Quality control protocol for in vitro micro-computed tomography

Authors


Rossella Stoico. Tel: +39-051-6366565; fax: +39-051-6366863; e-mail: stoico@tecno.ior.it

Summary

The aim of this work was to present and discuss a quality control protocol for in vitro micro-computed tomography (microCT), based on the adaptation of the quality control protocols for medical computed tomography. The importance of establishing a quality control protocol is related to the opportunity to identify problems on time comparing the microCT images acquired in different time points, and in this way to verify the performance of the device. The proposed quality control protocol was applied for a long-time monitoring period to verify the stability of the micro-tomographic system over time. The protocol proposed in this study was applied to the histomorphometric characterization of bone tissue, but it can be used on a wide range of in vitro microCT applications. Noise and uniformity tests, taken and adapted to micro-tomographic system by medical standard guidelines of quality control, were performed by the use of a water phantom. An accuracy test was designed and performed by the use of a morphometric calibrated phantom. All these tests were performed during a long-time monitoring period to control the stability of the system. Specific control charts and monitoring parameters for each test were used to represent the monthly measures collected during 20 months and an out of control condition was defined. The reference values (baseline), calculated to control the stability of micro-tomographic system over time, were calculated during acceptance/status test. During the period, no out of control conditions in noise, uniformity and accuracy tests were recorded. However, a changing condition was found in noise test, as showed by using statistical C (P < 0.01) and Kruskal–Wallis (P < 0.05) tests. In particular, a Wilcoxon rank sum test with Bonferroni correction (P < 0.0125) was applied in noise test to investigate which of the comparisons among first five acquisitions of year 2004 (group B.L.) and each group was significant (P < 0.0125). The noise showed a slight but significant increase over the years compared to baseline value; however, no out of control conditions were recorded. Nonetheless, a maintenance service to control the performance of mechanical components of microCT was required and performed.

Introduction

Quality control (QC) is a process applied to ensure a certain level of quality in a product or service. The level of quality of a measuring instrument is connected to the accuracy of the measurement. Therefore, quality control is a process employed to verify the accuracy of the measurements of the instrument over the time according to commonly accepted standards of quality.

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

Making a parallel with medical CT scanners used in the clinical practice, we notice that these machines are always subjected to QC procedures, not only to assure the safety for patients and medical professionals, but also to guarantee a sufficient accuracy of the measurements performed using the device (i.e. see European Directive 97/43/EURATOM).

Also in the research literature there is abundance of studies on new QC protocols for CT. Both the American Association of Physicists in Medicine (AAPM Report No. 74, 2002) and Institute of Physics and Engineering in Medicine (IPEM Report No. 91, 2005) propose QC protocols in medical CT based on international directive to ensure the quality of diagnostic accuracy. These QC protocols are about the control of the most important medical CT characteristics, such as image detail and noise (Sprawls, 1992), uniformity and linearity of CT numbers (Hounsfield Units), spatial and high/low contrast resolution and dose evaluation. The QC protocols are performed using phantoms with tissue-equivalent inserts designed according to specific clinical applications (Kalender & Suess, 1987; Kalender, 1992; Ruegsegger & Kalender, 1993; Kalender et al., 1995; Huda et al., 1997; Olerud et al., 1999; Birnbaum et al., 2002; Ko et al., 2003; Funama et al., 2005).

