peak velocity and flow quantification validation for sensitivity-encoded phase-contrast mr imaging

Peak Velocity and Flow Quantification Validationfor Sensitivity-Encoded Phase-Contrast

MR Imaging1

Calvin D. Lew, M.S., Marcus T. Alley, Ph.D., Roland Bammer, Ph.D., Daniel M. Spielman, Ph.D.,Frandics P. Chan, M.D., Ph.D.

Rationale and Objectives. Phase-contrast (PC) magnetic resonance imaging (MRI) technique has important clinical appli-cations in blood flow quantification and pressure gradient estimation by velocity measurement. Parallel imaging using sen-sitivity encoding (SENSE) may substantially reduce scan time. We demonstrate the utility of PC-MRI measurements ac-celerated by SENSE under clinical conditions.

Materials and Methods. Accuracy and repeatability of a SENSE-PC implementation was evaluated by comparison with acommercial PC sequence with five normal volunteers. Twenty-six patients were then scanned with SENSE-PC at reduc-tion factors (R � 1, 2, and 3). Blood flow and peak velocity were measured in the aorta and pulmonary trunk in 16 pa-tients and peak velocity was measured at the coarctation of 10 patients. Quantitative flow, shunt ratio, and peak velocitymeasurements obtained with different reduction factors were compared using correlation, linear regression, and Bland-Altman statistics. All studies were approved by an Institutional Review Board, and informed consent was acquired fromall subjects.

Results. The correlation coefficients for all comparisons were �0.962 and with high statistical significance (P � .01).Linear regression slopes ranged between 0.96 and 1.11 for flow and 0.88 to 1.05 for peak velocity. For flow, the Bland-Altman statistics yielded a total mean difference ranging from �0.02 to 0.05) L/minute with 2 standard of deviation limitsranging from �0.52 to 0.75 L/minute. For peak velocity, the total mean difference ranged from �0.10 to �0.004) milli-seconds with 2-SD limits ranging from �0.062 to 0.46 milliseconds. R � 3 to R � 1 comparisons had greater 2-SD lim-its than R � 2 to R � 1 comparisons.

Conclusion. SENSE PC-MRI measurements for flow and pressure gradient estimation were comparable to conventionalPC-MRI.

Key Words: Flow; parallel imaging; cardiac imaging© AUR, 2007

Acad Radiol 2007; 14:258–269

1 From the Lucas MRS/I Center, Stanford University, 1201 Welch Road,Stanford, CA 94305 (C.D.L., M.T.A., R.B., D.M.S., F.P.C.). Received August24, 2006; accepted November 17, 2006. Supported by GE Medical Sys-tems, Center of Advanced MR Technology at Stanford P41RR09784,R01EB002711, the Whitaker Foundation, and the Lucas Foundation.Address correspondence to: C.D.L. e-mail: [email protected]

©
AUR, 2007doi:10.1016/j.acra.2006.11.008
258

Phase-contrast magnetic resonance imaging (PC-MRI) is aproven noninvasive technique for quantifying velocity andblood flow in the great arteries (1–3). Measurements ofblood flow in the aorta and pulmonary trunk produce awealth of information, including cardiac outputs of theleft and right ventricles, regurgitant volumes and fractionof the aortic and pulmonary valves, and shunt ratio. Re-gurgitant fraction is a particularly important parameterthat determines the need for valvular repair or replace-
ment, whereas shunt ratio is an important parameter for

Academic Radiology, Vol 14, No 3, March 2007 SENSE PC-MRI VELOCITY AND FLOW STUDY

evaluating the need for closing shunt lesions because ofatrial septal and ventricular septal defects (4–8). Velocityof moving blood is related to the pressure gradient as de-scribed by the modified Bernoulli equation (9). This rela-tionship is used extensively by echocardiography to esti-mate pressure gradient across stenotic cardiovascularlesions. Using a similar approach, PC-MRI has been ap-plied to estimate the pressure gradient of coarctation ofthe ascending aorta (9). A PC-MRI scan usually requiresa breath-hold to minimize the movement of the vesselsthrough the imaging slice. However, conventional PC-MRIscans last on the order of 20–30 seconds. This duration maybe difficult for children and noncompliant patients.

Sensitivity encoding (SENSE) (10) is an importanttechnique that significantly reduces acquisition time. Im-plementations of SENSE for magnitude imaging are nowwidely available for clinical applications. A successfulimplementation of SENSE for PC-MRI promises short-ened scan time or increased temporal resolution in seg-mented k-space techniques, thereby improving patientcompliance in breath-hold imaging. An experimental ver-sion of SENSE PC-MRI has been tested on the measure-ment of shunt ratio (11,12). However, the potential appli-cations of SENSE PC-MRI have yet to be fully exploredin the measurement of flow velocity and the estimation ofpressure gradients.

