MihailA.

Boyanov




QualityAssuranceandControl
inDualenergyXrayAbsorptiometry

Sofia,2013
CentralMedicalLibrary


QualityAssuranceandControlinDualenergy
XrayAbsorptiometry
MihailA.Boyanov,2013
Firstedition
ISBN9789549318272

Mailingaddress:Prof.MihailBoyanov,MD,PhD,DSc
ClinicofEndocrinologyandMetabolism,
DepartmentInternalMedicine
MedicalUniversity
1G.Sofiyskist.,Sofia1431,Bulgaria
Phone:+35929230784,Fax:+35929230779
Email:mihailboyanov@yahoo.comor
boyanov@alexandrovskahospital.bg
Foreword

The Dual Energy Xray Absorptiometry (DXA) has become the gold
standard in thediagnosis of osteoporosis and low bone mass. The
widespreaduseofDXAeveninareaswithpreviouslydifficultaccess
to this technique precludes the need for proper education of the
technologists and physicians performing and reading DXA studies.
Qualityassurance(QA)andcontrolineachmeasurementmethodol
ogyareessentialasabasisforthecredibilityoftheyieldedresults.
This book is focused on QA and control in DXA. It is conceived and
writtenbyoneoftheleadingspecialistsinthefieldinourcountry,a
recognizedexpert,scientistandeducator.Itcontainsoriginalexperi
mentsanddatareflectinghisownexperience.Theworkrepresentsa
synthesis of some basic science and statistics and more advanced
conceptsofQAinbonedensitometrysuchascrosscalibrationoruse
ofmoresophisticatedtechniquesforprecisioncalculations.
Thelevelofscientificcontentishigh.Allthelatestdevelopmentsare
presentedanddiscussed.Thetextcaneasilybeunderstoodbyboth
DXAtechnologistsandphysicians,aswellasbyallspecialistsreading
DXAreportsintheireverydayclinicalpractice.Itaddssomenewin
sightsintotheproblemsinherenttoQA.Thusitcouldbeofgreathelp
forallthosededicatedtothefieldofbonedensitometryandDXA.
The structure of the book is coherent and balanced. The tables are
precise and informative. Well selected references are given at the
endofthework.
Iamsurethatthisexcellentcontemporarybookwillbewellreceived
andappreciated.Ihighlyrecommendit.

LjubomirDiankov
ProfessorofRadiology
 CONTENTS

Introductionanddefinitions.............................................................7

InstrumentQAprocedures .............................................................12

TheTechnologistQC ......................................................................24

Afewmorepracticalproblems ......................................................30
Manufacturersuppliedversusownprecisiondata..........................30
LongtermprecisionofDXA .............................................................34
Intraobserverversusinterobserverprecisionerrors......................34
Otherfactorsaffectingprecision .....................................................36
Somemorestatisticalconsiderations ..............................................40

Replacingorupgradingabonedensitometer.
CrosscalibrationofDXAdevicesfromdifferent
manufacturersordifferentmodelsofthesamemanufacturer....42
UpgradingcentralDXAequipment...................................................42
ReplacingcentralDXAequipment ....................................................43
QCandcrosscalibrationintheresearchsetting..............................49

Conclusion ........................................................................................50

Acknowledgments ...........................................................................50
References........................................................................................ 51
Abbreviations

BMC bonemineralcontent
BMD bonemineraldensity
CV coefficientofvariation
DXA dualenergyXrayabsorptiometry
GE GeneralElectric
ISCD InternationalSocietyforClinicalDensitometry
LSC LeastSignificantChange
MTI MonitoringTimeInterval
QA qualityassurance
QC qualitycontrol
QCT quantitativecomputedtomography
sBMD standardizedBMD
SD standarddeviation

Introductionanddefinitions
Theconceptofosteoporosis,asaquantifiablebonemetabolic
disease,reliesonmeasurementsofbonemineraldensity(BMD)
[1].Inaddition,BMDplaysacrucialroleinpredictingfracture
risk, as well as in monitoring antiresorptive treatment [2, 3].
DualenergyXrayabsorptiometry(DXA)isthegoldstandardfor
BMDtesting.Itisofutmostimportancethatthemeasurements
reflectthetruevalueofBMD(i.e.theinitialbonemineralcon
tent measurement should match the value of any repeated
measurements).Thesetwomajorrequirementsaretranslated
intotheconceptsoftruenessandprecision.Togethertheycon
tributetotheaccuracyofthetechnique.Severalyearsagothe
term accuracy was used instead of trueness. Nowadays the
term accuracy is used to depict the broader concept, which
incorporatesbothtruenessandprecision.
ThetermaccuracywasdefinedbytheInternationalStandard
Organization(ISO3534:3.11)astheabilityofameasurement
tomatchtheacceptedreferencevalue.Assuch,accuracycan
be affectedby two types of error: systematic error (trueness
error)andrandomerror(precisionerror)[46].
Trueness (ISO 3534:3.12) is the ability to measure a true
value of BMC or BMD using densitometry. When measuring
bonemineralcontent(BMC),themeasuredvaluecanbecom
paredtothemineralcontentofashedbone.Ideally,accuracy
shouldbeassessedonhumancadaversbutthisisrarelydone.
However, if that is the case, the region of interest (ROI) is
scanned;thepureboneintheROIisthenexcisedandinciner
atedinaspecialchamber.Theremnants,consistingmainlyof
minerals,areweighedandcomparedtotheBMCvaluemeas
ured by bone densitometry. This approach is very extensive

7
andmayposesomeproblems.Anotherpossibleapproachfor
BMC assessment is to measure equivalent material (e.g. hy
droxyappatitewithknowncalciumcontent)inartificialphan
toms [7]. Testing trueness of the bone densitometry proce
dureismainlythetaskofmanufacturers.Itisrarelydoneun
der clinical conditions. FDA 510K clearance requires an error
oflessthan10%forallBMDdevices.Mostmanufacturersin
bonedensitometryclaimthatthetruenessof thebonemass
measurementislowerthan56%[4,6,7].Thisisthesocalled
systematic error inherent in the technique. The trueness af
fects the correct use of DXA for the diagnosis of low bone
massandosteoporosis.
Precision(ISO3534:3.14)istheotherattributeofquantitative
measurement techniques. It refers to the ability of the tech
niquetoreproducethesamenumericalresultwhenthetestis
repeatedinanidenticalfashiononthesameobjectunderthe
assumption that there is no biological change [6, 8]. It is also
referred to as reproducibility or repeatability as it is based on
serialmeasurementcomparisons.Randomerrorsresultinscat
tered multiple measures. However, the average value should
still be true. In other words, precision shows whether re
peated measurements would yield approximately the same
resultirrespectiveofthetruevalue(i.e.oftheaccuracy).Preci
sion is important in bone densitometry monitoring (the so
calledserialmeasurements).Itreflectsboththeinherentpreci
sionerrorofthetechniqueandtheerrorintroducedbyincor
rectorinconsistentpatientpositioningandscananalysis[6,8].
Theconceptsoftruenessandprecisionareverywellillustrated
if each measurement is regarded as a dart aiming to hit the
samepoint,whichrepresentsthetruevalue.Truenessreflects

8
how close the hits are to the true value (the central point),
whileprecisionreflectswhetherthehitsareclosetoeachother
irrespective to their proximity to the central point (the true
value).ThisconceptisillustratedinFigure1.

goodtrueness,lowprecision
lowtrueness,goodprecision
goodtruenessandprecision(theidealmeasurementmethod)

Figure1.Theconceptoftruenessandprecisionisshown

Theconceptofprecisionwasfirstdevelopedinclinicalchemis
try.Theideaofhavingsomekindofqualitycontrolgrewwith
thetechnicaldevelopmentofDXA[913].Itwaspracticallyin
troduced in bone densitometry due to the seminal work of
C.Gleretal.[14,15].Intheirwork,Glueretal.presentedtypi
cal problems encountered during research studies and guide
lines for their solutions. Detailed descriptions of the precision
and trueness (accuracy) concepts can be found in a variety of
sources [5, 6, 8, 16, 17]. The mathematical construct behind
precisionisexplainedindetailinapaperwrittenbyS.L.Bon
nicketal.[8].Ifanobjectismeasuredntimes,themeanvalue
ofallmeasurementsisregardedasthetruevalueforeachob

9
ject (true value = (measurement 1 + measurement 2 + . +
measurementn)/n),wherenisthenumberofmeasurements.
Thedifferencesbetweeneachmeasurementandthemeanare
then calculated (difference 1 = measurement 1 mean value;
difference2=measurement2meanvalue;etc.).Thediffer
encesarethensquared,addedandthetotalisdividedbythe
numberofmeasurementsminus1(n1),wheren1represents
the degrees of freedom. The square root is then taken. This
value is the standard deviation (SD) for the set of n measure
ments.Itisexpressedintheunitsofmeasurement(e.g.,g/cm2
in the case of BMD) and represents the variability from the
meanofallnmeasurements.
n

(X X
i 1
i mean )
2

SD (1)
n1

whereXiisthevalueoftheithconsecutivemeasurement,Xmean
isthemeanvalueofallnmeasurementsandnisthenumberof
measurements.
In order to present the SD as a proportion of the mean, it is
divided by the mean, multiplied by 100 and expressed as a
percentage. This is known as the coefficient of variation
(%CV):
SD
%CV x 100 (2)
Mean
If m objects are scanned repeatedly, the average value, SD
and%CVcanbefoundforeachsetofrepeatedmeasurements
(i.e.foreachobject).Theprecisionvalueforallmeasurements
canthenbeexpressedastherootmeansquareSD(RMSSD)
or the rootmeansquare CV (RMSCV) of the calculated SD

10
andCVforeachobject.Theyarethencalculatedfortheentire
object group by squaring SD and CV for each patient, adding
these values, dividing them by the number of patients and
thentakingthesquareroot:
n

SD
i1
i
2

SDRMS (3)
n

n

CV
i1
i
2

CVRMS (4)
n

wherenisthenumberofobjects.
WhentheCVispresentedasapercentageofthemean(%CV)
itsvaluedependsonthemeanvalue.InthecaseofBMDthe
averagewoulddiffersubstantiallyinmeasurementsobtained
fromsubjectswithnormalversuslowBMD.Onewouldexpect
that the higher the average BMD, the lower the %CV (where
theCVisaproportionofthemean)[18].Inordertoavoidthis
sourceofbias,absoluteCVvalues(e.g.gramspersquarecen
timeter) are preferred. The RMSSD or RMSCV are respec
tivelytheappropriateexpressionsofthesocalledshortterm
precisionforbonedensitometryfacilities.
A precision study can be performed on a specific phantom (in
vitroprecisionstudy)oronrealpatients(invivoprecisionstudy).
Thefirsttypeofstudytestsforthereproducibilityofthemeas
urementsandtheprecisionerror,whicharegeneratedsolelyby
amachine.Thiserrorisduetothecalibrationandstabilityofthe
Xray source, the detectors and the software. An in vitro preci
sion study performed on a phantom is mandatory part of the
11
Instrument Quality Assurance and Control (Instrument QA, QC)
[19, 20]. Precision testing on real patients additionally depends
on proper patient positioning, scanning and data analysis. It
represents more clearly the real clinical situation and is part of
theTechnologistsQualityControl.

InstrumentQAprocedures
QAproceduresneedtobestandardproceduresinDXAfacili
ties,inordertohaveacontrolforoperatorandmachinevari
ability[1923].InstrumentQAmeans,thatthemachinesper
formance is monitored over time. One must keep in mind
thatDXAdevicesareXraybasedandmustmeetallradiation
safetyrequirements.PeriodicalmonitoringoftheXraysource,
thescatteredradiationandthepatientandtechnologistexpo
sure to radiation should be done. Radiation exposure and
safety procedures in DXA facilities are regulated through es
tablishedinternationalandlocalstandards.
Instrument QA procedures are performed according to the
manufacturers recommendations and include instrument
calibration as well as testing of some software features. The
instrument calibration is done differently by different manu
facturers.GEandNorlandbonedensitometersrequireanex
ternal calibration scan with a special manufacturerprovided
phantom (calibration standard) to be performed each day
(Figure 2). This calibration standard tests the mechanical op
erationandcalibrationoftheDXAmachine.Hologicmachines
perform a continuous internal calibration with each pixel
measurement by automatically using a rotating internal cali
brationwheelordrum.