MicroCT is based on the same physical principles of medical CT and is applied widely in medical research, especially in the histomorphometric characterization of bone tissue. To the best of the authors’ knowledge, only few applications of QC protocols for microCT have been reported in the literature, and their attention focus mostly on densitometry. They are based on the use of solid and liquid calibration phantoms, to evaluate the quality of the bone mineral density measures (Kazakia et al., 2008; Nazarian et al., 2008). Our attention focused mostly on the definition of a quality control protocol and its application over time in the morphometric characterization of bone tissue. Recently, a QC phantom was designed, to evaluate the performance of an in vivo micro-computed tomographic system, operated at 150 μm resolution (Du et al., 2007). The in vivo tomographic system is used to investigate small animals, such as mice or rats for pre-clinical analysis. The QC tests proposed in Du et al.'s study are the evaluation of spatial resolution, geometric accuracy, CT number accuracy, linearity, noise and image uniformity. The aim of the application of those tests is to assure that measurements are not affected by scanner drift. However, that phantom cannot be used for in vitro microCT evaluation of quality level because the inserts of that phantom have been chosen to control the most important parameters of pre-clinical analysis, with particular attention to tissue-density discrimination. On the contrary, in vitro microCT systems are used mostly for morphometric analysis. The phantom proposed by Du et al. does not contain inserts with different thickness and geometries and thus a morphometric phantom is necessary to monitor the stability of an in vitro micro-tomographic system for the specific application of the morphometric characterization of bone tissue. Moreover, a procedure for a long-time collecting data was not introduced by Du et al.

The aim of this study is to propose a QC protocol for in vitro microCT, by adapting the QC protocols of medical CT and to introduce a systematic procedure of collecting data to monitor the stability of the micro-tomographic system over the time. The proposed QC protocol was applied to the histomorphometric characterization of bone tissue but the QC tests proposed in this study can be adapted and applied to various applications of in vitro microCT.

Material and methods

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

Table 1.  The proposed QC protocol for in vitro microCT.
PhantomsNoise/uniformity test: Water or soft-tissue equivalent
Accuracy test: Morphometric calibration phantom (materials, dimensions and geometries known)
ProcedureNoise test: Circular region of interest (ROI) in proportion to the area of the water phantom positioned according IPEM procedure (IPEM Report No. 91, 2005) on the reconstructed grey-level images in the central part of the water phantom (Fig. 1)
Uniformity test: Central circular ROI in proportion of the area of the water phantom and four peripheral ones positioned according 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 value was used to calculate the parameters on reconstructed grey-level images
QC parametersNoise test: Standard deviation of grey levels of the ROI averaged on five spatially contiguous reconstructed slices
Uniformity 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 spatially contiguous reconstructed slices
Accuracy test: Average absolute error in thickness (average |ETh|) calculated as the average value of the absolute difference between measured values and baseline (B.L.)
BaselineNoise/uniformity/accuracy tests: B.L. calculated during acceptance/status test.
Data collecting methodNoise test: Chart with tolerance range according to the indication of IPEM guidelines (±20% of the B.L.) (IPEM Report No. 91, 2005)
Uniformity/accuracy test: Shewhart control chart (Montgomery, 2006)
Time monitoringNoise/uniformity/accuracy tests: Monthly
Maintenance serviceNoise/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 presented tests are based on the use of baseline values (baseline, B.L.) and control charts. The B.L. values are measured during an acceptance/status test, and represent the reference values for the control charts, which are subsequently used to monitor the specific parameters over time. The tests are based on the use of two specific calibration phantoms: a water phantom and a morphometric calibration phantom.

The application of the QC protocol for in vitro microCT is related to the particular use of the X-ray measuring instrument. In fact, the procedures and QC parameters in the noise and uniformity tests were directly inspired by IPEM indications and based on the use of a water phantom.

The IPEM indications are summarized as follows. In noise test, a circular region of interest (ROI) of 500 mm2 area is positioned in a centre of a reconstructed slice of a water phantom. The standard deviation of the average CT numbers (Hounsfield Unit, HU) is recorded. In the uniformity test, one circular ROI of 500 mm2 area is positioned in the centre and four ROIs are positioned in the peripheral region of a reconstructed slice of the water phantom, at the distance of 1 cm from the boundary. The difference in grey level between average density of central ROI and the average of average density of four peripheral ROIs is recorded. However, IPEM indications do not suggest any accuracy test for morphometric measurements. For the accuracy test, the calibration phantom had to reproduce materials, structure and dimensions of the object of investigation, that is, trabecular bone specimens. With this aim, a previously published physical morphometric calibration phantom was used (Perilli et al., 2006). Moreover, a procedure for the evaluation of the accuracy of morphometric measurements was introduced.