Feasibility of SENSE-accelerated PC-MRI has beendemonstrated by experimental sequences (11–13) for flowquantification although a clinically practical and efficientimplementation has not been reported. In particular, theSENSE paradigm requires an accurate, robust estimationof the coil sensitivities to unwrap the undersampled,aliased images. Hence, we have recently implemented aSENSE version of PC-MRI that uses a single, low-resolu-tion three-dimensional (3D) volume acquisition for calibra-tion. Data from this calibration scan are used to reconstructsubsequent, unlimited number of SENSE acceleratedPC-MRI scans, provided that the regions of interest liewithin the calibration volume. This implementation iscapable of variable reduction factor R. For convenience,we refer to a nonaccelerated (R � 1) scan as SENSE-PCx1, a reduction factor 2 (R � 2) scan as SENSE-PCx2,and a reduction factor 3 (R � 3) scan as SENSE-PCx3.

In this article, we first establish the accuracy of flowand velocity measurements of SENSE-PCx1 against astandard, commercial PC-MRI sequence (Fastcard-PC, GEHealthcare, Waukesha, WI, 1.5 T) in normal volunteers.We then examine the repeatability of SENSE-PC mea-
surements at reduction factors R � 1, 2, and 3. Finally,
we compare the results from SENSE-PC at various reduc-tion factors in a group of clinical patients. Specifically,we test the hypothesis that SENSE-PC at reduction fac-tors 2 and 3 produces flow and peak velocity measure-ments statistically similar to R � 1 values under clinicalconditions while simultaneously providing considerablyshorter scan times.

MATERIALS AND METHODS

MR Imaging

Volunteer population.—We examined five normal vol-unteers (weight 61–89 kg, mean age 34.8 years � 2.9years, 4 male/1 female) for PC-MRI study. The meanheart rate was 70 beats/minute (range 59–80). PC-MRIscans were performed with flow quantification from theslices transverse to the ascending aorta and pulmonarytrunk. Institutional Review Board approval was obtainedfor the study protocol, and informed consent was obtainedfrom all volunteers.

All imaging was done on a 1.5-T TwinSpeed MRIScanner (GE Healthcare) with 150 T/milliseconds maxi-mum slew rate and 40 mT/minute gradients. An eight-channel cardiac phased-array coil (GE Healthcare) wasused for all acquisitions. The regions of interest were cen-tered as much as possible within the field of view. Volun-teers were asked to hold their breath at deep inhalation atboth the calibration and SENSE acquisition scans.

A calibration scan was first acquired using a 3Dspoiled gradient-recalled echo (SPGR) sequence coveringthe full chest volume for all vessels of interest, namelythe aorta and the pulmonary trunk. The sequence was notcardiac gated. A low flip-angle of 5°, a low-resolutionmatrix of 64 � 64, and averaging of two excitations pro-duced high signal-to-noise ratio (SNR), low-contrast im-ages for the coil sensitivity maps. The total calibrationscan time was approximately 15 seconds for the 32 cali-bration slices.

After calibration, for each vessel of interest, Fast-card-PC scans were acquired twice, followed by twoscans acquired by SENSE-PC at reduction factors R � 1,2, and 3. The SENSE-PC sequence was implemented as atwo-dimensional fast GRE segmented k-space, cardiac-triggered, phase-contrast sequence with retrospective gat-ing (3,14). Imaging was performed with the slice perpen-dicular to the vessel lumen and flow encoding in thethrough-plane direction. For each vessel of interest, the
same slice prescription was used for all the SENSE-PC
259

LEW ET AL Academic Radiology, Vol 14, No 3, March 2007

scans. Scanning was performed with a 256 � 160 matrix,flip angle 15°, 10-mm slice, echo time 2.4 milliseconds,repetition time 4.6 milliseconds, and 62.5 kHz bandwidth.Ten views per segment were collected. Scan times wereapproximately 20–25 seconds for each Fastcard-PCand SENSE-PCx1 acquisition, 10–12 seconds for eachSENSE-PCx2 acquisition, and 6–8 seconds for eachSENSE-PCx3 acquisitions.

Patient population.—We recruited 26 patients (weight6–74 kg, mean age 11.8 years � 9.6 years, age range0.5–47 years) who were scheduled to undergo a routinecardiovascular MRI study. The mean heart rate was 91beats per minute (range 55–158). Table 1 shows the pa-tient demographics and relevant protocol parameters usedfor each patient. As in the volunteer population, thosepatients requiring flow quantification for the great vesselsunderwent PC-MRI scans from slices transverse to the

Table 1Patient Demographic and Scan Prescription

PatientNumber

Age(Years) Sex

Weight(kg) Venc (cm/s) Contrast?

