be observed over a narrow range (11). In addition, a large variation in blood flow exists because of pulsatility within a representative cardiac cycle, which can be appreciated by

examination

Blood neously

flowmeter

of Figure

flow

measured

with

MR

imaging

5.

simultaand

the

US

did

not vary significantly across a large range of average blood flows. The accuracy of the US flowmeter technique across the range of blood flows included in this study is within plus or minus 10% (12,13) of the measured value. The observed temporal and quantitative correlation of pulsatile blood flow throughout the cardiac cycle is excellent between the two methods. In cases of high heart rates, long TR, or both, MR measurements have limited temporal resolution. In this study, heart rates were 112-164 beats per minute, a range that produced a minimum of seven data points during each cardiac cycle before interpolation by the cine algorithm (with a worst-case TR of 50 msec). The number of sequences per heart cycle controls the temporal resolution of the data. If this number is too small, pulsatile flow waveforms will underestimate cyclical variations. In our data, underestimation of peak flow was less than 15%. Furthermore, limited temporal resolution should not produce errors in the measured mean flow rate. Our data support this position.

Our study examined blood large veins and medium-size

812

Radiology

#{149}

flow in arteries.

flow

studied.

in large

arteries

However,

the

was

principal

not limi-

tations in quantitation of blood flow with phase-contrast MR imaging are associated with the ratio of flow yelocity to Venc and to the number of pixels across the vascular diameter. Slow blood flow is more difficult to measure than fast blood flow because longer TE is required. Thus, the results are expected to be representative of what should be achieved in studies of large arteries if the pixel size and Venc are appropriately scaled. A limiting factor to the use of this technique is the requirement for a regular sinus rhythm. Our implementation is capable of discarding data from approximately 20 ectopic beats for the duration of the acquisition. Once this number is exceeded, the acquisition continues as if in a regular rhythm, and gating may be maccurate. When this occurs, the pulsatile nature of the flow waveform may be lost,

but

the

measured

mean

rate

References

2.

B -*-S

Frami

Figure

measured

4.

Nayler CL, Firmin DN, Longmore DB. Blood flow imaging by cine magnetic resonance. J Comput Assist Tomogr 1986; 101: 715-722. Spritzer CE, Pelc NJ, Lee JN, Evans AJ, Sostman HD, Riederer SJ. Rapid MR im-

aging

of blood

flow

(in milliliters

per

with

MR

imaging

and

the

flow-

5.

6.

frame

Pelc NJ, Herfkens RJ, Shimakawa A, Enzmann DR. Phase contrast cine magnetic resonance imaging. Magn Reson Q 1991; 7:229-254. Kondo C, Caputo GR, Semelka R, Foster E, Shimakawa A, Higgins CB. Right and left ventricular stroke volume measurements with velocity-encoded cine MR imaging: in vitro and in vivo validation. AJR 1991; 157: 9-16.

7.

Caputo

8.

Right and left lung perfusion: in vitro and in vivo validation with oblique-angle, yelocity-encoded cine MR imaging. Radiology 1991; 180:693-698. Firmin DN, Nayler GL, Klipstein SR. Un-

GR, Kondo

derwood

T, et al.

SR. Rees RSO, Longmore

vivo validation Comput Assist

9.

C, Masui

of MR velocity Tomogr 1987;

Pelc NJ, Shimakawa

A.

DB.

In

imaging. 11:751-756.

Reduced

pulsatil-

artifacts in magnetic resonance angiography (abstr). In: Book of abstracts: Society of Magnetic Resonance in Medicine 1991. Berkeley, Calif: Society of Magnetic Resonance in Medicine, 1991; 821. Pelc NJ, Sommer FG, Enzmann DR. Pelc ity

LR, Glover

11.

12.

13.

GH.

Accuracy

and precision

of phase-contrast MR flow measurements (abstr). Radiology 1991; 181(P):189. Armitage P. Statistical methods in medical research. Oxford, England: Blackwell Scientific, 1971; 186-213. Burton RG, Gorewit RC. Ultrasonic flowmeter uses wide-beam transit-time technique. Transonics Systems 1984; 15:68-73. Hartman JC, Koerner J, Lancaster L, Gorczynski R. In vivo calibration of a transit

time ultrasound cending macologist

system

for measuring

aorta volume flow 1985; 27:217.

(abstr).

asPhar-

flow with a phase-sensitive,

limited-flip-angle,

sequence: ogy 1990;

Blood

meter. Numbers on the x axis indicate number of composite cardiac cycle.

556.

3.

5.

minute) versus time during a composite cardiac cycle of blood flow in the iliac artery

10.

SingerJR, Crooks LE. Nuclear magnetic resonance blood flow measurements in the human brain. Science 1983; 221:654-656. Edelman RR, Mattle HP, Kleefield J, Silver MS. Quantffication ofblood flow with dynamic MR imaging and presaturation bolus tracking. Radiology 1989; 171:551-

M

-a--

of

blood flow should remain accurate. We have demonstrated that phasecontrast cine MR imaging can produce accurate estimates of arterial and venous blood flow in vivo over a large physiologic range. This technique is already being applied in a variety of clinical and experimental situations and represents a major advance in noninvasive measurement of blood flow. #{149}

1.

B

gradient

preliminary

recalled

experience.

pulse

Radiol-

176:255-262.

December

1992