12

Figure2.GELunarexternalcalibrationstandard

Asecondtypeofphantomisusedto:(1)testthemachineper
formanceovertimebydetectingchangesinBMDvalues,and
(2)monitorshorttermandlongtermprecisioninvitro.Those
phantoms with known BMC, area and BMD usually mimic a
regionofthehumanskeleton.Someofthemresemblelumbar
vertebrae or the hip and are termed anthropomorphic phan
toms.Theiraimistotestmachinefunctionwithinarangeof
bone densities (replicating normal, osteopenic and osteo
porotic bone) as well within a range of soft tissue densities.
The phantoms test edge detection (the ability to distinguish
bonefromsofttissue)andallowfortheassessmentofbones
thathaveanonuniformdensitythroughoutthetissuelength.
Boneisimitatedbyaluminiumorhydroxyappatiteofdifferent
densities, while soft tissue is simulated by embedded bone
equivalentsintoepoxide,resineorothermaterials.

Scanningofananthropomorphicphantomisrequiredforallsys
tems. The Hologic anthropomorphic phantom (see Figure 3 A)
consistsoffouranatomicallycorrectvertebraewithsimilarden

13
sities and areas, consisting of epoxyresin to simulate 60 % fat.
Thisphantomisoftenusedwithdevicesfromothermanufactur
ers.ThelackofaBMDvaluerange,however,makesitlesssuit
able for crosscalibration purposes than the newly developed
European Spine Phantom and Bona Fide Phantom (see below).
Hologichasdevelopedananthropomorphichipphantomaswell,
butitisconsiderablylessusedinclinicalmedicineandremains
reservedsolelyforresearchpurposes.Theuseofanoriginalstep
phantom for the calibration of whole body measurements in
Hologicmachinesisalsopossible,wherethephantomshouldbe
scannedatleastonceweekly(seeFigure3B).
A B

Figure3.Hologiccalibrationphantoms.(A)Hologicanthropomorphicspine
phantom.(B)Wholebodyscanningcalibrationphantom

TheLunarSpinePhantomrepresentsarectangularaluminum
bar mimicking all vertebrae at the level of L1L4, as well as
partsoftheTh12andL5vertebrae(seeFigure4).Eachverte
bra has a different area and density. To simulate soft tissue,
the phantom is submerged in water or it is embedded in an
epoxyresin block, which is the case in the newly developed
versionofthephantom.

14

Figure4.TheLunarspinephantom

AllNorlandscannersarecalibratedagainstaprimaryAPspine
phantom, composed of hydroxyapatite equivalent materials,
andaprimarybodycompositionphantom.
Peripheral DXA devices are calibrated using manufacturer
providedphantoms.TheDTX100,forinstance,usesanalumi
num phantom mimicking two rectangles for the radius and
ulna(Figure5).

Figure5.PhantomscanontheDTX100forearmbonedensitometer.Note
thatthephantomvalueisattheupperlimitofthe1.5%range

TheEuropeanSpinePhantomwasdevelopedinanattemptto
producetheperfectphantom.Itisasemianthropomorphic
phantom with three levels of BMD values intended for the

15
standardizationofanycentralbonedensitometer.Itisrather
expensiveandisusedprimarilyforresearchpurposes[24].
TheBoneFideSpinePhantomisacalciumhydroxyapatitestep
phantom embedded within an acrylic block. The acrylic mi
mickssofttissue,whilethestepwedgesreperesentarangeof
BMDvaluessimilartothosemeasuredinpatients.
The forearm presents more difficulties for standardization as
different devices measure different ROIs: ultradistal, distal,
middistal, site and 33% proximal site. A European semi
anthropomorphicphantomforthecrosscalibrationofperiph
eral bone densitometers was developed [25]. A universal
standardizationapproachwasalsotestedbuttheresultswere
notencouraging[26,27].
Inshort,QCphantomstestbothmachineparametersandsoft
warefeaturesofedgedetectionandanalysis.Thepurposeofthis
typeofQAscanningistodetectanymachinecalibrationfailures
ordriftsoccurringoveraperiodoftime,aswellaschangesre
sulting from maintenance and repair of the DXA machine. The
phantomshouldberescannedmultipletimes(1020times)af
tertheinstrumentinstallationtodefinethemeanvaluesofthe
measured BMD, BMCand area. These mean values are further
usedwhenmonitoringinvitroprecision(basedondailyphantom
measurements). The anthropometric phantom has to be
scannedoneachworkingdayandthephantomBMD,BMC,and
area should be plotted on graphs based on the, socalled, She
whartcontrolcharts.Shewhartisthenameofthescientistwho
developedtheQCproceduresinanalyticalchemistry[28].
There are different ways to monitor machine stability over
time using QC charts. One simple way to do this is by visual
inspection and the application of the, socalled, Shewhart

16
rules. They give simple recommendations about the kind of
BMD,BMCorareachangesthatshouldbecloselymonitored
and, which would require instrument recalibration or servi
cing. Firstly, it is necessary to establish a baseline value and
control limits. The baseline value represents the mean of 10
consecutivephantom measurementsonthesamedayorthe
mean of 15 to 25 measurements on consecutive days. The
standarddeviation(SD)forthesetofscansisthencalculated.
Control limits are expressed either as a percentage of the
meanorintermsofSD.Dailyscanvaluesarethenplottedon
agraphdepictingthebaselinevalueandthecontrollimits.If
anyoftheShewhartrulesareviolated,awarningmessageis
issuedrequiringspecificactions(fordetailsseeTable1).

Thephantommeasurement Whatshouldbedone
isfallingoutsiderange
1phantomvalueexceeds VisualinspectionoftheQCchart
theaverageby1%
1phantomvalueexceeds Instrumentrecalibrationshould
theaverageby1.5% beperformed*
2consecutivephantomvaluesexceedthe Instrumentrecalibrationshould
averageby1%range beperformed
Adifferencebetween2consecutivemeas Instrumentrecalibrationshould
urements>2%(andifbothareoutside beperformed
the1%range)
4consecutivephantomvaluesononeside Instrumentrecalibrationshould
and>+0.5%or0.5% beperformed
10consecutivemeasurementsfallonone Instrumentrecalibrationshould
sideoftheaverage beperformed
*Theinstrumentrecalibrationrequirestheobtainmentof5consecutivephan
tom measurements with repositioning. If the mean value differs by more than
1%fromthemanufacturerprovidedvalue(usuallyregisteredonthephantom),
theuserwillbepromptedtomaintenanceservicetheinstrument!
Table1.ShewhartrulesinQC.Theleftcolumnshowspossibleaberrations,
whiletherightcolumnshowstherequiredactions[6]

17
Usually,1SDiscloseto0.5%oftheaveragephantomvalue.
Therefore1SD=0.5%,2SD=1%,3SD=1.5%,4SD=2%.
TheShewhartrulesare,therefore,knownasthe3SDor1.5%
rule,the2SDor1%twicerule,therangeof4SDorrangeof
2%rule,thefour1SDorfour0.5%rule,andthemeanx
10rule.Theviolationofanyoftheserulesmustberegarded
asawarningformachinemalfunction.
However, the Shewhart rules tend to produce a high false
alarmrate.ThisisduetothefactthatevenDXAmachinesin
perfect condition may, from time to time, violate the She
whart rules. In the work by Lu et al. this was the case with
every39thscan[29].A,socalled,filtermaybeusedtoreduce
theoccurrenceoffalsealarms[29].Thisfilterrequiresthatthe
meanof10phantomscansiscalculatedafteraviolationofthe
Shewhartruleshasoccurred.Theviolationisconfirmedifthe
difference from the baseline average value exceeds 1 SD. Al
ternatively,onecansetthe3SDruleasthecrucialrule,whose
violationrequirestheimmediateapplicationoftheotherrules.
TheShewhartrulesareeasytouseandcandetectbothsud
den changes in machine calibration (i.e. a shift), as well as a
slowlygrowingbiasinmachineperformance(i.e.adrift).The
shift might be due to recent maintenance, changes in the X
ray tube or detectors, or relocations (and incorrect machine
calibration),aswellastosuddenmachinemalfunction.Adrift
(slowchanges)mightbeduetoanagingXraytubeordetec
tors, to change in the power supply or to slowly changing
roomconditions(e.g.temperature,humidity).
Figures6and7showtheQCchartsoftheDTX100forearm
bonedensitometer(seeFigure6)andtheHologicQDR4500

18
Adevice(seeFigure7)attheBoneMetabolicUnit.Figure8
illustratesthesameDTX100device(asinFigure6)duringa
prolonged malfunction period (servicing had been post
poned).

Figure6.Phantomstatistics(BMDchart)ofaforearmsingleXrayabsorp
tiometrydevice,theDTX100(OsteometerMeditech,USA).AllBMDvalues
arewithinthe1.5%values.Propermachinefunctioningisobserved.

Another way to monitor QC is by using CUSUM charts [30].

Theyaremoresophisticatedandsuitableforprofessionalpur
poses. The CUSUM charts reflect the random variation in
phantommeasurements.Themagnitudeanddirectionofthe
variation should remain relatively constant compared to the
average mean. The principle underlying the CUSUM charts is
inthatthedifferencesbetweenanactualphantommeasure
mentandthebaselineaveragevalue(fromthe10consecutive
scansonthesamedayorthe1525measurementsondiffer
entdays)shouldberandomlyscatteredaroundthezeroline.
This means that the cumulative sum of all of the differences
shouldbeclosetozeroifthereisnobiasintroduced.Tobuild
the CUSUM graph, each sequential value of the cumulative

19
sum is plotted against time. If the machine is functioning
properly, the values should be scattered in a horizontal
patternaroundthe0value.Otherwise,therewillbearisingor
falling pattern. Although the CUSUM charts can be inspected
visually,thereareotherwaystodetectsubtlechanges.Oneof
themisbydeterminingtheslopesofthearmsontheVmask.
Alternatively,tabularCUSUMcanbeused,inwhichupperand
lowercontrollimitsareintroduced.TheCUSUMchartsareless
intuitive than the Shewhart graph and require the use of a
softwareprogram.ThemainadvantageoftheCUSUMcharts
isintheirhighersensitivity.GoodexamplesforCUSUMcharts
canbefoundinref.6.

Figure7.Phantomstatistics(BMDchart)oftheHologicQDR4500Adevice
(Hologic Inc., Bedford, USA). The machine is functioning within the pre
specifiedrangeofBMD

20

Figure 8. Phantom statistics of the DTX100 forearm bone densitometer
during a period of prolonged malfunction. BMC values are scattered
aroundthemeanandsomeofthemexceedtheacceptablevariationrange.

The visual examination of QC charts has been compared to

processcontrol charts (CUSUM charts) [31]. The regular
Shewhart chart, with its associated sensitizing rules, has a
highfalsealarmrate.Introducinganadditionalfilterreduces
the false alarm rate and overall sensitivity. The CUSUM
method offers the best sensitivity/specificity ratio and can
estimate when (i.e.the date) a change in DXA scanner per
formance occurred [31]. In another study, the ROC analysis
usingtheShewhartmultirulechartswasshowntobeasef
fective, if not better, than the one done using the CUSUM
chart. However, the stringency of the rules must be tight
enedtoobtainanoptimalperformance[32,33].
Aninterestingapproachistouse,notonlythebaselinephan
tomvalue(frommultiplescans),buttoalsograduallyincor
poratethephantomvaluesduringperiodsofpropermachine
functioning.Thisiscalledthemovingaverage[17].

21
A detailed description of how in vitro QC charts are to be
handled can be found in the Canadian ISCD standards for
technologists [34]. Hologic software supplies these charts
automatically, NorlandCooper Surgical maintains BMD
charts only, while GE Lunar operators need to create and
maintainthesechartsmanually.Practicallyallmanufacturers
provide sophisticated software programs for calculating
phantom value variations and displaying drifts in machine
performance.

A phantom log of all phantom scans should be maintained.

Servicelogsshouldalsobemaintained(includingprintoutsof
error messages), as should all service, government, and ra
diationsurveyreports.Afteranypreventivemaintenanceor
service, the phantom should be scanned 10 times without
movingthephantom.IfthemeanBMDexceedsthemeanof
thedailyphantomscansthatwereperformedduringthe10
days prior to maintenance or servicing, the machine should
berecalibratedandanewmeanshouldbeestablishedbased
on new 10 scan set of the phantom. In this scenario, a 1%
change in the readings is considered within normal limits
followingasoftwarechange.