1 3 M 18 250 Yes L-tran2 12 M 37 250 Yes Coarc3 10 F 24 500 Yes Coarc4 19 F 73 300 Yes Takay5 13 M 57 250 Yes Coarc6 17 M 74 250 Yes Coarc7 9 M 19 200 Yes Tetralo8 1.5 F 10 250 Yes Alagill9 5 M 16 200 No Ebste

10 0.5 M 6 400 Yes Pulmo11 3 F 13 200 Yes Tricus12 12 M 43 200 Yes D-tran13 10 F 50 200 Yes Heart14 21 F 64 400 No D-tran15* 2 M 10 250 Yes Tetralo16 47 M 61 250 Yes Ross17 10 M 37 250 Yes Tetralo18 16 M 59 450 Yes Coarc19 1 F 8 450 Yes Coarc20 24 F 57 350 Yes Coarc21 13 M 52 450 Yes Bicusp22 15 M 60 350 Yes Tetralo23 6 M 22 250 No Pulmo

plac24 10 F 32 250 Yes Trunc25 12 M 30 350 Yes Coarc26 15 F 68 150 Yes Marfa

Those with contrast were injected at 0.2 mmol/kg of gadopentet*Four-coil array was used for this patient.

ascending aorta and pulmonary trunk. Patients with coarc-

260

tation were scanned transverse to the narrowest part ofthe stenosis. One patient had severe signal loss at the pul-monary trunk because of turbulent flow from pulmonarystenosis. For this patient, PC-MRI was performed at theright and left pulmonary arteries independently. For theflow quantification, this patient’s pulmonary flow wascalculated as the sum of the right and left pulmonary arte-rial flows. For peak velocity quantification, the right andleft pulmonary arteries were separately measured andcompared. In summary, the following measurements wereobtained: aortic flow (n � 17), pulmonary flow (n � 16),shunt ratio (n � 16), aortic peak velocity (n � 17), pul-monary peak velocity (n � 17), and coarctation peak ve-locity (n � 10). Institutional Review Board approval wasobtained for the study protocol, and informed consent wasobtained from all patients.

All imaging was done on a 1.5-T TwinSpeed MRI

Clinical History

ition of the great arteries and pulmonary stenosis

ortitis with distal arch stenosisrepair

, Shone syndromef Fallot repaireddromenomalyhypertension, coarctation repairtresia, Glenn shuntition, status post-Damus procedureplantition of the great arteries, status post Mustard/Senning proceduref Fallot repairdure, pulmonary stenosisf Fallout repaired

rtic valve, coarctationf Fallot repairedand tricuspid regurgitation, post pulmonary valve prosthesis

ntteriosus repaired, pulmonary stenosis

drome, aortic root aneurysm

imeglumine (Magnevist).

spostationtationasu atationtationgy o

e synin’s anarypid aspostranssposgy o

procegy o

tationtationtationid aogy onaryeme

us artationn syn

ate d

Scanner (GE Healthcare) with 150 T/milliseconds maxi-


mum slew rate and 40 mT/minute gradient strength. Thisscanner was sited in a children’s hospital for routine clini-cal studies. An eight-channel cardiac phased-array coil(GE Healthcare) was used for all acquisitions but one.The exception used a four-channel torso phased-array coil(GE Healthcare).

Initially, 24 of the 26 patients received intravenousgadopentetate dimeglumine (Magnevist, Berlex, Wayne,NJ) at a dose of 0.2 mmol/kg as needed by the protocolfor the conventional clinical scan. After the completion ofthe standard clinical MRI images, additional scans wereperformed for the purpose of this study. Patients whowere able were asked to hold their breath at deep inhala-tion at both the calibration scan and SENSE-PC scan. Forthose unable to breath-hold, they were allowed to takeshallow breaths during both the calibration scan and theSENSE-PC scan. The calibration scan protocol was iden-tical to that used for the volunteer population.

For each vessel of interest, SENSE-PCx1 was acquiredfirst, followed by SENSE-PCx2 and SENSE-PCx3 (Fig 1).Scanning was performed with a 256 � 160 matrix, flipangle 15–30°, 5- to 7-mm slice, echo time approximately2 milliseconds, repetition time approximately 4 millisec-onds, and 32.5- to 125-kHz bandwidth. One to six viewsper segment were collected. Scan times were approxi-mately 20–30 seconds for the reference set at R � 1,12–20 seconds for SENSE at R � 2, and 7–10 secondsfor SENSE at R � 3.

Image ReconstructionBoth volunteer and patient studies used the following

data processing algorithm. The coil sensitivity maps usedfor the SENSE reconstruction were computed by first in-terpolating the 3D calibration scan onto the same pre-scribed plane as the SENSE-PC acquisition. The interpo-lated images were smoothened by a 3 � 3 spatial kernel.The final sensitivity map for each coil was then computedby pixel-wise complex division by sum of squares of allcoil images, as described by Pruessman (10). AfterSENSE reconstruction, the phase images were calculatedas the phase difference of the flow echoes. Postprocessingwas also performed to eliminate concomitant gradienteffects from the Maxwell terms (15). Phase error fromeddy currents was corrected by subtracting the phasevalue of adjacent static soft tissue.