In the presented QC protocol, in the acceptance/status test both phantoms were scanned five consecutive times to calculate the reference value (B.L.). The B.L. was used as the reference value for data collection charts. Each scan was performed on a different day for 5 consecutive days, and the phantom was repositioned in the microCT each time. The QC phantoms were scanned maintaining constant the scanner settings, 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 for each specific adopted chart, it is defined as out of control. When this happens, the QC protocol here proposed recommends an intervention, e.g. maintenance service.

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

MicroCT scanner settings and image processing

The QC protocol was applied on a Skyscan in vitro microCT model 1072 (Skyscan, Kontich, Belgium) composed by a micro-focus X-ray tube (spot size <5 μm) and a 1024 × 1024 12-bit charge-coupled device camera with maximum field of view of 25 mm. The scanning parameters were 50 kVp, 200 μA and rotation step 0.45°. A 1-mm aluminium filter was used to reduce the beam-hardening effect. Each frontal X-ray image was averaged over two frames, each single frame having 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 of 20 × 20 mm. The size of the pixel corresponds to the isotropic voxel size in the reconstructed slices.

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

A global thresholding procedure was chosen to convert the reconstructed cross sections from grey-level images (8-bit format, 256 grey levels) into binary images. A unique threshold value was used for all acquisitions. It was defined as the threshold value that minimizes the accuracy error in comparison with nominal thickness declared by manufacturer of the aluminium (Al)-inserts of the morphometric calibration phantom (Perilli et al., 2006).

Application of the in vitro microCT QC protocol

Acceptance/status test.  The acceptance/status test was performed on both the water phantom and the morphometric phantom in 2004, from which the B.L. was calculated for each parameter.

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

Noise test

The noise test was performed with the water phantom. The phantom consists of a cylindrical plastic vessel with an 18-mm outer diameter, 14-mm inner diameter and 44-mm height. A series of five spatially contiguous cross sections were reconstructed at 22-mm height from the bottom of the water phantom.

According to IPEM indications, the ROI area corresponds to 500 mm2 because a standard-size soft tissue-equivalent phantom is used for the application of QC protocols in medical CT. In in vitro microCT systems, the water phantom dimensions can be chosen according to the available field of view size. In this case, the ROI size was chosen to be 10% of the area of the inner water phantom area. This percentage was the same suggested in IPEM indications, but in that case it was calculated as ROI diameter (25 mm) and water phantom diameter (25 cm) ratio. The percentage, calculated as ratio of areas, represented a good balance between statistics and the ROI size for the application of protocol. The ROI was positioned in the centre of each of the five spatially contiguous reconstructed slices.

The QC parameter of noise test was calculated as the standard deviation of the grey levels averaged on five spatially contiguous cross sections (average SD). The B.L. was used to establish the quality control chart with an upper noise limit and lower noise limit as ±20% of B.L. respectively, according to IPEM indications (IPEM Report No. 91, 2005).

Image-Pro Plus v.4.5.1.22 (The Proven Solution, Media Cybernetics, Inc.) image analysis software was used to calculate the average SD in grey levels in the circular ROI for the noise test (Fig. 1) as described in field Procedure in Table 1.

Figure 1.

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

Uniformity test

For the uniformity test, the water phantom and the five spatially consecutive grey-level images of the noise test were used. A circular ROI, the same as used for the noise test, was positioned in the centre of the water phantom and four circular ones, with the same dimensions, were positioned in the peripheral location (Fig. 2).

Figure 2.

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

The QC parameter of the uniformity test was calculated as the difference in grey levels (GL) between average density (inline image) of the central circular ROI (ROIa) and the average of average density of all four circular peripheral ones (ROIb, ROIc, ROId, ROIe) (Eq. 1, Fig. 2). This difference was averaged on five spatially contiguous cross sections.

image(1)

where i is the circular ROI reference index and GL is the grey levels.

Image-Pro Plus v.4.5.1.22 (The Proven Solution, Media Cybernetics, Inc.) image analysis software was used to calculate the average difference in grey levels for the uniformity test (Fig. 2) as described in the field Procedure in Table 1.

