- Email: [email protected]
Reanalysis precision of 3D quantitative computed tomography (QCT) of the spine
Bone 44 (2009) 566–572
Contents lists available at ScienceDirect
Bone j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / b o n e
Reanalysis precision of 3D quantitative computed tomography (QCT) of the spine Klaus Engelke a,b,⁎, André Mastmeyer a, Valérie Bousson d, Thomas Fuerst c, Jean-Denis Laredo d,e, Willi A. Kalender a a
Institute of Medical Physics, University of Erlangen, Germany Synarc Inc, Hamburg, Germany Synarc Inc, San Francisco, CA, USA d Service de Radiologie Ostéo-Articulaire, Hôpital Lariboisière, Paris, France e Department of Radiology, Hospital of the University of Pennsylvania, Philadelphia, PA, USA b c
a r t i c l e
i n f o
Article history: Received 11 June 2008 Revised 21 October 2008 Accepted 6 November 2008 Available online 25 November 2008 Edited by: H. Genant Keywords: 3D QCT Spine Precision Bone mineral density Integral Cortical and trabecular bone
a b s t r a c t Purpose: To evaluate the precision of 3D QCT of the spine. Methods: Interoperator analysis reproducibility of two different 3D QCT analysis systems (QCTPro from Mindways Software Inc and MIAF-Spine from the Institute of Medical Physics, University of Erlangen) was evaluated in 29 postmenopausal women. For each analysis system four different trained operators analyzed all scans independently. Results of the vertebrae L1 and L2 were averaged. With QCTPro BMD of the central trabecular elliptical VOI was analyzed. With MIAF-Spine integral, trabecular and cortical BMD, BMC and volume were analyzed in the total vertebral body, the elliptical cylinder and the Osteo VOIs that were further subdivided into superior, mid and inferior subVOIs, each. Results: Precision errors (%CVrms) for the central trabecular VOI that is also used in the traditional single slice QCT techniques were 1.7± 2.2% and 0.6 ± 0.6% for QCTPro and MIAF-Spine, respectively. For MIAF-Spine integral BMD precision errors were lowest in the total and mid Osteo subVOIs (0.5 ± 0.5%). Trabecular BMD precision errors were lowest in the mid subVOIs (0.6 ± 0.6%). For trabecular BMD there were no differences among the total vertebral body, elliptical cylinder and Osteo VOIs. Cortical BMD precision errors were lowest in the mid total vertebral body subVOI (2.1 ± 1.9%) and slightly higher in the mid of the Osteo subVOI. Precision errors in the superior and inferior subVOIs were typically 50% to 100% higher compared to the mid subVOIs. Discussion: Compared to QCTPro MIAF-Spine uses an automated 3D segmentation and an anatomic vertebral coordinate system to position a variety of analysis VOIs. This results in better precision than the more manually assisted analysis used by QCTPro. In-vivo precision errors will be approximately 0.5% higher compared to the analysis precision errors reported here (13). The results demonstrate that with 3D QCT invivo precision errors of about 1%-1.5% for trabecular and 2.5% to 3% for cortical bone can be obtained in postmenopausal women. © 2008 Elsevier Inc. All rights reserved.
Introduction QCT of the lumbar spine is one of the standard procedures in bone densitometry used widely for diagnosis and fracture risk assessment as well as for monitoring changes in trabecular bone. Dual X-ray absorptiometry (DXA) is still the most widely used method to determine bone mineral density (BMD), however, QCT offers several advantages over DXA because it measures a physical density in g/cm3 instead of an areal density in g/cm2 by DXA. The projected area analyzed in DXA contains the cortical and trabecular bone of the vertebral body and considerable amounts of cortical bone from the spinal processes and articular facets. Also in particular in the posterior anterior projection degenerative changes often cannot be identified in ⁎ Corresponding author. Institute of Medical Physics, University of ErlangenNürnberg, Henkestr. 92, D-91052 Erlangen, Germany. Fax: +49 9131 8522824. E-mail address: [email protected] (K. Engelke). 8756-3282/$ – see front matter © 2008 Elsevier Inc. All rights reserved. doi:10.1016/j.bone.2008.11.008
DXA images and in elderly subjects falsely result in increased BMD values [9]. In contrast with QCT trabecular and cortical bone can be separately assessed and geometrical parameters related for example to bone strength such as moments of inertia can be measured. Also in the spine trabecular BMD can be investigated locally, for example in the mid, superior or anterior part of the vertebral body. These advanced possibilities recently have sparked renewed interest to integrate QCT in particular in clinical trials of osteoporosis as changes in areal BMD as measured by DXA can only partly explain the fracture reduction under pharmaceutical intervention [2]. Historically, 2D single 8 to 10 mm thick midvertebral slices were acquired and analyzed for three to four vertebrae between T12 and L4. In single slice QCT the main output of all analysis programs is a midvertebral trabecular volume using either an elliptical region of interest (ROI) or a Pacman-shaped ROI (also called the ‘Osteo’ ROI) in which bone mineral density (BMD) is determined [1,6,11,20]. Several commercial analysis packages exist that are either integrated in the CT
K. Engelke et al. / Bone 44 (2009) 566–572 Table 1 Published short term in-vivo precision values of spinal trabecular BMD as measured by QCT Author
Technique
Cohort
Short-term in-vivo precision CV [%]
Genant [6]
Single slice L1-L3? Single slice L1-L3? Single slice L1-L3 Single slice T12-L3 Single slice L1+L2 Single slice T13-L3 3D QCT L1+L2
n = ? premenopausal females n = 10 females?
1.6 1.6
n = 20
2.4
n = 9 females age = 21–73
1.4
Rosenthal [19] Louis [14] Steiger [20]
Gudmundsdottir [8] Lang [13]
1.5 n = 22 females age not specified n = 10 postmenopausal females
1.9
567
Table 2 CT scanner types used in the study. For all scans 100 mAs were used Scanner type
Number of scans
Voltage [kV]
Slice thickness [mm]
Reconstruction kernel
GE LightSpeed QX/i GE LightSpeed Ultra GE HiSpeed CT/i Philips Mx8000D Siemens Sensation 4 Siemens Emotion Duo
7
120
1.25
Standard
1
120
1.25
Standard
8
120
1.0
Standard
1
120
1.0
B
4
120
1.0
B40 s
5
110
1.0
B40 s
1.3
scanner software or that are implemented as stand-alone software on workstations to which the acquired CT scans are transferred and analyzed. The introduction of spiral multidetector CT (MDCT) technology provided the possibility to rapidly scan a larger volume than with the earlier sequential technique and consequently 3D QCT of the spine was developed [12,13,15]. QCT in general can be split into the scan acquisition and reconstruction process and the scan analysis process, in which densitometric and sometimes geometrical parameters are extracted from the CT images. Spiral CT simplifies the scan acquisition because complete vertebrae are scanned allowing various ROIs to be defined in the image analysis step. This represents an improvement over traditional QCT in which single slices are acquired in each vertebra, the location of which has to be carefully positioned relative to the vertebral body during the acquisition. Several approaches that will be discussed in detail in this paper have been developed for the analysis of the acquired 3D scans. They range from a largely manual placement of a midvertebral trabecular volume of interest (VOI) that in size and location is similar to the volume analyzed in the single slice techniques to a sophisticated 3D segmentation of the scanned vertebral bodies with an automatic placement of a variety of anatomically defined VOIs. In this paper we specifically assessed the precision of two different 3D QCT analysis techniques. One is the commercial software QCTPro (Mindways Software, Inc., Austin, TX), the other the MIAF-Spine software developed at the University of Erlangen (MIAF: Medical Image Analysis Framework). Low precision errors are important for monitoring BMD changes. In-vivo short term precision data for single slice QCT have been published in a few early papers listed in Table 1. In-vivo precision values for 3D QCT of the spine so far have only been published by Lang and coauthors [13].
for all scans (100 mAs). An example image is shown in Fig. 1. For all subjects L1 and L2 were scanned and a reconstruction field of view of 360 mm was used. The first system used for analysis was QCTPro™ (version 4.1.3) from Mindways Software Inc. (Austin, TX). QCTPro places an elliptical cylinder in the anterior trabecular portion of each vertebral body. As a first step the operator has to rotate each vertebra to optimally align it with a reference frame (Fig. 2) in order to facilitate the automatic placement of the analysis VOI. Each vertebra has to be aligned separately. In the aligned vertebrae QCTPro analyzes horizontal line profiles and determines the position of the cortical edges which are then used to calculate the size and position of the elliptical VOI (Fig. 3). By default its vertical extension (the Z-axis dimension) is set to 9 mm which is close to the 10 mm slice thickness most often used in single slice QCT. The operator can manually adjust the size and location of the automatically determined VOI if necessary. Finally average bone mineral density is calculated within the VOI. The second analysis system was MIAF-Spine, software developed at the University of Erlangen. It is based on an automatic 3D segmentation of the vertebral bodies. VOIs are automatically positioned relative to an anatomic vertebral coordinate system (VCS) that is aligned along the principal axes of the vertebral body. Some operator interaction is required for the identification of the vertebrae and their separation. Details have been published before [15]. In addition to the elliptical trabecular VOI used by QCTPro, with MIAFSpine a large variety of VOIs can be analyzed. The three main VOI types
Methods In this study we evaluated the analysis reproducibility of two different 3D QCT techniques. Baseline scans of 30 subjects (mean age: 62.9 ± 5.7 years) were randomly selected from a multi-center study of 150 postmenopausal (N5 years) women with lumbar spine BMD T-score between −2 and −3.5 and total femur T-score of larger than −3.5 as measured by DXA. Exclusion criteria were any bone disease other than osteopenia, osteoporosis, Paget's bone disease, osteomalacia, or osteogenesis imperfecta. All scans were analyzed by four different operators for each analysis system. The scans were acquired with six different CT scanners. As can be seen from Table 2 acquisition and reconstruction parameters depended slightly on the scanner type but were selected to be as equivalent as possible; for example, a standard body reconstruction kernel was used for all scanners. With one exception the voltage was set to 120 kV and the mAs was constant
Fig. 1. Example CT image: One slice from a stack of images is shown. The phantom below the patient is used to derive BMD values from Hounsfield units displayed the CT image.