Measurements for Flow and Pressure GradientAll flow measurements were calculated using the CV

Flow Analysis software package in an Advantage Work-

station (GE Healthcare). For every cardiac phase, theaorta and pulmonary trunk were segmented manually.Flow rate was calculated by summing flow within thevessel lumen per cardiac phase and then averaging overthe cardiac cycle (Fig 2). The shunt flow was calculatedas the ratio of the average pulmonary flow to the averageaortic flow.

The peak pressure gradient, �P, is related to the peakvelocity, Vp, by the modified Bernoulli equation, �P �

4Vp2 (9). Peak velocity measurements were performed

using the Matlab software program (Mathworks, Natick,MA). To minimize the effect of outlier values from noise,the peak velocity was calculated as the average of the top10% velocities from contiguous pixels within the vessels.The final peak velocity was calculated as the maximum ofthe peak velocities at systole.

Statistical Analysis

Volunteer population.—To establish the accuracy ofour SENSE-PC implementation, Bland-Altman statistics(16) were calculated for the difference between SENSE-PCx1 and Fastcard-PC separately for flow and peak ve-locity measurements. Data from both the aorta and thepulmonary trunk are included. Bland-Altman comparisonbased on absolute difference measures is appropriate be-cause phase error in PC-MRI does not increase with themeasured phase value. The noise in PC-MRI image re-construction is phase-independent as long as the noise inthe real and imaginary channels is uncorrelated (17).Therefore, the derived velocity error and flow error werenot expected to change with the measured velocity. Toevaluate consistency in repeated measurements, Bland-Altman statistics were calculated for the difference be-tween repeated measurements separately for Fastcard-PC,SENSE-PCx1, SENSE-PCx2, and SENSE-PCx3. Data forcomparing Fastcard-PC and SENSE-PC were derivedfrom the means of the repeated studies. Bland-Altmanlimits of agreement were calculated as the totalmean � 2 SD.

Patient population.—To establish the accuracy of mea-surements obtained with SENSE acceleration, statisticalcomparisons between SENSE-PCx1 and SENSE-PCx2,and between SENSE-PCx1 and SENSE-PCx3 were per-formed for flow and peak velocity measurements. For theflow measurements, average aortic flow and pulmonaryflow, and shunt ratio were separately compared for each
patient. Peak velocity for the aorta, pulmonary artery, and
261

Fig 1


coarctation were separately compared. Pearson correla-tion coefficients and their significance were computed.For this study, a statistic was considered significant ifP � .05. Linear regression analysis was computed.Bland-Altman statistics for absolute differences were

Figure 1. Representative magnitude (left) and velocity (right) imagtivity encoding (SENSE) phase-contrast (PC)x1, (b) SENSE-PCx2. (

also calculated.

262

g-Factor CalculationIn the SENSE formalism, the coil geometry factor, or

g-factor, is a measure of the effect of noise influenced bythe physical aspects of the coil receiver array and mayinfluence the precision of the phase-contrast measure-

f the aorta (arrow) for a patient with Ebstein’s anomaly. (a) Sensi-continues)

es o

ments. The coil geometry factor was computed to deter-

imag


mine its variation in clinical scans and thus its possibleeffects on the measurements. The g-factor affects the SNRof the magnitude reconstruction by this equation (10):

SNRSENSE �SNRconventional

Figure 1 (continued). (c) SENSE-PCx3. Noise from coil sensitivitiedirection is in the anteroposterior direction, whereas flow is along tthe baseline vessel signal still provides reliable flow estimation.

Figure 2. Flow profile of the aortic scanagreement despite noise in the magnitude

g�R

where R is the SENSE reduction factor and g is theg-factor. For velocity images, the standard deviation ofthe noise in the velocity measurement is (3):

�2VENC

n be seen in the SENSE reconstructions. The phase-encodingperior-inferior direction. The noise in the background is high, but

n in Fig 1. The flow profiles show goodes.

s cahe su

show

�velocity ��SNR

263

ble

2ti

stic

alC

om

par

iso

nsB

etw

een

Fast

card

-PC

and

SE

NS

E-P

Cfo

rFl

ow

inN

orm

alV

olu

ntee

rs

n

Fast

card

-PC

Rep

eata

bili

tyS

EN

SE

-PC

x1R

epea

tab

ility

SE

NS

E-P

Cx2

Rep

eata

bili

tyS

EN

SE

-PC

x3R

epea

tab

ility

Fast

card

-PC

vsS

EN

SE

-PC

x1S

EN

SE

-PC

x1vs

SE

NS

E-P

Cx2

SE

NS

E-P

Cx1

vsS

EN

SE

-PC

x310

1010

1010

1010

nd-A

ltman

Tota

lmea

nd

iffer

ence

(L/m

inut

e)0.

07�

0.10

�0.

03�

0.26

�0.

06�

0.15

�0.

