Journal of Microscopy, 2009 doi: 10.1111/j.1365-2818.2009.03338.x

Received 13 March 2009; accepted 9 September 2009

Quality control protocol for in vitro micro-computed tomography

R . S T O I C O ∗, S . T A S S A N I ∗,†, E . P E R I L L I‡,§, F . B A R U F F A L D I ∗

& M . V I C E C O N T I ∗∗Laboratorio di Tecnologia Medica, Istituto Ortopedico Rizzoli, Bologna, Italy

†Engineering Department, University of Bologna, Bologna, Italy

‡Bone and Joint Research Laboratory, SA Pathology and Hanson Institute, Adelaide, Australia

§Discipline of Pathology, University of Adelaide, Adelaide, Australia

Key words. Phantoms, quality control protocol, X-ray microtomography.

Summary

The aim of this work was to present and discuss a qualitycontrol protocol for in vitro micro-computed tomography(microCT), based on the adaptation of the quality controlprotocols for medical computed tomography. The importanceof establishing a quality control protocol is related to theopportunity to identify problems on time comparing themicroCT images acquired in different time points, and in thisway to verify the performance of the device. The proposedquality control protocol was applied for a long-time monitoringperiod to verify the stability of the micro-tomographic systemover time. The protocol proposed in this study was appliedto the histomorphometric characterization of bone tissue,but it can be used on a wide range of in vitro microCTapplications. Noise and uniformity tests, taken and adaptedto micro-tomographic system by medical standard guidelinesof quality control, were performed by the use of a waterphantom. An accuracy test was designed and performed bythe use of a morphometric calibrated phantom. All thesetests were performed during a long-time monitoring period tocontrol the stability of the system. Specific control charts andmonitoring parameters for each test were used to represent themonthly measures collected during 20 months and an out ofcontrol condition was defined. The reference values (baseline),calculated to control the stability of micro-tomographic systemover time, were calculated during acceptance/status test.During the period, no out of control conditions in noise,uniformity and accuracy tests were recorded. However, achanging condition was found in noise test, as showed byusing statistical C (P < 0.01) and Kruskal–Wallis (P < 0.05)tests. In particular, a Wilcoxon rank sum test with Bonferronicorrection (P < 0.0125) was applied in noise test to investigatewhich of the comparisons among first five acquisitions ofyear 2004 (group B.L.) and each group was significant

Correspondence to: Rossella Stoico. Tel: +39-051-6366565; fax: +39-051-

6366863; e-mail: stoico@tecno.ior.it

(P < 0.0125). The noise showed a slight but significantincrease over the years compared to baseline value; however,no out of control conditions were recorded. Nonetheless, amaintenance service to control the performance of mechanicalcomponents of microCT was required and performed.

Introduction

Quality control (QC) is a process applied to ensure a certainlevel of quality in a product or service. The level of quality ofa measuring instrument is connected to the accuracy of themeasurement. Therefore, quality control is a process employedto verify the accuracy of the measurements of the instrumentover the time according to commonly accepted standards ofquality.

X-ray micro-computed tomography (microCT) is a non-destructive investigation method with widespread use indifferent research and industrial fields, including biomedicalresearch. It is a three-dimensional (3D) imaging methodwith high spatial resolution that does not require specialpreparation of the specimen, and that is used in mostcases to perform spatial measurements of the microstructure(morphometry) or of the X-ray attenuation (densitometry).The accuracy needed in these microCT investigations suggestsa systematic application of QC protocols, but no establishedprotocols are available, yet.

Making a parallel with medical CT scanners used in theclinical practice, we notice that these machines are alwayssubjected to QC procedures, not only to assure the safety forpatients and medical professionals, but also to guarantee asufficient accuracy of the measurements performed using thedevice (i.e. see European Directive 97/43/EURATOM).

Also in the research literature there is abundance of studieson new QC protocols for CT. Both the American Associationof Physicists in Medicine (AAPM Report No. 74, 2002) andInstitute of Physics and Engineering in Medicine (IPEM ReportNo. 91, 2005) propose QC protocols in medical CT based

C© 2009 The AuthorsJournal compilation C© 2009 The Royal Microscopical Society

2 R . S T O I C O E T A L .

on international directive to ensure the quality of diagnosticaccuracy. These QC protocols are about the control of themost important medical CT characteristics, such as imagedetail and noise (Sprawls, 1992), uniformity and linearityof CT numbers (Hounsfield Units), spatial and high/lowcontrast resolution and dose evaluation. The QC protocolsare performed using phantoms with tissue-equivalent insertsdesigned according to specific clinical applications (Kalender& Suess, 1987; Kalender, 1992; Ruegsegger & Kalender,1993; Kalender et al., 1995; Huda et al., 1997; Olerudet al., 1999; Birnbaum et al., 2002; Ko et al., 2003; Funamaet al., 2005).

MicroCT is based on the same physical principles of medicalCT and is applied widely in medical research, especially inthe histomorphometric characterization of bone tissue. Tothe best of the authors’ knowledge, only few applicationsof QC protocols for microCT have been reported in theliterature, and their attention focus mostly on densitometry.They are based on the use of solid and liquid calibrationphantoms, to evaluate the quality of the bone mineral densitymeasures (Kazakia et al., 2008; Nazarian et al., 2008). Ourattention focused mostly on the definition of a quality controlprotocol and its application over time in the morphometriccharacterization of bone tissue. Recently, a QC phantomwas designed, to evaluate the performance of an in vivomicro-computed tomographic system, operated at 150 μmresolution (Du et al., 2007). The in vivo tomographic systemis used to investigate small animals, such as mice or ratsfor pre-clinical analysis. The QC tests proposed in Du et al.’sstudy are the evaluation of spatial resolution, geometricaccuracy, CT number accuracy, linearity, noise and imageuniformity. The aim of the application of those tests is toassure that measurements are not affected by scanner drift.However, that phantom cannot be used for in vitro microCTevaluation of quality level because the inserts of that phantomhave been chosen to control the most important parametersof pre-clinical analysis, with particular attention to tissue-density discrimination. On the contrary, in vitro microCTsystems are used mostly for morphometric analysis. Thephantom proposed by Du et al. does not contain inserts withdifferent thickness and geometries and thus a morphometricphantom is necessary to monitor the stability of an in vitromicro-tomographic system for the specific application of themorphometric characterization of bone tissue. Moreover, aprocedure for a long-time collecting data was not introducedby Du et al.