Concerning the data collection method, two different approaches were considered to choose the most suitable monitoring data method for the proposed QC protocol. In one case, the upper uniformity limit and lower uniformity limit were represented by ±1.5% of B.L. respectively, which are the tolerance limits defined by IPEM guidelines (IPEM Report No. 91, 2005). In the second case, the upper control limit and the lower control limit were represented as ±3 × mean moving range divided by 1.128, which are the tolerance limits defined by the Shewhart control chart for single measures (Montgomery, 2006). The moving range is the absolute difference between each pair of consecutive measures. The 1.128 value is the reference value for measures with a sample size equal to two (Montgomery, 2006). The mean moving range was calculated by the first five acquisitions in 2004.

Accuracy test

IPEM guidelines do not contain any accuracy test for morphometric measurements. However, the research interest for in vitro microCT was actually in morphometric characterization of trabecular bone tissue. Thus, a dedicated protocol was developed and applied, based on a morphometric calibration phantom. The main characteristic of a phantom for accuracy test in morphometry is to reproduce the typical 3D structure of the analyzed specimen, e.g. the trabecular bone framework.

A previously published physical 3D phantom with calibrated Al-inserts was chosen (Perilli et al., 2006) for the application of the proposed QC protocol (Table 1). The phantom was designed in cylindrical shape (13-mm diameter × 23-mm height). The Al-inserts were of different geometries (foils, wires, meshes and spheres) and calibrated thickness to reproduce the typical thickness of trabecular bone structure (Perilli et al., 2006). The objects of the calibration phantom are four foils of 20 ± 3, 50 ± 7.5, 100 ± 10 and 250 ± 25 μm in thickness, four wires of 20 ± 2, 50 ± 5, 125 ± 12.5 and 250 ± 25 μm in thickness, a small horizontal and a vertical mesh composed of 100 ± 10 μm thick wires, a horizontal mesh of bigger external size and four spheres of 1000 ± 50 μm in diameter embedded in polymethylmethacrylate (Fig. 3). The tolerance in nominal thickness of the Al-inserts declared by the manufacturer (Goodfellow Cambridge Limited, Huntingdon, U.K.) was 5–15%, depending on type of inserts. However, the nominal thickness of the Al-inserts was verified by performing repeated measurements in the laboratory on samples of the same batch (Perilli et al., 2006). The foils and spheres were measured by using a screw–thread micrometer with digital display (1-μm resolution, Mytutoyo, Kawasaki, Kanagawa, Japan), whereas wires and mesh by using optical microscope (400× magnification, 0.346 μm pixel size; DC300, Leica Microsystems, Wetzlar, Germany). The Al material was used because its X-ray attenuation coefficient is very similar to the bone tissue [μAl(30 keV) = 3.04 cm−1; μBone(30 keV) = 2.56 cm−1] (Hubbell et al., 1996).

Figure 3.

Morphometric calibration phantom with materials, geometries and dimensions of aluminium inserts known (measures in mm).

A stack of 861 reconstructed slices was chosen to include all the Al-inserts. For the binarization of the cross-sectional images, a uniform thresholding procedure defined in previously published work was used (Perilli et al., 2006). The thinnest Al-inserts such as 20 μm, 50-μm-thick wires and 20-μm-thick foil, although visible in the grey level cross-sectional images, were not properly segmented. This might be related to beam-hardening artefacts and more likely to partial volume effects. The nominal thickness of the thinnest inserts correspond to approximately one to two times the pixel size (19.5 μm). Thus, making their segmentation difficult, specially for thin cylindrical objects such as wires. Some of the binarized cross sections of the 20-μm- and 50-μm-thick wires and the 20-μm-thick foils lacked in segmented structure, which makes the calculation of thickness of these objects challenging or meaningless. Thus, these three inserts were excluded from calculation (Table 2).