Apracticalquestion,onebegstoaskiswhocanensurethat
theQCataparticularbonedensitometrycenterisperformed
correctly. Large imaging facilities, such as BioImaging or Sy
narc,requiresourceinformationfromparticipatingDXAunits
andreviewactualQCdata.Centralizedcontrolisusuallyper
formedinmulticenterclinicaltrials[24,3537].Alternatively,
one can have centralized control for all bone densitometry
unitsinaparticularcountryanapproachthatmightbeim

22
plemented in the future. It is very difficult to decide who
should perform this centralized control in this case. The QC
procedures could also be made mandatory through the Na
tional Standards/Guidelines for Clinical Densitometry in a
givencountry.Inthiscase,allhealthcarefundingbodieswill
have the right to check for the implementation of the QC
procedures before any payment for bone densitometry
occurs.

The concept of centralized quality control of bone densi

tometrywasappliedinanationalhealthsurvey(NHANESIII)
using three mobile examination centers [37]. A small per
centage (3.5%) of all scans was rejected, but 33% of the re
maining required reanalysis at the QC center. Precision in
spineandhipphantomwasbelow1%.Invivoprecisionfor
the femur neck BMD was 3.2% (CV) and was found to be
5.1 % for Wards triangle BMD [34]. A few similar trials can
be found in the literature [35, 36]. In Bulgaria, a National
Osteoporosis Program 20062010 was implemented. During
this program, an epidemiology study on the prevalence of
osteoporosis usingtheDXAtechnologywasperformed [38].
SixDXAcentersweresubjectedtoacentralizedreviewofQA
andQCproceduresandtheresultswerepublishedelsewhere
[38]. The QC study was done on 5 pencil beam DXA densi
tometers one Hologic QDR 1000, one Hologic QDR 1500,
two Lunar DPXIQ and one DPXL, as well as one fanbeam
Hologic QDR 4500 A. Table 2 shows in vitro precision errors
of the machines that occurred while scanning the Hologic
anthropomorphicphantom20timesconsecutively.

23

DXAUnit Nr1 Nr2 Nr3 Nr4 Nr5 Nr6

SD(g/cm2) 0.0034 0.0039 0.004 0.007 0.011 0.004

%CV(%) 0.334 0.307 0.377 0.524 0.848 0.428

Table2.ShortterminvitroprecisionerrorsofthesixDXAcenterspartici
patingintheNationalOsteoporosisPrograminBulgaria20062010[39]

ThedatafromTable2showthatDXAmachinenumber5dif
fered significantly from all others. The data from the in vivo
precisionstudyarepresentedinthetextbelow(seealsoTable
7).ThisexampledisplaystheimportanceofreviewingQClogs
andchartsataparticularDXAcenteraspartofsitecertifica
tionandaccreditation.

TheTechnologistQC
At least 30 degrees of freedom are needed for a precision
study to be valid [40, 41]. Thirty degrees of freedom ensure
thattheupperlimitforthe95%confidenceinterval(CI)ofthe
calculated precision value does not exceed 34%. Table 3
representsthepossiblecombinationsofobjects(i.e.patients)
andscansperpatienttoachieve30degreesoffreedom.

No.ofpatients(objects) No.ofscansperpatient
1 31
10 4
15 3
30 2

Table 3. Number of patients and scans per patient ensuring at least 30

degreesoffreedominashorttermprecisionstudy

24
Inclinicalpractice,30patientsarescannedtwiceover2weeks
to 1 month. This is called a shortterm precision study. In
longterm precision studiestheobjectsarefollowedover an
extended period of time (> 1 month). Longterm precision
studies are easily performed with phantoms as their mineral
contentdoesnotchangeovertime.Thistypeofprecisioner
roriscalledlongtermprecisioninvitro.Itreflectsthecalibra
tionandproperfunctioningofthebonedensitometer.Inthe
caseofprecisionstudiesonrealpatients,amajorconfounding
factoristheBMDchangeovertime.Thislogisticdifficultyre
quirestheuseoflinearregressiontoaccountforthisbiologi
cal change and the calculation of standard error of the esti
mate(SEE)inordertoaccuratelyexpressprecision.Thelong
term precision errors are expected to exceed the shortterm
precisionestimates.Inclinicalpractice,invivoshorttermpre
cisionstudiesareusuallyperformed,whereRMSSDandRMS
CV(inabsolutevaluesor%)arecalculated.
Iftheinvivoprecisionerrorataparticularsiteisknown,the
magnitudeofBMDchangemeasuredindicatingrealbiological
changeatthatsitecanbedetermined.Thisiscalledtheleast
significant change (LSC). One has to choose which level of
statisticalconfidencewillbeusedforcalculatingLSC.Accord
ingtothelevelofstatisticalconfidence,aspecificcoefficient,
as shown in Table 4, should be used to calculate the LSC as
follows:
LSC=2xcoefficientxprecisionerror (5),
wherethecoefficientcanbefoundinTable4accordingtothe
desired level of statistical significance and precision is ex
pressedaseitherRMSCVorRMSCV.

25
 Levelofconfidence Coefficientvalue
99 2.58
95 1.96
90 1.65
80 1.28

Table4.CoefficientsusedinthecalculationofLSCaccordingtothedesired
levelofstatisticalsignificance

Therefore,ifa95%significanceisrequiredtheLSC95=1.44x
1.96xprecision=2.77xprecision(6).For80%confidence,the
LSC80 = 1.44 x 1.28 x precision = 1.84 x precision (7). Usually
the95%confidencelevelisusedandtheprecisionerrorhasto
bemultipliedby2.77toobtaintheLSC.
The International Society for Clinical Densitometry has pro
vided a useful tool for the calculation of %CV and the LSC at
www.iscd.org. The user must insert the absolute values of
consecutivemeasurements(30patientsmeasuredtwiceor15
patientsmeasuredtrice)andthevaluesfor%CV,SD,RMSSD,
RMSCVandLSCarethencalculatedautomatically.
In 2005, the International Society for Clinical Densitometry
(ISCD)issuedstandardsfortheminimumacceptableprecision
foranindividualtechnologist[2,42]:
Lumbarspine:1.9%(LSC=5.3%)
Totalhip:1.8%(LSC=5.0%)
Femoralneck:2.5%(LSC=6.9%)

If a technologists precision is worse than these values, re

trainingisrequired[2,41].

26
InordertodetectrealchangesinBMD,aserialscanshouldbe
repeated when the biological change is expected to exceed
LSC.Thetimeintervalneededdependsonboththeexpected
rateofchangeperyearandtheLSC:

Timeinterval=LSC/expectedrateofchangeperyear (6)

The greater the LSC and the smallest the annual change, the
longer the time interval should be. A simple approach in de
finingtheinterrelationshipbetweenprecisionerrorandantici
patedrateofchange,inestablishingthetimeintervalforserial
measurementsneededtoexceedtheLSC,isgiveninTable5.

Precisionas%CV Changeperyear(%) Timeinterval(yr)

1 2.77
1.0 3 0.92
5 0.55
1 4.16
1.5 3 1.39
5 0.83
1 5.54
2.0 3 1.85
5 1.11

Table5.Interrelationshipbetweenprecisionandanticipatedrateof
changeindeterminingthetimeintervalneededtoexceedLSC[8]

For example, the change in BMD is rapid in corticosteroid

induced osteoporosis, and repeated measurements can be
performedwithin912months.Intheusualosteoporosispa
tienttreatedwithantiresorptivedrugs,thebiologicalchanges
inBMDcouldbedetectedafteratleast23years.However,it
issafertoobtainthefirstBMDmeasurementearlyafterther

27
apy initiation as nonresponders to therapy can be identified
earlier. Typically, the first repeat scan might be performed a
year after treatment has started. If BMD stability (i.e. lack of
significantchange)orimprovementarenoted,thenextscans
should be performed at longer time intervals (23 years) [3].
Repeated measurements should be performed on the same
device and, preferably, by the same operator as outlined in
thecrosscalibrationsection[3,42].
AWhitePaperonprecisionassessmentandradiationsafety
for DXA was published [22]. A number of important ques
tionswereexhausted:thedefinitionandimportanceofaccu
racy and precision, methods of DXA precision assessment,
commonfactorsaffectingprecisionandothers.Accordingto
thispaperclinicalDXAprecisionisinfluencedbyacombina
tion of short and longterm variability of the scanner, pa
tient motion, body habitus, and operator dependentfactors
such as patient positioning and scan analysis. Patient and
operator related sources of variability are more important
than the scanner variability itself with operator related fac
tors having the most influence on the overall precision of
DXA measurements. The White Paper strongly recom
mended that each technologist and bone density facility
conduct a precision assessment. Technologists should per
formaprecisionassessmentaftertheyhavescannedatleast
100patients.Precisionassessmentdoesnotneedtobere
peatedaslongasthereisnoreasontobelievethattherehas
beenachangeinthetechnologistslevelofcompetenceorin
themachinesoftwareandhardware[3,21].Usingmanufac
turerprovided precision data was not endorsed. The ques
tionofdefiningameanprecisionerrorforaDXAcenterwith

28
multipletechnologistswassolvedwiththeaveragingoftheir
individualprecisionerrors.Thisisapplicableiftheconfidence
intervalsoftheprecisionerrorsforeachtechnologistoverlap,
indicatingthatthesmallintertechnologistdifferencesmight
not represent true differences [3]. As the White Paper also
reviewed Radiation Safety, a short paragraph indicated that
for the need of precision studies no special consent form
would be required from the patient, but he/she should be
informedaboutthepurposeoftheduplicatescansandgiven
theopportunitytorefuseparticipation.
Standardsandguidelinesfortechnologistsperformingcentral
DXA were published in 2007 as a part of the ISCD Canadian
standards[34].AnumberofimportantpracticalpointsinDXA
precisionassessmentwerediscussed:
1. Precision studies in a particular center should be done on
patients who are typical for that center. Translated into
clinical terms, this means that all kinds of subjects with
normal,loworosteoporoticBMDshouldbeincludedinthe
precisionstudy.TheinfluenceoftheabsoluteBMDvalueon
theprecisionvalueisdiscussedfurtherinthischapter.
2. Sameday precision is generally better than different day
precision, with intraobserver variability being lower than
interobserver variability. To reflect clinical reality in the
bestway,arepeatscanafewdaysafterthefirstscancan
beadvocated,whileevenhavingthetwoscansperformed
bydifferenttechnologists.
3. A monitoring time interval (MTI) is defined as the time re
quired to achieve the LSC given an assumed rate of BMD
changeforthatpatientandtheparticularskeletalsite.

29
TheISCDCanadianstandardsalsointroducedtheconceptof
qualityassuranceprocedures.Qualityassurance(QA)con
trolsforbothoperatorandmachinevariability,thusconsist
ing of two distinct parts: Instrument Quality Control and
Technologist Quality Control. Instrument QA is achieved by
monitoring the machines performance over time (as de
scribed below), while the Technologist QA displays the re
producible positioning, performing and analysis of patient
scans(asdescribedabove)[34].

Afewmorepracticalproblems
The concept of precision is very important in real life as it is
the foundation of DXA credibility as a precise technology. It
canalsobeappliedtovariousDXAapplicationssuchaslateral
spine, morphometric Xray absorptiometry, hip structural
analysisorbodycompositionanalysis[4348].

Some real life developments and research in the field of QC

andQAincentralDXAaredescribedbelow.

Manufacturersuppliedversusownprecisiondata

TheISCDdoesnotrecommendusingthemanufacturersupplied
precision data [42]. Instead, each technologist must establish
hisownsetofprecisiondata.Theconditionsreproducedbythe
manufacturer could be regarded as ideal, while the condi
tions in real life are far from ideal. Table 6 summarizes the
manufacturersuppliedprecisiondataderivedfromref.49.