568
K. Engelke et al. / Bone 44 (2009) 566–572
Fig. 2. QCTPro alignment of vertebral bodies. (Top) before, (bottom) after manual rotation applied to L1. The procedure has to be repeated for L2.
are the vertebral body, the elliptical cylinder, and the Osteo VOI (Fig. 4). Integral, trabecular and cortical compartments can be separated as well as superior, mid and inferior portions of the vertebral body. Due to the limited spatial resolution of the clinical CT scanners of about 1 mm at the settings we used (360 mm FoV resulting in an inplane pixel size of 0.7 mm, slice thickness 1–1.25 mm according to Table 2), the cortical VOI includes some subcortical bone. In this study after the separation of the trabecular and cortical VOIs, the trabecular VOI was peeled by two additional millimeters to ensure that no subcortical bone is included (Fig. 5). Four different operators analyzed all 30 datasets independently after having received appropriate training. Due to logistical reasons two different groups of operators performed the QCTPro and the MIAF-Spine analysis. Results of L1 and L2 were averaged and the coefficient of variation (% CV) among the four operators was calculated for each patient. Then the root mean square coefficient of variation (CVrms) was calculated for all patients. Analysis differences among
operators were analyzed by one way ANOVA using absolute and percentage BMD differences from the patient average BMD values of the four operators. One way ANOVA was also used to identify scanner specific differences in precision errors. Paired t-tests were used to investigate precision differences between VOIs. Results Twenty nine out of the 30 subjects were included in the analysis. Mean BMD of L1-L4 as measured by DXA was 0.759 g/cm2, and for L1 and L2 (which were measured by QCT) was 0.705 g/cm2. One scan was excluded because the vertebrae were not fully included in the reconstructed field of view and therefore the 3D segmentation used by MIAF-Spine did not work properly. Table 3 lists the average BMD values derived with MIAF-Spine for the subjects included in the analysis for the various analysis VOIs. It is obvious that the cortical VOIs (range of average BMD from 223 to 327 mg/cm3) include subcortical bone as pure cortical BMD is in the range of 1000 mg/cm3.
Fig. 3. Elliptical analysis VOIs used by QCTPro. (Left) VOIs automatically determined; (right) after operator adjustment; in particular the computer determined position of the VOI in L1 was not in the center of the vertebra (see sagittal view of L1 in left image).
K. Engelke et al. / Bone 44 (2009) 566–572
569
Fig. 4. MIAF-Spine: trabecular VOIs used in this study.
For QCTPro the corresponding average BMD value of the elliptical VOI was 72.8 ± 19.4 mg/cm3, which is almost identical to the MIAF-Spine result for the mid trabecular compartment of the elliptical cylinder
VOI. Neither MIAF-Spine nor QCTPro showed any significant interoperator differences in BMD, bone mineral content (BMC), or VOI volume. Analysis precision for MIAF-Spine is shown as SDrms for BMD in Table 4 and as CVrms in Table 5 through Table 7 for BMD, BMC and volume, respectively. Absolute precision errors are in the order of 1 to 2 mg/cm3 for integral and trabecular BMD and 5 to 10 mg/cm3 for cortical BMD with the exception of the total and inferior cortical VOIs of the elliptical cylinder. As shown in Table 6 overall higher precision errors were observed for BMC than for BMD although the relationships of errors among VOIs Table 3 BMD ± SD in mg/cm3 averaged over all patients using MIAF-spine VOI shape
VOI
Integral BMD
Trabecular BMD
Cortical BMD
Vertebral body
Total Superior Mid Inferior Total Superior Mid Inferior Total Superior Mid Inferior
152 ± 22 164 ± 23 120 ± 23 173 ± 28 94 ± 20 110 ± 21 NA 113 ± 22 140 ± 22 152 ± 23 108 ± 23 162 ± 27
81 ± 21 80 ± 20 74 ± 23 93 ± 23 70 ± 20 80 ± 20 68 ± 22 70 ± 21 77 ± 21 77 ± 20 70 ± 23 88 ± 23
296 ± 28 293 ± 28 327 ± 28 285 ± 32 231 ± 35 240 ± 34 NA 223 ± 42 283 ± 30 279 ± 29 315 ± 35 275 ± 33
Elliptical cylinder
Osteo
Fig. 5. MIAF-Spine: cortical and trabecular compartments of the total vertebral body VOI. The trabecular VOI is derived by peeling the endosteal cortical bone by 2 mm. The integral VOI includes the cortical and the trabecular VOIs as well as the 2 mm peeled subcortical VOI that is neither included in the cortical or the trabecular VOIs.
The trabecular VOIs are depicted in Fig. 4. The VOIs for the measurement of integral, trabecular and cortical are shown for the total vertebral body VOI in Fig. 5 and are defined for the other VOIs accordingly. Note. for the mid VOI of the elliptical cylinder there are no integral and cortical values.
570
K. Engelke et al. / Bone 44 (2009) 566–572
Table 4 Analysis precision (SDrms in mg/cm3) of MIAF-Spine for BMD
Table 6 Analysis precision (CVrms in %) of MIAF-Spine for BMC
VOI shape
VOI
Integral BMD
Trabecular BMD
Cortical BMD
VOI shape
VOI
Integral BMC
Trabecular BMC
Cortical BMC
Vertebral body
Total Superior Mid Inferior Total Superior Mid Inferior Total Superior Mid Inferior
1.0 ± 1.1 1.9 ± 2.3 1.0 ± 1.1 1.0 ± 1.3 0.5 ± 0.6 1.5 ± 2.1 NA 0.8 ± 0.8 0.7 ± 0.7 1.3 ± 1.2 0.5 ± 0.5 1.0 ± 0.8
0.7 ± 0.9 1.1 ± 1.5 0.4 ± 0.4 2.1 ± 2.6 0.8 ± 0.7 1.5 ± 2.7 0.4 ± 0.4 2.8 ± 2.8 0.8 ± 1.0 1.3 ± 1.7 0.4 ± 0.5 2.2 ± 2.8
5.3 ± 5.1 6.3 ± 5.9 6.9 ± 6.2 7.6 ± 7.6 13.8 ± 13.3 9.1 ± 11.8 NA 23.5 ± 23.0 5.6 ± 5.4 5.9 ± 7.0 8.1 ± 7.0 8.5 ± 8.6
Vertebral body
Total Superior Mid Inferior Total Superior Mid Inferior Total Superior Mid Inferior
1.5 ± 1.7 2.3 ± 2.5 1.2 ± 1.2 2.0 ± 2.4 1.6 ± 1.6 2.4 ± 2.8 NA 2.1 ± 2.1 1.8 ± 1.9 2.3 ± 2.2 1.3 ± 1.5 2.2 ± 2.4
1.6 ± 1.7 2.4 ± 2.6 0.8 ± 0.7 3.5 ± 6.2 2.2 ± 2.2 3.8 ± 4.8d 1.0 ± 1.1d 5.6 ± 6.2c 2.4 ± 3.2d 3.0 ± 3.3 1.7 ± 1.8d,b 4.1 ± 7.1d,c
3.0 ± 3.0 3.5 ± 4.0 3.2 ± 3.0 3.9 ± 3.8 5.2 ± 5.7d 3.6 ± 4.9 NA 9.2 ± 9.7d 3.4 ± 3.3d,b 3.1 ± 3.3 3.8 ± 3.5c 4.3 ± 4.4b
Elliptical cylinder
Osteo
The trabecular VOIs are depicted in Fig. 4. The VOIs for the measurement of integral, trabecular and cortical are shown for the total vertebral body VOI in Fig. 5 and are defined for the other VOIs accordingly. Note: for the mid VOI of the elliptical cylinder there are no integral and cortical values.
were similar compared to those obtained for BMD. For QCTPro the BMD precision error of the central elliptical cylinder, which is comparable to the mid elliptical cylinder VOI of the MIAF-Spine analysis, was 1.7 ± 2.2%. Cortical thickness was only determined by MIAF-Spine. Analysis precision (CVrms) of cortical thickness of the total vertebral body was 4.7 ± 5.2%. For the mid section of the vertebral body it was 5.6 ± 5.2%. Again there were no significant differences among operators. Neither for MIAF-Spine nor for QCTPro there were any scanner specific differences in precision errors for any of the variables analyzed. Least significant changes (LSC) for the mid trabecular VOIs were 1.1 mg/cm3, and 1.5 to 3 mg/cm3 for the other trabecular and integral VOIs. For cortical BMD the LSC varied from 13 to 25 mg/cm3 with even higher LSC for the VOIs of the elliptical cylinder (11.8 to 30.7 mg/cm3). Discussion In this study we determined inter-operator analysis precision errors for 3D QCT of the spine in 29 postmenopausal women using a variety of commercial whole-body CT scanners. Compared to other studies this is a larger investigation reflecting the clinical settings in which QCT is typically applied. Also in addition to the universal midvertebral trabecular VOI, precision was determined for a large variety of additional VOIs that can be analyzed with 3D QCT. As can be seen from Table 3 the study subjects had low trabecular BMD values ranging from about 70 to 90 mg/cm3 characteristic of an osteoporotic population where most clinical QCT measurements are performed.