58

Low

erlim

itof

agre

emen

t(L

/min

ute)

*

�0.

65�

0.98

�1.

28�

1.16

�0.

75�

2.03

�1.

67

Up

per

limit

ofag

reem

ent

(L/m

inut

e)†

0.79

0.79

1.21

0.64

0.63

1.74

0.50

C:

pha

seco

ntra

st;

SE

NS

E,

sens

itivi

tyen

cod

ing.

Low

erlim

itof

agre

emen

tre

pre

sent

sth

em

ean

diff

eren

ce–

2st

and

ard

dev

iatio

ns.

Up

per

limit

ofag

reem

ent

rep

rese

nts

the

mea

nd

iffer

ence

�2

stan

dar

dd

evia

tions

.


For each scan of the aorta, pulmonary trunk, and co-arctation, the vessel lumen was segmented, and the meanand standard deviation of the g-factor were computedwithin that lumen.

RESULTS

Tables 2 and 3 summarize the Bland-Altman upperlimit, mean, and lower limit for the difference betweenFastcard-PC and SENSE-PCx1 and for the differencesbetween the repeated studies. The mean differences be-tween Fastcard-PC and SENSE-PCx1 were –0.06L/minute for flow and 0.0053 milliseconds for peak ve-locity. For flow repeatability studies, the maximum limitsof agreement were 0.98 L/minute for SENSE-PCx1, 1.28L/minute for SENSE-PCx2, and 1.16 L/minute forSENSE-PCx3. For peak velocity repeatability studies, themaximum limits of agreement were 0.12 milliseconds forFastcard-PC, 0.11 milliseconds for SENSE-PCx1, 0.15milliseconds for SENSE-PCx2, and 0.17 milliseconds forSENSE-PCx3.

Tables 4 and 5 summarize the correlation coefficients,linear regression, and Bland-Altman statistics for flow andpeak velocity measurements, respectively. Figure 3 showsthe Bland-Altman plot for the difference in aortic flowbetween accelerated SENSE-PC and nonacceleratedSENSE-PC. Figures 4 and 5 show similar plots for pul-monary flow and coarctation peak velocity.

The correlation coefficients for all the comparisonswere above 0.962 and were all statistically significant(P � .01). For both aortic and pulmonary flow, the maxi-mum mean difference was 0.12 L/minute and the maxi-mum limit of agreement was 0.80 L/minute. Generally,the pulmonary flow showed a greater difference in meanand in limit of agreement than the aortic flow. For thecalculated shunt ratio, the maximum mean difference was0.04 and the maximum limit of agreement was 0.40. Forpeak velocity, the maximum mean difference was 0.1milliseconds and the maximum limit of agreement was0.62 milliseconds. For both peak velocity and flow, ahigher reduction factor at R � 3 generally showed anincrease in limit of agreement when compared to a lowerreduction factor at R � 2.

The mean and standard deviation of the g-factor areshown in Table 6. A g-factor of 1 corresponds to nonoise amplification from the coil geometry array. The SNRdecreases with increasing g-factor. The mean aorta g-factors
were the highest followed by the pulmonary artery g-factors.
264

Ta

Sta

Bla P * †

ble

3ti

stic

alC

om

par

iso

nsB

etw

een

Fast

card

-PC

and

SE

NS

E-P

Cfo

rP

eak

Vel

oci

tyin

No

rmal

Vo

lunt

eers

n

Fast

card

-PC

Rep

eata

bili

tyS

EN

SE

-PC

x1R

epea

tab

ility

SE

NS

E-P

Cx2

Rep

eata

bili

tyS

EN

SE

-PC

x3R

epea

tab

ility

Fast

card

-PC

vsS

EN

SE

-PC

x1S

EN

SE

-PC

x1vs

SE

NS

E-P

Cx2

SE

NS

E-P

Cx1

vsS

EN

SE

-PC

x310

1010

1010

1010

nd-A

ltman

Tota

lmea

nd

iffer

ence

(mill

isec

ond

s)0.

0098

�0.

029

0.04

1�

0.00

550.

0053

�0.

071

�0.

041

Low

erlim

itof

agre

emen

t(m

illis

econ

ds)

*

�0.

099

�0.

11�

0.06

8�

0.17

�0.

077

�0.

17�

0.22

Up

per

limit

ofag

reem

ent

(mill

isec

ond

s)†

0.12

0.05

70.

150.

160.

088

0.02

90.

14

C:

pha

seco

ntra

st;

SE

NS

E,

sens

itivi

tyen

cod

ing.

Low

erlim

itof

agre

emen

tre

pre

sent

sth

em

ean

diff

eren

ce–

2st

and

ard

dev

iatio

ns.

Up

per

limit

ofag

reem

ent

rep

rese

nts

the

mea

nd

iffer

ence

�2

stan

dar

dd

evia

tions

.