Table 2.  The different ROI sizes and shapes and specific volume of interest (VOI) used for thickness calculation of the segmented Al-inserts.
Al-insertsROI size (mm2) − pixelROI shapeVOI (ROI size × no. of slices)
Wire (20 μm)Not segmentedNot segmentedNot segmented
Foil (20 μm)Not segmentedNot segmentedNot segmented
Wire (50 μm)Not segmentedNot segmentedNot segmented
Foil (50 μm)5.4 × 1.6 − 275 × 80Rectangular275(80 × 281)
Mesh (small horizontal)5.8 × 5.8 − 300 × 300Square300(300 × 31)
Mesh (big horizontal)9.7 × 9.7 − 500 × 500Square500(500 × 26)
Mesh (small vertical)5.4 × 1.6 − 275 × 80Rectangular275(80 × 191)
Foil (100 μm)5.4 × 1.6 − 275 × 80Rectangular275(80 × 281)
Wire (125 μm)0.9 × 0.9 − 48 × 48Square48(48 × 576)
Foil (250 μm)5.4 × 1.6 − 275 × 80Rectangular275(80 × 281)
Wire (250 μm)0.9 × 0.9 − 48 × 48Square48(48 × 576)
Spheres5.8 × 5.8 − 300 × 300Square300(300 × 61)

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 program is based on the method described by Hildebrand & Ruegsegger (1997). In that method, the thickness (Th) is obtained by fitting maximal spheres to each point in the 3D structure (Hildebrand & Ruegsegger, 1997). ROIs of various shapes and sizes were chosen according to the different geometry and size of the Al-inserts to calculate the thickness (Table 2). For each single Al-insert a specific ROI was drawn, which had to include the whole single insert in each cross section (Fig. 4). The thickness was calculated separately for each Al-insert over the specific volume of interest (Table 2).

Figure 4.

Example of reconstructed slices of (a) spheres, (b) large horizontal mesh, (c) foils and (d) vertical mesh. The measures are compared to the nominal values of Al-inserts.

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

image(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 previously described. The average |ETh| was obtained by calculating the average value of the absolute error of the thickness of all the segmented Al-insert. The |ETh|* value was calculated as average of |ETh| values on the 20 months for each single segmented Al-insert.

At each measuring time point, each accuracy measurement required nine |ETh| measures, corresponding to the nine segmented Al-inserts (Table 2). Not segmented Al-inserts were excluded in the calculation of the error, as mentioned previously. For data collection, the use of the Shewhart control chart for single measures was not suitable, as the sample size (9) is greater than two. In this case, the Shewhart control chart (Montgomery, 2006) was more appropriate, and hence was used to collect the monthly average |ETh| values on the morphometric phantom. The standard deviation was used to establish the range of the tolerance limits. The upper control limit and the lower control limit of the chart were calculated as ±3 × average standard deviation (inline image) on the segmented Al-inserts. The intermediate tolerance limits, corresponding to ±2inline image and ±1inline image, were also reported.

Statistical analysis

All the measures of the QC protocol were processed by the statistical test C, with the null hypothesis that the collected data were randomly distributed. This test is very flexible for verifying the null hypothesis, however without considerations about the alternative hypothesis (Eq. 3) (Young, 1941).

image(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, with each group containing the measures of 5 months to reproduce the same group dimension of collecting data method during acceptance/status test. Apart from the reference group of the first five measures of year 2004 (group B.L.), the other four groups were group 1 composed by the five consecutive monthly measures starting from August 2007, group 2 composed by five consecutive monthly measures starting from January 2008, group 3 composed by five consecutive monthly measures starting from June 2008 and group 4 composed by five consecutive monthly measures starting from November 2008. A Kruskal–Wallis test was applied to evaluate if there were significant differences in the measures between all the monthly acquisition groups, including the B.L. group (total of five groups). If a measure was found significantly different, a Wilcoxon rank sum test with Bonferroni correction was applied to investigate which of the comparisons among first five acquisitions of year 2004 (group B.L.) and each group was significant (P < 0.0125). Wilcoxon rank sum test is a statistical test with a P= 0.05 as level of significance. The Bonferroni correction states if n independent hypothesis on a set of data are tested, each individual hypothesis will be tested with a statistical significant level of 1/n times. In this study, we have four independent hypotheses when each group, composed of five monthly measures, and group B.L. are compared. Thus, each hypothesis can be tested with a level of significance P/n= 0.05/4 = 0.0125.

