J A C C : C A R D I O V A S C U L A R I M A G I N G V O L . 1 0 , N O . 5 , 2 0 1 7

ª 2 0 1 7 B Y T H E AM E R I C A N C O L L E G E O F C A R D I O L O G Y F O U N D A T I O N

P U B L I S H E D B Y E L S E V I E R

I S S N 1 9 3 6 - 8 7 8 X / $ 3 6 . 0 0

h t t p : / / d x . d o i . o r g / 1 0 . 1 0 1 6 / j . j c m g . 2 0 1 6 . 0 9 . 0 1 9

Routine Clinical Quantitative RestStress Myocardial Perfusion forManaging Coronary Artery DiseaseClinical Relevance of Test-Retest Variability

Danai Kitkungvan, MD,a Nils P. Johnson, MD, MS,a Amanda E. Roby, PET, CNMT, RT(N),b

Monika B. Patel, MD,a Richard Kirkeeide, PHD,c K. Lance Gould, MDd

ABSTRACT

Fro

Mc

We

Te

Ho

Ho

Ath

Ph

an

rec

ap

pro

rel

Ma

OBJECTIVES Positron emission tomography (PET) quantifies stress myocardial perfusion (in cc/min/g) and coronary

flow reserve to guide noninvasively the management of coronary artery disease. This study determined their test-retest

precision within minutes and daily biological variability essential for bounding clinical decision-making or risk stratifi-

cation based on low flow ischemic thresholds or follow-up changes.

BACKGROUND Randomized trials of fractional flow reserve–guided percutaneous coronary interventions established

an objective, quantitative, outcomes-driven standard of physiological stenosis severity. However, pressure-derived

fractional flow reserve requires invasive coronary angiogram and was originally validated by comparison to

noninvasive PET.

METHODS The time course and test-retest precision of serial quantitative rest-rest and stress-stress global myocardial

perfusion by PETwithinminutes and days apart in the same patient were compared in 120 volunteers undergoing serial 708

quantitative PET perfusion scans using rubidium 82 (Rb-82) and dipyridamole stress with a 2-dimensional PET-computed

tomography scanner (GE DST 16) and University of Texas HeartSee software with our validated perfusion model.

RESULTS Test-retest methodological precision (coefficient of variance) for serial quantitative global myocardial

perfusion minutes apart is �10% (mean DSD at rest �0.09, at stress �0.23 cc/min/g) and for days apart is �21%

(mean DSD at rest �0.2, at stress �0.46 cc/min/g) reflecting added biological variability. Global myocardial perfusion at

8 min after 4-min dipyridamole infusion is 10% higher than at standard 4 min after dipyridamole.

CONCLUSIONS Test-retest methodological precision of global PET myocardial perfusion by serial rest or stress PET

minutes apart is �10%. Day-to-different-day biological plus methodological variability is �21%, thereby establishing

boundaries of variability on physiological severity to guide or follow coronary artery diseasemanagement. Maximum stress

increases perfusion and coronary flow reserve, thereby reducing potentially falsely low values mimicking ischemia.

(J Am Coll Cardiol Img 2017;10:565–77) © 2017 by the American College of Cardiology Foundation.

m the aDivision of Cardiology, Department of Medicine and Weatherhead PET Center for Preventing Atherosclerosis,

Govern Medical School and Memorial Hermann Hospital, Houston, Texas; bPET Imaging, Department of Medicine and

atherhead PET Center for Preventing Atherosclerosis, McGovern Medical School and Memorial Hermann Hospital, Houston,

xas; cDepartment of Medicine and Weatherhead PET Center for Preventing Atherosclerosis, McGovern Medical School,

uston, Texas; and the dWeatherhead PET Center for Preventing and Reversing Atherosclerosis, McGovern Medical School,

uston, Texas. Dr. Johnson has received internal funding from the Weatherhead PET Center for Preventing and Reversing

erosclerosis; has received significant institutional research support from St. Jude Medical (for NCT02184117) and Volcano/

ilips Corporation (for NCT02328820), makers of intracoronary pressure and flow sensors; and has an institutional licensing

d consulting agreement with Boston Scientific for the smart minimum fractional flow reserve (FFR) algorithm. Dr. Gould has

eived internal funding from the Weatherhead PET Center for Preventing and Reversing Atherosclerosis; and is the 510(k)

plicant for CFR Quant (K113754) and HeartSee (K143664), software packages for cardiac positron emission tomography image

cessing and analysis, including absolute flow quantification. All other authors have reported that they have no

ationships relevant to the contents of this paper to disclose.

nuscript received April 26, 2016; revised manuscript received September 6, 2016, accepted September 8, 2016.

FIGUR

Solid h

tomog

CFR ¼

ABBR EV I A T I ON S

AND ACRONYMS

ANOVA = analysis of variance

CAD = coronary artery disease

CFR = coronary flow reserve

COV = coefficient of variance

ECG = electrocardiography

KS = Kolmogorov-Smirnov

LV = left ventricle

PET = positron emission

tomography

PRP = pressure rate product

RCA = right coronary artery

Kitkungvan et al. J A C C : C A R D I O V A S C U L A R I M A G I N G , V O L . 1 0 , N O . 5 , 2 0 1 7

Precision and Variability of Quantitative Myocardial Perfusion M A Y 2 0 1 7 : 5 6 5 – 7 7

566

P ositron emission tomography (PET)quantifies stress myocardial perfu-sion in units of cc/min/g and cor-

onary flow reserve (CFR) (1–3) asnoninvasive guides to management of coro-nary artery disease (CAD) paralleling indi-rect relative flow reserve of invasivepressure-derived fractional flow reserve(4,5) originally validated by comparison toPET (6). Therefore, their test-retest precisionand daily biological variability are essentialfor clinical decision-making based onthresholds of low myocardial perfusioncausing ischemia.