The distal descending aorta g-factors corresponding to thecoarctation scans were the lowest. The g-factors at R � 3were higher than those at R � 2, as expected.

DISCUSSION

PC-MRI is an established, routinely applied techniquefor noninvasive, clinical assessment of cardiovascularphysiology. Most implementations of PC-MRI employsegmented k-space techniques to shorten imaging times topermit breath-held imaging, which in turn minimize respi-ratory artifacts. However, this is done at the expense oftemporal resolution. SENSE can improve scan time, tem-poral resolution, spatial resolution, or a combination ofthese factors. Increasing temporal resolution improvesclarity of rapidly moving parts, such as the heart valves.Increasing spatial resolution helps imaging small struc-tures and reduces partial volume effects. Decreasing scantime also helps improve patient compliance. Therefore,the incorporation of SENSE with PC-MRI is a logical andpowerful step. Despite demonstration of feasibility(11–13), a clinically practical and efficient implementa-tion has yet to be reported. Although the accuracy ofSENSE PC-MRI has been assessed in calculating shunt ratio(11,12), it has yet to be evaluated in velocity measurements,which are needed for pressure gradient estimation.

In this study, we explored the approach of using a sin-gle, low-resolution, low-contrast, high-SNR, 3D-SPGRsequence to acquire the calibration scan for all subsequentSENSE-PC scans. For the case of shunt ratio quantifica-tion using conventional PC-MRI, the total scan timewould include the time for two scans of the aorta andpulmonary artery. If a nonaccelerated PC-MRI sequencetakes 20 seconds per slice, then the total scan time for twoslices would be about 40 seconds. In our approach, thesingle 3D calibration scan takes about 10 seconds and thetotal scan time is 30 seconds, close to a 25% savings. Inclinical practice, the savings in scan time is likely greaterbecause additional flow planes are routinely acquired.Alternatively, the savings in scan time can be traded forimproved temporal resolution, spatial resolution, or both.To maintain the focus of this study, we did not explorethese additional possibilities.

In our volunteer study, we showed that the perfor-mance of our SENSE-PC implementation is generallycompatible with an established, nonaccelerated PC-MRIsequence. The direct comparison between SENSE-PCx1
and Fastcard-PC of blood flow values showed a meanT
aS

ta

Bla P * †

265

anda

anda


difference of 0.06 L/minute, which is insignificant com-pared to a normal cardiac output of 5 L/minute or a peaksystolic aortic flow of 10 L/minute. Similarly, the meandifference for peak velocity was 0.0053 milliseconds,which is insignificant compared to a normal aortic valvu-lar velocity of 1–2 milliseconds.

Our repeatability study showed that consistency fromscan to scan worsens with SENSE acceleration. This isreflected by an increased limit of agreements from 0.98L/minute at R � 1 to 1.16 L/minute at R � 3 for flow,and from 0.11 milliseconds at R � 1 to 0.17 millisecondsat R � 3. These changes can be expected from the in-creased noise and decreased SNR ratio in SENSE acceler-ated images.

In the patient study, there is good agreement in flow

Table 4Statistical Comparisons Between PC-MRI and SENSE PC-MRI

n

SENSE Reduction Factor

Bland-Altman Total mean difference (L/minute)Lower limit of agreement (L/minute)* �

Upper limit of agreement (L/minute)†

Pearson Correlation coefficientP value

linear regression Slopey-intercept (L/minute) �

PC-MRI: phase-contrast magnetic resonance imaging; SENSE: s*Lower limit of agreement represents the mean difference – 2 sta†Upper limit of agreement represents the mean difference � 2 st

Table 5Statistical Comparisons Between PC-MRI and SENSE PC-MRI

n

SENSE Reduction Factor

Bland-Altman Total mean difference (milliseconds) �

Lower limit of agreement (milliseconds)* �

Upper limit of agreement (milliseconds)†

Pearson Correlation coefficientP value

linear regression Slopey-intercept (milliseconds) �

PC-MRI: phase-contrast magnetic resonance imaging; SENSE: s*Lower limit of agreement represents the mean difference – 2 sta†Upper limit of agreement represents the mean difference � 2 st

and peak velocity for a reduction factors up to 3. A small

266

positive bias, up to 0.12 L/minute, was detected for flowand a small negative bias, up to 0.083 milliseconds, wasdetected for peak velocity, but both are small comparedto the normal physiologic ranges. Beerbaum (11,12) re-ported close agreement between SENSE-accelerated PC-MRI compared with nonaccelerated PC-MRI for aorticflow, pulmonary flow, and shunt ratio. They employed alog-transformed Bland-Altman method to compare theratio of two techniques and reported a 3% mean differ-ence and a 24% maximum limit of agreement. In ourstudy, we believe that a comparison of absolute differenceis more appropriate because there is no expectation thaterror increases with the magnitude of the velocity or flowdata. Although our results qualitatively agree with theirassessments concerning flow, a quantitative comparison of

low in Patients

Aorta Pulmonary Artery Shunt Ratio

17 16 16

3 2 3 2 3

0.04 0.12 �0.02 0.04 0.03�0.51 �0.41 �0.80 �0.16 �0.34

0.59 0.65 0.75 0.25 0.400.985 0.994 0.986 0.992 0.969.01 .01 .01 .01 .01

1.065 1.016 0.955 1.091 1.11�0.16 0.06 0.14 �0.063 �0.103

ivity encoding.d deviations.rd deviations.