Table 5 Analysis precision (CVrms in %) of MIAF-Spine for BMD VOI shape
VOI
Integral BMD
Trabecular BMD
Cortical BMD
Vertebral body
Total Superior Mid Inferior Total Superior Mid Inferior Total Superior Mid Inferior
0.7 ± 0.8 1.2 ± 1.4 0.9 ± 1.0 0.6 ± 1.0 0.6 ± 0.8 1.3 ± 1.6 NA 0.7 ± 0.8 0.5 ± 0.5c 0.9 ± 0.9a 0.5 ± 0.5d 0.6 ± 0.6
1.0 ± 1.3 1.5 ± 1.9 0.6 ± 0.6 2.4 ± 3.4 1.3 ± 1.7 2.3 ± 3.3c 0.6 ± 0.7 4.4 ± 5.3c 1.2 ± 1.5d 1.9 ± 2.5c 0.6 ± 0.7 2.7 ± 3.9c, a
1.8 ± 1.8 2.2 ± 2.1 2.1 ± 1.9 2.7 ± 2.7 6.0 ± 5.6d 4.2 ± 5.9c NA 10.9 ± 10.5d 2.0 ± 2.0b 2.2 ± 2.7a 2.6 ± 2.2c 3.1 ± 3.1d,b
Elliptical cylinder
Osteo
a
Significant differences between osteo and elliptical cylinder VOIs (p b 0.05). Significant differences between osteo and elliptical cylinder VOIs (p b 0.01). Significant differences between osteo/elliptical cylinder VOIs and vertebral body VOI (p b 0.05). d Significant differences between osteo/elliptical cylinder VOIs and vertebral body VOI (p b 0.01). b c
Elliptical cylinder
Osteo
a
Significant differences between osteo and elliptical cylinder VOIs (p b 0.05). Significant differences between osteo and elliptical cylinder VOIs (p b 0.01). Significant differences between osteo/elliptical cylinder VOIs and vertebral body VOI (p b 0.05). d Significant differences between osteo/elliptical cylinder VOIs and vertebral body VOI (p b 0.01). b c
Using MIAF-Spine the RMS-precision error in the classical midvertebral trabecular VOI was 0.4 mg/cm3 or 0.6% independent of whether the elliptical cylinder VOI, the Osteo VOI or the vertebral body VOI was used. Using QCTPro, a precision error of 1.7% was determined. Compared to the precision values of the classical single slice based QCT listed in Table 1 it should be pointed out that here we report CVrms compared to the ‘non-RMS’ CV values shown in the table. RMS-CV errors are typically higher by 25% to 40% compared with average CVs [7]. For example, corresponding ‘non-RMS’ CVs in our analysis were 0.5% for MIAF-Spine and 1.4% for QCTPro, respectively. Also BMD values in most of the studies listed in Table 1 were most likely higher than in our study as younger subjects were investigated. However, for a given absolute error measured as the standard deviation the corresponding %CVs are larger if absolute BMD values decrease. The standard deviation in our study was 0.34 mg/cm3 for MIAF-Spine and 0.8 mg/cm3 for QCTPro corresponding to 0.5 to 1 Hounsfield Units, which are excellent results. On the other hand the values in Table 1 represent true in-vivo precision measurements, in which subjects were measured twice with repositioning, whereas in this study only the inter-operator analysis precision was determined. Obviously in 3D QCT repositioning has a lower impact on precision errors than in the traditional 2D slice based technique where the location of the slice relative to the vertebral body is determined at the time of acquisition using the scout view taken before the actual CT scan. In 3D QCT the complete volume of the vertebra is scanned and the selection of analysis VOIs is no longer influenced by the scout view. In an earlier publication by Lang and coworkers [13] of 3D QCT of the spine, for the midvertebral elliptical cylinder VOI the difference between in-vivo (1.3% CVrms) and interoperator analysis precision (0.7%) was 0.6% derived in repeated scans of ten subjects with a mean trabecular BMD of 88 ± 20 mg/cm3 compared to our 68 ± 22 mg/cm3. Thus to assess true measurement imprecision, approximately half a percent has to be added to the values we derived in our study, although we speculate that for MIAFSpine the difference between in-vivo and inter-observer analysis precision will be smaller than the 0.6% reported by Lang [13]. The reason is the use of the vertebral coordinate system that only depends on the segmentation, which for MIAF-spine is highly automated. In comparison, in QCTPro the alignment of the vertebral body has to be performed manually. The software used in Lang's study [13] is slightly more automated than QCTPro but the user still had to manually determine the main vertebral axes on axial, sagittal, and coronal reformations of the CT dataset. The comparison of precision errors in different compartments of the vertebral body must be restricted to the MIAF-Spine results as these values do not exist for QCTPro. As can be seen from Table 5,
K. Engelke et al. / Bone 44 (2009) 566–572
precision errors for trabecular BMD were lowest in the mid VOIs independent of VOI shape. Precision errors in the total superior and inferior VOIs were usually at least twice as high. The same result was found for trabecular volume although to a lesser degree as precision errors for the total VOIs differed only by 20% to 70% from those of the mid VOIs. In QCT volume and BMD are the independent measurements, BMC is a derived quantity. The very low precision error of the mid VOI indicates that this VOI can be very reproducibly relocated and that there is a smaller BMD variation than in the other regions of the vertebral body. Even the total VOIs, despite being significantly larger showed higher precision errors for trabecular bone. It is interesting to note that in the inferior VOIs precision errors were by a factor of at least four higher compared to the mid VOIs and by a factor of at least two compared to the total VOIs although precision errors for trabecular volume were higher in the superior compared to the inferior VOIs indicating a high variation of trabecular BMD in the inferior VOIs. With respect to VOI shape, precision errors for trabecular bone were lower (for the mid VOIs: comparable) in the vertebral body than in elliptical cylinder or Osteo VOIs which may be explained by the larger size of the VOI. With a few exceptions the observations discussed for the trabecular VOIs are also true for the integral VOIs. The most obvious exception is that the precision errors in the inferior VOIs for integral bone were comparable to the other integral VOIs. Also there were no significant differences for integral BMD between the vertebral body and the elliptical cylinder VOIs, whereas some differences in precision errors between the elliptical cylinder and the Osteo VOIs remained. The precision errors for integral BMD of the total Osteo or the total vertebral body VOI (Table 5) are equivalent to those reported for DXA of comparable populations [4,10,17] even if a 0.5% error due to repositioning is added to the QCT values. The determination of a cortical VOI in the spine is ambiguous as the spatial resolution of clinical CT scanners is not adequate to accurately image the thin cortex [18]. Thus the cortical VOI shown in Fig. 5 also includes subcortical bone. This is similar to the classical single slice based analysis [11] but in contrast the results shown here are based on a 3D segmentation. Precision errors for cortical BMD, BMC and volume are also listed in Tables 5 to 7. The errors are lowest in the vertebral body VOIs which is understandable as here the analyzed cortical volume is highest. Given the limitations of the imaging procedure a precision error of about 2%-2.5% for cortical BMD is a surprisingly good result, in particular as the 2 mm peeled subcortical bone (Fig. 5) was not included in the cortical compartment. Similar to the trabecular VOIs, precision errors were significantly higher in the inferior VOIs for cortical BMD but not for cortical volume. The cortical contributions of Table 7 Analysis precision (CVrms in %) of MIAF-spine for volume VOI shape
VOI
Integral vol
Trabecular vol
Cortical vol
Vertebral body
Total Superior Mid Inferior Total Superior Mid Inferior Total Superior Mid Inferior
1.2 ± 1.4 1.8 ± 1.7 0.5 ± 0.5 1.9 ± 2.3 1.3 ± 1.4 1.9 ± 2.0 NA 1.9 ± 1.8 1.6 ± 1.8 2.1 ± 2.2d 1.1 ± 1.1d 2.0 ± 2.5
0.7 ± 1.0 1.7 ± 1.7 0.4 ± 0.4 1.5 ± 2.3 1.2 ± 1.3c 2.4 ± 2.3c 0.9 ± 1.1 1.9 ± 1.7 1.4 ± 1.6d 2.2 ± 2.1 1.2 ± 1.3d,b 1.7 ± 2.6d
4.1 ± 4.3 4.3 ± 4.7 3.9 ± 3.5 4.7 ± 4.8 5.0 ± 6.4 4.2 ± 5.8 NA 6.7 ± 8.1c 4.5 ± 5.0d 4.6 ± 5.0 5.1 ± 4.6d 4.9 ± 5.6c
Elliptical cylinder
Osteo
a
Significant differences between osteo and elliptical cylinder VOIs (p b 0.05). Significant differences between osteo and elliptical cylinder VOIs (p b 0.01). Significant differences between osteo/elliptical cylinder VOIs and vertebral body VOI (p b 0.05). d Significant differences between osteo/elliptical cylinder VOIs and vertebral body VOI (p b 0.01). b c
571
Fig. 6. Dependency of precision errors of the total vertebral body integral BMD on the reconstruction field of view. Data are from a different study of premenopausal women by Mastmeyer et al. [15]. As outlined in the Methods section above the study on postmenopausal women reported in this contribution used a field of view (FOV) of 360 mm.