30

Manufacturer DXAmodel %CVBMD %CVBMD %CVBMD InvitroQCprocedures
APspine totalhip whole
body
Hologic,Inc., QDR4500A <1.0% <1.0% <1.0%
Bedford,MA, QDR4500C <1.0% <1.0%
USA QDR4500SL <1.0% <1.0% SelfcalibratingwithHologicAutomaticInternal
QDR4500W <1.0% <1.0% <1.0% Referencesystemandautomatedqualitycontrol
Delphi <1.0% <1.0% program
Discovery <1.0% <1.0% <1.0%
Explorer <1.0% <1.0%
GEHealthcare DPXBravo <1.0% <1.0% Blockphantomandaluminumspinephantom;
Madison,WI, DPXDuo <1.0% <1.0% automatedQAprogramwithdailyprecision
USA monitoring
DPXIQ 0.5% 1.0% 0.5% Blockphantomandaluminumspinephantom
DPXMD 1.0% 0.7% 0.5% Automatictestprogram
DPXNT 1.0% 1.0% 1.0% AutomatedQAprogram
Prodigy 1.0% <1.0% <1.0% AutomatictestprogramwithQAtrending
ExpertXL <1.0% <1.0% <1.0% Internalhydroxyapatite;AutomatedQAprogram
withspinephantom
iDXA 1.1% 0.7% 0.5% Automated6pointcalibrationandquality
assurancewithQAtrending
CooperSurgical Excell 1.0% 1.2% Automatedwith77stepcalibrationstandardand
Norland, Excellplus 1.0% 1.2% QCphantom
Trumball,CT, XR46 1.0% 1.2% 1.0% Automaticwithsuppliedcalibrationstandardand
USA QCphantom
XR600 1.0% 1.4% Automaticwithsuppliedcalibrationstandardand
XR800 1.0% 1.4% 0.8% QCphantom;automatedShewhartChartanalysis

Table6.PrecisiondatafortheFDAapproveddensitometrydevices[49]
The precision data from the literature are so abundant that
onlyafewillustrativesetswillbecitedbelow.Astudytested3
GEHealthcareProdigyand3HologicDelphidevicesandfound
that in vivo averaged %CV for the AP spine, total hip and
femoralnecktobe1.0%,0.9%and1.5%respectivelyforthe
Prodigyand1.2%,1.3%and1.9%respectivelyfortheHologic
devices[50].ThisstudysuggestedthatshorttermBMDpreci
sionerrorswereskeletalsiteandmanufacturerspecific.

The introduction of fanbeam technology required precision

studies.Insuchastudy,theprecisionerroroffanbeamDXA
ofthespine,hipandforearmwascalculated[51].TheCVat
thelumbarspine was 1.1 %, atthe total hip1.2 %, atthe
total forearm 0.5 % [51]. The authors concluded that fan
beamDXAofcentralsitesandtheforearmwereaspreciseas
scansperformedinslowerscanningmodes.Whenusingthe
Comparefunction,theyfoundasignificantdecreaseinpreci
sionattheforearmandnoimprovementatthespineorhip
[51]. Howerer, the use of the Compare function whenever
possibleisadvocated.
WiththeintroductionofnewDXAdevicesintoclinicalprac
tice,precisionstudiesareusuallyperformed.Veryoftenthey
arecomparedtooldermodelsfromthesamemanufacturer,
aswasthecasewithGELunarProdigyandiDXA[52].Anew
pencilbeam bone densitometer (the DMS Stratos) was as
sessed and the RMSCV values were found to be 1.22%,
1.38%,2.11%and0.86%forthelumbarspine(L1L4),lumbar
spine (L2L4), femoral neck and total hip respectively [53].
AnotherresearchgrouptestedthefirstconebeamDXAsys
tem(theDMSLexxos)andfoundaprecisionof1.3%fortotal
hip,2.0%forfemoralneckand2.3%forspineBMD[54].
32
Inourownprecisionstudy,wetestedtheshortterminvivo
precisionofourHologicQDR4500Abonedensitometer.The
results were published elsewhere [55] and were within the
prespecifiedISCDlimits(seeTable7).
Preci Leastsignificant Leastsignifi
sion change(LSC) cantchange
%CV absolutevalue (LSC)in%
BMDL1L4(g/cm2) 1.35% 0.026g/cm2 3.57%
2
TotalhipBMD(g/cm ) 0.95% 0.019g/cm2 2.52%
2 2
FemoralneckBMD(g/cm ) 1.29% 0.023g/cm 3.71%
2 2
TrochanterBMD(g/cm ) 1.31% 0.019g/cm 3.59%
WardsBMD(g/cm2) 3.14% 0.036g/cm2 8.76%
Table7.ShortterminvivoprecisionoftheHologicQDR4500AattheBone
MetabolicUnitoftheUniversityHospitalAlexandrovska,Sofia[55]

DuringtheimplementationoftheNationalOsteoporosisPro
gram20062010inBulgaria,sixDXAcentersweresubjectedto
acentralizedreviewoftheirprecisiondata(sTable8).One
ofthemshowedseveredeviationsfromtheISCDrecommen
dations.Thedatageneratedbythiscenterwerenotincluded
inthefinalanalysis.
DXACenter Nr1 Nr2 Nr3 Nr4 Nr5 Nr6
%CV 0.604 0.982 1.880 3.589 0.440 1.622
LSC(%) 1.670 2.720 5.210 9.940 1.220 4.493
a 2
LSCI (g/cm ) 0.017 0.027 0.052 0.094 0.012 0.045
LSCIIb(g/cm2) 0.015 0.024 0.047 0.096 0.013 0.043
acalculatedmanually
bcalculatedusingtheISCDcalculationtool
Table 8. Theinvivo%CVsforthelumbarspinemeasurementsofsixDXA
centers,whichparticipatedintheepidemiologystudyoftheNationalOs
teoporosisProgram,areshown[39].
33
LongtermprecisionofDXA
ThelongtermprecisionofDXAhasbeentestedinafewstud
ies. It was assessed over 7 years in forty postmenopausal
women [56]. Using trimmed data (untrimmed data in brack
ets) the calculated %CV was found to be 1.12 % (1.65 %) for
the AP spine, 1.32 % (1.57 %) for the total hip, and 2.21 %
(2.48%)forthefemoralneck.Thelongtermprecisionofcen
tral DXA was studied in 64 postmenopausal women over a
period of 10 years [57]. The long term precision errors ex
pressedinSDor%CVwere0.018g/cm2(1.9%)forthespine,
0.016g/cm2(1.7%)forthetotalhipand0.017g/cm2(2.3%)
forthefemoralneckBMD.Inanotherstudy,usingtwoHologic
DXA scanners, the longterm precision %CV was found to be
2.4%forlumbarspine,2.3%fortotalhipand2.7%forfemo
ral neck [58]. These values were almost twice as high when
compared to those obtained from the shortterm precision
data: 1.3 % (AP spine), 1.2 % (total hip) and 1.4 % (femoral
neck).Theexpectedlongterminvivoprecisionwascompared
to expected annual bone loss rates in men and women in a
subsetfromtheMalmOPRAstudy,whereseveralyearswere
needed to detect changes (e.g. LSC/median range of BMD
change)[59].

Intraobserverversusinterobserverprecisionerrors
Theintroductionofasecondtechnologistisusuallyexpected
toincreasetheprecisionerrorasbothtechnologistsmayposi
tionpatientsandanalyzethescansinadifferentmanner.
The in vivo intrarater and interrater precision of measuring
BMDusingsupinelateralDXAwasassessedintwocohortsa
younger(withhighBMD)andanolderone(withpresumably

34
lower BMD) [43]. Intrarater precision was better in the
younger cohort it ranged in %CV from 0.503.68 % for the
youngcohort,and1.465.30%fortheoldercohort.Interrater
precisionwaslowerwith%CVof1.112.35%fortheyounger
cohortand1.854.32%fortheoldercohort.
Inourownprecisionstudy,wetestedtheinvivoprecisionofa
single Xray forearm bone densitometer the DTX100 device
(OsteometerMeditech,USA)andcomparedittopreviouspub
lications [60, 61]. On this specific device the distal region of
interest, ROI, begins at the 8 mm separation point between
radiusandulnaandthencontinuesproximallyforadistanceof
24 mm. It contains around 6070 % cortical bone and has a
more cylindrical and, thus, reproducible shape. The ultradistal
ROI extends from the radial endplate proximally to the 8 mm
point and is built primarily by trabecular bone (6070 %). The
radial endplate shows much more anatomical variations and,
thus,resultsinlessprecisemeasurements.Overthedurationof
the study, the shortterm in vitro precision %CV was 0.58 %.
Theshortterminvivoprecision%CVsforasingleortwodiffer
ent technologists, as well as the longterm in vivo preci
sion %CVs (same patients rescanned 3 months later) are
showninTable9.

Distalsite Ultradistalsite
intraobservershortterm%CV 1.65% 2.03%
interobservershortterm%CV 1.86% 2.54%
intraobserverlongterm%CV 1.92% 2.13%

Table9.Theinvivointraobserverandinterobserverprecision%CVonthe
DTX100forearmsingleXraydevice[60,61]

35
Our results demonstrate that the magnitude of the precision
errorvarieswiththeanatomicalsite,aswellaswiththeintro
duction of a second technologist. That is why the ISCD rec
ommends the use of an average value for all technologists
performingDXAataparticularDXAsite[3].

Otherfactorsaffectingprecision
A number of nonmodifiable factors are affecting precision:
typeandmodeloftheequipmentused,siteofmeasurement,
software version, range of BMD values, soft tissue thickness
(especially in body composition studies) and others. Among
themodifiablefactorsarethepropertrainingoftechnologists
andgoodpatientcompliance[62].
W. Leslie investigated the factors affecting shortterm BMD
precision assessment during the Manitoba Bone Density Pro
gram [63]. There were no significant differences in precision
errorswhenscanpairswereacquiredbyasingletechnologist
ortwotechnologists.Scannermode(pencilbeamversusfan
beam),age,heightandboneareaalsodidnotaffectthepreci
sionerror.Weight,BMIandTscoresshowedinconsistentdif
ferences.Thesinglefactoraffectingprecisionsignificantlywas
acquiring the scanpairs on different days. The author con
cludedthatthemostwidelyusedprocedureforperforminga
BMD precision assessment (same technologist on the same
day) might systematically underestimate precision error and
wouldleadtoovercategorizationofchangeinmanypatients
[63]. A similar finding was published in a study where a sub
group of patients were rescanned 3 to 10 days later, and a
second study where patients wererescanned 188 days later
onaverage[64].Theshorttermprecisionfortheleftfemoral
36
neck with immediate rescanning was 0.007 g/cm2, 0.017
g/cm2 with rescanning within days and 0.024 g/cm2 with re
scanning within months. This study underlined the fact that
shortterm precision of immediate replicate scans might un
derestimatetheerrorovertime[64].
Different measurement sites yield different precision errors.
TheAPspineandthetotalhipusuallyshowthebestprecision:
AP spine Total hip > Femoral neck > Wards. The expected
biological changes at the spine (mostly trabecular bone) are,
however,greaterthanthoseatthetotalhip(mixedbone)and
thatiswhyitistheprimaryrecommendedsiteforserialbone
density measurements [2, 3]. The total hip (either single or
dualfemur)mightalsobeusedformonitoring.Noothersite
shouldbeused.Theforearmsitehasexcellentprecisionbutis
ratherunresponsivetotreatment.Thelateralspineislesspre
cisethantheAPspine,hipandtotalbodyDXA.Thishypothesis
wasproveninastudybyGMBlakeetal.[65].Theresultswere
normalized for the ratio of treatment effect/precision. The
authors concluded that, although the treatment effect was
largerforlateralratherthanforPAspineBMD,thisadvantage
wasoffsetbythegreaterprecisionerrors[65].
PrecisiondataareaffectedbytherangeofBMDvaluesifthey
are expressed in percent of the mean. This is very well illus
trated in a number of studies [18, 66]. One of these studies
highlightedthatcalculatedCVshowsaprogressivefallinvalue
as BMD rises. Therefore, the SD should be used to calculate
significantBMDchanges[66].Anotherstudyinvestigatedthe
fractionsofpatientswhoseBMDchangesexceededLSCinthe
ManitobaBoneDensityprogram[18].Spectrumbiaswasob
servedwhenBMDmonitoringwasbaseduponrelativechange

37
(in percent) rather than absolute measurements (g/cm2).
Therefore, it advocated the categorization of change in BMD
baseduponabsolutevalues[18].Theprecisionerrordepend
enceontheabsolutevalueiswhypatients,representativeof
thepopulationbeingusuallyreferredtoaDXAcenter,should
beusedinprecisionstudies.
TheeffectofboneareaonshorttermBMDmeasurementerror
was investigated during the Manitoba Bone Density Program
[67]. Differences in bone area exceeding the 2% rule were
common and also caused greater BMD measurement errors.
This finding was corroborated in another study showing that
greaterinterpretableboneareaimprovedprecision[68].
Software advances may additionally improve precision. This
was the case with theHologic Apex v2.0 analysis software in
comparison with the Hologic Delphi v11.2 software. A study
foundthattheprecisionofApexandDelphiwasrespectively
1.0%and1.2%(L1L4spine),1.1%and1.3%(totalhip),1.6%
and1.9%(femoralneck)and0.7%and0.9%(dualtotalfemur)
[69]. The same was true for the Encore software used by GE
Healthcare.Softwareadvancesalsoprovedusefulinreducing
the precision errors in body composition studies. In a study
comparingtwoversionsofsoftware(HologicV8.1aandV8.21),
the reproducibility of DXA in obese women was tested [70].
The CVs for fat and lean weight and BMC were 1.2%, 1.1%,
and1.7%respectivelyforthenewsoftwareV8.21,compared
to1.3%,1.3%,and2.1%respectivelyfortheolderV8.1a[70].
The effect of femoral rotation on hip BMD was tested in a
number of older and more recent studies [7173]. External
rotationby5to10degreeswasabletochangeBMDbymore