The aim of this study is to propose a QC protocol for invitro microCT, by adapting the QC protocols of medical CTand to introduce a systematic procedure of collecting datato monitor the stability of the micro-tomographic systemover the time. The proposed QC protocol was applied to thehistomorphometric characterization of bone tissue but the QCtests proposed in this study can be adapted and applied tovarious applications of in vitro microCT.

Material and methods

A QC protocol for in vitro microCT is proposed. It was inspiredby QC protocols used in medical CT (IPEM Report No. 91,2005), which ensures a level of quality suitable for clinicalpractice. The proposed QC protocol is summarized in Table 1.Three tests were proposed: (1) noise and (2) uniformity testsperformed by the use of a water phantom; (3) accuracy testperformed by the use of a morphometric calibrated phantom.

The presented tests are based on the use of baselinevalues (baseline, B.L.) and control charts. The B.L. values aremeasured during an acceptance/status test, and represent thereference values for the control charts, which are subsequentlyused to monitor the specific parameters over time. The testsare based on the use of two specific calibration phantoms: awater phantom and a morphometric calibration phantom.

The application of the QC protocol for in vitro microCTis related to the particular use of the X-ray measuringinstrument. In fact, the procedures and QC parameters inthe noise and uniformity tests were directly inspired by IPEMindications and based on the use of a water phantom.

The IPEM indications are summarized as follows. In noisetest, a circular region of interest (ROI) of 500 mm2 areais positioned in a centre of a reconstructed slice of a waterphantom. The standard deviation of the average CT numbers(Hounsfield Unit, HU) is recorded. In the uniformity test, onecircular ROI of 500 mm2 area is positioned in the centreand four ROIs are positioned in the peripheral region of areconstructed slice of the water phantom, at the distance of1 cm from the boundary. The difference in grey level betweenaverage density of central ROI and the average of averagedensity of four peripheral ROIs is recorded. However, IPEMindications do not suggest any accuracy test for morphometricmeasurements. For the accuracy test, the calibration phantomhad to reproduce materials, structure and dimensions of theobject of investigation, that is, trabecular bone specimens.With this aim, a previously published physical morphometriccalibration phantom was used (Perilli et al., 2006). Moreover, aprocedure for the evaluation of the accuracy of morphometricmeasurements was introduced.

In the presented QC protocol, in the acceptance/statustest both phantoms were scanned five consecutive times tocalculate the reference value (B.L.). The B.L. was used asthe reference value for data collection charts. Each scan wasperformed on a different day for 5 consecutive days, and thephantom was repositioned in the microCT each time. The QCphantoms were scanned maintaining constant the scannersettings, the same used to scan bone biopsies (Perilli et al.,2007). A periodic monitoring time of 1 month was proposed,to control whether the measurements showed any drift.

If a measure exceeds the upper or lower limits defined foreach specific adopted chart, it is defined as out of control. Whenthis happens, the QC protocol here proposed recommends anintervention, e.g. maintenance service.

C© 2009 The AuthorsJournal compilation C© 2009 The Royal Microscopical Society, Journal of Microscopy

Q U A L I T Y C O N T R O L P R O T O C O L F O R I N V I T R O M I C R O C T 3

Table 1. The proposed QC protocol for in vitro microCT.

Phantoms Noise/uniformity test: Water or soft-tissue equivalentAccuracy test: Morphometric calibration phantom (materials, dimensions and geometries known)

Procedure Noise test: Circular region of interest (ROI) in proportion to the area of the water phantom positioned according IPEMprocedure (IPEM Report No. 91, 2005) on the reconstructed grey-level images in the central part of the waterphantom (Fig. 1)

Uniformity test: Central circular ROI in proportion of the area of the water phantom and four peripheral ones positionedaccording to IPEM Procedure (2005) on the reconstructed grey-level images in the central part of the water phantom(Fig. 2)

Accuracy test: Shape and dimension of the ROI had to be selected to contain the whole object of interest; a threshold valuewas used to calculate the parameters on reconstructed grey-level images

QC parameters Noise test: Standard deviation of grey levels of the ROI averaged on five spatially contiguous reconstructed slicesUniformity test: Difference in grey levels calculated between average density of the circular central ROI and the average

of average density of the four circular peripheral ones; the difference in grey levels averaged on the same five spatiallycontiguous reconstructed slices

Accuracy test: Average absolute error in thickness (average |ETh|) calculated as the average value of the absolutedifference between measured values and baseline (B.L.)

Baseline Noise/uniformity/accuracy tests: B.L. calculated during acceptance/status test.

Data collecting method Noise test: Chart with tolerance range according to the indication of IPEM guidelines (±20% of the B.L.) (IPEM ReportNo. 91, 2005)

Uniformity/accuracy test: Shewhart control chart (Montgomery, 2006)

Time monitoring Noise/uniformity/accuracy tests: Monthly

Maintenance service Noise/uniformity/accuracy tests: Intervention, e.g. maintenance service request in case of out of control situations:measure out of upper and lower control limits of the specific chart adopted

The data collection method for the noise test was inspiredby the IPEM indications, whereas for the uniformity andaccuracy tests a new method was proposed. In the noise test,IPEM indications suggested that the tolerance range of thestandard deviation of the average CT number (HU) in thecentral ROI has to be ±20% of the B.L. value. In particular,for the uniformity test two approaches were considered: thedata collection method inspired by IPEM indication and anovel method. These methods were applied and comparedto evaluate the most suitable data collection method for theQC protocol. In the uniformity test, IPEM indications suggestedthat the tolerance range of difference in grey level betweencentral ROI and peripheral ones has to be ±1.5% of the B.L.value. The novel methods for uniformity and accuracy testswere described further on.