the elliptical cylinder VOI are only the elliptical parts of the endplates cut out by the cylinder thus this VOI should not be used for cortical measurements. Some authors advocate the use of BMC or of the product of BMD and cortical thickness for cortical measurements [3,16] arguing that the partial volume artifacts may be partly compensated. However, BMC precision errors in the vertebral body and Osteo VOIs were approximately 50% higher than those for BMD. Analysis precision errors for cortical thickness were only analyzed in the total and mid vertebral body VOIs and were 4.7% and 5.6%, respectively, which is an acceptable result. The clinical usefulness of QCT has recently been summarized in the official positions of the International Society of Clinical Densitometry (ICSD) [5]. These positions are still primarily based on trabecular BMD of the mid section of the elliptical cylinder VOI, although with 3D QCT now a large variety of additional VOIs is available and regional differences in BMD can be determined across subjects or monitored longitudinally. Our analysis has shown that the lowest precision errors are achieved in the classical mid vertebral VOI. However, precision errors are reduced by about 50% compared to the classical technique if a true 3D segmentation in combination with an anatomic coordinate system is used. In comparison, a 3D data acquisition without real 3D image processing and a largely manually driven analysis is less advantageous. It must be cautioned that the comparison of precision errors obtained in this study with the values in Table 1 is limited by differences in imaging protocols and study populations used in the cited studies but a more detailed analysis is beyond the scope of this paper. For parameters other than the mid trabecular VOI, the use of the vertebral body or the Osteo VOIs are advantageous compared to the elliptical cylinder VOI. From the perspective of precision errors, the Osteo VOI should be used for the analysis of integral BMD and the vertebral body VOI for the measurement of cortical BMD. In both cases the total and mid VOIs gave better results than inferior and superior VOIs. For cortical measurements the elliptical cylinder VOIs are inadequate. As we have shown previously for the MIAF-Spine software, precision can be further improved by reducing the reconstruction field of view of the CT images, which increases the spatial resolution and will improve the 3D segmentation and the accuracy of the vertebral coordinate system [15]. As an example Fig. 6 shows the field of view dependency of the %CVrms of the integral total vertebral body VOI. The data were generated from a population of premenopausal women applying the same scan protocol used here but with a lower tube current of 60 mAs. Thus the absolute %CVs differ slightly from our study but the impact of the reconstruction field of view on precision errors of BMD and volume will be similar. In summary, in 3D QCT of the spine precision errors equivalent to those reported for DXA can be achieved if advanced analysis techniques are applied to the QCT scans exploiting their volumetric nature. Without a 3D segmentation and the use of anatomic
572
K. Engelke et al. / Bone 44 (2009) 566–572
coordinate systems precision errors are higher by about a factor of two. Lowest precision errors were obtained for the midvertebral elliptical VOI that is equivalent to the VOI used in classical single slice QCT. For the investigation of trabecular BMD in the inferior and superior parts of the vertebrae either the Osteo or the vertebral body VOIs should be used, for cortical measurements the vertebral body VOI is suited best. Acknowledgments This work was partly supported by the European Research Grant EU QCK6-CT-2002-02440. Preliminary results of this study have been presented at the 28th Annual Meeting of the American Society for Bone and Mineral Research, Philadelphia, (2006) and at the 17th International Bone Densitometry Workshop, Kyoto, Japan (2006). Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.bone.2008.11.008.
References [1] Cann CE, Genant HK, Kolb FO, Ettinger B. Quantitative computed tomography for prediction of vertebral fracture risk. Bone 1985;6:1–7. [2] Delmas PD, Li Z, Cooper C. Relationship between changes in bone mineral density and fracture risk reduction with antiresorptive drugs: some issues with metaanalyses. J Bone Miner Res 2004;19:330–7. [3] Dougherty G, Newman D. Measurement of thickness and density of thin structures by computed tomography: a simulation study. Med Phys 1999;26:1341–8. [4] El Maghraoui A, Do Santos Zounon AA, Jroundi I, Nouijai A, Ghazi M, Achemlal L, et al. Reproducibility of bone mineral density measurements using dual X-ray absorptiometry in daily clinical practice. Osteoporos Int 2005;16:1742–8. [5] Engelke K, Adams JE, Armbrecht G, Augat P, Bogado CE, Bouxsein ML, et al. Clinical use of quantitative computed tomography and peripheral quantitative computed tomography in the management of osteoporosis in adults: the 2007 ISCD official positions. J Clin Densitom 2008;11:123–62.
[6] Genant HK, Cann CE, Ettinger B, Gordan GS. Quantitative computed tomography of vertebral spongiosa: a sensitive method for detecting early bone loss after oophorectomy. Ann Intern Med 1982;97:699–705. [7] Glüer CC, Blake G, Lu Y, Blunt BA, Jergas M, Genant HK. Accurate assessment of precision errors: how to measure the reproducibility of bone densitometry techniques. Osteoporos Int 1995;5:262–70. [8] Gudmundsdottir H, Jonsdottir B, Kristinsson S, Johannesson A, Goodenough D, Sigurdsson G. Vertebral bone density in Icelandic women using quantitative computed tomography without an external reference phantom. Osteoporos Int 1993;3:84–9. [9] Guglielmi G, Floriani I, Torri V, Li J, van Kuijk C, Genant HK, et al. Effect of spinal degenerative changes on volumetric bone mineral density of the central skeleton as measured by quantitative computed tomography. Acta Radiol 2005;46:269–75. [10] Henzell S, Dhaliwal S, Pontifex R, Gill F, Price R, Retallack R, et al. Precision error of fan-beam dual X-ray absorptiometry scans at the spine, hip, and forearm. J Clin Densitom 2000;3:359–64. [11] Kalender WA, Klotz E, Süss C. Vertebral bone mineral analysis: an integrated approach with CT. Radiology 1987;164:419–23. [12] Lang TF, Guglielmi G, van Kuijk C, De Serio A, Cammisa M, Genant HK. Measurement of bone mineral density at the spine and proximal femur by volumetric quantitative computed tomography and dual-energy x-ray absorptiometry in elderly women with and without vertebral fractures. Bone 2002;30: 247–50. [13] Lang TF, Li J, Harris ST, Genant HK. Assessment of vertebral bone mineral density using volumetric quantitative CT. J Comput Assist Tomogr 1999;23:130–7. [14] Louis O, Luypaert R, Kalender W, Osteaux M. Reproducibility of CT bone densitometry: operator versus automated ROI definition. Eur J Radiol 1988;8:82–4. [15] Mastmeyer A, Engelke K, Fuchs C, Kalender WA. A hierarchical 3D segmentation method and the definition of vertebral body coordinate systems for QCT of the lumbar spine. Med Image Anal 2006;10:560–77. [16] Newman DL, Dougherty G, al Obaid A, al Hajrasy H. Limitations of clinical CT in assessing cortical thickness and density. Phys Med Biol 1998;43:619–26. [17] Phillipov G, Seaborn CJ, Phillips PJ. Reproducibility of DXA: potential impact on serial measurements and misclassification of osteoporosis. Osteoporos Int 2001; 12:49–54. [18] S. Prevhral, K. Engelke, and W.A. Kalender The influence of object size and CT imaging parameters on the measurement accuracy of cortical width and density. 13th International Bone Densitometry Workshop 8, pp. 519. Lake Lawn, WI: Osteoporos Int. [19] Rosenthal DI, Ganott MA, Wyshak G, Slovik DM, Doppelt SH, Neer RM. Quantitative computed tomography for spinal density measurement; factors affecting precision. Invest Radiol 1985;20:306–10. [20] Steiger P, Block JE, Steiger S, Heuck A, Friedlander A, Ettinger B, et al. Spinal bone mineral density by quantitative computed tomography: effect of region of interest, vertebral level, and technique. Radiology 1990;175:537–43.