38
than5%[71].Inanotherstudy,theauthorsinvestigatedthe
effect of subtle positioning flaws on hip BMD measurements
[72]. Overall, there was no significant difference between all
right or all left hips. In hip pairs, in which one hip was posi
tionedflawlesslyandthecontralateralhipwasflawed(verti
cally,rotationallyorboth),themeasuredBMDofthelatterhip
was not predictably greater than the former. The conclusion
wasthatsubtlepositioningflawsdonotgeneratepredictable
changes in measured BMD at any hip ROI [72]. The effect of
femur rotation on BMD precision was further confirmed in a
methodologicalstudyofthedistalfemurfollowingtotalknee
arthroplasty[73].
The negative effect of increasing weight on shortterm preci
sion of DXA was tested on a GE Lunar Prodigy scanner [74].
The %CVs for the lumbar spine, total hip and total body in
creased by approximately 1/3 with increasing BMI [74]. The
effectofweightandweightchangeonthelongtermprecision
of spine and hip DXA measurements was studied in 64 post
menopausal women over a period of 10 years [57]. The au
thorsconcludedthatbodyweighthadasmalleffectonpreci
sion expressed in absolute values of BMD, but not on %CV,
and this was true only for weight changes of over 5 kg. An
other study simulated change in body fatness and registered
somechangesinHologicQDR4500Awholebodyandcentral
DXAbonemeasures[75].Theauthorsconcludedthat,onav
erage, simulated weight change minimally impacted bone
measures. However, individual variability in measurement
errorwasnoteworthy[75].
The effect of weight is more pronounced in studies of body
composition.AreviewonthetruenessandprecisionofDXA

39
intheassessmentofbodycompositionhasbeensummarized
inanumberofstudiesshowingtruenesserrorsofDXA%fat
estimates between 5.3 % and +2.9 % [76]. For the Hologic
QDR 4500A, the %CV for lean and fat mass of 1.3 % was
found, for the GE Lunar Prodigy it was found to be 0.7
1.0 % for lean and 1.2 % for fat mass, and for the GE Lunar
iDXAitwasfoundtobe0.40.5%and0.70.8%respec
tively[76].

Somemorestatisticalconsiderations
The sample size required for bone density precision assess
ment is a matter of discussion. A few authors have hypothe
sizedthatasamplesizeof30mightbetoosmallforarobust
precision estimate, and compared it to a pool of approxi
mately200replicatescans[41].ThepooledspineRMSSDwas
found to be 0.017 and pooled hip RMSSD was found to be
0.009g/cm2.Samplesizesof30gavearangeofRMSSDesti
mates from 0.012 to 0.021 for the spine and from 0.008 to
0.012g/cm2forthehip.Theconclusionwasthatasamplesize
of 30 is insufficient to reliably characterize precision error or
changeduringclinicalmonitoring[41].
The validity of the LSC can be undermined by a longterm
gradual shift in themachine performance. A research group
used a specially designed phantom to monitor machine sta
bilityovertime[77].TwoHologicscannerswereobservedto
causeclinicallyinsignificantBMDshifts(maximumof0.34%),
whereas two GE/Lunar scanners revealed higher BMD shifts
(1.5%and2.1%).Asaresult,usingtheLSCcalculationsbased
only on shortterm in vivo precision studies might not be

40
validincaseofpoorerlongtermmachinestability[77].Ina
similar study, when testing the longterm performance of a
6year oldbone densitometer,the manufacturers QC failed
to detect a 2% shift in the phantom BMD values, which re
quired regular measurements of the Lunar aluminum phan
tom in addition to the daily QC measurement of the tissue
equivalentblock[78].
In addition to LSC, the measurement error can be detected
using Bland and Altmans plots [79]. They are also used in
comparisons of different technologies measuring similar tis
sues[80].AgooddescriptionofBlandandAltmansplotscan
be found in ref. [81]. Precision expressed using this method
gives an absolute estimate of random measurement error,
whichisknownasSDD.Iftherearetwoobservationsforeach
subject,thestandarddeviationofthedifferences(SDdiff)esti
mates within measurement variability. Most disagreements
between measurements are expected to be in the so called
limits of agreement between 1.96 x SDdiff and + 1.96 x
SDdiff(95%CIlimitsofagreement).Atestiscapabletodetect
a difference of at least that established by the magnitude of
the limits of agreement. The authors recommend the use of
theSDDadditionallyto,orinsteadof,theLSC.Intheirstudy,
the SDD was 0.0218 g/cm2 when both femurs were mea
sured, and 0.0339 g/cm2 when only one femur was meas
ured.Thus,theyadvocatedtheuseofthedualfemurfeature
[81].
Alternativeapproachestolongitudinalstudieshavealsobeen
proposed.Tofindoutliersmoreeasilyoftheprimaryoutcome,
aBayesianmethodwasproposedforcalculatingaprediction
intervalthatcanincorporateexternalprocessinformation[40].

41
The authors believed that this approach outperformed the
least squares method. Another research group evaluated the
empirical Monte Carlo simulation method for estimating the
significance of an observed change in BMD, which simulta
neously considers the magnitude of the change [82]. Using
this approach they identified a progressive increase in the
ability to detect BMD change using larger precision study
samplesizes[82].

Replacingorupgradingabonedensitometer.Cross
calibrationofDXAdevicesfromdifferentmanufacturers
ordifferentmodelsofthesamemanufacturer
Theindustryisconstantlydevelopingnewbonedensitometry
devices.SoitisnotuncommonforaBoneUnittoupgradean
old device or replace it with a new, more sophisticated one.
The ISCD has published guidelines for how to proceed in dif
ferentsituationsofsystemupgradeorchange[2,3,83].The
topicisalsoextensivelypresentedintheliterature.

UpgradingcentralDXAequipment
Inthecaseofsoftwareupgrade,themanufacturerhastobe
asked as to whether the new software has implemented
changesintheedgedetectionalgorithmsorthecalculationof
BMD, T and Zscores (e.g. by introducing a new reference
database). The first possibility can be tested by scanning the
phantom10timesandestablishinganewbaselineBMD.The
difference to the old phantom baseline BMD should not
exceed 1 %. The BMD, T and Zscores calculations can be

42
testedbyreanalyzingscansdonewiththeoldsoftware(with
out changing the region of interest). The numerical results
shouldnotchange.Unintentionalconsequencescan,however,
occur with software changes (84). An example is a study of
body composition analysis, in which measurements by a GE
Lunar DPXL machine differed between two (standard and
extended)analysismodes[85].
Inthecaseofhardware changes or upgrades,atechnologist
should perform 10 phantom scans with repositioning before
and after the hardware change. If a difference of more than
1%inBMDisfound,themanufacturershouldtobecontacted
forservicing.

ReplacingcentralDXAequipment
The transfer of patients to a completely new device for
measurementsusuallycreatesasetofspecificproblems.They
areexhaustedindetailinthepublicationsbytheISCD,aswell
asthosebyotherauthors[2,3,83,8688].DifferentDXAsys
temsfromthesamemanufacturermayusedifferentacquisi
tion methods, software and reference databases. Interdevice
variability is expected to be 2 %. Different manufacturers
usedifferenttechnologiestoproduceandcollimatetheXray
beam. They also use different calibration methods, edge
detection software, regions of interest, and normative data
bases.Asaresult,theinterdevicevariabilityisexpectedtobe
57%.Interdevicevariabilityisconsideredtoohighforprac
tical purposes and data are not interchangeable. Cross
calibrationisneededinthesecases.

43
Whenreplacingabonedensitometer,therearethreepossible
outcomes:
1. Thenewdeviceisfromthesamemanufacturerandmodel,
butexhibitssometechnicalandprogramimprovements.
2. The new device is from the same manufacturer but the
technology differs substantially. A good example is the
transitionfromdualphotontoXrayabsorptiometry(from
DPA to DXA). The DPA technology is rather oldfashioned
andthistypeoftransitionisveryrarelyseen.Thatiswhyit
will not be discussed further. Another example is the re
placementofapencilbeamwithafanbeamdevice.
3. Thenewdeviceisfromadifferentmanufacturer

Inthefirstscenariocrosscalibrationshouldbeperformedby
scanning the phantom 10 times with repositioning on the
new device. A new baseline BMD and intrasystem LSC is es
tablished. If there is more than 1% difference in phantom
BMD between the two devices, the manufacturer has to be
contactedforservicing.Ifthedifferencecannotbecorrected,
crosscalibrationbyscanningpatientsismandatory.
Inthesecond(e.g.replacingpencilbeamwithfanbeamtech
nologyfromthesamemanufacturer)andthirdscenarios(when
changingtoasystemmadebyadifferentmanufacturer)cross
calibrationisneeded.Crosscalibrationisan approachusedto
overcomedifferencesbetweendifferentDXAmachinesbycre
ating a regression line based on BMD values of the same pa
tientsmeasuredusingbothdevices.Thisprocessisdescribedin
detail in the Crosscalibration and Minimum Precision Stan
dardsforDXA:The2005ISCDOfficialPositions[42].

44
One approach to crosscalibration is to scan 30 subjects, who
represent the facilitys patient population, on the initial system
and then scan them twice on the new system within 60 days
(lumbar spine and hip are measured). The subjects should be
scannedonbothmachinespreferablyonthesameday.Thisap
proachcreatestheinconvenienceof:(1)keepingbothmachines
attheDXACenterand(2)ofmultiplescanningforthepatients.
Thedatacollectedareusedinregressionanalysestogeneratea
slopeandinterceptforeachregionofinterest.TheISCDhasde
veloped a specific CrossCalibration Tool (www.iscd.org). The
userhastosimplyinserthisBMDdatafromthetwomachines
andallofthecalculationsaredoneautomatically.Usingthistool,
theaverageBMDrelationship(fromtheregressionline)andLSC
forthenewandoldmachinecanbecalculated.
Intersystem LSC for two densitometers of different technol
ogyisexpectedtobe2to3timeshigherthanthatofonesin
gle machine. Some authors have suggested using the root
meansquare error (RMSE) twice for the LSC between meas
ures,forthemeasuredchangetoreachstatisticalsignificance
[89].TheintersystemLSCincorporatesthreesourcesofvaria
tion:(1)theLSCoftheinitialmachine,(2)theLSCofthenew
machine and (3) the SEE of the regression equations. The
newly established LSC on the new system, which is derived
fromtheduplicatescans,shouldbeusedfurther.
Others have advocated for the use of a general LSC [90, 91].
This approach requires 30 patients to be scanned on the old
systemforaprecisionstudy,30patientstobescannedonthe
new system for the same purpose, and 30 patients to be
scannedonbothmachinesinordertobuildaregressionline.
If two machines are not crosscalibrated, any directcompari
sonsarepronetoinherenterrors,asdescribedabove,andare
45
notallowed.Thus,itisnotpossibletoquantitativelycompare
BMDfromdifferentmachineswithoutcrosscalibration.
Extensiveworkhasbeendoneinthefieldofcrosscalibration.
Most of the older studies comparing devices from different
manufacturerswereperformedonLunar,NorlandandHologic
devices. The different machines were calibrated differently
andBMDing/cm2wasnotidentical(i.e.universal)onanyone
machine. It was readily seen that Hologic machines usually
provide lower values for BMD [88, 92]. On the other hand,
evenmachinesfromthesamemanufacturerdidnotproduce
identicalresults[9396].Twowayswereproposedtopartially
correct for this problem: (1) to crosscalibrate different ma
chines (as described above); or (2) to introduce a universal
kind of BMD (called standardized BMD or sBMD), which
mightbecalculatedfromtherawdataofthemachine.
Thefirstapproach(thecrosscalibration)isthegoldstandard
today.Itcancreatedifficultiesbecauseofthegreatvarietybe
tweendifferentDXAdevices.Anattemptwasmadeinthepast
to overcome this diversity by introducing conversion formulas
based on differentmanufacturers.A number of studiesdevel
opedconversionformulasfortheQDR1000,theLunarDPXand
theNorlandXR26[9799].Withtodaysvariety,withatleast15
central DXA devices, each machine should be crosscalibrated
againsteachother.ThatiswhytheInternationalCommitteefor
Standards in Bone Measurement supported a study, in which
100 women underwent PA spine and proximal femur meas
urements on 3 devices the Hologic QDR2000, the Norland
XR26MarkIIandtheLunarDPXL[100].Equationsforthecon
version of PA lumbar spine BMD from device to device were
published.Inthesamestudy,theEuropeanSpinePhantomwas