MicroCT scanner settings and image processing

The QC protocol was applied on a Skyscan in vitro microCTmodel 1072 (Skyscan, Kontich, Belgium) composed by amicro-focus X-ray tube (spot size <5 μm) and a 1024 × 102412-bit charge-coupled device camera with maximum fieldof view of 25 mm. The scanning parameters were 50 kVp,200 μA and rotation step 0.45◦. A 1-mm aluminium filterwas used to reduce the beam-hardening effect. Each frontalX-ray image was averaged over two frames, each single framehaving an exposure time of 5.9 s. The magnification was set

to 16× to obtain a pixel size of 19.5 μm and a field of view of20 × 20 mm. The size of the pixel corresponds to the isotropicvoxel size in the reconstructed slices.

A filtered back-projection Feldkamp algorithm (Feldkampet al., 1984) was used for cross-section reconstruction(software Cone-Rec v.2.9, Skyscan). The X-ray projectionimages were obtained in 12-bit grey levels, stored in a 16-bit format. The reconstructed tomographic images were savedin 8-bit format (256 grey levels) and 1024 × 1024 pixelsin size. The grey-level values close to 0 grey level were setequivalent to the bone and the ones close to 255 grey levelwere set equivalent to the air.

A global thresholding procedure was chosen to convertthe reconstructed cross sections from grey-level images(8-bit format, 256 grey levels) into binary images. A uniquethreshold value was used for all acquisitions. It was definedas the threshold value that minimizes the accuracy error incomparison with nominal thickness declared by manufacturerof the aluminium (Al)-inserts of the morphometric calibrationphantom (Perilli et al., 2006).

Application of the in vitro microCT QC protocol

Acceptance/status test. The acceptance/status test wasperformed on both the water phantom and the morphometricphantom in 2004, from which the B.L. was calculated for eachparameter.

C© 2009 The AuthorsJournal compilation C© 2009 The Royal Microscopical Society, Journal of Microscopy

4 R . S T O I C O E T A L .

Periodic time monitoring. The periodic time monitoringstarted 3 years after the acceptance/status test to verify ifthe measurements showed any scanner drift. Three yearstime corresponds to approximately 1800 work-hours of theX-ray tube (elapsed time tool, TomoNT acquisition softwarever. 3L.5, Kontich, Belgium). All the measures described inthe QC protocol were performed monthly, for 20 months.

Noise test

The noise test was performed with the water phantom. Thephantom consists of a cylindrical plastic vessel with an18-mm outer diameter, 14-mm inner diameter and 44-mmheight. A series of five spatially contiguous cross sections werereconstructed at 22-mm height from the bottom of the waterphantom.

According to IPEM indications, the ROI area correspondsto 500 mm2 because a standard-size soft tissue-equivalentphantom is used for the application of QC protocols inmedical CT. In in vitro microCT systems, the water phantomdimensions can be chosen according to the available field ofview size. In this case, the ROI size was chosen to be 10% ofthe area of the inner water phantom area. This percentagewas the same suggested in IPEM indications, but in that case itwas calculated as ROI diameter (25 mm) and water phantomdiameter (25 cm) ratio. The percentage, calculated as ratioof areas, represented a good balance between statistics andthe ROI size for the application of protocol. The ROI waspositioned in the centre of each of the five spatially contiguousreconstructed slices.

The QC parameter of noise test was calculated as thestandard deviation of the grey levels averaged on five spatiallycontiguous cross sections (average SD). The B.L. was used toestablish the quality control chart with an upper noise limitand lower noise limit as ±20% of B.L. respectively, accordingto IPEM indications (IPEM Report No. 91, 2005).

Image-Pro Plus v.4.5.1.22 (The Proven Solution, MediaCybernetics, Inc.) image analysis software was used tocalculate the average SD in grey levels in the circular ROIfor the noise test (Fig. 1) as described in field Procedure inTable 1.

Uniformity test

For the uniformity test, the water phantom and the five spatiallyconsecutive grey-level images of the noise test were used. Acircular ROI, the same as used for the noise test, was positionedin the centre of the water phantom and four circular ones,with the same dimensions, were positioned in the peripherallocation (Fig. 2).

The QC parameter of the uniformity test was calculated asthe difference in grey levels (GL) between average density (G L )of the central circular ROI (ROIa) and the average of average

Fig. 1. Circular ROI area, positioning for the noise test. This procedurewas inspired by IPEM guidelines (IPEM Report No. 91, 2005).

Fig. 2. The uniformity test procedure inspired by IPEM guidelines. Imagesin grey levels were used. The ROI was the same as that used for the noise testand was positioned in the central part (a) and in four peripheral locations(b, c, d, e) of the water phantom.

C© 2009 The AuthorsJournal compilation C© 2009 The Royal Microscopical Society, Journal of Microscopy

Q U A L I T Y C O N T R O L P R O T O C O L F O R I N V I T R O M I C R O C T 5

density of all four circular peripheral ones (ROIb, ROIc, ROId,ROIe) (Eq. 1, Fig. 2). This difference was averaged on fivespatially contiguous cross sections.

Difference (G L ) = G L ROIa −

⎛⎜⎜⎝

e∑i=b

G L ROIi

4

⎞⎟⎟⎠ (1)

where i is the circular ROI reference index and GL is the greylevels.

Image-Pro Plus v.4.5.1.22 (The Proven Solution, MediaCybernetics, Inc.) image analysis software was used tocalculate the average difference in grey levels for the uniformitytest (Fig. 2) as described in the field Procedure in Table 1.