SEE PAGE 578

Despite the long history of dipyridamole stressoriginated by Gould et al. (7,8), we critically exam-ined our technology and biological variability ofstress perfusion and CFR to determine whethermethodology variability needed technical impro-vement or day-to-day biological variability werelimitations to guiding or following management ofCAD as recently reported for regadenoson (9).Therefore, this study aimed to establish the essentiallink between comprehensive quantification ofperfusion measurements and their variability that

E 1 Protocols for Quantitative Myocardial Perfusion Imaging

STRESS PET

PET 1

PET 1

PET 1

PET 1

PET 1

PE

39

PET 2

8-13 pair

8-14 pair

8-15 pair

8-16 pair

7-12 pair

4 min PET Imageor dipyridamole infusion

REST Dipyridamole PET

Minutes

-4 0 4 7 8 12

stress

eavy horizontal bars indicate 4 min of dipyridamole infusion. Num

raphy (PET) scans for timing protocols. Small parallel lines indica

coronary flow reserve.

sets bounds for clinical decision-making or riskstratification.

METHODS

From November 2014 to August 2015, subjects40 years or older were recruited at Weatherhead PETCenter for Preventing and Reversing Atherosclerosisof University of Texas Medical School. Writteninformed consent was obtained from volunteers(commonly with risk factors), patients referred forclinical PET who did not have insurance coverage, orclinic patients who desired PET follow-up.

Exclusion criteria included contraindication todipyridamole, pregnancy, active breastfeeding, clin-ical instability, and inability to undergo 2 PET pro-tocols (Figure 1) within 2 days to 3 weeks apart inwhich early-late sequences within minutes wererandomized before the first PET scan.CARDIAC PET ACQUISITION AND ANALYSIS. Subjectswere instructed to fast for 4 h and to abstain fromcaffeine, theophylline, and cigarettes for at least 24 h.Cardiac PET used the Discovery ST 16-slice PET-computed tomography scanner (GE Healthcare,Waukesha, Wisconsin) in 2-dimensional mode aspreviously reported (2,3,7–12).

Emission images were obtained over 4 min atstarting intravenous injection of 30 to 50 mCi of

by PET Using Rb-82

T 2

34

32

47

20

PET 2

PET 2

PET 2

13 14 15 16

bers under each solid bar are the number of paired positron emission

te a time break in x-axis for different paired protocols.

FIGURE 2 Plot of Stress Perfusion (Flow) and CFR for Each of 1,344 Perfusion Pixels of LV

Plot determining unique, unequivocal color code for stress perfusion and CFR for each pixel that is back-projected onto coronary flow capacity

map with histogram for percentage of all pixels in LV in each color-coded range of stress flow and CFR of each pixel. For illustration, only

lateral views of stress flow and CFR are shown, and a single pixel is tracked through the pixel color-coding process. Quadrant mean perfusion

values are calculated as average of all pixels in that quadrant. Global perfusion is calculated as average of all pixel values in the LV.

CFR ¼ coronary flow reserve; LV ¼ left ventricle.

J A C C : C A R D I O V A S C U L A R I M A G I N G , V O L . 1 0 , N O . 5 , 2 0 1 7 Kitkungvan et al.M A Y 2 0 1 7 : 5 6 5 – 7 7 Precision and Variability of Quantitative Myocardial Perfusion

567

generator-produced rubidium 82 (Rb-82) (BraccoDiagnostics, Princeton, New Jersey). The first 2-minemission acquisition comprised arterial input images.The last 2-min emission acquisition comprisedmyocardial uptake images. Pharmacological stressused dipyridamole infusion (0.56 mg/kg) over 4 min(0.142 mg/kg/min).

An experienced PET cardiologist administereddipyridamole and monitored every patientthroughout imaging, followed by 75 mg of intrave-nous aminophylline. Angina was treated with intra-venous aminophylline, metoprolol, or sublingualnitroglycerin. Continuous heart rate, blood pressure,and 12-lead electrocardiographic (ECG) monitoringduring stress identified significant >1-mm ST-segment depression.

QUANTITATIVE PET ANALYSIS. Computed tomogra-phy scans for attenuation correction were acquiredbefore rest and after the last stress emission imagingat reduced radiation dose as previously reported(2,3,7–12). Coregistration was optimized for every

image by shifting PET to fit attenuation data andreconstructed as previously reported (10).

As previously reported (2,3,7–12), for each radialsegment of every short-axis slice, absolute myocardialperfusion (in cc/min/g) was quantified for each of 1,344pixels in the left ventricular (LV) image with a 5-pixelsmoothing noise reduction algorithm using HeartSeesoftware (University of Texas-Houston, Houston,Texas, FDA K14366) approximating reconstructedscanner resolution of 1.5-cm full width one-halfmaximum with filters. This software incorporates ourvalidated model for Rb-82 (13), reported by others to“have higher sensitivity for detection and localizationof abnormal flow” (14) than multicompartmentalmodels using time-activity curves derived from serialshort (15-s) noisy images. Optimal arterial inputs werecustomized for individual patients from among aorticand left atrial locations as previously reported todetermine rest and stress flow (15).

CFR was computed as stress-to-rest ratio for eachof 1,344 pixels, synonymous with myocardial perfu-sion reserve to emphasize physiological concepts.

TABLE 1 Patient Characteristics

Serial StressProtocol (n ¼ 89)

Serial RestProtocol (n ¼ 31)

Clinical characteristics

Age, yrs 58.5 � 9.9 56.3 � 9.7

BMI, kg/m2 29.2 � 5.2 27.9 � 4.8

Male 58 (65.2) 19 (61.3)

Hypertension 42 (47.2) 9 (29.0)

Dyslipidemia 61 (68.5) 23 (74.2)

Diabetes 11 (12.4) 2 (6.5)

Active smoking 5 (5.6) 2 (6.5)

Prior PCI 8 (9.0) 2 (6.5)

Prior CABG 2 (2.3) 2 (6.5)

Prior myocardial infarction 4 (4.5) 0 (0.0)

Symptoms before cardiac PET

Angina 0 (0.0) 1 (3.2)

Dyspnea 2 (2.3) 5 (16.1)

Medications

Statin 34 (38.2) 12 (38.7)

Antiplatelet 42 (47.2) 12 (38.7)

Beta-blocker 19 (21.4) 6 (19.4)

ACEI or ARB 31 (34.8) 6 (19.4)

Calcium-channel blocker 10 (11.2) 3 (9.7)

Diuretics 15 (16.7) 7 (22.6)

Nitrate 2 (2.3) 0 (0.0)

Values are mean � SD or n (%).