46
scannedoneachofthethreemachines.Thedatawereusedto
developformulasforconversion ofrawBMD dataintosBMD,
which would not differ by more than 35 % on the different
machines. These formulas were approved in 1994 [101]. The
sBMDisexpressedinmg/cm2,noting/cm2,tobereadilydistin
guished by the reader. Two years later, formulas for the total
femur(totalhip)wereapproved[102].TheformulasforsBMD
arepresentedbelow[101,102]:
sBMDSPINE=1000(1.0761xNorlandXR26BMDSPINE)
sBMDSPINE=1000(0.9522xLunarDPXLBMDSPINE)
sBMDSPINE=1000(1.0755xHologicQDR2000BMDSPINE)

sBMDTOTALFEMUR=1000[(1.008xHologicBMDTOTALFEMUR)+0.006]
sBMDTOTALFEMUR=1000[(0.979xLunarBMDTOTALFEMUR)0.031]
sBMDTOTALFEMUR=1000[(1.012xNorlandBMDTOTALFEMUR)+0.026]

AfewyearslaterLuetal.[103]developedequationsforsBMD
ofthefemoralneck,trochanterandWardsarea.Theformulas
forthefemoralneckarepresentedbelow:
sBMDFEMORALNECK=1000[(1.087xHologicBMDFEMORALNECK)+0.019]
sBMDFEMORALNECK=1000[(0.939xLunarBMDFEMORALNECK)0.023]
sBMDFEMORALNECK=1000[(0.985xNorlandBMDFEMORALNECK)+0.006]

sBMD is calculated automatically using the software and is
part of the printout of all stateoftheart DXA machines.
However, one should keep in mind that: (1) the conversion
equations used data from older technologies; and (2) the
meanerrorintroducedbyusingsBMDcanreach5%.Itissu
perimposedontheprecisionerrorof12%and,therefore,has
adeterioratingabilitytodetectearlychangesinBMD.Thatis
whysBMDshouldnotbeusedforindividualpatients[24,42].

47
TheEuropeanSpinePhantomwasusedforcrosscalibrationin
theNOREPOSstudyandwasavalidsubstitutewhenassessing
theagreementbetweenscannersofthesamemodel,notbe
tween different models by the same manufacturer [104].
Therefore,crosscalibrationofspecificDXAdevices,whilecre
ating associated regression lines and BlandAltmann plots, is
theonlyvalidwayofcomparingBMDdatabetweendevices.
Inacomparativestudy,crosscalibrationwasdonebetweenden
sitometers of different manufacturers [88]. As expected, BMD
measuredontheGELunarmachinewasabout15%higherthan
thatontheHologic.Furthermore,usingsBMDtocrosscalibrate
gaveameandifferenceof3%atthelumbarspine.Theinvivo
calibrationgavebetteragreementthanusingsBMD[88].
An example for the impact of DXA technology on crossca
librationresultsisthecomparisonofpencilbeamandfanbeam
technology. The effect of bone density, scan mode and tissue
depthonspinemeasurementswasinvestigatedinaninvivoand
phantom study using GE Lunar MD and Prodigy densitometers
[105].Thisstudydemonstratedthatthechoiceofscanmodewas
themostimportantfactorfortheProdigyandforsubjectswho
werethin,obeseorwhohadalowBMD[105].Anotherpublica
tiondiscussedthecrosscalibrationandvariabilityoftheGELu
narDPXLandProdigybonedensitometers[106].GELunariDXA
andProdigybonedensitometershavebeencrosscalibratedina
numberofstudies,buttwoofthemdemonstratedverywellthe
techniqueofcrosscalibration[52,107].Inoneofthesestudies,
BMDwasmeasuredin3groups,eachconsistingof30subjects,
by using a crossover design and both a GE Lunar iDXA and a
Prodigy[107].Crosscalibrationequationsdecreasedthesystem
atic errors between the Prodigy and the iDXA by 0.4 % at the

48
spine,0.8%atthefemoralneck,and0.1%atthetotalfemur.In
anotherstudy,thegreatestmeanBMDdifferencewasonly0.007
g/cm2andwas,therefore,ofnoclinicalconsequence[52].
The forearm site has been the subject of extensive research
focusedontryingtofindawayofstandardization.Duetothe
great variety of devices, technologies and forearm regions of
interest,theattemptofuniversalforearmstandardizationhas
obviouslyfailed[26,108,109].Norealconversionfromdevice
todeviceisreliable.

QCandcrosscalibrationintheresearchsetting
TheincreasingeffortstofillallofthegapsinthefieldofDXAQC
andcrosscalibrationhaveleadtotheinclusionofdifferentpopu
lationgroups.AcandidatefortestingthelimitsofDXAtechnol
ogyisthepediatricpopulation[110,111].Childrenandadoles
centshavesmallerbones,whosebonedensityisratherlow.Ex
cellentmachineperformanceisneededtoovercomethesediffi
culties. Special pediatric phantoms have been developed [112].
Furthermore,newphantomsfortestingbodycompositionanaly
sis and morphometric Xray absorptiometry have also been in
troduced[113116].
AnotherinterestingdirectionofresearchisintestingDXAper
formance in small animals such as mice, rats, piglets and
others[117119].
TheconceptofQCmightalsobeexploredinmoreadvanced
DXAapplications,suchasmorphometricXrayabsorptiometry
orhipstructuralanalysis,asstatedearlier[120].Itisalsoused
in other methods, which contribute to the quantification of
bonemass,suchasQCTorpQCT[121].

49
Conclusion
DXAtechnologyisnowbelievedtoprovidearobustbasisfor
thediagnosisofosteoporosis,theassessmentoffracturerisk
and monitoring of bone mass changes over time. The diag
nostic classification in osteoporosis is based on arbitrary
Tscores,whicharehighlydependentontheprecisionofthe
measurement technique [122, 123]. Low precision can yield
misleadingresults,especiallywhencombinedwiththedele
teriouseffectoflowaccuracy[124].Thus,bothaccuracyand
precision are crucial for the validity of the DXA technique.
There are attempts that are aimed to continue to improve
ourunderstandingofbothtypesoferrorandtoreducetheir
impactontheeverydayclinicalpracticeinbonedensitome
try [125127]. Proper QA and QC in bone densitometry has
becomeastandardtoolinclinicalpractice,aswellasaman
datorycriterionforaDXAsitecertificationandaccreditation
[128,129].

Acknowledgments
The author would like to thank Dr. Margarita Blajeva for im
provingtheEnglishlanguageofthismanuscript

50
References
[1] KanisJA,MeltonLJ,III,ChristiansenC,JohnstonCC,KhaltaevN.The
diagnosis of osteoporosis. Journal of Bone and Mineral Research
1994;9(8)11371141.
[2] BinkleyN,BilezikianJP,KendlerDL,LeibES,LewieckiEM,PetakSM.
Official Positions of the InternationalSociety forClinicalDensitome
try and Executive Summary of the 2005 ISCD Position Development
Conference.JournalofClinicalDensitometry2006;9(1)414.
[3] Baim S, Binkley N., Bilezikian JP, Kendler DL, Hans DB, Lewiecki EM,
Silverman S. Official Positions of the International Society for Clinical
DensitometryandExecutiveSummaryofthe2007ISCDPositionDevel
opmentConference.JournalofClinicalDensitometry2008;11(1)7591.
[4] BlakeGM,WahnerHW,FogelmanIF.Assessmentofinstrumentper
formance: precision, installation of new equipment and radiation
dose.In:BlakeGM,WahnerHW,FogelmanIF(ed.).Theevaluationof
osteoporosis. Dualenergy Xray absorptiometry and ultrasound in
clinicalpractice.London:MartinDunitz;1999,p147172.
[5] KoltaS,RouxC.Controledequalitendensitomtrieosseuse.Jour
naldeRadiologie1999;80(1)9398.
[6] BonnickSL.Chapter4:Qualitycontrol.In:BonnickSL(ed.)Boneden
sitometry in clinical practice application and interpretation, 3rd ed.
TotowaNJ:HumanaPress;2010,p105126.
[7] Dunn WL, Wahner HW. Instrument evaluation and routine quality
controlprocedures.In:WahnerHW,FogelmanI(eds.)Theevaluation
ofosteoporosis:DualenergyXrayabsoprtiometryinclinicalpractice.
MartinDunitz,London1994,p4974.
[8] BonnickSL,JohnstonCC,Jr.,KleerekoperM,LindsayR,MillerP,Sher
wood L, Siris E. Importance of precision in bone density measure
ments.JournalofClinicalDensitometry2001;4(2)105110.
[9] KellyTL,SlovikDM,NeerRM.Calibrationandstandardizationofbone
mineral densitometers. Journal of Bone and Mineral Research
1989;4(5)663669.
[10] LilleyJ,WaltersBG,HeathDA,DrolcZ.Invivoandinvitroprecision
for bone density measured by dualenergy Xray absorption. Osteo
porosisInternational1991;1(3)141146.

51
[11] OrwollES,OviattSK,BiddleJA.Precisionofdualenergyxrayabsorp
tiometry: development of quality control rules and their application
in longitudinal studies. Journal of Bone and Mineral Research
1993;8(6)693699.
[12] Kelly T, Grane G, Baran D. Single Xray Absorptiometry of the Fore
arm: Precision, Corelation, and Reference Data. Calcified Tissue In
ternational1994;54(2)212218.
[13] BassettCA.Qualityassuranceforbonedensitometryresearchstudies.
OsteoporosisInternational1994;4(5)283(comment).
[14] GlerCC,FaulknerKG,EstiloMJ,EngelkeK,RosinJ,GenantHK.Qual
ity assurance for bone densitometry research studies: concept and
impact.OsteoporosisInternational1993;3(5)227235.
[15] Gler CC, Blake GM, Lu Y, Blunt BA, Jergas M, Genant HK. Accurate
assessment of precision errors: How to measure the reproducibility
of bone densitometry techniques. Osteoporosis International
1995;5(3)262270.
[16] MillerCG.InstrumentqualitycontrolandphantomsforDXAscanners.
In:Theevaluationofosteoporosis:dualenergyXrayabsorptiometry
in clinical practice. Wahner HW, Fogelman I (eds.) London: Martin
Dunitz;1994,p17393.
[17] Njeh CF, Hans D. Instrument Evaluation. In: The Physical measure
mentofBone.LangtonCM,NjehCF(eds.)IoP,2004,p91122.
[18] Leslie WD. The importance of spectrum bias on bone density moni
toringinclinicalpractice.Bone2006;39(2)361368.
[19] Damilakis J, Guglielmi G. Quality assurance and dosimetry in bone
densitometry.RadiologyClinicsofNorthAmerica2010;48(3)629640.
[20] SheahanNF,DowlingA,O'ReillyG,MaloneJF.Commissioningandqual
ityassuranceprotocolfordualenergyXrayabsorptiometry(DEXA)sys
tems.RadiationProtectionDosimetry2005;117(13)288290.
[21] BaimS,WilsonCR,LewieckiEM,LuckeyMM,DownsRWJr,LentleBC.
Precision assessment and radiation safety for dualenergy Xray ab
sorptiometry.JournalofClinicalDensitometry2005;8(4)371378.
[22] HawkinsonJ,TiminsJ,AngeloD,ShawM,TakataR,HarshawF.Tech
nical white paper: bone densitometry. Journal of the American Col
legeofRadiology2007;4(5)320327.