Concerning the data collection method, two differentapproaches were considered to choose the most suitablemonitoring data method for the proposed QC protocol. In onecase, the upper uniformity limit and lower uniformity limitwere represented by ±1.5% of B.L. respectively, which arethe tolerance limits defined by IPEM guidelines (IPEM ReportNo. 91, 2005). In the second case, the upper control limit andthe lower control limit were represented as±3×mean movingrange divided by 1.128, which are the tolerance limits definedby the Shewhart control chart for single measures (Montgomery,2006). The moving range is the absolute difference betweeneach pair of consecutive measures. The 1.128 value is thereference value for measures with a sample size equal to two(Montgomery, 2006). The mean moving range was calculatedby the first five acquisitions in 2004.

Accuracy test

IPEM guidelines do not contain any accuracy test formorphometric measurements. However, the research interestfor in vitro microCT was actually in morphometriccharacterization of trabecular bone tissue. Thus, a dedicatedprotocol was developed and applied, based on a morphometriccalibration phantom. The main characteristic of a phantomfor accuracy test in morphometry is to reproduce the typical 3Dstructure of the analyzed specimen, e.g. the trabecular boneframework.

A previously published physical 3D phantom withcalibrated Al-inserts was chosen (Perilli et al., 2006) forthe application of the proposed QC protocol (Table 1).The phantom was designed in cylindrical shape (13-mmdiameter × 23-mm height). The Al-inserts were of differentgeometries (foils, wires, meshes and spheres) and calibratedthickness to reproduce the typical thickness of trabecular bonestructure (Perilli et al., 2006). The objects of the calibrationphantom are four foils of 20 ± 3, 50 ± 7.5, 100 ± 10and 250 ± 25 μm in thickness, four wires of 20 ± 2, 50 ±5, 125 ± 12.5 and 250 ± 25 μm in thickness, a smallhorizontal and a vertical mesh composed of 100 ± 10 μmthick wires, a horizontal mesh of bigger external size and

Fig. 3. Morphometric calibration phantom with materials, geometriesand dimensions of aluminium inserts known (measures in mm).

four spheres of 1000 ± 50 μm in diameter embedded inpolymethylmethacrylate (Fig. 3). The tolerance in nominalthickness of the Al-inserts declared by the manufacturer(Goodfellow Cambridge Limited, Huntingdon, U.K.) was 5–15%, depending on type of inserts. However, the nominalthickness of the Al-inserts was verified by performing repeatedmeasurements in the laboratory on samples of the samebatch (Perilli et al., 2006). The foils and spheres weremeasured by using a screw–thread micrometer with digitaldisplay (1-μm resolution, Mytutoyo, Kawasaki, Kanagawa,Japan), whereas wires and mesh by using optical microscope(400× magnification, 0.346 μm pixel size; DC300, LeicaMicrosystems, Wetzlar, Germany). The Al material was usedbecause its X-ray attenuation coefficient is very similar tothe bone tissue [μAl(30 keV) = 3.04 cm−1; μBone(30 keV) =2.56 cm−1] (Hubbell et al., 1996).

A stack of 861 reconstructed slices was chosen toinclude all the Al-inserts. For the binarization of the cross-sectional images, a uniform thresholding procedure definedin previously published work was used (Perilli et al., 2006).The thinnest Al-inserts such as 20 μm, 50-μm-thick wiresand 20-μm-thick foil, although visible in the grey level cross-sectional images, were not properly segmented. This mightbe related to beam-hardening artefacts and more likely topartial volume effects. The nominal thickness of the thinnest

C© 2009 The AuthorsJournal compilation C© 2009 The Royal Microscopical Society, Journal of Microscopy

6 R . S T O I C O E T A L .

Table 2. The different ROI sizes and shapes and specific volume of interest (VOI) used for thickness calculation of the segmentedAl-inserts.

Al-inserts ROI size (mm2) − pixel ROI shape VOI (ROI size × no. of slices)

Wire (20 μm) Not segmented Not segmented Not segmentedFoil (20 μm) Not segmented Not segmented Not segmentedWire (50 μm) Not segmented Not segmented Not segmentedFoil (50 μm) 5.4 × 1.6 − 275 × 80 Rectangular 275(80 × 281)Mesh (small horizontal) 5.8 × 5.8 − 300 × 300 Square 300(300 × 31)Mesh (big horizontal) 9.7 × 9.7 − 500 × 500 Square 500(500 × 26)Mesh (small vertical) 5.4 × 1.6 − 275 × 80 Rectangular 275(80 × 191)Foil (100 μm) 5.4 × 1.6 − 275 × 80 Rectangular 275(80 × 281)Wire (125 μm) 0.9 × 0.9 − 48 × 48 Square 48(48 × 576)Foil (250 μm) 5.4 × 1.6 − 275 × 80 Rectangular 275(80 × 281)Wire (250 μm) 0.9 × 0.9 − 48 × 48 Square 48(48 × 576)Spheres 5.8 × 5.8 − 300 × 300 Square 300(300 × 61)

inserts correspond to approximately one to two times the pixelsize (19.5 μm). Thus, making their segmentation difficult,specially for thin cylindrical objects such as wires. Some ofthe binarized cross sections of the 20-μm- and 50-μm-thickwires and the 20-μm-thick foils lacked in segmented structure,which makes the calculation of thickness of these objectschallenging or meaningless. Thus, these three inserts wereexcluded from calculation (Table 2).

A 3D calculator program (software 3D calculator v.0.9,Skyscan) was used to calculate the thickness of the all Al-inserts. The algorithm implemented in 3D calculator programis based on the method described by Hildebrand & Ruegsegger(1997). In that method, the thickness (Th) is obtained by fittingmaximal spheres to each point in the 3D structure (Hildebrand& Ruegsegger, 1997). ROIs of various shapes and sizes werechosen according to the different geometry and size of the Al-inserts to calculate the thickness (Table 2). For each singleAl-insert a specific ROI was drawn, which had to include thewhole single insert in each cross section (Fig. 4). The thicknesswas calculated separately for each Al-insert over the specificvolume of interest (Table 2).