ACEI ¼ angiotensin-converting enzyme inhibitor; ARB ¼ angiotensin receptorblocker; BMI ¼ body mass index; CABG ¼ coronary artery bypass graft; PCI ¼percutaneous coronary intervention; PET ¼ positron emission tomography.

Kitkungvan et al. J A C C : C A R D I O V A S C U L A R I M A G I N G , V O L . 1 0 , N O . 5 , 2 0 1 7

Precision and Variability of Quantitative Myocardial Perfusion M A Y 2 0 1 7 : 5 6 5 – 7 7

568

Coronary flow capacity maps display stress perfusionand CFR for each pixel as a percentage of LV asillustrated in Figure 2, which maps each of 1,344 LVpixels with its stress perfusion and CFR value at itsregional location (12). The histogram distribution of1,344 pixel severities defined by both stress flow andCFR then provides a pixel-level analysis of the entirerange, size, and distribution of stress perfusion andCFR for assessing the effects of maximal or submax-imal stress. Global values are the average of 1,344pixels; quadrant values are averages of all pixelvalues (336) in nonoverlapping quadrants.

REST AND DIPYRIDAMOLE STRESS IMAGING

PROTOCOLS. Rest perfusion precision. Sequential rest-rest-stress PET perfusion imaging was performed at10-min intervals on day 1 and repeated 1 to 3 weekslater on day 2. The serial 10-min rest-rest scans quan-tified test-retest methodological precision formeasuring resting perfusion (in cc/min/g), whereasrepeat rest scans on different days assessed dailybiological plus methodological variability in the samepatient.

Stress perfusion precision. Quantifying test-retestprecision of stress perfusion (in cc/min/g) and CFRrequires stable constant stress perfusion for

approximately 15 min to allow serial acquisition ofstress perfusion images. Because the time course ofmyocardial hyperemia after standard 4-min dipyr-idamole infusion has not been defined, the protocolsshown in Figure 1 were implemented in repeat pairedPET studies in the same patient at minutes and daysintervals.

After dipyridamole injection starting at time 0, thefirst RB-82 generator activation occurred at 7 or 8 min(day 1 PET 1) and again at 12, 13, 14, 15, or 16 min onday 2 (day 1 PET 2). Radiotracer delivery, PET scanneracquisition, image processing, and quantitativeperfusion software remained the same for all PETscans. Subjects returned for the second PET scan (day2 PET 1 and PET 2) within 3 weeks with similar orrandomized different time intervals.

STATISTICAL ANALYSIS. Precision and variabilitywere determined for global, average quadrant, andindividual pixel values of stress flow and CFR.R version 3.1.0 (R Foundation for StatisticalComputing, Vienna, Austria) and standard summarystatistical tests were used. Applicable tests were2-tailed, and p < 0.05 was considered statisticallysignificant. Linear regression is reported between restperfusion and rest pressure rate product (PRP).Analysis of variance (ANOVA) compared characteris-tics among timing protocols. Paired or unpaired Stu-dent t test was used to evaluate continuous variableswhere appropriate. The Pitman-Morgan F test wasused to test for differences in variability of stressperfusion between 2 groups. An ANOVA model withmixed effects (to account for repeated measurementsfrom the same subject) compared absolute flow andCFR among various timing sequences. Because anoverall ANOVA p value was significant, a Tukey all-pair comparison was applied to determine whichtiming conditions provided a different response. Tocompare the histogram distribution between groupsof stress perfusion and CFR for each of 1344 pixels aspercentage of LV in color-coded ranges of coronaryflow capacity, we used the Kolmogorov-Smirnov (KS)test for differences in histogram distribution.

RESULTS

One hundred twenty subjects underwent 708 PETquantitative PET perfusion scans. Thirty-one subjectsunderwent rest-rest-stress protocol on day 1 and 1 to 3weeks later on day 2. Eighty-nine subjects underwentrest-stress-stress protocol on days 1 and 2. The me-dian between paired PETs was 16 days (mean 22 � 15days) with no change in medical status or medica-tions. Baseline characteristics of subjects are listed inTable 1. Four day-2 PET sessions were not obtained

FIGURE 3 Stress PET Images of Relative Myocardial Uptake of Rb-82

(A) Stress PET images of relative myocardial uptake of Rb-82 after standard 4-min infusion of dipyridamole scaled by color bar from 100% for

maximum relative uptake (white), with red being next highest, progressively graded to yellow, green, and blue-purple for severe relative

defect as previously reported (2,3,7–12). Quadrant views of LV follow generic coronary artery distributions indicated by shadow overlays.

(B) Stress perfusion (in cc/min/g). (C) Coronary flow reserve. (D) Coronary flow capacity map integrating both stress flow and CFR into

a single 4-view display of the LV. See text for color-coded regional severity as percentage of LV based on color-coded steps in Figure 2.

AV ¼ atrioventricular node artery; D1 ¼ first diagonal; D2 ¼ second diagonal; LAD ¼ left anterior descending; LCx ¼ left circumflex;

OM1 ¼ first obtuse marginal; OM2 ¼ second obtuse marginal; PET ¼ positron emission tomography; RCA ¼ right coronary artery; RI ¼ ramus

intermedius; other abbreviations as in Figure 2.

J A C C : C A R D I O V A S C U L A R I M A G I N G , V O L . 1 0 , N O . 5 , 2 0 1 7 Kitkungvan et al.M A Y 2 0 1 7 : 5 6 5 – 7 7 Precision and Variability of Quantitative Myocardial Perfusion

569

because subjects withdrew, and 5 PET sessions wereexcluded because of technical issues of scanneroperations or settings, intravenous infusion of Rb-82or dipyridamole, or venous abnormalities that inva-lidated arterial input.

Because definitive regional quantitative myocar-dial perfusion by PET for guiding management ofCAD is not widely recognized, Figure 3 illustratesrelative stress myocardial perfusion images in com-plex CAD in which angiograms did not provideguidance for revascularization compared to medicalmanagement.