Results

Noise test

The average value in grey levels of the water corresponds to 233.51 and it was calculated during acceptance/status test with the application of the noise test of the present QC protocol. Figure 5 shows the test chart that was used to monitor the average SD in grey levels by using the water phantom, with the tolerance range taken from IPEM quality indications. The chart shows no out of control points. However, the statistical test C shows that collected data are not randomly distributed (Table 3). The Kruskal–Wallis test found that there were significant differences between the five groups (Table 3). In particular, the Wilcoxon rank sum test with Bonferroni correction showed statistically significant differences between each group and the B.L. (Table 4). The average SD value of each group increased compared to the B.L. value.

Figure 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 was equal 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.

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 1Group 2Group 3Group 4Test C PKruskal–Wallis P
  1. 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 and the Kruskal–Wallis statistical test (P= 0.05, level of significance) are shown.

Water phantom
Noise test:
Average SD in grey levels3.35 ± 0.023.52 ± 0.053.56 ± 0.023.51 ± 0.053.52 ± 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 ± 0.090.820.31
Al-calibration phantom
Accuracy test:
Average |ETh| (μm)19.6 ± 5.320.1 ± 1.320.3 ± 1.819.8 ± 2.720.6 ± 0.40.910.78
Table 4.  The P-value of Wilcoxon rank sum test with Bonferroni correction.
 Noise test: Average SD in grey levels
  1. This test was performed to identify which group is different from baseline value for noise test. The P-value was deemed significant if P < 0.0125. In noise test, the Kruskal–Wallis test shows statistically significant differences among five groups (group B.L., group 1, group 2, group 3 and group 4).

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

Uniformity test

Figure 6 shows the Shewhart control chart for single measures used to monitor the difference in grey levels, with two different approaches for data collection. By following the tolerance limits given by IPEM guidelines (upper uniformity limit, lower uniformity limit =±1.5% of the B.L., dotted lines narrowed to the B.L), 18 out of control points were found. If, however, the tolerance limits of the Shewhart control chart for single measures were considered, no out of control points were found. The statistical test C showed that the collected data were randomly distributed (Table 3), and the Kruskal–Wallis test showed no statistically significant differences when comparing the five groups of measures (Table 3).

Figure 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, lower control 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.

Accuracy test

Figure 7 shows the application of Shewhart control chart to monitor the average |ETh| parameter of the segmented Al-inserts of the morphometric calibration phantom. The chart shows no out of control points. The statistical test C showed that the collected data were randomly distributed (Table 3), and the Kruskal–Wallis test showed no statistically significant differences between the five groups of measures (Table 3). Table 5 shows the B.L. values and the |ETh|* values of each segmented Al-inserts. |ETh|* values were calculated as average of |ETh| values of 20 months collecting data.

Figure 7.

Accuracy test chart: Shewhart control chart for the average |ETh| parameter calculated by using the Al-calibration phantom. The UCL and LCL were calculated as B.L. ± 3inline image, upper control limit (UCL) = B.L. + 3 × 5.3, lower control limit (LCL) = B.L. − 3 × 5.3. The baseline value was equal to 19.6 ± 5.3 μm.