The QC parameter considered in this study was the averageabsolute error of the thickness. The absolute error of thethickness |ETh| for each segmented Al-insert was calculatedas the absolute difference between the measured Thj and thereference value Thref (Montgomery, 2006) (Eq. 2).

|ETh| = |T hj − T href | (2)

The reference value Thref was the nominal value of the Al-inserts declared by the manufacturer (Perilli et al., 2006)and verified by repeated optical measurements, as previouslydescribed. The average |ETh| was obtained by calculatingthe average value of the absolute error of the thickness ofall the segmented Al-insert. The |ETh|∗ value was calculatedas average of |ETh| values on the 20 months for each singlesegmented Al-insert.

Fig. 4. Example of reconstructed slices of (a) spheres, (b) large horizontalmesh, (c) foils and (d) vertical mesh. The measures are compared to thenominal values of Al-inserts.

At each measuring time point, each accuracy measurementrequired nine |ETh| measures, corresponding to the ninesegmented Al-inserts (Table 2). Not segmented Al-insertswere excluded in the calculation of the error, as mentionedpreviously. For data collection, the use of the Shewhart controlchart for single measures was not suitable, as the sample size(9) is greater than two. In this case, the Shewhart controlchart (Montgomery, 2006) was more appropriate, and hencewas used to collect the monthly average |ETh| values on themorphometric phantom. The standard deviation was usedto establish the range of the tolerance limits. The uppercontrol limit and the lower control limit of the chart werecalculated as ±3 × average standard deviation (σavg |ETh|) onthe segmented Al-inserts. The intermediate tolerance limits,corresponding to ±2σavg |ETh| and ±1σavg |ETh|, were alsoreported.

C© 2009 The AuthorsJournal compilation C© 2009 The Royal Microscopical Society, Journal of Microscopy

Q U A L I T Y C O N T R O L P R O T O C O L F O R I N V I T R O M I C R O C T 7

Statistical analysis

All the measures of the QC protocol were processed by thestatistical test C, with the null hypothesis that the collecteddata were randomly distributed. This test is very flexible forverifying the null hypothesis, however without considerationsabout the alternative hypothesis (Eq. 3) (Young, 1941).

C = 1 −∑n−1

i−1 (Xi − Xi+1)2

2∑n

i=1 (Xi − X̄)2 (3)

where Xi, Xi+1, . . . , Xn = measurements.The measures were collected monthly over 20 months for

each test. The data were then divided into four groups, witheach group containing the measures of 5 months to reproducethe same group dimension of collecting data method duringacceptance/status test. Apart from the reference group ofthe first five measures of year 2004 (group B.L.), the otherfour groups were group 1 composed by the five consecutivemonthly measures starting from August 2007, group 2composed by five consecutive monthly measures starting fromJanuary 2008, group 3 composed by five consecutive monthlymeasures starting from June 2008 and group 4 composed byfive consecutive monthly measures starting from November2008. A Kruskal–Wallis test was applied to evaluate if therewere significant differences in the measures between all themonthly acquisition groups, including the B.L. group (totalof five groups). If a measure was found significantly different,a Wilcoxon rank sum test with Bonferroni correction wasapplied to investigate which of the comparisons among firstfive acquisitions of year 2004 (group B.L.) and each group wassignificant (P<0.0125). Wilcoxon rank sum test is a statistical

test with a P = 0.05 as level of significance. The Bonferronicorrection states if n independent hypothesis on a set of dataare tested, each individual hypothesis will be tested with astatistical significant level of 1/n times. In this study, we havefour independent hypotheses when each group, composed offive monthly measures, and group B.L. are compared. Thus,each hypothesis can be tested with a level of significanceP/n = 0.05/4 = 0.0125.

Results

Noise test

The average value in grey levels of the water correspondsto 233.51 and it was calculated during acceptance/statustest with the application of the noise test of the present QCprotocol. Figure 5 shows the test chart that was used tomonitor the average SD in grey levels by using the waterphantom, with the tolerance range taken from IPEM qualityindications. The chart shows no out of control points. However,the statistical test C shows that collected data are not randomlydistributed (Table 3). The Kruskal–Wallis test found that therewere significant differences between the five groups (Table 3).In particular, the Wilcoxon rank sum test with Bonferronicorrection showed statistically significant differences betweeneach group and the B.L. (Table 4). The average SD value ofeach group increased compared to the B.L. value.

Uniformity test

Figure 6 shows the Shewhart control chart for single measuresused to monitor the difference in grey levels, with two different

Fig. 5. Noise test chart: Noise test chart of the average SD in grey levels of an ROI area in proportion of the water phantom area. The tolerance range[upper noise limit (UNL) = +20% B.L., lower noise limit (LNL) = −20% B.L.] was calculated in accordance with IPEM guidelines. The baseline value wasequal to 3.35 ± 0.02 grey levels. The average density in grey levels of water corresponds to 233.51. It was calculated during acceptance/status test.

C© 2009 The AuthorsJournal compilation C© 2009 The Royal Microscopical Society, Journal of Microscopy

8 R . S T O I C O E T A L .

Table 3. Mean and SD of the most important selected parameters for the QC protocol application on water and Al-calibration phantoms.