The patient shown in Figure 3 is a 65-year-old manwith risk factors and a right coronary artery(RCA) stent inserted in 2009 who was referredfor PET because of abnormal stress test andangiogram showing in-stent RCA occlusion with

distal collaterals, moderate stenosis of the first andsecond diagonal branches, and moderate stenosisof the first and second obtuse marginal branches.

Detailed review of images for this patient providesinsight into diffuse and focal physiological severityunfamiliar to most readers and its precision. Restingrelative perfusion images showed a basal inferiortransmural scar composing 10% of the LV (notshown). Relative stress perfusion images show a se-vere defect involving 22% of inferior LV in the RCAdistribution, indicating large border zones of viablemyocardium with reduced CFR composing an addi-tional 14% of LV (Figure 3A). ECG-gated perfusionimages showed ejection fraction of 63% at rest and67% during dipyridamole stress, with stress-inducedinferior hypokinesis without angina or ECG changes.In view of multiple stenosis by angiogram, relative

FIGURE 4 Serial PET of Same Patient With Complex CAD in 2 of 4 Quadrant Views

Test-retest precision of serial stress quantitative PET perfusion measured within minutes and on different days for day-to-day biological

variability. CAD ¼ coronary artery disease; PET ¼ positron emission tomography.

Kitkungvan et al. J A C C : C A R D I O V A S C U L A R I M A G I N G , V O L . 1 0 , N O . 5 , 2 0 1 7

Precision and Variability of Quantitative Myocardial Perfusion M A Y 2 0 1 7 : 5 6 5 – 7 7

570

images do not identify or quantify the extent of“balanced stenosis,” low flow ischemia, or diffusedisease essential for management decisions.

Quantitative stress perfusion (in cc/min/g) wasmoderately reduced diffusely and severely reduced ininferior, inferolateral, and inferoseptal quadrants ofLV according to the color bar scale for stress perfusion(flow) (Figure 3B). All quantitative color bars arecoded red for 125 healthy young volunteers youngerthan 40 years without risk factors; orange for healthysubjects with risk factors but no known CAD; yellowfor patients with known CAD with or without revas-cularization; blue for patients with stress perfusiondefects, angina, and/or >1-mm ST-segment depres-sion on ECG during dipyridamole stress; and green fora stress defect and either angina or ST-segmentdepression but not both, as previously reported(2,3,7–12). CFR outside the stress defect is mildly tomoderately reduced diffusely but above low flowischemic levels reflecting diffuse, nonischemic coro-nary atherosclerosis in those regions (Figure 3C). Inthe stress defect, CFR is severely reduced inferiorly tobelow resting perfusion in a small segment, indi-cating myocardial steal (dark blue) associated withcollaterals beyond the RCA chronic total occlusion.

These 2 primary flow metrics, stress perfusion (incc/min/g) and CFR, completely define physiologicalseverity but are complex to interpret independently.Accordingly, as previously reported (2,3,7–12),they are integrated into a coronary flow capacitymap (Figure 3D) with color-coded pixels for thesame ranges of patient groups according to the2-dimensional plot in Figure 2. Coronary flow capacityis severely reduced inferiorly in 26% of the LV, ofwhich 10% is transmural scar and 14% viable borderzones with reduced flow capacity. The remaining LVoutside border zones of the inferior stress defect(green) has mild diffusely reduced coronary flow ca-pacity (yellow) due to diffuse coronary atheroscle-rosis but is adequate above ischemic low flow. Thereferring cardiologist and consulting PET cardiologyfaculty concluded that medical treatment was a validoption without bypass surgery because of: 1) reducedbut adequate coronary flow capacity withoutischemia in left anterior descending coronary arteryand left circumflex distributions; 2) collaterals toapproximately 16% of viable inferior border zonesadequate for the patient’s daily activities withoutangina; 3) absence of angina or heart failure; and 4)normal rest and stress ejection fraction.

FIGURE 5 Serial PET in Single Quadrant Views of 2 Different Patients

PET from a healthy young volunteer with high coronary flow capacity and a patient with risk factors only and mildly reduced coronary flow

capacity, illustrating test-retest comparison of serial stress quantitative PET perfusion measured within minutes for a wide range of perfusion

values. PET ¼ positron emission tomography.

TABLE 2 Minute-to-Minute and Day-to-Day Test-Retest Precision of Rest Perfusion in

the Same Patient

Rest PET Serial No.Mean Rest,cc/min/g Bias SD of D

Paired Studentt Test p Value

Pitman-Morganp Value

COVRest

Rest 1 day 1 0.87 � 0.22

Rest 2 day 1 0.86 � 0.19

Rest 1 vs. rest 2* 0.001 �0.093 0.93 0.06 10.7%

Rest day 1–rest day 2 0.07 �0.20 0.13 21.1%

Rest flow minutes D vs. days D <0.001

Values are mean � SD unless otherwise indicated. Pitman-Morgan F test for coefficient of variance (COV) ¼ SD ofdifferences/mean of the 2 measurements. *Rest 1 and rest 2 acquired 10 min apart.

PET ¼ positron emission tomography.

J A C C : C A R D I O V A S C U L A R I M A G I N G , V O L . 1 0 , N O . 5 , 2 0 1 7 Kitkungvan et al.M A Y 2 0 1 7 : 5 6 5 – 7 7 Precision and Variability of Quantitative Myocardial Perfusion

571

For clinical reliability, this comprehensive quanti-fication of complex physiological severity of focal anddiffuse CAD is highly reproducible, as shown by serialPETs shown in lateral and inferior views for simplicityin Figure 4. Subjects were studied over the full range,from low rest perfusion or severe stress impairmentto high coronary flow capacity in healthy young vol-unteers with no risk factors (Figure 5). These exam-ples also illustrate that stress perfusion at 13 min washigher than at 8 min and at 15 min was lower than at 8min (Figure 5).

TEST-RETEST PRECISION IN THE SAME PATIENT OF

REST PERFUSION. Table 2 lists test-retest precision ofserial resting perfusion (in cc/min/g) over minutes andday-to-day variability in the same patient. For same-day rest 1–rest 2 perfusion minutes apart in the samepatient, the SD of differences was �0.093 cc/min/g,and the coefficient of variance (COV) was 10.7% (0.093of 0.87). For the day-to-different-day difference in restperfusion in the same patient, the SD of differenceswas 0.2 and COV was 21.1%, significantly higher byPitman-Morgan F test (p < 0.001).