Contents lists available at ScienceDirect
Bone j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / b o n e
Reanalysis precision of 3D quantitative computed tomography (QCT) of the spine Klaus Engelke a,b,⁎, André Mastmeyer a, Valérie Bousson d, Thomas Fuerst c, Jean-Denis Laredo d,e, Willi A. Kalender a a
Institute of Medical Physics, University of Erlangen, Germany Synarc Inc, Hamburg, Germany Synarc Inc, San Francisco, CA, USA d Service de Radiologie Ostéo-Articulaire, Hôpital Lariboisière, Paris, France e Department of Radiology, Hospital of the University of Pennsylvania, Philadelphia, PA, USA b c
a r t i c l e
i n f o
Article history: Received 11 June 2008 Revised 21 October 2008 Accepted 6 November 2008 Available online 25 November 2008 Edited by: H. Genant Keywords: 3D QCT Spine Precision Bone mineral density Integral Cortical and trabecular bone
a b s t r a c t Purpose: To evaluate the precision of 3D QCT of the spine. Methods: Interoperator analysis reproducibility of two different 3D QCT analysis systems (QCTPro from Mindways Software Inc and MIAF-Spine from the Institute of Medical Physics, University of Erlangen) was evaluated in 29 postmenopausal women. For each analysis system four different trained operators analyzed all scans independently. Results of the vertebrae L1 and L2 were averaged. With QCTPro BMD of the central trabecular elliptical VOI was analyzed. With MIAF-Spine integral, trabecular and cortical BMD, BMC and volume were analyzed in the total vertebral body, the elliptical cylinder and the Osteo VOIs that were further subdivided into superior, mid and inferior subVOIs, each. Results: Precision errors (%CVrms) for the central trabecular VOI that is also used in the traditional single slice QCT techniques were 1.7± 2.2% and 0.6 ± 0.6% for QCTPro and MIAF-Spine, respectively. For MIAF-Spine integral BMD precision errors were lowest in the total and mid Osteo subVOIs (0.5 ± 0.5%). Trabecular BMD precision errors were lowest in the mid subVOIs (0.6 ± 0.6%). For trabecular BMD there were no differences among the total vertebral body, elliptical cylinder and Osteo VOIs. Cortical BMD precision errors were lowest in the mid total vertebral body subVOI (2.1 ± 1.9%) and slightly higher in the mid of the Osteo subVOI. Precision errors in the superior and inferior subVOIs were typically 50% to 100% higher compared to the mid subVOIs. Discussion: Compared to QCTPro MIAF-Spine uses an automated 3D segmentation and an anatomic vertebral coordinate system to position a variety of analysis VOIs. This results in better precision than the more manually assisted analysis used by QCTPro. In-vivo precision errors will be approximately 0.5% higher compared to the analysis precision errors reported here (13). The results demonstrate that with 3D QCT invivo precision errors of about 1%-1.5% for trabecular and 2.5% to 3% for cortical bone can be obtained in postmenopausal women. © 2008 Elsevier Inc. All rights reserved.
Introduction QCT of the lumbar spine is one of the standard procedures in bone densitometry used widely for diagnosis and fracture risk assessment as well as for monitoring changes in trabecular bone. Dual X-ray absorptiometry (DXA) is still the most widely used method to determine bone mineral density (BMD), however, QCT offers several advantages over DXA because it measures a physical density in g/cm3 instead of an areal density in g/cm2 by DXA. The projected area analyzed in DXA contains the cortical and trabecular bone of the vertebral body and considerable amounts of cortical bone from the spinal processes and articular facets. Also in particular in the posterior anterior projection degenerative changes often cannot be identified in ⁎ Corresponding author. Institute of Medical Physics, University of ErlangenNürnberg, Henkestr. 92, D-91052 Erlangen, Germany. Fax: +49 9131 8522824. E-mail address: [email protected] (K. Engelke). 8756-3282/$ – see front matter © 2008 Elsevier Inc. All rights reserved. doi:10.1016/j.bone.2008.11.008
DXA images and in elderly subjects falsely result in increased BMD values [9]. In contrast with QCT trabecular and cortical bone can be separately assessed and geometrical parameters related for example to bone strength such as moments of inertia can be measured. Also in the spine trabecular BMD can be investigated locally, for example in the mid, superior or anterior part of the vertebral body. These advanced possibilities recently have sparked renewed interest to integrate QCT in particular in clinical trials of osteoporosis as changes in areal BMD as measured by DXA can only partly explain the fracture reduction under pharmaceutical intervention [2]. Historically, 2D single 8 to 10 mm thick midvertebral slices were acquired and analyzed for three to four vertebrae between T12 and L4. In single slice QCT the main output of all analysis programs is a midvertebral trabecular volume using either an elliptical region of interest (ROI) or a Pacman-shaped ROI (also called the ‘Osteo’ ROI) in which bone mineral density (BMD) is determined [1,6,11,20]. Several commercial analysis packages exist that are either integrated in the CT
K. Engelke et al. / Bone 44 (2009) 566–572 Table 1 Published short term in-vivo precision values of spinal trabecular BMD as measured by QCT Author
Technique
Cohort
Short-term in-vivo precision CV [%]
Genant [6]
Single slice L1-L3? Single slice L1-L3? Single slice L1-L3 Single slice T12-L3 Single slice L1+L2 Single slice T13-L3 3D QCT L1+L2
n = ? premenopausal females n = 10 females?
1.6 1.6
n = 20
2.4
n = 9 females age = 21–73
1.4
Rosenthal [19] Louis [14] Steiger [20]
Gudmundsdottir [8] Lang [13]
1.5 n = 22 females age not specified n = 10 postmenopausal females
1.9
567
Table 2 CT scanner types used in the study. For all scans 100 mAs were used Scanner type
Number of scans
Voltage [kV]
Slice thickness [mm]
Reconstruction kernel
GE LightSpeed QX/i GE LightSpeed Ultra GE HiSpeed CT/i Philips Mx8000D Siemens Sensation 4 Siemens Emotion Duo
7
120
1.25
Standard
1
120
1.25
Standard
8
120
1.0
Standard
1
120
1.0
B
4
120
1.0
B40 s
5
110
1.0
B40 s
1.3
scanner software or that are implemented as stand-alone software on workstations to which the acquired CT scans are transferred and analyzed. The introduction of spiral multidetector CT (MDCT) technology provided the possibility to rapidly scan a larger volume than with the earlier sequential technique and consequently 3D QCT of the spine was developed [12,13,15]. QCT in general can be split into the scan acquisition and reconstruction process and the scan analysis process, in which densitometric and sometimes geometrical parameters are extracted from the CT images. Spiral CT simplifies the scan acquisition because complete vertebrae are scanned allowing various ROIs to be defined in the image analysis step. This represents an improvement over traditional QCT in which single slices are acquired in each vertebra, the location of which has to be carefully positioned relative to the vertebral body during the acquisition. Several approaches that will be discussed in detail in this paper have been developed for the analysis of the acquired 3D scans. They range from a largely manual placement of a midvertebral trabecular volume of interest (VOI) that in size and location is similar to the volume analyzed in the single slice techniques to a sophisticated 3D segmentation of the scanned vertebral bodies with an automatic placement of a variety of anatomically defined VOIs. In this paper we specifically assessed the precision of two different 3D QCT analysis techniques. One is the commercial software QCTPro (Mindways Software, Inc., Austin, TX), the other the MIAF-Spine software developed at the University of Erlangen (MIAF: Medical Image Analysis Framework). Low precision errors are important for monitoring BMD changes. In-vivo short term precision data for single slice QCT have been published in a few early papers listed in Table 1. In-vivo precision values for 3D QCT of the spine so far have only been published by Lang and coauthors [13].
for all scans (100 mAs). An example image is shown in Fig. 1. For all subjects L1 and L2 were scanned and a reconstruction field of view of 360 mm was used. The first system used for analysis was QCTPro™ (version 4.1.3) from Mindways Software Inc. (Austin, TX). QCTPro places an elliptical cylinder in the anterior trabecular portion of each vertebral body. As a first step the operator has to rotate each vertebra to optimally align it with a reference frame (Fig. 2) in order to facilitate the automatic placement of the analysis VOI. Each vertebra has to be aligned separately. In the aligned vertebrae QCTPro analyzes horizontal line profiles and determines the position of the cortical edges which are then used to calculate the size and position of the elliptical VOI (Fig. 3). By default its vertical extension (the Z-axis dimension) is set to 9 mm which is close to the 10 mm slice thickness most often used in single slice QCT. The operator can manually adjust the size and location of the automatically determined VOI if necessary. Finally average bone mineral density is calculated within the VOI. The second analysis system was MIAF-Spine, software developed at the University of Erlangen. It is based on an automatic 3D segmentation of the vertebral bodies. VOIs are automatically positioned relative to an anatomic vertebral coordinate system (VCS) that is aligned along the principal axes of the vertebral body. Some operator interaction is required for the identification of the vertebrae and their separation. Details have been published before [15]. In addition to the elliptical trabecular VOI used by QCTPro, with MIAFSpine a large variety of VOIs can be analyzed. The three main VOI types
Methods In this study we evaluated the analysis reproducibility of two different 3D QCT techniques. Baseline scans of 30 subjects (mean age: 62.9 ± 5.7 years) were randomly selected from a multi-center study of 150 postmenopausal (N5 years) women with lumbar spine BMD T-score between −2 and −3.5 and total femur T-score of larger than −3.5 as measured by DXA. Exclusion criteria were any bone disease other than osteopenia, osteoporosis, Paget's bone disease, osteomalacia, or osteogenesis imperfecta. All scans were analyzed by four different operators for each analysis system. The scans were acquired with six different CT scanners. As can be seen from Table 2 acquisition and reconstruction parameters depended slightly on the scanner type but were selected to be as equivalent as possible; for example, a standard body reconstruction kernel was used for all scanners. With one exception the voltage was set to 120 kV and the mAs was constant
Fig. 1. Example CT image: One slice from a stack of images is shown. The phantom below the patient is used to derive BMD values from Hounsfield units displayed the CT image.