52
[23] Lewiecki EM, Binkley N, Petak SM. DXA quality matters. Journal of
ClinicalDensitometry2006;9(4)388392.
[24] Kolta S, Ravaud P, Fechtenbaum J, Dougados M, Roux C. Accuracy
and precision of 62 bone densitometers using a European Spine
Phantom.OsteoporosisInternational1999;10(1)1419.
[25] Pearson J, Ruegsegger P, Dequeker J, Henley M, Bright J, Reeve J,
Kalender W, Felsenberg D, LavalJeantet AM, Adams JE, et al. Euro
peansemianthropomorphicphantomforthecrosscalibrationofpe
ripheral bone densitometers: assessment of precision accuracy and
stability.BoneMineral1994;27(2)109120.
[26] ShepherdJA,ChengXG,LuY,NjehC,ToschkeJ,EngelkeK,Grigorian
M,GenantHKUniversalstandardizationofforearmbonedensitome
try.JournalofBoneandMineralResearch2002;17(4)734745
[27] PrevrhalS,LuY,GenantHK,ToschkeJO,ShepherdJA.Towardsstan
dardization of dual Xray absorptiometry (DXA) at the forearm: a
commonregionofinterest(ROI)improvesthecomparabilityamong
DXAdevices.CalcifiedTissueInternational2005;76(5)348354.
[28] Shewhart WA. On the Measurement of a Physical Quantity Whose
MagnitudeisInfluencedbyPrimaryCausesbeyondtheControlofthe
Observer and on the Method of Determining the Relation between
Two Such Quantities. Procedures National Academy of Science USA
1922;8(8)248251.
[29] Lu Y, Mathur AK, Blunt BA, Gluer CC, Will AS, Fuerst TP, Jergas MD,
Andriano KN, Cummings SR, Genant HK. Dual Xray absorptiometry
qualitycontrol:comparisonofvisualexaminationandprocesscontrol
charts.JournalofBoneandMineralResearch1998;11(5)626637.
[30] Westgard JO, Groth T, Aronsson T, Falk H, de Verdier CH. Perform
ancecharacteristicsofrulesforinternalqualitycontrol:probabilities
forfalserejectionanderrordetection.ClinicalChemistry1977;23(10)
18571867.
[31] Westgard JO, Groth T, Aronsson T, de Verdier CH. Combined She
whartcusum control chart for improved quality control in clinical
chemistry.ClinicalChemistry1977;23(10)18811187.
[32] GarlandSW,LeesB,StevensonJC.DXAlongitudinalqualitycontrol:a
comparisonofinbuiltqualityassurance,visualinspection,multirule

53
 Shewhart charts and Cusum analysis. Osteoporosis International
1997;7(3)2317.
[33] PearsonD,CawteSA.LongtermqualitycontrolofDXA:acomparison
of Shewart rules and Cusum charts. Osteoporosis International
1997;7(4)338343.
[34] Khan AA, ColquhounA, Hanley DA, Jankowski LG, Josse RG, Kendler
DL,LentleB,LeslieWD,LewieckiEM,O'neillE,RobertsonS,SyedZA,
Tanner SB, Webster D. Standards and guidelines for technologists
performing central dualenergy Xray absorptiometry. Journal of
ClinicalDensitometry2007;10(2)189195.
[35] Faulkner KG, McClung MR. Quality control of DXA instruments in
multicentertrials.OsteoporosisInternational1995;5(4)218227.
[36] KhanKM,HenzellSL,BroderickC,PrinceRL,SaulA,LommanJ,Wark
JD.InstrumentperformanceinbonedensitytestingatfiveAustralian
centres.AustralianandNewZealandJournalofMedicine1997;27(5)
526530.
[37] Wahner HW, Looker A, Dunn WL, Walters LC, Hauser MF, Novak C.
Quality control of bone densitometry in a national helath survey
(NHANESIII)usingthreemobileexaminationcenters.JournalofBone
andMineralResearch1994;9(6)951960.
[38] Borissova AM, Rashkov R, Boyanov M, Shinkov A, Popivanov P,
Temelkova N, Vlahov J, Gavrailova M. Femoral neck bone mineral
densityand10yearabsolutefractureriskinanationalrepresentative
sampleofBulgarianwomenaged50yearsandolder.ArchivesofOs
teoporosis2011;6(12)189195.
[39] Slavchev A, Boyanov M. Impact of precision on longterm patient
monitoringduringosteoporosistreatment.In:DosselO,SchlegelWC.
(eds.),WC2009,IFMBEProceedings2009;25(7)352355.
[40] Ambrosius WT, Hui SL. A quality control measure for longitudinal
studieswithcontinuousoutcomes.StatisticsinMedicine2000;19(10)
13391362.
[41] LeslieWD,MoayerriA.Minimumsamplesizerequirementsforbone
densityprecisionassessmentproduceinconsistencyinclinicalmoni
toring.OsteoporosisInternational2006;17(11)16731680.

54
[42] Shepherd JA, Lu Y, Wilson K, Fuerst T, Genant H, Hangartner TN,
Wilson C, Hans D, Leib ES; International Society for Clinical Densi
tometry Committee on Standards of Bone Measurement.Cross
calibration and minimum precision standards for DualEnergy Xray
absorptiometry: The 2005 ISCD Official Positions. Journal of Clinical
Densitometry2006;9(1)3136.
[43] BriggsAM,WarkJD,KantorS,TehR,GreigAM,FazzalariNL,Bennell
KL.Invivointraraterandinterraterprecisionofmeasuringapparent
bone mineral density in vertebral subregions using supine lateral
dualenergy Xray absoprtiometry. Journal of Clinical Densitometry
2005;8(3)314319.
[44] FerrarL,JiangG,EastellR.ShorttermprecisionformorphometricX
rayabsorptiometry.OsteoporosisInternational2001;12(9)710715.
[45] FerrarL,JiangG,EastellR.Longitudinalevaluationofmorphometric
Xray absorptiometry for the identification of vertebral deformities.
Osteoporos.Int.2001;12(8);661671.
[46] HindK,OldroydB,TruscottJG.InvivoprecisionoftheGELunariDXA
densitometerforthemeasurementoftotalbodycompositionandfat
distribution in adults. European Journal of Clinical Nutrition
2011;65(1)140142.
[47] OldroydB,SmithAH,TruscotJG.CrosscalibrationofGE/Lunarpencil
andfanbeamdualenergydensitometersbonemineraldensityand
body composition studies. European Journal of Clinical Nutrition
2003;57(8)977987.
[48] KhooBC,BeckTJ,QiaoQH,ParakhP,SemanickL,PrinceRL,SingerKP,
Price RI. In vivo shortterm precision of hip structure analysis vari
ables in comparison with bone mineral density using paired dual
energy Xray absorptiometry scans from multicenter clinical trials.
Bone2005;37(1)112121.
[49] BonnickSL.Chapter15:FDAapprovedbonedensitometrydevices.In:
BonnickSL(ed.)Bonedensitometryinclinicalpracticeapplicationand
interpretation,3rded.TotowaNJ:HumanaPress;2010,p371425.
[50] Shepherd JA,Fan B, Lu Y, Lewiecki EM,Miller PD, Genant HK. Com
parison of BMD precision for Prodigy and Delphi spine and femur
scans.OsteoporosisInternational2006;17(9)13031308.

55
[51] Henzell, S., S. Dhaliwal, R. Pontifex, Gill F, Price R, Retallack R,
Prince R. Precision error of fanbeam dual Xray absorptiometry
scansatthespine,hip,andforearm.JournalofClinicalDensitome
try2000;3(4)359364.
[52] Krueger D, VallartaAst N, Checovich M, Gemar D, Binkley N. BMD
measurement and precison: a comparison of GE Lunar Prodigy and
iDXADensitometers.JournalofClinicalDensitometry2012;15(1)2125.
[53] Nalda E, Mahadea KK, Demate C, Kotzki PO, Pouget JP, Boudousq
V.AssessmentoftheStratos:anewpencilbeambonedensitometer:
dosimetry, precision, and cross calibration Journal of Clinical Densi
tometry2011;14(4)395406.
[54] BlakeGM,KnappKM,FogelmanI.DualXrayabsorptiometry:clinical
evaluationofanewconebeamsystem.CalcifiedTissueInternational
2005;76(2)113120.
[55] Boyanov M, Shinkov A, Nestorova R. Quantitativer Ultraschall am
Kalkaneus und am distalen Radius: eine Vergleichsstudie mit der
ZweiSpektren RntgenAbsorptiometrie der Lendenwirbelsule und
des proximalen Femurs. Deutsche Medizinische Wochenschrift
2007;132(16)869873(articleinGerman).
[56] PatelR,BlakeGM,RymerJ,FogelmanI.LongtermprecisionofDXA
scanningassessedoversevenyearsinfortypostmenopausalwomen.
OsteoporosisInternational2000;11(1)6875.
[57] RajamanoharaR,RobinsonJ,RymerJ,PatelR,FogelmanI,BlakeGM.
Theeffectofweightandweightchangeonthelongtermprecisionof
spine and hip DXA measurements. Osteoporosis International
2011;22(5)15031512.
[58] TothillP,HannanWJ.Precisionandaccuracyofmeasuringchangesin
bonemineraldensitybydualenergyXrayabsorptiometry.Osteopo
rosisInternational2007;18(11)15151523.
[59] Lenora J, Akesson K, Gerdhem P. Effect of precision on longitudinal
followup of bone mineral density measurements in elderly women
andmen.J.Clin.Densitom.2010;13(4);407412.
[60] Boyanov M, Prodanova A, Popivanov P. Reproducibility of forearm
Single Xray absoprtiometry. Rentgenologia i Radiologia 1999;38(2)
1519(articleinBulgarian).

56
[61] Boyanov M. Quality control and standardization of forearm Xray
bone densitometry. Rentgenologia i Radiologia 2000;39(2);95100
(articleinBulgarian).
[62] Frimeth J, Galiano E, Webster D. Some physical and clinical factors
influencing the measurement of precision errors, least significant
change,andbonemineraldensityindualenergyXrayabsorptiome
try.JournalofClinicalDensitometry2010;13(1)2935.
[63] LeslieWD,fortheManitobaBoneDensityProgram.Factorsaffecting
shorttermbonedensityprecisionassessmentandtheeffectonpa
tient monitoring. Journal of Bone and Mineral Research 2008;23(2)
199204.
[64] Kiebzak GM, Morgan SL. Longterm versus shortterm precision of
dualenergyXrayabsorptiometryscansandtheimpactofinterpret
ing change in bone mineral density at followup. Journal of Clinical
Densitometry2011;14(2)108115.
[65] BlakeGM,HerdRJ,FogelmanIF.Alongitudinalstudyofsupinelateral
DXAofthelumbarspine:acomparisonwithposteroanteriorspine,hip
andtotalbodyDXA.OsteoporosisInternational1996;6(6)462470.
[66] Wong JC, Griffiths MR. Precision of bone densitometry measure
ments:whenischangetruechangeanddoesitvaryacrossboneden
sityvalues?AustralasianRadiology2003;47(3)236239.
[67] Leslie WD, for the Manitoba Bone Density Program. The impact of
bone area on shortterm bone density precision. Journal of Clinical
Densitometry2006;9(2)150153.
[68] Blank RD, Malone DG, Christian RC, VallartaAst NL, Krueger DC,
DreznerMK,BinkleyNC,HansenKE.Patientvariablesimpactlumbar
spine dualenergy Xray absorptiometry precision. Osteoporosis In
ternational2006;17(5)768774.
[69] Fan B, Lewiecki EM, Sherman M, Lu Y, Miller PD, Genant HK,
ShepherdJA.ImprovedprecisionwithHologicApexsoftware.Osteo
porosisInternational2008;19(11)15971602.
[70] CorderoMacIntyre ZR, Peters W, Libanati CR, Espaa RC, Abila SO,
Howell WH, Lohman TG.Reproducibility of DXA in obese women.
JournalofClinicalDensitometry2002;5(1)3544.

57
[71] Lekamwasam S, Lenora RSJ. The effect of leg rotation on hip bone
mineral density measurements. Journal of Clinical Densitometry
2003;6(4)331336.
[72] McKiernan F, Washington W. Effect of subtle positioning flaws on
measuredbonemineraldensityofthehip.JournalofClinicalDensi
tometry2005;8(3)330334.
[73] TherboM,PetersenMM,SchrderHM,NielsenPK,ZerahnB,LundB.
The precision and influence of rotation for measurements of bone
mineraldensityofthedistalfemurfollowingtotalkneearthtroplasty:
a methodological study using DXA. Acta Orthopaedica Scandinavica
2003;74(6)677682.
[74] Knapp KM, Welsman JR, Hopkins SJ, Fogelman I, Blake GM. Obesity
increasesprecisionerrorsindualenergyXrayabsorptiometrymeas
urements.JournalofClinicalDensitometry2012;15(3)315319.
[75] EvansEM,MojtahediMC,KessingerRB,MisicMM.Simulatedchange
inbodyfatnessaffectsHologicQDR4500AwholebodyandcentralDXA
bonemeasures.JournalofClinicalDensitometry2006;9(3)315322.
[76] ToombsRJ,DucheG,ShepherdJA,DeSouzaMJ.Theimpactofrecent
technological advances on the trueness and precision of DXA to as
sessbodycomposition.Obesity2012;20(1)3039.
[77] Hangartner TN. A study of longterm precision of dualenergy Xray
absorptiometrybonedensitometersandimplicationsforthevalidity
oftheleastsignificantchangecalculation.OsteoporosisInternational
2007;18(4)513523.
[78] WellsJ,RyanPJ.ThelongtermperformanceofDXAbonedensitome
ters.BritishJournalofRadiology2000;73(871)737739.
[79] Feldmann U, Schneider B, Klinkers H, Haeckel R. A multivariate ap
proachforthebiometriccomparisonofanalyticalmethodsinclinical
chemistry. Journal of Clinical Chemistry and Clinical Biochemistry
1981;19(3)121137.
[80] Boneva Z, Boyanov M. Body composition analysis by legtoleg bio
electrical impedance and by dualenergy Xray absorptiometry in
nonobese and obese individuals. Diabetes Obesity Metabolism
2008;10(10)10121018.