Group B.L. Group 1 Group 2 Group 3 Group 4 Test C P Kruskal–Wallis P

Water phantomNoise test:Average SD in grey levels 3.35 ± 0.02 3.52 ± 0.05 3.56 ± 0.02 3.51 ± 0.05 3.52 ± 0.08 <0.01 <0.05

Uniformity test:Difference in grey levels −0.98 ± 0.06 −0.90 ± 0.10 −0.96 ± 0.04 −0.90 ± 0.12 −0.96 ± 0.09 0.82 0.31

Al-calibration phantomAccuracy test:Average |ETh| (μm) 19.6 ± 5.3 20.1 ± 1.3 20.3 ± 1.8 19.8 ± 2.7 20.6 ± 0.4 0.91 0.78

The mean of each group composed of 5 consecutive monthly measures and B.L. was calculated. The P-value (P = 0.01, level of significance) of test C andthe Kruskal–Wallis statistical test (P = 0.05, level of significance) are shown.

Table 4. The P-value of Wilcoxon rank sum test with Bonferronicorrection.

Noise test:Average SD in grey levels

Group B.L. vs. Group 1 <0.0125Group B.L. vs. Group 2 <0.0125Group B.L. vs. Group 3 <0.0125Group B.L. vs. Group 4 <0.0125

This test was performed to identify which group is different from baselinevalue for noise test. The P-value was deemed significant if P < 0.0125.In noise test, the Kruskal–Wallis test shows statistically significantdifferences among five groups (group B.L., group 1, group 2, group 3and group 4).

approaches for data collection. By following the tolerancelimits given by IPEM guidelines (upper uniformity limit, loweruniformity limit = ±1.5% of the B.L., dotted lines narrowed tothe B.L), 18 out of control points were found. If, however, the

tolerance limits of the Shewhart control chart for single measureswere considered, no out of control points were found. Thestatistical test C showed that the collected data were randomlydistributed (Table 3), and the Kruskal–Wallis test showed nostatistically significant differences when comparing the fivegroups of measures (Table 3).

Accuracy test

Figure 7 shows theapplication of Shewhart control chart tomonitor the average |ETh| parameter of the segmented Al-inserts of the morphometric calibration phantom. The chartshows no out of control points. The statistical test C showedthat the collected data were randomly distributed (Table 3),and the Kruskal–Wallis test showed no statistically significantdifferences between the five groups of measures (Table 3).Table 5 shows the B.L. values and the |ETh|∗ values of eachsegmented Al-inserts. |ETh|∗ values were calculated as averageof |ETh| values of 20 months collecting data.

Fig. 6. Uniformity test chart: Shewhart control chart for single measures of the difference in grey levels, upper control limit (UCL) = B.L. + 3 × 0.09, lowercontrol limit (LCL) = B.L. − 3 × 0.09, procedure indicated by IPEM. The dotted lines [upper uniformity limit (UUL) = +1.5% B.L., lower uniformity limit(LUL) = −1.5% B.L.] narrowed to B.L. were the tolerance limits suggested by IPEM guidelines. The baseline value was equal to −0.98 ± 0.07 grey levels.

C© 2009 The AuthorsJournal compilation C© 2009 The Royal Microscopical Society, Journal of Microscopy

Q U A L I T Y C O N T R O L P R O T O C O L F O R I N V I T R O M I C R O C T 9

Fig. 7. Accuracy test chart: Shewhart control chart for the average |ETh| parameter calculated by using the Al-calibration phantom. The UCL and LCL werecalculated as B.L. ± 3σavg |ETh|, upper control limit (UCL) = B.L. + 3 × 5.3, lower control limit (LCL) = B.L. − 3 × 5.3. The baseline value was equal to19.6 ± 5.3 μm.

Discussion

In this study, a QC protocol for in vitro microCT was definedand applied. It was inspired by QC protocols commonly usedin medical CT. The proposed QC protocol was applied in aparticular medical research field, which is the morphometriccharacterization of trabecular bone specimens. For thispurpose, suitable phantoms were used. In the accuracytest, a previously published phantom was used. This wasdesigned to be reproducible by other laboratories, and thus,it was looked for standard calibrated aluminium objects thatwere commercially available, with given tolerances (Perilliet al., 2006). In a first instance, the percentage values ofthe tolerances appear rather large for calibration purposes.

However, repeated optical measurements were done in asample of the same batch (Perilli et al., 2006). These showedthat the variation in thickness from the nominal valueswere 2–10 times smaller than the tolerance limits given bythe manufacturer and also than the pixel size used duringmicroCT scans in this paper (19.5 μm). Although the nominaltolerances could represent a caveat of this study, the opticalmeasurements show that the aluminium inserts used for thephantom can be considered appropriate for the purpose of thisstudy.

Quality control charts were used to collect and monitor themonthly measures of selected QC parameters. The procedure ofnoise and uniformity tests was taken from IPEM guidelines andadapted to microCT. The adaptation was related to the choice

Table 5. The nominal value of each Al-insert, Thref , the deviation of the measured thickness of each insertcalculated at baseline, |ETh|, and the deviation of each insert averaged over 20 month, |ETh|∗.

Al-inserts Thref (μm) Baseline (μm) |ETh|∗ (μm)

Wire (20 μm) 20 ± 2 Not segmented Not segmentedFoil (20 μm) 20 ± 3 Not segmented Not segmentedWire (50 μm) 50 ± 5 Not segmented Not segmentedFoil (50 μm) 50 ± 7.5 1.9 ± 1.3 3.1 ± 2.4Mesh (small horizontal) 100 ± 10 3.4 ± 1.2 7.8 ± 2.3Mesh (big horizontal) 100 ± 10 10.8 ± 1.0 7.3 ± 2.4Mesh (small vertical) 100 ± 10 18.5 ± 0.5 19.7 ± 2.5Foil (100 μm) 100 ± 10 27.8 ± 1.0 28.5 ± 1.9Wire (125 μm) 125 ± 12.5 19.3 ± 6.2 24.7 ± 9.2Foil (250 μm) 250 ± 25 39.7 ± 0.8 40.4 ± 1.7Wire (250 μm) 250 ± 25 8.3 ± 0.8 7.7 ± 2.6Spheres 1000 ± 50 46.7 ± 3.3 42.5 ± 8.0