Table 5.  The nominal value of each Al-insert, Thref, the deviation of the measured thickness of each insert calculated at baseline, |ETh|, and the deviation of each insert averaged over 20 month, |ETh|*.
Al-insertsThref (μm)Baseline (μm)|ETh|* (μm)
Wire (20 μm)20 ± 2Not segmentedNot segmented
Foil (20 μm)20 ± 3Not segmentedNot segmented
Wire (50 μm)50 ± 5Not segmentedNot segmented
Foil (50 μm)50 ± 7.51.9 ± 1.33.1 ± 2.4
Mesh (small horizontal)100 ± 103.4 ± 1.27.8 ± 2.3
Mesh (big horizontal)100 ± 1010.8 ± 1.07.3 ± 2.4
Mesh (small vertical)100 ± 1018.5 ± 0.519.7 ± 2.5
Foil (100 μm)100 ± 1027.8 ± 1.028.5 ± 1.9
Wire (125 μm)125 ± 12.519.3 ± 6.224.7 ± 9.2
Foil (250 μm)250 ± 2539.7 ± 0.840.4 ± 1.7
Wire (250 μm)250 ± 258.3 ± 0.87.7 ± 2.6
Spheres1000 ± 5046.7 ± 3.342.5 ± 8.0
AllAverage |ETh| = 19.6 ± 5.3Average |ETh|*= 20.2 ± 1.6

Discussion

In this study, a QC protocol for in vitro microCT was defined and applied. It was inspired by QC protocols commonly used in medical CT. The proposed QC protocol was applied in a particular medical research field, which is the morphometric characterization of trabecular bone specimens. For this purpose, suitable phantoms were used. In the accuracy test, a previously published phantom was used. This was designed to be reproducible by other laboratories, and thus, it was looked for standard calibrated aluminium objects that were commercially available, with given tolerances (Perilli et al., 2006). In a first instance, the percentage values of the tolerances appear rather large for calibration purposes. However, repeated optical measurements were done in a sample of the same batch (Perilli et al., 2006). These showed that the variation in thickness from the nominal values were 2–10 times smaller than the tolerance limits given by the manufacturer and also than the pixel size used during microCT scans in this paper (19.5 μm). Although the nominal tolerances could represent a caveat of this study, the optical measurements show that the aluminium inserts used for the phantom can be considered appropriate for the purpose of this study.

Quality control charts were used to collect and monitor the monthly measures of selected QC parameters. The procedure of noise and uniformity tests was taken from IPEM guidelines and adapted to microCT. The adaptation was related to the choice of ROI size because the ROI dimension suggested by IPEM (10% of water phantom diameter) was not considered suitable for the water phantom dimension in microCT, however the percentage suggested by IPEM (10%) was kept. The ROI area was calculated as 10% of the area of water phantom. It represented a good balance for statistics and ROI dimension for the application of protocol. A procedure for accuracy test by using morphometric measurements was introduced. In the noise test, the same collecting data method suggested by IPEM indications was used. In the uniformity test, a new collecting data method was proposed. It was based on new tolerance range defined by using a Shewhart control chart for single measures. The data and analysis over a 20 months application of the QC protocol were presented.

Noise test

Monthly noise measures were in accordance with tolerance limits defined by international guidelines for medical CT (±20% of B.L.). However, the collected data were not randomly distributed (Table 3), and were in the upper part of the tolerance range (Fig. 5). Moreover, a statistically significant difference between the five groups was found by using Kruskal–Wallis test (Table 3). In particular, statistically significant differences were found between B.L. and each group of five consecutive monthly measures (Table 4). The noise (average SD in grey levels) was found to increase after 3 years, compared to B.L. These results can be considered an indication of a changing condition. Although the values were not found out of the upper and lower noise limits of the chart, a maintenance service was required and performed to verify the cause of these results. The maintenance service reproduces the same tests performed during acceptance/status tests to verify the performances of the mechanical components of microCT, the X-ray tube and the charge-coupled device camera after 3 years’ working. During maintenance service, apart from the increased noise, no major problems were identified, suggesting a probable cause being related to the normal ageing of the X-ray source. A minor problem was identified, a misalignment between X-ray tube and charge-coupled device camera. It was resolved with a calibration procedure. However, further monthly measures will be performed according QC protocol and used to verify the performance of the micro-tomographic system.