Therefore, the 21.1% variability observed onrepeat rest perfusion measurements on differentdays is partly due to 10.7% imprecision of method-ology with comparable additional contributionfrom biological variability. Resting perfusion waspreviously reported to have a modest direct linearrelation to PRP (11,12). Similarly, in this study restperfusion and rest PRP were linearly related: rest

TABLE 3 Stress Myocardial Perfusion in Paired Sequential PET at Minute Intervals After Baseline PET

No. ofStudies

Rest Flow,cc/min/g

Stress Flow (cc/min/g) CFR

Stress Flow PET 1 Stress Flow PET 2 p ValueDFlow Stress2–Stress 1 CFR PET 1 CFR PET 2 p Value

DCFR Stress2–Stress 1

7- to 12-min pair 39 0.90 � 0.35 2.07 � 0.61 2.31 � 0.68 <0.001 0.24 � 0.29 2.43 � 0.59 2.72 � 0.66 <0.001 0.28 � 0.33

8- to 13-min pair 34 0.87 � 0.33 2.13 � 0.58 2.38 � 0.59 <0.001 0.25 � 0.30 2.64 � 0.73 2.92 � 0.65 <0.001 0.29 � 0.38

8- to 14-min pair 32 0.84 � 0.26 2.12 � 0.47 2.21 � 0.52 0.018 0.09 � 0.21 2.65 � 0.64 2.75 � 0.67 0.063 0.10 � 0.29

8- to 15-min pair 47 0.93 � 0.23 2.39 � 0.57 2.39 � 0.57 0.90 0.02 � 0.26 2.63 � 0.50 2.64 � 0.50 0.70 0.02 � 0.29

8- to 16-min pair 20 0.87 � 0.31 2.04 � 0.70 1.93 � 0.57 0.14 –0.12 � 0.34 2.48 � 0.76 2.33 � 0.64 0.16 –0.15 � 0.45

Values are mean� SD unless otherwise indicated. Bold indicates a significant difference between the paired PETs in those rows. Bold italics indicates that the 8- to 15-min paired PETs differed by only 0.02�0.26, that is not significant and reflects test-retest precision.

CFR ¼ coronary flow reserve; PET ¼ positron emission tomography.

FIGURE 6 Time Co

2.0

1.5

1.0

0.5

0.0

7 or 8 mi

R

Rat

io t

o 7

or

8 M

inu

te F

low

100%

Changes in stress flow

PET 2/PET 1 at 7 and 8

indicate 1 SD.

Kitkungvan et al. J A C C : C A R D I O V A S C U L A R I M A G I N G , V O L . 1 0 , N O . 5 , 2 0 1 7

Precision and Variability of Quantitative Myocardial Perfusion M A Y 2 0 1 7 : 5 6 5 – 7 7

572

PRP ¼ 3,565 � rest perfusion þ 4,196 and R2 ¼ 0.25,indicating that PRP accounted for 25% of variation inrest perfusion.

However, differences between day 1 and day 2PETs in resting heart rate (1 � 6 beats/min), systolicblood pressure (3 � 11 mm Hg), and diastolic bloodpressure (0 � 6 mm Hg) were small. In contrast,some patients may have greater differences inpressure rate product because of anxiety, medica-tions, caffeine, and labile blood pressure alteringresting perfusion.

urse of Stress Perfusion During Dipyridamole Stress

n 12 min 13 min 14 min 15 min 16 min

b-82 Injection Timing After Dipyridamole Infusion

110% 109% 103% 102% 94%

(in cc/min/g) at different time intervals expressed as ratio of

min.Heavy solid red horizontal bars indicate average. Thin red bars

TIME COURSE OF MYOCARDIAL PERFUSION AFTER

4-MIN DIPYRIDAMOLE INFUSION. With time 0 forstarting the standard 4-min dipyridamole infusionand stress PET 1 as the reference, stress perfusion ofPET 2 at 12, 13, or 14 min was significantly higherthan PET 1 at 7 or 8 min; was highest at 12 and 13min; and decreased at 14, 15, and 16 min with nosignificant difference between perfusion at 8 versus15 min (Table 3). Stress perfusion at each timeinterval as a ratio to perfusion at minute 7 or 8is graphed in Figure 6, which show a peak of 1.10at minute 12 to 13 falling thereafter to below1.0 at minute 16.

Stress perfusion changes significantly over timeafter the 4-min dipyridamole injection, with a globalp < 0.001 by the mixed-effects ANOVA model forstress flow, accounting for repeated measurements inthe same subjects. Because the global ANOVA test issignificant, pairwise comparisons at different timeintervals can be compared using the Tukey test formultiple comparisons. Table 4 lists the Tukey test formultiple comparisons of differences among PET 1–PET 2 stress perfusion (cc/min/g) for the row minusthe column for minute intervals between stress PET1 and stress PET 2. For example, the 12-min dipyr-idamole perfusion is 0.243 cc/min/g higher than the8-min dipyridamole perfusion (p ¼ 0.0044). The16-min perfusion is 0.362 cc/min/g lower than the12-min dipyridamole perfusion (p # 0.02). All bolddifferences are significant with p # 0.022; all othervalues (italics or roman) are not significantlydifferent.

TEST-RETEST PRECISION IN THE SAME PATIENT OF

STRESS PERFUSION. The subjects with 2 8- to 15-minstress pairs provide the most compelling analysis ofpure methodological test-retest precision of stressperfusion. For 15 to 8 min PETs on day 1, the meandifference was 0.02 � 0.26 cc/min/g with COV of10.8% (0.259 of 2.393) (Table 5). For 15- to 8-min PETs

TABLE 4 Tukey Test for Multiple Comparisons of Differences Among PET 1–PET 2 Stress

Perfusion (cc/min/g) for the Row Minus the Column for Minute Intervals Between Stress

P1 and Stress PET 2

Stress Flow D 7 8 12 13 14 15

8 0.014 p £ 0.022

12 0.256 0.243 p < 0.20

13 0.229 0.216 �0.027

14 0.104 0.091 �0.152 �0.125

15 0.042 0.029 �0.214 �0.187 �0.062

16 �0.106 �0.119 L0.362 L0.335 �0.210 �0.148

All bold differences between PET pairs defined by the columns and rows are significant with p # 0.022 indicatingthat their differences listed in the table are significant where a positive sign indicates a higher value on PET2 thanPET1 and a negative sign indicates a lower value on PET2 than on PET1. Outside the bold cells, no differences aresignificant; italics indicate a substantial negative difference with a lower value on PET2 that, however, is notsignificantly different.