All – Average |ETh| = 19.6 ± 5.3 Average |ETh|∗ = 20.2 ± 1.6

C© 2009 The AuthorsJournal compilation C© 2009 The Royal Microscopical Society, Journal of Microscopy

1 0 R . S T O I C O E T A L .

of ROI size because the ROI dimension suggested by IPEM(10% of water phantom diameter) was not considered suitablefor the water phantom dimension in microCT, however thepercentage suggested by IPEM (10%) was kept. The ROI areawas calculated as 10% of the area of water phantom. Itrepresented a good balance for statistics and ROI dimensionfor the application of protocol. A procedure for accuracy testby using morphometric measurements was introduced. In thenoise test, the same collecting data method suggested by IPEMindications was used. In the uniformity test, a new collectingdata method was proposed. It was based on new tolerancerange defined by using a Shewhart control chart for singlemeasures. The data and analysis over a 20 months applicationof the QC protocol were presented.

Noise test

Monthly noise measures were in accordance with tolerancelimits defined by international guidelines for medical CT(±20% of B.L.). However, the collected data were notrandomly distributed (Table 3), and were in the upper partof the tolerance range (Fig. 5). Moreover, a statisticallysignificant difference between the five groups was found byusing Kruskal–Wallis test (Table 3). In particular, statisticallysignificant differences were found between B.L. and each groupof five consecutive monthly measures (Table 4). The noise(average SD in grey levels) was found to increase after 3years, compared to B.L. These results can be considered anindication of a changing condition. Although the values werenot found out of the upper and lower noise limits of the chart, amaintenance service was required and performed to verify thecause of these results. The maintenance service reproduces thesame tests performed during acceptance/status tests to verifythe performances of the mechanical components of microCT,the X-ray tube and the charge-coupled device camera after 3years’ working. During maintenance service, apart from theincreased noise, no major problems were identified, suggestinga probable cause being related to the normal ageing of the X-ray source. A minor problem was identified, a misalignmentbetween X-ray tube and charge-coupled device camera. Itwas resolved with a calibration procedure. However, furthermonthly measures will be performed according QC protocoland used to verify the performance of the micro-tomographicsystem.

Uniformity test

Two different approaches to collect the data were applied. Thefirst approach was based on the tolerance range suggestedby IPEM guidelines (±1.5% of B.L.). However, this was notsuitable to monitor uniformity data stored in 8-bit format(256 grey levels) images. In fact, the tomographic images inmedical CT, for which the IPEM guidelines are designed, arestored in a 12-bit format, which give a much higher grey-

level range (4096 grey levels). The tolerance range of thedifference in grey levels between the central ROI (233.52 GL)and peripheral ROIs (234.50 GL) was a limitation becauseof the small dispersion of the grey levels around B.L. (upperuniformity limit = −0.97, lower uniformity limit = −0.99;Fig. 6). The consequence was to observe 18 out of control pointsin the uniformity chart. But statistical C and Kruskal–Wallistests (Table 3) showed no statistically significant differences.To save images in a larger bit format might be a solution.By contrast, the concern in this paper was to observe howthe variability of the average SD (noise test) and differencein grey levels (uniformity test) influenced the morphometricmeasurements. For this purpose, the grey-level images for noiseand uniformity tests were to be reconstructed in the same bitformat of grey-level images for accuracy test (8 bit).

A second approach for data collection in uniformitymeasurements was introduced and applied, which was basedon the Shewhart control chart for single measures. This chartshowed no out of control points (Fig. 6).

Considering the outcomes of the statistical test C, the datawere randomly distributed (Table 3), and there were nostatistically significant differences between the five groups(Kruskal–Wallis test, P < 0.05; Table 3). This outcome isconsistent with that of the Shewhart control chart for singlemeasurements, which can be considered the most suitableapproach to collect the data of uniformity test.

Accuracy test

The choice to combine Al-inserts with different sizes andshapes for the morphometric calibration phantom was relatedto the aim to reproduce the geometries and thickness of atypical 3D structure of trabecular bone specimens. Over theseinserts, a single threshold value was used. This was chosento minimize the average |ETh| calculated on all the segmentedAl-inserts (at B.L., average |ETh| = 19.6 μm), and was thesame threshold value as in a previous published study (Perilliet al., 2006). Then, a unique Shewhart control chart was used tomonitor the average |ETh| for the QC parameters calculated onall the segmented Al-inserts of the morphometric calibrationphantom over time. No out of control points were observed. Onaverage, the deviations from the nominal values in thickness(average |ETh|∗ = 20.2 μm; Table 5) were about the pixel sizeused for the tomographic acquisition (19.5 μm). The accuracyerror for the single segmented Al-insert was about one to twotimes the pixel size used in this study (19.5 μm) or much lower.In general, all the monthly measures were within one standarddeviation (Fig. 7), randomly distributed (Table 3), with nosignificant differences between five groups of measurements(Table 3). Thus, the results of the morphometric accuracy testshows that the performance of the microCT system can beconsidered not out of control. A Shewhart control chart for thesingle Al-insert, although feasible for specific purposes, was notintroduced here into the QC protocol, both to not complicate

C© 2009 The AuthorsJournal compilation C© 2009 The Royal Microscopical Society, Journal of Microscopy

Q U A L I T Y C O N T R O L P R O T O C O L F O R I N V I T R O M I C R O C T 1 1

the monthly application of the protocol and because themonitoring of the accuracy error of a single insert might not berepresentative of the accuracy error calculated for the complex3D structure to investigate, that is trabecular bone specimen,which contains a mixture of geometries. However, monitoringthe accuracy of a single specific measurement is possible, andthe presented protocol can be used and easily adapted to thespecific purpose.