568
K. Engelke et al. / Bone 44 (2009) 566–572
Fig. 2. QCTPro alignment of vertebral bodies. (Top) before, (bottom) after manual rotation applied to L1. The procedure has to be repeated for L2.
are the vertebral body, the elliptical cylinder, and the Osteo VOI (Fig. 4). Integral, trabecular and cortical compartments can be separated as well as superior, mid and inferior portions of the vertebral body. Due to the limited spatial resolution of the clinical CT scanners of about 1 mm at the settings we used (360 mm FoV resulting in an inplane pixel size of 0.7 mm, slice thickness 1–1.25 mm according to Table 2), the cortical VOI includes some subcortical bone. In this study after the separation of the trabecular and cortical VOIs, the trabecular VOI was peeled by two additional millimeters to ensure that no subcortical bone is included (Fig. 5). Four different operators analyzed all 30 datasets independently after having received appropriate training. Due to logistical reasons two different groups of operators performed the QCTPro and the MIAF-Spine analysis. Results of L1 and L2 were averaged and the coefficient of variation (% CV) among the four operators was calculated for each patient. Then the root mean square coefficient of variation (CVrms) was calculated for all patients. Analysis differences among
operators were analyzed by one way ANOVA using absolute and percentage BMD differences from the patient average BMD values of the four operators. One way ANOVA was also used to identify scanner specific differences in precision errors. Paired t-tests were used to investigate precision differences between VOIs. Results Twenty nine out of the 30 subjects were included in the analysis. Mean BMD of L1-L4 as measured by DXA was 0.759 g/cm2, and for L1 and L2 (which were measured by QCT) was 0.705 g/cm2. One scan was excluded because the vertebrae were not fully included in the reconstructed field of view and therefore the 3D segmentation used by MIAF-Spine did not work properly. Table 3 lists the average BMD values derived with MIAF-Spine for the subjects included in the analysis for the various analysis VOIs. It is obvious that the cortical VOIs (range of average BMD from 223 to 327 mg/cm3) include subcortical bone as pure cortical BMD is in the range of 1000 mg/cm3.
Fig. 3. Elliptical analysis VOIs used by QCTPro. (Left) VOIs automatically determined; (right) after operator adjustment; in particular the computer determined position of the VOI in L1 was not in the center of the vertebra (see sagittal view of L1 in left image).
K. Engelke et al. / Bone 44 (2009) 566–572
569
Fig. 4. MIAF-Spine: trabecular VOIs used in this study.
For QCTPro the corresponding average BMD value of the elliptical VOI was 72.8 ± 19.4 mg/cm3, which is almost identical to the MIAF-Spine result for the mid trabecular compartment of the elliptical cylinder
VOI. Neither MIAF-Spine nor QCTPro showed any significant interoperator differences in BMD, bone mineral content (BMC), or VOI volume. Analysis precision for MIAF-Spine is shown as SDrms for BMD in Table 4 and as CVrms in Table 5 through Table 7 for BMD, BMC and volume, respectively. Absolute precision errors are in the order of 1 to 2 mg/cm3 for integral and trabecular BMD and 5 to 10 mg/cm3 for cortical BMD with the exception of the total and inferior cortical VOIs of the elliptical cylinder. As shown in Table 6 overall higher precision errors were observed for BMC than for BMD although the relationships of errors among VOIs Table 3 BMD ± SD in mg/cm3 averaged over all patients using MIAF-spine VOI shape
VOI
Integral BMD
Trabecular BMD
Cortical BMD
Vertebral body
Total Superior Mid Inferior Total Superior Mid Inferior Total Superior Mid Inferior
152 ± 22 164 ± 23 120 ± 23 173 ± 28 94 ± 20 110 ± 21 NA 113 ± 22 140 ± 22 152 ± 23 108 ± 23 162 ± 27
81 ± 21 80 ± 20 74 ± 23 93 ± 23 70 ± 20 80 ± 20 68 ± 22 70 ± 21 77 ± 21 77 ± 20 70 ± 23 88 ± 23
296 ± 28 293 ± 28 327 ± 28 285 ± 32 231 ± 35 240 ± 34 NA 223 ± 42 283 ± 30 279 ± 29 315 ± 35 275 ± 33
Elliptical cylinder
Osteo
Fig. 5. MIAF-Spine: cortical and trabecular compartments of the total vertebral body VOI. The trabecular VOI is derived by peeling the endosteal cortical bone by 2 mm. The integral VOI includes the cortical and the trabecular VOIs as well as the 2 mm peeled subcortical VOI that is neither included in the cortical or the trabecular VOIs.
The trabecular VOIs are depicted in Fig. 4. The VOIs for the measurement of integral, trabecular and cortical are shown for the total vertebral body VOI in Fig. 5 and are defined for the other VOIs accordingly. Note. for the mid VOI of the elliptical cylinder there are no integral and cortical values.
570
K. Engelke et al. / Bone 44 (2009) 566–572
Table 4 Analysis precision (SDrms in mg/cm3) of MIAF-Spine for BMD
Table 6 Analysis precision (CVrms in %) of MIAF-Spine for BMC
VOI shape
VOI
Integral BMD
Trabecular BMD
Cortical BMD
VOI shape
VOI
Integral BMC
Trabecular BMC
Cortical BMC
Vertebral body
Total Superior Mid Inferior Total Superior Mid Inferior Total Superior Mid Inferior
1.0 ± 1.1 1.9 ± 2.3 1.0 ± 1.1 1.0 ± 1.3 0.5 ± 0.6 1.5 ± 2.1 NA 0.8 ± 0.8 0.7 ± 0.7 1.3 ± 1.2 0.5 ± 0.5 1.0 ± 0.8
0.7 ± 0.9 1.1 ± 1.5 0.4 ± 0.4 2.1 ± 2.6 0.8 ± 0.7 1.5 ± 2.7 0.4 ± 0.4 2.8 ± 2.8 0.8 ± 1.0 1.3 ± 1.7 0.4 ± 0.5 2.2 ± 2.8
5.3 ± 5.1 6.3 ± 5.9 6.9 ± 6.2 7.6 ± 7.6 13.8 ± 13.3 9.1 ± 11.8 NA 23.5 ± 23.0 5.6 ± 5.4 5.9 ± 7.0 8.1 ± 7.0 8.5 ± 8.6
Vertebral body
Total Superior Mid Inferior Total Superior Mid Inferior Total Superior Mid Inferior
1.5 ± 1.7 2.3 ± 2.5 1.2 ± 1.2 2.0 ± 2.4 1.6 ± 1.6 2.4 ± 2.8 NA 2.1 ± 2.1 1.8 ± 1.9 2.3 ± 2.2 1.3 ± 1.5 2.2 ± 2.4
1.6 ± 1.7 2.4 ± 2.6 0.8 ± 0.7 3.5 ± 6.2 2.2 ± 2.2 3.8 ± 4.8d 1.0 ± 1.1d 5.6 ± 6.2c 2.4 ± 3.2d 3.0 ± 3.3 1.7 ± 1.8d,b 4.1 ± 7.1d,c
3.0 ± 3.0 3.5 ± 4.0 3.2 ± 3.0 3.9 ± 3.8 5.2 ± 5.7d 3.6 ± 4.9 NA 9.2 ± 9.7d 3.4 ± 3.3d,b 3.1 ± 3.3 3.8 ± 3.5c 4.3 ± 4.4b
Elliptical cylinder
Osteo
The trabecular VOIs are depicted in Fig. 4. The VOIs for the measurement of integral, trabecular and cortical are shown for the total vertebral body VOI in Fig. 5 and are defined for the other VOIs accordingly. Note: for the mid VOI of the elliptical cylinder there are no integral and cortical values.
were similar compared to those obtained for BMD. For QCTPro the BMD precision error of the central elliptical cylinder, which is comparable to the mid elliptical cylinder VOI of the MIAF-Spine analysis, was 1.7 ± 2.2%. Cortical thickness was only determined by MIAF-Spine. Analysis precision (CVrms) of cortical thickness of the total vertebral body was 4.7 ± 5.2%. For the mid section of the vertebral body it was 5.6 ± 5.2%. Again there were no significant differences among operators. Neither for MIAF-Spine nor for QCTPro there were any scanner specific differences in precision errors for any of the variables analyzed. Least significant changes (LSC) for the mid trabecular VOIs were 1.1 mg/cm3, and 1.5 to 3 mg/cm3 for the other trabecular and integral VOIs. For cortical BMD the LSC varied from 13 to 25 mg/cm3 with even higher LSC for the VOIs of the elliptical cylinder (11.8 to 30.7 mg/cm3). Discussion In this study we determined inter-operator analysis precision errors for 3D QCT of the spine in 29 postmenopausal women using a variety of commercial whole-body CT scanners. Compared to other studies this is a larger investigation reflecting the clinical settings in which QCT is typically applied. Also in addition to the universal midvertebral trabecular VOI, precision was determined for a large variety of additional VOIs that can be analyzed with 3D QCT. As can be seen from Table 3 the study subjects had low trabecular BMD values ranging from about 70 to 90 mg/cm3 characteristic of an osteoporotic population where most clinical QCT measurements are performed.