PET ¼ positron emission tomography.

TABLE 5 Minute-to-Minute and Day-to-Day Test-Retest Precision of Stress Perfusion in

Same Patient

Stress PET SequenceMean D,cc/min/g

p Value by PairedStudent t Test

Pitman-Morganp Value COV

Day 1 15- to 8-min PET 0.02 � 0.26 0.74 10.6%

Day 2 15- to 8-min PET �0.03 � 0.23 0.49 9.6%

Day 1–day 2 for 8-min PET 0.02 � 0.46 0.81 19.3%

Day 1–day 2 for 15-min PET 0.08 � 0.50 0.48 20.7%

Stress flow minutes D vs. days D #0.011

Coefficient of variance (COV) ¼ SD of differences/mean of the 2 measurements.

PET ¼ positron emission tomography.

J A C C : C A R D I O V A S C U L A R I M A G I N G , V O L . 1 0 , N O . 5 , 2 0 1 7 Kitkungvan et al.M A Y 2 0 1 7 : 5 6 5 – 7 7 Precision and Variability of Quantitative Myocardial Perfusion

573

on day 2, the difference was –0.03 � 0.23 with COV of9.6% (0.23 of 2.393).

The day-to-different-day test-retest reproducibilityof stress perfusion for day 1–day 2 in the same patientfor the 15-min PETs (Table 5) had COV of 19% to 21%that is significantly greater than minute differencesbetween PET 1–PET 2 having COV of 9.6% to 10.6%, asignificant difference with Pitman-Morgan p # 0.011.Therefore, daily variability on serial stress-stressperfusion measurements is due to approximately�10% methodological imprecision with an additionalcomparable component of biological variability. Therest day 1–day 2 differences (SD of D ¼ �0.10) (Table 2)were smaller than the stress day 1–day 2 difference (SDof D ¼ �0.50) (Table 5) with Pitman-Morgan p < 0.001but the COVs were similar, 21% for both rest and stressflows. COV for minute and day differences for CFR aresimilar to those for stress perfusion.

Table 6 lists minute-to-minute precision and day-to-day variability for regional average quadrantvalues of stress perfusion and CFR that are compa-rable to global values because both are determinedfrom the primary 1,344 pixel flows averaged for336 pixels in each quadrant and 1,344 pixels forthe entire LV.

Figure 7 illustrates the clinical relevance ofmaximal versus submaximal stress. Relative stressimages show little difference (Figures 7A and 7B).With submaximal stress, global stress perfusion was1.5 cc/min/g and CFR was 1.8 compared to maximumstress with global stress perfusion of 2.2 cc/min/g andCFR of 2.7.

Submaximal stress lowers stress perfusion andCFR, with a larger percentage of LV having lowerflows by pixel colored percentage of LV, which mightbe interpreted as abnormal for both regional anddiffuse CAD (Figures 7C and 7D). Maximal stress in-creases flows into higher ranges of color-codedperfusion, thereby reducing apparent severitycompared to submaximal stress and hence reducingpotential false-positive results due to inadequatestress. Therefore, submaximum stress (Figure 7C)erroneously suggests more severe focal and diffusedisease (yellow and green) than true coronary flowcapacity with maximal stress (red and orange)(Figure 7D).

In addition to comparable precision for globaland regional quadrants, Figure 8 shows mean indi-vidual pixel distribution of stress perfusion bycomparing histogram distribution of all 1,344 perfu-sion pixels in LV for all serial PET histograms ac-quired within minutes. There is no difference inpixel distributions in the 2 histograms by KS statistic(KS ¼ 0.06; p ¼ 0.30).

However, in Figure 9, the mean histogram of 1,344pixel distribution for all subjects with submaximumstress is significantly different than for all subjectswith maximum stress (KS statistic ¼ 0.18; p <

0.0001). Submaximal stress incurs a higher percent-age of LV in the middle perfusion ranges (yellow andorange) and a lower percentage of LV in the highestrange of perfusion (red) compared to maximal stresswith a lower percentage of LV in middle ranges andgreatest percent of LV in the highest range ofperfusion. Therefore, submaximum stress causinglower stress perfusion (yellow) may be mis-interpreted as showing diffuse CAD, small vesseldisease, or even ischemia compared to higherperfusion with maximum stress that in some pa-tients may bracket the low flow ischemic thresholdas shown in Figure 7.

DISCUSSION

Test-retest methodological precision of serialmyocardial perfusion in the same patient on serialimaging minutes apart without daily biological or

TABLE 6 Minute-to-Minute and Day-to-Day Precision of SP and CFR by LV Quadrant

PET Stress Sequence

Anterior Quadrant Average Inferior Quadrant Average Lateral Quadrant Average Septal Quadrant Average

DSD p Value COV DSD p Value COV DSD p Value COV DSD p Value COV

8- to 15-min SP 0.26 0.70 11% 0.27 0.60 12% 0.21 1.00 9% 0.28 0.50 12%

8- to 15-min CFR 0.31 0.60 11% 0.32 0.90 12% 0.24 1.00 9% 0.31 0.40 12%

Day 1–day 2 SP 0.50 0.50 20% 0.49 0.70 22% 0.49 0.40 19% 0.49 0.50 21%

Day 1–day 2 CFR 0.57 0.70 21% 0.49 0.50 19% 0.46 0.30 18% 0.52 0.20 20%

Coefficient of variance (COV) ¼ SD of differences/mean of the 2 measurements.