In conclusion, a QC protocol in in vitro microCT wasproposed and applied successfully. The protocol, inspired byIPEM guidelines, was adapted to monitor the performanceof in vitro microCT in a common application field, which istrabecular bone histomorphometry. The noise and uniformitytests were performed with a water phantom. For the accuracytest, a physical phantom for the calibration of morphometricmeasurements in 3D was used. Three years after measuringB.L. values, the application of the QC protocol showed thatthe quality level of the microCT scanner used was not outof control concerning accuracy and uniformity. The noiseshowed a slight but significant increase over the years, whichnonetheless can be considered negligible, as it had no effect onthe outcome in histomorphometry. However, the systematicincrease in monthly measures compared to B.L. can beinterpreted as a changing condition in the performance of theX-ray tube and charge-coupled device camera of the microCTsystem. A maintenance service to control the performance ofmechanical components of microCT after 3 years’ working wasrequired and performed. The importance to apply a periodicQC protocol is related both to observe and resolve punctualproblems and to monitor the stability of micro-tomographicsystem over years to maintain a high-quality level of themeasurements.

Acknowledgements

The authors thank Paolo Erani and Mauro Ansaloni fortechnical support, Luigi Lena for the pictures and Keith Smithfor English language review. This work has been partiallysupported by the Region—University Research Program2007–2009 and by the VPHOP Project project (FP-223865).

References

The American Association of Physicists in Medicine (AAPM), DiagnosticX-ray Imaging Committee Task Group #12, Report No. 74 (2002) Qualitycontrol in diagnostic radiology, 74 pp.

Institute of Physics and Engineering in Medicine (IPEM), Report No.91 (2005) Recommended standards for the performance testing ofdiagnostic X-ray imaging systems, 112 pp.

Birnbaum, B.A., Maki, D.D., Chakraborty, D.P., Jacobs, J.E. & Babb,J.S. (2002) Renal cyst pseudoenhancement: evaluation with ananthropomorphic body CT phantom. Radiology 225, 83–90.

Du, L.Y., Umoh, J., Nikolov, H.N., Pollmann, S.I., Lee, T.Y. & Holdsworth,D.W. (2007) A quality assurance phantom for the performanceevaluation of volumetric micro-CT systems. Phys. Med. Biol. 52, 7087–7108.

Feldkamp, L.A., Davis, L.C. & Kress, J.W. (1984) Practical cone-beamalgorithm. J. Opt. Soc. Am. A. 1, 612–619

Funama, Y., Awai, K., Nakayama, Y. et al. (2005) Radiationdose reduction without degradation of low-contrast detectabilityat abdominal multisection CT with a low-tube voltage technique:phantom study. Radiology 237, 905–910.

Hildebrand, T. & Ruegsegger, P. (1997) A new method for the model-independent assessment of thickness in three-dimensional images. J.Microsc. 185, 67–75.

Hubbell, J.H. & Selzer, S.M. (1996) Tables of X-Ray Mass AttenuationCoefficients. NISTIR 5632. NIST, Gaithersburg, MD, USA.

Huda, W., Atherton, J.V., Ware, D.E. & Cumming, W.A. (1997) Anapproach for the estimation of effective radiation dose at CT in pediatricpatients. Radiology 203, 417–422.

Kalender, W.A. (1992) A phantom for standardization and qualitycontrol in spinal bone mineral measurements by QCT and DXA: designconsiderations and specifications. Med. Phys. 19, 583–586.

Kalender, W.A. & Suess, C. (1987) A new calibration phantom forquantitative computed tomography. Med. Phys. 14, 863–866.

Kalender, W.A., Felsenberg, D., Genant, H.K., Fischer, M., Dequeker,J. & Reeve, J. (1995) The European Spine Phantom – a toolfor standardization and quality control in spinal bone mineralmeasurements by DXA and QCT. Eur. J. Radiol. 20, 83–92.

Kazakia, G.J., Burghardt, A.J., Cheung, S. & Majumdar, S. (2008)Assessment of bone tissue mineralization by conventional x-raymicrocomputed tomography: comparison with synchrotron radiationmicrocomputed tomography and ash measurements. Med. Phys. 35,3170–3179.

Ko, J.P., Rusinek, H., Jacobs, E.L., Babb, J.S., Betke, M., McGuinness, G. &Naidich, D.P. (2003) Small pulmonary nodules: volume measurementat chest CT–phantom study. Radiology 228, 864–870.

Montgomery, D.C. (2006) Introduction to Statistical Quality Control, TheMcGraw-Hill Companies, S.r.l., Publishing Group Italia, Milano.

Nazarian, A., Snyder, B.D., Zurakowski, D. & Muller, R. (2008)Quantitative micro-computed tomography: a non-invasive method toassess equivalent bone mineral density. Bone 43, 302–311.

Olerud, H.M., Olsen, J.B. & Skretting, A. (1999) An anthropomorphicphantom for receiver operating characteristic studies in CT imaging ofliver lesions. Br. J. Radiol. 72, 35–43.

Perilli, E., Baruffaldi, F., Bisi, M.C., Cristofolini, L. & Cappello, A. (2006)A physical phantom for the calibration of three-dimensional X-raymicrotomography examination. J. Microsc. 222, 124–134.

Perilli, E., Baruffaldi, F., Visentin, M., Bordini, B., Traina, F., Cappello, A. &Viceconti, M. (2007) MicroCT examination of human bone specimens:effects of polymethylmethacrylate embedding on structural parameters.J. Microsc. 225, 192–200.

Ruegsegger, P. & Kalender, W.A. (1993) A phantom for standardizationand quality control in peripheral bone measurements by PQCT andDXA. Phys. Med. Biol. 38, 1963–1970.

Sprawls, P. (1992) AAPM tutorial. CT image detail and noise.Radiographics 12, 1041–1046.

Young, L.C. (1941) On randomness in ordered sequences. Ann. Math. Stat.12, 293–300.

C© 2009 The AuthorsJournal compilation C© 2009 The Royal Microscopical Society, Journal of Microscopy