Table 5 Analysis precision (CVrms in %) of MIAF-Spine for BMD VOI shape
VOI
Integral BMD
Trabecular BMD
Cortical BMD
Vertebral body
Total Superior Mid Inferior Total Superior Mid Inferior Total Superior Mid Inferior
0.7 ± 0.8 1.2 ± 1.4 0.9 ± 1.0 0.6 ± 1.0 0.6 ± 0.8 1.3 ± 1.6 NA 0.7 ± 0.8 0.5 ± 0.5c 0.9 ± 0.9a 0.5 ± 0.5d 0.6 ± 0.6
1.0 ± 1.3 1.5 ± 1.9 0.6 ± 0.6 2.4 ± 3.4 1.3 ± 1.7 2.3 ± 3.3c 0.6 ± 0.7 4.4 ± 5.3c 1.2 ± 1.5d 1.9 ± 2.5c 0.6 ± 0.7 2.7 ± 3.9c, a
1.8 ± 1.8 2.2 ± 2.1 2.1 ± 1.9 2.7 ± 2.7 6.0 ± 5.6d 4.2 ± 5.9c NA 10.9 ± 10.5d 2.0 ± 2.0b 2.2 ± 2.7a 2.6 ± 2.2c 3.1 ± 3.1d,b
Elliptical cylinder
Osteo
a
Significant differences between osteo and elliptical cylinder VOIs (p b 0.05). Significant differences between osteo and elliptical cylinder VOIs (p b 0.01). Significant differences between osteo/elliptical cylinder VOIs and vertebral body VOI (p b 0.05). d Significant differences between osteo/elliptical cylinder VOIs and vertebral body VOI (p b 0.01). b c
Elliptical cylinder
Osteo
a
Significant differences between osteo and elliptical cylinder VOIs (p b 0.05). Significant differences between osteo and elliptical cylinder VOIs (p b 0.01). Significant differences between osteo/elliptical cylinder VOIs and vertebral body VOI (p b 0.05). d Significant differences between osteo/elliptical cylinder VOIs and vertebral body VOI (p b 0.01). b c
Using MIAF-Spine the RMS-precision error in the classical midvertebral trabecular VOI was 0.4 mg/cm3 or 0.6% independent of whether the elliptical cylinder VOI, the Osteo VOI or the vertebral body VOI was used. Using QCTPro, a precision error of 1.7% was determined. Compared to the precision values of the classical single slice based QCT listed in Table 1 it should be pointed out that here we report CVrms compared to the ‘non-RMS’ CV values shown in the table. RMS-CV errors are typically higher by 25% to 40% compared with average CVs [7]. For example, corresponding ‘non-RMS’ CVs in our analysis were 0.5% for MIAF-Spine and 1.4% for QCTPro, respectively. Also BMD values in most of the studies listed in Table 1 were most likely higher than in our study as younger subjects were investigated. However, for a given absolute error measured as the standard deviation the corresponding %CVs are larger if absolute BMD values decrease. The standard deviation in our study was 0.34 mg/cm3 for MIAF-Spine and 0.8 mg/cm3 for QCTPro corresponding to 0.5 to 1 Hounsfield Units, which are excellent results. On the other hand the values in Table 1 represent true in-vivo precision measurements, in which subjects were measured twice with repositioning, whereas in this study only the inter-operator analysis precision was determined. Obviously in 3D QCT repositioning has a lower impact on precision errors than in the traditional 2D slice based technique where the location of the slice relative to the vertebral body is determined at the time of acquisition using the scout view taken before the actual CT scan. In 3D QCT the complete volume of the vertebra is scanned and the selection of analysis VOIs is no longer influenced by the scout view. In an earlier publication by Lang and coworkers [13] of 3D QCT of the spine, for the midvertebral elliptical cylinder VOI the difference between in-vivo (1.3% CVrms) and interoperator analysis precision (0.7%) was 0.6% derived in repeated scans of ten subjects with a mean trabecular BMD of 88 ± 20 mg/cm3 compared to our 68 ± 22 mg/cm3. Thus to assess true measurement imprecision, approximately half a percent has to be added to the values we derived in our study, although we speculate that for MIAFSpine the difference between in-vivo and inter-observer analysis precision will be smaller than the 0.6% reported by Lang [13]. The reason is the use of the vertebral coordinate system that only depends on the segmentation, which for MIAF-spine is highly automated. In comparison, in QCTPro the alignment of the vertebral body has to be performed manually. The software used in Lang's study [13] is slightly more automated than QCTPro but the user still had to manually determine the main vertebral axes on axial, sagittal, and coronal reformations of the CT dataset. The comparison of precision errors in different compartments of the vertebral body must be restricted to the MIAF-Spine results as these values do not exist for QCTPro. As can be seen from Table 5,
K. Engelke et al. / Bone 44 (2009) 566–572
precision errors for trabecular BMD were lowest in the mid VOIs independent of VOI shape. Precision errors in the total superior and inferior VOIs were usually at least twice as high. The same result was found for trabecular volume although to a lesser degree as precision errors for the total VOIs differed only by 20% to 70% from those of the mid VOIs. In QCT volume and BMD are the independent measurements, BMC is a derived quantity. The very low precision error of the mid VOI indicates that this VOI can be very reproducibly relocated and that there is a smaller BMD variation than in the other regions of the vertebral body. Even the total VOIs, despite being significantly larger showed higher precision errors for trabecular bone. It is interesting to note that in the inferior VOIs precision errors were by a factor of at least four higher compared to the mid VOIs and by a factor of at least two compared to the total VOIs although precision errors for trabecular volume were higher in the superior compared to the inferior VOIs indicating a high variation of trabecular BMD in the inferior VOIs. With respect to VOI shape, precision errors for trabecular bone were lower (for the mid VOIs: comparable) in the vertebral body than in elliptical cylinder or Osteo VOIs which may be explained by the larger size of the VOI. With a few exceptions the observations discussed for the trabecular VOIs are also true for the integral VOIs. The most obvious exception is that the precision errors in the inferior VOIs for integral bone were comparable to the other integral VOIs. Also there were no significant differences for integral BMD between the vertebral body and the elliptical cylinder VOIs, whereas some differences in precision errors between the elliptical cylinder and the Osteo VOIs remained. The precision errors for integral BMD of the total Osteo or the total vertebral body VOI (Table 5) are equivalent to those reported for DXA of comparable populations [4,10,17] even if a 0.5% error due to repositioning is added to the QCT values. The determination of a cortical VOI in the spine is ambiguous as the spatial resolution of clinical CT scanners is not adequate to accurately image the thin cortex [18]. Thus the cortical VOI shown in Fig. 5 also includes subcortical bone. This is similar to the classical single slice based analysis [11] but in contrast the results shown here are based on a 3D segmentation. Precision errors for cortical BMD, BMC and volume are also listed in Tables 5 to 7. The errors are lowest in the vertebral body VOIs which is understandable as here the analyzed cortical volume is highest. Given the limitations of the imaging procedure a precision error of about 2%-2.5% for cortical BMD is a surprisingly good result, in particular as the 2 mm peeled subcortical bone (Fig. 5) was not included in the cortical compartment. Similar to the trabecular VOIs, precision errors were significantly higher in the inferior VOIs for cortical BMD but not for cortical volume. The cortical contributions of Table 7 Analysis precision (CVrms in %) of MIAF-spine for volume VOI shape
VOI
Integral vol
Trabecular vol
Cortical vol
Vertebral body
Total Superior Mid Inferior Total Superior Mid Inferior Total Superior Mid Inferior
1.2 ± 1.4 1.8 ± 1.7 0.5 ± 0.5 1.9 ± 2.3 1.3 ± 1.4 1.9 ± 2.0 NA 1.9 ± 1.8 1.6 ± 1.8 2.1 ± 2.2d 1.1 ± 1.1d 2.0 ± 2.5
0.7 ± 1.0 1.7 ± 1.7 0.4 ± 0.4 1.5 ± 2.3 1.2 ± 1.3c 2.4 ± 2.3c 0.9 ± 1.1 1.9 ± 1.7 1.4 ± 1.6d 2.2 ± 2.1 1.2 ± 1.3d,b 1.7 ± 2.6d
4.1 ± 4.3 4.3 ± 4.7 3.9 ± 3.5 4.7 ± 4.8 5.0 ± 6.4 4.2 ± 5.8 NA 6.7 ± 8.1c 4.5 ± 5.0d 4.6 ± 5.0 5.1 ± 4.6d 4.9 ± 5.6c
Elliptical cylinder
Osteo
a
Significant differences between osteo and elliptical cylinder VOIs (p b 0.05). Significant differences between osteo and elliptical cylinder VOIs (p b 0.01). Significant differences between osteo/elliptical cylinder VOIs and vertebral body VOI (p b 0.05). d Significant differences between osteo/elliptical cylinder VOIs and vertebral body VOI (p b 0.01). b c
571
Fig. 6. Dependency of precision errors of the total vertebral body integral BMD on the reconstruction field of view. Data are from a different study of premenopausal women by Mastmeyer et al. [15]. As outlined in the Methods section above the study on postmenopausal women reported in this contribution used a field of view (FOV) of 360 mm.