CFR ¼ coronary flow reserve; LV ¼ left ventricle; p ¼ p value by paired Student t test; PET ¼ positron emission tomography; DSD ¼ standard deviation of the difference in paired PETs; SP ¼ stressperfusion.

Kitkungvan et al. J A C C : C A R D I O V A S C U L A R I M A G I N G , V O L . 1 0 , N O . 5 , 2 0 1 7

Precision and Variability of Quantitative Myocardial Perfusion M A Y 2 0 1 7 : 5 6 5 – 7 7

574

intersubject variability is approximately �10%,accounting for about one-half of the approximately�20% day-to-day variability due to added biologicalvariability. Our results show that myocardial perfu-sion is maximal at 8 min after completing the 4-mindipyridamole infusion, averaging 10% higher than

FIGURE 7 Relative Stress Perfusion Images of a Patient With Subm

(A) Relative stress images showing a small, mild-to-moderate, basal ante

mild-to-moderate, basal inferolateral defect in the distal left circumflex

defects are slightly worse on maximal stress, increasing from 7% to 10%

flow capacity map combining both stress perfusion (in cc/min/g) and CFR

stress (D). Submaximum stress (C) erroneously suggests more severe foc

capacity with maximal stress (D) (orange and red). CFR ¼ coronary flow

perfusion imaging at the standard 4 min after dipyr-idamole. For some patients with CAD severity nearthe low flow ischemic threshold, this 10% increasedstress flow may substantially change the relativeimportance of focal and diffuse CAD for integratedphysiological severity as illustrated in Figure 7.

aximum Stress and Maximum Stress

rior stress defect in a small diagonal branch distribution and a small,

distribution, both confirmed by angiogram. (B) Small stress relative

of the left ventricle below 60% of maximum. However, the coronary

with submaximal stress (C) is substantially improved with maximal

al and diffuse disease (yellow and green) than true coronary flow

reserve.

FIGURE 8 Methodological Histogram Reproducibility

0.60

0.50

0.40

0.30

0.20

0.10

0.00

Severe Moderate Mild Minimal Good

Fra

ctio

n o

f L

eft

Ven

tric

le

Young Healthy Volunteers

Minimally Reduced

Mildly Reduced

Moderately Reduced

Severely Reduced

PET 1

PET 2

Histograms of the average fractions of left ventricle in color-coded severity ranges of coronary flow capacity of 2 serial stress PETs acquired

minutes apart accounting for all 1,344 pixels in each PET. PET ¼ positron emission tomography.

J A C C : C A R D I O V A S C U L A R I M A G I N G , V O L . 1 0 , N O . 5 , 2 0 1 7 Kitkungvan et al.M A Y 2 0 1 7 : 5 6 5 – 7 7 Precision and Variability of Quantitative Myocardial Perfusion

575

PRIOR LITERATURE. Minute-to-minute methodolog-ical precision of quantitative myocardial stressperfusion by PET has not been previously reported.Our day-to-day variability of dipyridamole stressperfusion compares with day-to-day variability in ourregadenoson study (9).

FIGURE 9 Histogram Differences Between Maximum and Submaxim

0.60

0.50

0.40

0.30

0.20

0.10

0.00

Severe Moderate

Fra

ctio

n o

f L

eft

Ven

tric

le

Young Healthy Volunteers

Minimally Reduced

Mildly Reduced

Moderately Reduced

Severely Reduced

Submaximum Stress

Histograms of the average fractions of left ventricle in color-coded seve

submaximum (dashed line) and maximum (solid line) vasodilator stress

CLINICAL RELEVANCE. Our reported low stress flowischemic threshold of 0.9 cc/min/g is the stress flowthat maximizes the area under the receiver-operatingcharacteristic curve (98%), predicting ischemiadefined as significant regional stress defect with ECGdepression >1 mm or moderate-to-severe angina

um Stress

Mild Minimal Good

Maximum Stress

rity ranges of coronary flow capacity for 2 serial PETs with

obtained minutes apart. PET ¼ positron emission tomography.

PERSPECTIVES

COMPETENCY IN MEDICAL KNOWLEDGE:

Personalized medical management increasingly

depends on complex quantitative measurements

integrated with traditional history, examination, blood

tests, functional testing, or visually interpreted images.

For laboratory testing, the test-retest precision and

biological variability are well defined and indeed are

criteria for laboratory accreditation. However, for most

cardiac imaging, particularly quantitative myocardial

perfusion, this systematic, objective, definitive mea-

surement of test-retest methodology precision and

day-to-day biological variability are rarely determined

because of the time, expense, and complexity involved

in doing so, replaced with an emotional bias that “our

methods are the best,—are adequate,—are satisfactory

for our purposes.”As quantitativemyocardial perfusion

increasingly guides management of CAD for optimal

personalized outcomes, this study sets the standard for

variability of quantitative perfusion to guide invasive

procedures that may help or harm our patients: �10%

test-retest methodology precision plus �10% biolog-

ical day-to-day biological variability for �20% total

variability on repeat measurements on different days

for severity thresholds or after changes to guide man-

agement of CAD.

TRANSLATIONAL OUTLOOK: For guiding clinical

management, quantitative myocardial perfusion

imaging needs validation by its integration with

management to predict personalized optimal out-

comes, which is the greatest benefit with the least

harm based on some objective evidence-based,

threshold, or risk-to-benefit balance documented for

the size and severity of quantitative abnormalities.

However, measurement variability profoundly affects

correlation with subsequent MACE and, hence,

optimal management based on MACE predictions. This

study on the “mundane” methodology of quantitative

myocardial perfusion is the essential link between

comprehensive integration of all myocardial perfusion

measurements, their variability, and their bounds for

predicting MACE or for clinical decision-making

affected by measured variability. It provides the

essential technical basis for variability-dependent,

evidenced-based prediction of MACE, which requires

different future studies.

Kitkungvan et al. J A C C : C A R D I O V A S C U L A R I M A G I N G , V O L . 1 0 , N O . 5 , 2 0 1 7

Precision and Variability of Quantitative Myocardial Perfusion M A Y 2 0 1 7 : 5 6 5 – 7 7

576

requiring aminophylline reversal (11,12). Thisthreshold of 0.9 cc/min/g reflects a high probability ofischemia during dipyridamole stress in a large groupof patients, including methodology and day-to-daybiological variability in the same patient and be-tween different patients (11,12). The methodologyplus biological variability of �20% indicates that fortrue flow of 0.9 cc/min/g, repeated measurementswould give a range of flows between 0.9 � 20% or 0.72to 1.08, averaging 0.9 cc/min/g.