Uniformity test

Two different approaches to collect the data were applied. The first approach was based on the tolerance range suggested by IPEM guidelines (±1.5% of B.L.). However, this was not suitable to monitor uniformity data stored in 8-bit format (256 grey levels) images. In fact, the tomographic images in medical CT, for which the IPEM guidelines are designed, are stored in a 12-bit format, which give a much higher grey-level range (4096 grey levels). The tolerance range of the difference in grey levels between the central ROI (233.52 GL) and peripheral ROIs (234.50 GL) was a limitation because of the small dispersion of the grey levels around B.L. (upper uniformity limit =−0.97, lower uniformity limit =−0.99; Fig. 6). The consequence was to observe 18 out of control points in the uniformity chart. But statistical C and Kruskal–Wallis tests (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 how the variability of the average SD (noise test) and difference in grey levels (uniformity test) influenced the morphometric measurements. For this purpose, the grey-level images for noise and uniformity tests were to be reconstructed in the same bit format of grey-level images for accuracy test (8 bit).

A second approach for data collection in uniformity measurements was introduced and applied, which was based on the Shewhart control chart for single measures. This chart showed no out of control points (Fig. 6).

Considering the outcomes of the statistical test C, the data were randomly distributed (Table 3), and there were no statistically significant differences between the five groups (Kruskal–Wallis test, P < 0.05; Table 3). This outcome is consistent with that of the Shewhart control chart for single measurements, which can be considered the most suitable approach to collect the data of uniformity test.

Accuracy test

The choice to combine Al-inserts with different sizes and shapes for the morphometric calibration phantom was related to the aim to reproduce the geometries and thickness of a typical 3D structure of trabecular bone specimens. Over these inserts, a single threshold value was used. This was chosen to minimize the average |ETh| calculated on all the segmented Al-inserts (at B.L., average |ETh| = 19.6 μm), and was the same threshold value as in a previous published study (Perilli et al., 2006). Then, a unique Shewhart control chart was used to monitor the average |ETh| for the QC parameters calculated on all the segmented Al-inserts of the morphometric calibration phantom over time. No out of control points were observed. On average, the deviations from the nominal values in thickness (average |ETh|*= 20.2 μm; Table 5) were about the pixel size used for the tomographic acquisition (19.5 μm). The accuracy error for the single segmented Al-insert was about one to two times the pixel size used in this study (19.5 μm) or much lower. In general, all the monthly measures were within one standard deviation (Fig. 7), randomly distributed (Table 3), with no significant differences between five groups of measurements (Table 3). Thus, the results of the morphometric accuracy test shows that the performance of the microCT system can be considered not out of control. A Shewhart control chart for the single Al-insert, although feasible for specific purposes, was not introduced here into the QC protocol, both to not complicate the monthly application of the protocol and because the monitoring of the accuracy error of a single insert might not be representative of the accuracy error calculated for the complex 3D structure to investigate, that is trabecular bone specimen, which contains a mixture of geometries. However, monitoring the accuracy of a single specific measurement is possible, and the presented protocol can be used and easily adapted to the specific purpose.

In conclusion, a QC protocol in in vitro microCT was proposed and applied successfully. The protocol, inspired by IPEM guidelines, was adapted to monitor the performance of in vitro microCT in a common application field, which is trabecular bone histomorphometry. The noise and uniformity tests were performed with a water phantom. For the accuracy test, a physical phantom for the calibration of morphometric measurements in 3D was used. Three years after measuring B.L. values, the application of the QC protocol showed that the quality level of the microCT scanner used was not out of control concerning accuracy and uniformity. The noise showed a slight but significant increase over the years, which nonetheless can be considered negligible, as it had no effect on the outcome in histomorphometry. However, the systematic increase in monthly measures compared to B.L. can be interpreted as a changing condition in the performance of the X-ray tube and charge-coupled device camera of the microCT system. A maintenance service to control the performance of mechanical components of microCT after 3 years’ working was required and performed. The importance to apply a periodic QC protocol is related both to observe and resolve punctual problems and to monitor the stability of micro-tomographic system over years to maintain a high-quality level of the measurements.

Acknowledgements

The authors thank Paolo Erani and Mauro Ansaloni for technical support, Luigi Lena for the pictures and Keith Smith for English language review. This work has been partially supported by the Region—University Research Program 2007–2009 and by the VPHOP Project project (FP-223865).

Ancillary

Advertisement