the elliptical cylinder VOI are only the elliptical parts of the endplates cut out by the cylinder thus this VOI should not be used for cortical measurements. Some authors advocate the use of BMC or of the product of BMD and cortical thickness for cortical measurements [3,16] arguing that the partial volume artifacts may be partly compensated. However, BMC precision errors in the vertebral body and Osteo VOIs were approximately 50% higher than those for BMD. Analysis precision errors for cortical thickness were only analyzed in the total and mid vertebral body VOIs and were 4.7% and 5.6%, respectively, which is an acceptable result. The clinical usefulness of QCT has recently been summarized in the official positions of the International Society of Clinical Densitometry (ICSD) [5]. These positions are still primarily based on trabecular BMD of the mid section of the elliptical cylinder VOI, although with 3D QCT now a large variety of additional VOIs is available and regional differences in BMD can be determined across subjects or monitored longitudinally. Our analysis has shown that the lowest precision errors are achieved in the classical mid vertebral VOI. However, precision errors are reduced by about 50% compared to the classical technique if a true 3D segmentation in combination with an anatomic coordinate system is used. In comparison, a 3D data acquisition without real 3D image processing and a largely manually driven analysis is less advantageous. It must be cautioned that the comparison of precision errors obtained in this study with the values in Table 1 is limited by differences in imaging protocols and study populations used in the cited studies but a more detailed analysis is beyond the scope of this paper. For parameters other than the mid trabecular VOI, the use of the vertebral body or the Osteo VOIs are advantageous compared to the elliptical cylinder VOI. From the perspective of precision errors, the Osteo VOI should be used for the analysis of integral BMD and the vertebral body VOI for the measurement of cortical BMD. In both cases the total and mid VOIs gave better results than inferior and superior VOIs. For cortical measurements the elliptical cylinder VOIs are inadequate. As we have shown previously for the MIAF-Spine software, precision can be further improved by reducing the reconstruction field of view of the CT images, which increases the spatial resolution and will improve the 3D segmentation and the accuracy of the vertebral coordinate system [15]. As an example Fig. 6 shows the field of view dependency of the %CVrms of the integral total vertebral body VOI. The data were generated from a population of premenopausal women applying the same scan protocol used here but with a lower tube current of 60 mAs. Thus the absolute %CVs differ slightly from our study but the impact of the reconstruction field of view on precision errors of BMD and volume will be similar. In summary, in 3D QCT of the spine precision errors equivalent to those reported for DXA can be achieved if advanced analysis techniques are applied to the QCT scans exploiting their volumetric nature. Without a 3D segmentation and the use of anatomic
572
K. Engelke et al. / Bone 44 (2009) 566–572
coordinate systems precision errors are higher by about a factor of two. Lowest precision errors were obtained for the midvertebral elliptical VOI that is equivalent to the VOI used in classical single slice QCT. For the investigation of trabecular BMD in the inferior and superior parts of the vertebrae either the Osteo or the vertebral body VOIs should be used, for cortical measurements the vertebral body VOI is suited best. Acknowledgments This work was partly supported by the European Research Grant EU QCK6-CT-2002-02440. Preliminary results of this study have been presented at the 28th Annual Meeting of the American Society for Bone and Mineral Research, Philadelphia, (2006) and at the 17th International Bone Densitometry Workshop, Kyoto, Japan (2006). Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at doi:10.1016/j.bone.2008.11.008.
References [1] Cann CE, Genant HK, Kolb FO, Ettinger B. Quantitative computed tomography for prediction of vertebral fracture risk. Bone 1985;6:1–7. [2] Delmas PD, Li Z, Cooper C. Relationship between changes in bone mineral density and fracture risk reduction with antiresorptive drugs: some issues with metaanalyses. J Bone Miner Res 2004;19:330–7. [3] Dougherty G, Newman D. Measurement of thickness and density of thin structures by computed tomography: a simulation study. Med Phys 1999;26:1341–8. [4] El Maghraoui A, Do Santos Zounon AA, Jroundi I, Nouijai A, Ghazi M, Achemlal L, et al. Reproducibility of bone mineral density measurements using dual X-ray absorptiometry in daily clinical practice. Osteoporos Int 2005;16:1742–8. [5] Engelke K, Adams JE, Armbrecht G, Augat P, Bogado CE, Bouxsein ML, et al. Clinical use of quantitative computed tomography and peripheral quantitative computed tomography in the management of osteoporosis in adults: the 2007 ISCD official positions. J Clin Densitom 2008;11:123–62.
[6] Genant HK, Cann CE, Ettinger B, Gordan GS. Quantitative computed tomography of vertebral spongiosa: a sensitive method for detecting early bone loss after oophorectomy. Ann Intern Med 1982;97:699–705. [7] Glüer CC, Blake G, Lu Y, Blunt BA, Jergas M, Genant HK. Accurate assessment of precision errors: how to measure the reproducibility of bone densitometry techniques. Osteoporos Int 1995;5:262–70. [8] Gudmundsdottir H, Jonsdottir B, Kristinsson S, Johannesson A, Goodenough D, Sigurdsson G. Vertebral bone density in Icelandic women using quantitative computed tomography without an external reference phantom. Osteoporos Int 1993;3:84–9. [9] Guglielmi G, Floriani I, Torri V, Li J, van Kuijk C, Genant HK, et al. Effect of spinal degenerative changes on volumetric bone mineral density of the central skeleton as measured by quantitative computed tomography. Acta Radiol 2005;46:269–75. [10] Henzell S, Dhaliwal S, Pontifex R, Gill F, Price R, Retallack R, et al. Precision error of fan-beam dual X-ray absorptiometry scans at the spine, hip, and forearm. J Clin Densitom 2000;3:359–64. [11] Kalender WA, Klotz E, Süss C. Vertebral bone mineral analysis: an integrated approach with CT. Radiology 1987;164:419–23. [12] Lang TF, Guglielmi G, van Kuijk C, De Serio A, Cammisa M, Genant HK. Measurement of bone mineral density at the spine and proximal femur by volumetric quantitative computed tomography and dual-energy x-ray absorptiometry in elderly women with and without vertebral fractures. Bone 2002;30: 247–50. [13] Lang TF, Li J, Harris ST, Genant HK. Assessment of vertebral bone mineral density using volumetric quantitative CT. J Comput Assist Tomogr 1999;23:130–7. [14] Louis O, Luypaert R, Kalender W, Osteaux M. Reproducibility of CT bone densitometry: operator versus automated ROI definition. Eur J Radiol 1988;8:82–4. [15] Mastmeyer A, Engelke K, Fuchs C, Kalender WA. A hierarchical 3D segmentation method and the definition of vertebral body coordinate systems for QCT of the lumbar spine. Med Image Anal 2006;10:560–77. [16] Newman DL, Dougherty G, al Obaid A, al Hajrasy H. Limitations of clinical CT in assessing cortical thickness and density. Phys Med Biol 1998;43:619–26. [17] Phillipov G, Seaborn CJ, Phillips PJ. Reproducibility of DXA: potential impact on serial measurements and misclassification of osteoporosis. Osteoporos Int 2001; 12:49–54. [18] S. Prevhral, K. Engelke, and W.A. Kalender The influence of object size and CT imaging parameters on the measurement accuracy of cortical width and density. 13th International Bone Densitometry Workshop 8, pp. 519. Lake Lawn, WI: Osteoporos Int. [19] Rosenthal DI, Ganott MA, Wyshak G, Slovik DM, Doppelt SH, Neer RM. Quantitative computed tomography for spinal density measurement; factors affecting precision. Invest Radiol 1985;20:306–10. [20] Steiger P, Block JE, Steiger S, Heuck A, Friedlander A, Ettinger B, et al. Spinal bone mineral density by quantitative computed tomography: effect of region of interest, vertebral level, and technique. Radiology 1990;175:537–43.