Therefore, day-to-day variability of �20% does notnegate this threshold but rather reinforces its validitywithin objective probability bounds, which has notbeen previously demonstrated for coronary bloodflow. The �10% due to methodology and the added10% due to biological variability leave somewhatlimited opportunity for further improvement,thereby defining the probability bounds formeasuring low perfusion (in cc/min/g) as the ischemicthreshold to guide, follow, or risk-stratify CADmanagement.

STUDY LIMITATIONS. Although data are from a sin-gle experienced center, our results exemplify a stan-dard for variability of routine quantitative myocardialperfusion imaging by PET to guide or follow CADmanagement. The total radiation dose for 2 3-PETsequences was 16 mSv, comparable to an averageTc-99m sestamibi rest stress study (16).

CONCLUSIONS

Methodological test-retest precision of serial quanti-tative myocardial perfusion by PET within minutesapart in the same patient is �10% and days apartis �20% due to added biological variability, there-by establishing the boundaries of variability for phys-iological severity to guide or follow CADmanagement.Therefore, this study provides an essential link be-tween comprehensive quantification of perfusionmeasurements and their variability that set the boundsfor clinical decision-making or risk stratificationaffected by measurement variability.

ADDRESS FOR CORRESPONDENCE: Dr. K. LanceGould, Weatherhead PET Center for Preventing andReversing Atherosclerosis, McGovern Medical School,University of Texas Health Science Center at Houston,6431 Fannin Street, Room MSB 4.256, Houston,Texas 77030. E-mail: k.lance.gould@uth.tmc.edu.

J A C C : C A R D I O V A S C U L A R I M A G I N G , V O L . 1 0 , N O . 5 , 2 0 1 7 Kitkungvan et al.M A Y 2 0 1 7 : 5 6 5 – 7 7 Precision and Variability of Quantitative Myocardial Perfusion

577

RE F E RENCE S

1. Taqueti VR, Hachamovitch R, Murthy VL, et al.Global coronary flow reserve is associated withadverse cardiovascular events independently ofluminal angiographic severity, and modifies theeffect of early revascularization. Circulation 2015;131:19–27.

2. Gould KL, Johnson NP, Kaul S, et al. Patientselection for elective revascularization to reducemyocardial infarction and mortality: new lessonsfrom randomized trials, coronary physiology, andstatistics. Circ Cardiovasc Imaging 2015;8:e003099.

3. Gould KL, Johnson NP, Bateman TM, et al.Anatomic versus physiologic assessment of coro-nary artery disease: role of CFR, FFR, and PETimaging in revascularization decision-making.J Am Coll Cardiol 2013;62:1639–53.

4. van Nunen LX, Zimmermann FM, Tonino PA,et al. Fractional flow reserve versus angiographyfor guidance of PCI in patients with multivesselcoronary artery disease (FAME): 5-year follow-upof a randomised controlled trial. Lancet 2015;386:1853–60.

5. De Bruyne B, Fearon WF, Pijls NH, et al., FAME 2Trial Investigators. Fractional flow reserve-guidedPCI for stable coronary artery disease. N Engl JMed 2014;371:1208–17.

6. De Bruyne B, Baudhuin T, Melin JA, et al. Cor-onary flow reserve calculated from pressuremeasurements in humans. Validation with positron

emission tomography. Circulation 1994;89:1013–22.

7. Gould KL, Westt JR, Hamilton GW. Noninvasiveassessment of coronary stenoses by myocardialimaging during coronary vasodilation. II. Clinicalmethodology and feasibility. Am J Cardiol 1978;41:279–89.

8. Gould KL, Goldstein RA, Mullani NA, et al.Noninvasive assessment of coronary stenoses bymyocardial perfusion imaging during pharmaco-logic coronary vasodilation. V. Detection of 47%diameter coronary stenosis with intravenous13NH4þ and emission computed tomography inintact dogs. Am J Cardiol 1979;43:200–8.

9. Johnson NP, Gould KL. Regadenoson versusdipyridamole hyperemia for cardiac positronemission tomography (PET) imaging. J Am CollCardiol Img 2015;8:438–47.

10. Gould KL, Pan T, Loghin C, et al. Frequentdiagnostic errors in cardiac PET-CT due tomisregistration of CT attenuation and emissionPET images: a definitive analysis of causes, con-sequences and corrections. J Nucl Med 2007;48:1112–21.

11. Johnson NP, Gould KL. Physiological basis forangina and ST-segment change: PET-verifiedthresholds of quantitative stress myocardialperfusion and coronary flow reserve. J Am CollCardiol Img 2011;4:990–8.

12. Johnson NP, Gould KL. Integrating noninvasiveabsolute flow, coronary flow reserve, and ischemicthresholds into a comprehensive map of physio-logic severity. J Am Coll Cardiol Img 2012;5:430–40.

13. Yoshida K, Mullani N, Gould KL. Coronaryflow and flow reserve by positron emission to-mography simplified for clinical application usingRb-82 or N-13 ammonia. J Nucl Med 1996;37:1701–12.

14. deKemp RA, Yoshinaga K, Beanlands RSB. Will3-dimensional PET-CT enable the routine quanti-fication of myocardial blood flow? J Nucl Cardiol2007;14:380–97.

15. Vasquez AF, Johnson NP, Gould KL. Variationin quantitative myocardial perfusion due to arterialinput selection. J Am Coll Cardiol Img 2013;6:559–68.

16. Jerome SD, Tilkemeier PL, Farrell MB, Shaw LJ.Nationwide laboratory adherence to myocardialperfusion imaging radiation dose reduction prac-tices: a report from the Intersocietal AccreditationCommission data repository. J Am Coll Cardiol Img2015;8:1170–6.

KEY WORDS coronary artery disease, PETimaging, quantitative myocardial perfusion,vasodilator stress