eak Velocity in Patients

Aorta Pulmonary Artery Coarctation

17 17 10

3 2 3 2 3

9 �0.041 �0.0040 �0.083 �0.10 �0.080�0.30 �0.25 �0.62 �0.40 �0.45

0.22 0.24 0.46 0.19 0.293 0.984 0.993 0.962 0.986 0.974

.01 .01 .01 .01 .013 0.904 1.048 0.885 0.883 0.87464 0.057 �0.071 0.11 0.13 0.20

ivity encoding.d deviations.rd deviations.

for F

2

0.050.520.620.985.01

1.0750.19

ensitndar

for P

2

0.020.220.160.99.01

0.980.00

ensitndar

results was not possible.


We note in our study that accuracy decreases in pul-monary trunk with an increase in the reduction factor.However, this trend is not as prominent in the aorta orthe coarctation. This phenomenon cannot be explained byg-factor difference because the g-factor is greater at theaorta as compared with the pulmonary trunk. One possi-ble explanation may be the difference in orientation of theimage planes for the aorta and the pulmonary trunk. Forthe aorta, the transverse image lies mostly in the axialplane, whereas the anteroposterior dimension of the chestis small. In contrast, the transverse image for the pulmo-nary trunk lies mostly in the coronal plane, where thebody fills the image. When SENSE undersamples k-space,

Figure 3. Bland-Altman statistics for aortic flow. (a) Sensitivityencoding (SENSE) phase-contrast (PC)x2 vs SENSE-PCx1.(b) SENSE-PCx3 vs. SENSE-PCx1. The dashed lines representthe mean � 2 SD.

the wrapping of mostly empty space in the axial plane

produces fewer artifacts compared with scanning in thecoronal plane. At higher reduction factors, separatingaliasing artifacts with SENSE in the axial plane producesless error and artifacts than in the coronal plane.

Despite its benefits, SENSE PC-MRI introduces sev-eral problems not encountered in conventional PC-MRI.The fidelity of SENSE reconstruction is strongly depen-dent on the appropriate coil design and the proper matchbetween the phase-array coils and the body size of thesubject. The primary eight-channel cardiac coil that wasused for this study is designed for adult body size and isnot optimized for small pediatric patients, especially in

Figure 4. Bland-Altman statistics for pulmonary artery flow.(a) Sensitivity encoding (SENSE) phase-contrast (PC)x2 vs.SENSE-PCx1. (b) SENSE-PCx3 vs SENSE-PCx1. The dashedlines represent the mean � 2 SD.

the left-right direction. Errors in coil sensitivity maps in

267


the chest wall, where fat signals are strong, may addition-ally contribute to inaccuracies.

One complication regarding SENSE PC-MRI is alias-

Figure 5. Bland-Altman statistics for coarctation peak velocity.(a) Sensitivity encoding (SENSE) phase-contrast (PC)x2 vs.SENSE-PCx1. (b) SENSE-PCx3 vs. SENSE-PCx1. The dashedlines represent the mean � 2 SD.

Table 6g-Factor Statistics

Average Size(mm2)

MeanStandardDeviation

R � 2 R � 3 R � 2 R � 3

Aorta 137.06 1.322 2.103 0.324 1.191Pulmonary artery 162.45 1.161 1.554 0.181 0.305Coarctation 94.75 1.159 1.409 0.472 0.589

ing in full field-of-view image reconstructions (18). When

268

the coil sensitivity maps do not fully correct for aliasingin the full field of view, artifacts are produced. In thisstudy, we carefully avoided the prescription of aliasing infull field of view as much as possible. For those patientsin whom we could not avoid this complication, we didnot encounter appreciable changes in the flow and peakvelocity measurements.

For estimating peak velocities, the SENSE PC-MRImeasurements introduce a slight negative bias. One possi-ble explanation is that when peak blood flow velocities ap-proach the maximum encoded velocity, additive noise be-yond the maximum velocity would be registered as analiased, negative velocity. This would introduce a negativebias to the estimation. Because SNR decreases with greaterreduction factor, this negative bias may also become greater.

Finally, our approach of scanning a calibration scanonce, although efficient, may introduce misregistrationartifacts when the patient moves or breathes differentlyover the course of the study. When this type of artifactbecomes noticeable, the calibration scan must be repeated.