58
[81] ElMaghraouiA,AchemlalL,BezzaA.MonitoringofdualenergyXray
absorptiometry measurement in clinical practice. Journal of Clinical
Densitometry2006;9(3)281286.
[82] LeslieWD,MoayyeriA,SadatsafaviM,WangL.Anewapproachfor
quantifyingchangeandtestprecisioninbonedensitometry.Journal
ofClinicalDensitometry2007;10(4)365369.
[83] WritingGroupfortheISCDPositionDevelopmentConference.Tech
nicalstandardizationfordualenergyXrayabsorptiometry.Journalof
ClinicalDensitometry2004;7(1)2736.
[84] BinkleyN,KiebzakGM,LewieckiEM,KruegerD,GangnonRE,Miller
PD,ShepherdJA,DreznerMK.RecalculationoftheNHANESDatabase
SD improves TScore agreement and reduces osteoporosis preva
lence.JournalofBoneandMineralResearch2005;20(2)195201.
[85] NakataY,TanakaK,MizukiT,YoshidaT.Bodycompositionmeasure
ments by dualenergy Xray absorptiometry differ between two
analysismodes.JournalofClinicalDensitometry2004;7(4)443447.
[86] OrmicaC.StandardizationofBMDmeasurements.OsteoporosInter
national1998;8(1)13.
[87] Bonnick SL. Chapter 6: Bone density data among technologies and
manufacturers. In: Bonnick SL (ed.) Bone densitometry in clinical
practiceapplicationandinterpretation,3rded.TotowaNJ:Humana
Press;2010,p141161.
[88] PearsonD,HortonB,GreenDJ.CrosscalibrationofDXAaspartofan
equipment replacement program. Journal of Clinical Densitometry
2006;9(3)287294.
[89] Blake GM, Harrison EJ, Adams JE. Dual Xray absorptiometry: cross
calibration of a new fanbeam system. Calcified Tissue International
2004;75(1)714.
[90] ShepherdJA,LuY.Ageneralizedleastsignificantchangeforindividu
als measured on different DXA systems. Journal of Clinical Densi
tometry2007;10(3)249258.
[91] ShepherdJA,MorganSL,LuY.ComparingBMDresultsbetweentwo
similar DXA systems using the generalized least significant change.
JournalofClinicalDensitometry2008;11(2)237242.

59
[92] HbschP,SchneiderB,SeidlG,KalchhauserG,KlaushoferK,Popovic
R. [DEXA (dualenergy xray absorptiometry) and DPA (dual photon
absorptiometry) in densitometry of the femoral neck: correlation of
themeasurementsofthreecommerciallyavailableinstruments].Ak
tuelleRadiologie1991;1(4)16973.(ArticleinGerman).
[93] AbrahamsenB,GramJ,HansenTB,BeckNilesenH.Crosscalibration
ofQDR2000andQDR1000dualenergyXrayabsorptiometrydensi
tometers for bone mineral and soft tissue measurements. Bone
1995;16(3)385390.
[94] BartheN,BraillonP,DucassouD,BasseCathalinatB.Comparisonof
twoHologicDXAsystems(QDR1000andQDR4500A).BritishJournal
ofRadiology1997;70(835)728739.
[95] Saarelainen J, Honkanen R, Vanninen E, Krger H, Tuppurainen M,
Niskanen L, Jurvelin JS. Crosscalibration of Lunar DPXIQ and DPX
dualenergy Xray absorptiometry densitometers for bone mineral
measurements in women Journal of Clinical Densitometry 2005;8(3)
320329.
[96] Choi D, Kim DY, Han CS, Kim S, Bok HS, Huh W, Ko JW, Hong SH.
Measurements of bone mineral density in the lumbar spine and
proximal femur using Lunar Prodigy and the new pencilbeam dual
energy Xray absorptiometry. Skeletal Radiology 2010;39(11) 1109
1116.
[97] Gundry CR, Miller CW, Ramos E, Moscona A, Stein JA, Mazess RB,
Sartoris DJ, Resnick D. Dualenergy radiographic absorptiometry of
thelumbarspine:clinicalexperiencewithtwodifferentsystems.Ra
diology.1990;174(2)53941.
[98] Lai KC, Goodsitt MM, Murano R, Chesnut CH. A comparison of two
dualenergy Xray absorptiometry systems for spinal bone mineral
measurement.CalcifiedTissueInternational1992;50(3)203208.
[99] Pocock NA, Sambrook PN, Nguyen T, Kelly P, Freund J, Eisman JA.
AssessmentofspinalandfemoralbonedensitybydualXrayabsorp
tiometry: comparison of Lunar and Hologic instruments. Journal of
BoneandMineralResearch1992;7(9)10811084.

60
[100] GenantHK,GramppS,GluerCC,FaulknerKG,JergasM,EngelkeK,
HagiwaraS,VanKuijkC.UniversalstandardizationfordualXrayab
sorptiometry:patientandphantomcrosscalibrationresults.Journal
ofBoneandMineralResearch1994;9(10)15031514.
[101] Steiger P. Standardization ofspine BMD measurements. Journal of
BoneandMineralResearch1995;10(10)16021603.
[102] HansonJ.StandardizationoffemurBMD.JournalofBoneandMin
eralResearch1997;12(8)13161317.
[103] Lu Y, Fuerst T, Hui S, Genant HK. Standardization of bone mineral
densityatfemoralneck,trochanter,andWardstriangle.Osteopo
rosisInternational2001;12(6)438444.
[104] OmslandTK,EmausN,GjesdalCG,FalchJA,TellGS,ForsenL,Berntsen
GK,MeyerHE.Invivoandinvitrocomparisonofdensitometersinthe
NOREPOSstudy.JournalofClinicalDensitometry2008;11(2)276282.
[105] LaskeyMA,MurgatroydPR,PrenticeA.Comparisonofnarrowangle
fan beam and pencil beam densitometers. In vivo and Phantom
studyoftheeffectofbonedensity,scanmodeandtissuedepthon
spine measurements. Journal of Clinical Densitometry 2004;7(3)
341348.
[106] GagnonDR,McLeanRR,HannanMT,CupplesLA,HoganM,KielDP.
Crosscalibration and comparison of variability of two bone densi
tometers in research setting: the Framingham experience. Journal
ofClinicalDensitometry2010;13(2)210218.
[107] Choi YJ, Lee BJ, Lim HC, Chung YS. CrossCalibration of iDXA and
Prodigy on Spine and Femurs Scans in Korean Adults. Journal of
ClinicalDensitometry2009;12(4)450455.
[108] HagiwaraS,EngelkeK,YangSO,DhillonMS,GuglielmiG,NelsonDL,
Genant HK. Dual xray absorptiometry forearm software: accuracy
and intermachine relationship. Journal of Bone and Mineral Re
search1994;9(9)14251427.
[109] Forsn L, Bernsten GK, Meyer HE, Tell GS, Fnneb V, NOREPOS
CoreResearchGroup.Differencesinprecisioninbonemineralden
sitymeasuredbySXAandDXA:theNOREPOSstudy.EuropeanJour
nalofEpidemiology2008;23(9)615624.

61
[110] Litaker MS, Barbeau P, Humphries MC, Gutin B. Comparison of
Hologic QDR1000/W and 4500W DXA scanners in 13 to 18year
olds.ObesityResearch2003;11(12)15451552.
[111] LeonardCM,RozaMA,BarrRD,WebberCE.PeproducibilityofDXA
measurements of bone mineral density and body composition in
children.PediatricRadiology2009;39(2)148154.
[112] PicaudJC,DuboeufF,VeyMartyV,DelamsP,ClarisO,SalleBL,Rigo
J. First allsolid pediatric phantom for dualenergy Xray absorpti
ometry measurements in infants. Journal of Clinical Densitometry
2003;6(1)1724.
[113] DiesselE,ShepherdJA,FuerstT,GonzalezM,GenantHK,CarewB,
LookerA.Comparisonoftwophantomsforbodycompositionwith
dualenergyXrayabsorptiometry.AnnalsoftheNewYorkAcademy
ofScience2000;904107110.
[114] EllisKJ,ShypailoRS,SteinbergFM,LewisRD,YoungRL,WongWW.
ReproducibilityoffanbeamDXAmeasurementsinadultsandphan
toms.JournalofClinicalDensitometry2004;7(4)413418.
[115] ReaJA,BlakeGM,FogelmanI.Developmentofaphantomformor
phometric Xray absorptiometry. British Journal of Radiology
2001;74(880)341350.
[116] SteelSA,ThorpeJA,WalkerR,HoweyS,LangtonCM.Development
andevaluationofaphantomformorphometricXrayabsorptiome
try.OsteoporosisInternational1999;9(1)3844.
[117] IidaKleinA,LuSS,YokoyamaK,DempsterDW,NievesJW,LindsayR.
Precision,accuracyandreproducibilityofdualenergyXrayabsorp
tiometry measurements in mice in vivo. Journal of Clinical Densi
tometry2003;6(1)2534.
[118] HortonJA,MurrayGM,SpadaroJA,MarguliesBS,AllenMJ,Damron
TA.PrecisionandaccuracyofDXAandpQCTfordensitometryofthe
ratfemur.JournalofClinicalDensitometry2003;6(4)381390.
[119] Fink C, Cooper HJ, Huebner JL, Guilak F, Kraus VB. Precision and
accuracy of a transportable dualenergy Xray absorptiometry unit
forbonemineralmeasurementsinguineapigs.CalcifiedTissueIn
ternational2002;70(3)164169.

62
[120] HindK,OldroydB,PrajapatiA,RhodesL.Invivoprecisionofdual
energyXrayabsorptiometryderivedhipstructureanalysisinadults.
JournalofClinicalDensitometry2012;15(3)302307.
[121] EngelkeK,LibanatiC,LiuY,WangH,AustinM,FuerstT,StampaB,
TimmW,GenantHK.Quantitativecomputedtomography(QCT)of
the forearm using general purpose spiral wholebody CT scanners:
accuracy,precisionandcomparisonwithdualenergyXrayabsorp
tiometry(DXA).Bone2009;45(1)110118.
[122] BlakeGM,ShepherdJA.EffectofrandomBMDmeasurementerror
ondiagnosticclassificationusingTscores.JournalofClinicalDensi
tometry2007;10(4)415416.
[123] KiebzakGM,FaulknerKG,WackerW,HamdyR,SeierE,WattsNB.
EffectofprecisionerroronTscoresandthediagnosticclassification
ofbonestatus.JournalofClinicalDensitometry2007;10(3)239243.
[124] KiebzakGM,FaulknerKG,WackerW,HamdyR,WattsNB.Accuracy
error versus precision error and the effect on Tscores. Journal of
ClinicalDensitometry2007;10(4)417419.
[125] Bonnick SL. Monitoring changes in bone density. Womens Health
(LondonEngl).2008;4(1)8997.
[126] EngelkeK,GlerCC.Qualityperformancemeasuresinbonedensi
tometry: part 1: errors and diagnosis. Osteoporosis International
2006;17(9)12831289.
[127] ProvynS,ClarysJP,WallaceJ,ScafoglieriA,ReillyT.Qualitycontrol,
accuracy,andpredictioncapacityofdualenergyXrayabsorptiome
try variables and data acquisition. Journal of Physiological Anthro
pology2008;27(6)317323.
[128] AvecillaLS.Professionalcertificationandsiteaccreditationinbone
densitometry.JournalofClinicalDensitometry1998;1(1)8189.
[129] Bennett HS, Dienstfrey A, Hudson LT, Oreskovic T, Fuerst T, Shep
herd J. Standards and measurements for assessing bone health.
JournalofClinicalDensitometry2006;9(4)399405.

63