Some of the variations observed in this study may becaused by physiologic changes. Beerbaum (19) reported5.3% variability in shunt ratio quantification in scans per-formed three times with pediatric populations. Powell(20) reported �0.2 limits of agreement for normal volun-teers with expected shunt ratios of 1.0, corresponding to20% variability. It is expected that random patient move-ment or physiologic changes occur which may affect flowmeasurements seen in this study.

In summary, a practical and efficient SENSE PC-MRIsequence has been implemented and tested in patientsunder clinical conditions. Flow and velocity data derivedfrom SENSE PC-MRI were found compatible with con-ventional PC-MRI in the imaging of the great arteries.The efficiency gained from SENSE may be traded forshorter scan time, increased temporal resolution, increasedspatial resolution, or combinations thereof. Thus SENSEis a valuable and powerful tool that enhances the perfor-mance of clinical PC-MRI.

ACKNOWLEDGMENT

We wish to thank Dr. Lars Wigstrom for helping withinitial construction of peak velocity calculations.

REFERENCES

1. Szolar DH, Hajime S, Higgins CB. Cardiovascular applications of mag-netic resonance flow and velocity measurements. J Magn Reson Imag-
ing 1996; 1:78–89.


2. Higgins CB, Sakuma H. Heart disease: functional evaluation with MRimaging. Radiology 1996; 199:307–315.

3. Pelc NJ, Herfkens RJ, Shimakawa A, et al. Phase contrast cine mag-netic resonance imaging. Magn Reson Q 1991; 7:229–254.

4. Hundley WG, Li HF, Lange RA, et al. Assessment of left-to-right intra-cardiac shunting by velocity-encoded, phase-difference magnetic reso-nance imaging: a comparison with oximetric and indicator diffusiontechniques. Circulation 1995; 91:2955–2960.

5. Bremerich J, Reddy GP, Higgins CB. MRI of supracristal ventricularseptal defects. J Comput Assist Tomogr 1999; 23:13–15.

6. Parsons JM, Baker EJ, Anderson RH, et al. Morphological evaluation ofatrioventricular septal defects by magnetic resonance imaging. BrHeart J 1990; 64:138–145.

7. Korperich H, Gieseke J, Barth P, et al. Flow volume and shunt quantifi-cation in pediatric congenital heart disease by real-time magnetic reso-nance velocity mapping: a validation study. Circulation 2004;109:1987–1993.

8. Brenner LD, Caputo GR, Mostbeck G, et al. Quantification of left toright atrial shunts with velocity-encoded cine nuclear magnetic reso-nance imaging. J Am Coll Cardiol 1992; 20:1246–1250 1.4–3.9

9. Mohiaddin RH, Kilner PT, Rees S, et al. Magnetic resonance volumeflow and jet velocity mapping in aortic coarctation. J Am Coll Cardiol1993; 22:1515–1521.

10. Pruessman KP, Weiger M, Scheidegger MB, et al. SENSE: SensitivityEncoding for Fast MRI. Magn Reson Med 1999; 42:952–962.

11. Beerbaum P, Korperich H, Gieseke J, et al. Rapid left-to-right shuntquantification in children by phase-contrast magnetic resonance imag-

ing combined with sensitivity encoding. Circulation 2003;108:1355–1361.

12. Beerbaum P, Korperich H, Gieseke J, et al. Blood flow quantification inadults by phase-contrast MRI combined with SENSE—a validationstudy. J Cardio MR 2005; 7:361–369.

13. Thunberg P, Karlsson M, Wigstrom L. Accuracy and reproducibility inphase contrast imaging using SENSE. Magn Reson Med 2003;50:1061–1068.

14. Nayler GL, Firmin DN, Longmore DB. Blood flow imaging by cine mag-netic resonance. J Comput Assist Tomogr 1986; 10:715–722.

15. Bernstein MA, Zhou XJ, Polzin JA, et al. Concomitant gradient terms inphase contrast MR: analysis and correction. Magn Reson Med 1998;38:300–308.

16. Bland JM, Altman DG. Statistical methods for assessing agreementbetween two methods of clinical measurement. Lancet 1986; 1:307–310.

17. Conturo TE, Gregory D. Signal-to-noise in phase angle reconstruction:dynamic range extension using phase reference offsets. Magn ResMed 1990; 15:420–437.

18. Griswold MA, Kannengiesser S, Heidemann RM, et al. Field-of-viewlimitations in parallel imaging. Magn Res Med 2004; 52:1118–1126.

19. Beerbaum P, Korperich H, Barth P, et al. Noninvasive quantification ofleft-to-right shunt in pediatric patients. Circulation 2001; 103:2476.

20. Powell AJ, Maier SE, Chung T, et al. Phase-velocity cine magnetic res-onance imaging measurement of pulsatile blood flow in children and
young adults: in vitro and in vivo validation. Pediatr Cardiol 2000; 21:104–110.
269

peak velocity and flow quantification validation for sensitivity-encoded phase-contrast mr imaging

Documents