Osteoporosis is a common disease characterized by low bone mass and changes in the microarchitectural environment of affected bone. The implications of this and other metabolic bone diseases are of great consequence, particularly with regard to the increased risk of fragility fractures1. In 2000, the incidence of osteoporotic fractures was approximately 9 million, of which 1.4 million were vertebral fractures2. The burden of this disease and its economic implications have led to the routine use of dual x-ray absorptiometry measurements of bone mineral density as a screening examination. Dual x-ray absorptiometry is currently considered to be the gold standard for bone mineral density quantification and has been shown to correlate with fracture risk and the efficacy of treatment3. Approximately 62 million computed tomography scans are performed annually in the United States alone4. Bone mineral density measurements obtained with computed tomography have been shown in vitro to be predictive of fracture patterns and failure loads5. However, this information is rarely considered or evaluated on clinical computed tomography examinations. The standardized linear attenuation coefficient of tissue, measured in Hounsfield units (HU), may provide information regarding bone quantity that is readily available on computed tomography scans without added costs, radiation, or the use of phantoms. A Hounsfield unit represents a normalized index of x-ray attenuation based on a scale of -1000 defined for air and 0 for water at standard pressure and temperature. For clinical computed tomography examinations, the total linear x-ray absorption coefficient (µ) is calibrated to the x-ray attenuation of water (w), generating a standardized HU value based on the formula: HU = ([µ — µw]/µw) × 1000, with µ defined as the linear x-ray attenuation coefficient of the selected voxel and µw defined as the attenuation coefficient of distilled water at room temperature and pressure. The HU value for bone typically ranges from 300 to 30006. Metal devices markedly attenuate the radiation beam and demonstrate high HU values, typically measuring >5000.
Quantitative computed tomography requires the use of calibrating phantoms or phantomless calibration based on internal standards to account for differences in computed tomography scanners, detector or x-ray tube drift, and attenuation differences based on the patient's body habitus. The majority of studies evaluating computed tomography and quantitative computed tomography for determining bone mineral density were performed prior to the advent of automatic exposure control, which adjusts the tube current on the basis of the amount of attenuation detected, accounting for the patient's body habitus. The computed tomography scanner utilizes data from the scout view as well as real-time feedback from the detectors to determine the necessary exposure time. Exposure times decrease for less dense regions of the body, whereas they increase for denser portions of the body. Automatic exposure control is aimed at shortening the x-ray exposure time for the patient. This technology will result in a more homogenous energy spectrum of the x-ray beam encountered by the spine, making the HU value primarily dependent on the composition of the targeted tissue. As a patient's body-mass index accounts for a large portion of attenuation differences, the use of automatic exposure control theoretically could eliminate the use of phantoms in determining bone mineral density. Thin-section scanning also minimizes partial volume averaging effects, which can affect attenuation coefficients as well.
The goal of the present study was to determine if a correlation exists between the results of dual x-ray absorptiometry bone mineral density measurements and the HU value obtained from computed tomography scans using automatic exposure control. On the basis of a literature search, to the best of our knowledge, evaluation of this relationship has not been investigated. Our hypothesis is that the HU value may serve as a surrogate marker for bone mineral density and may be of value for estimating regional bone density. Normative data were obtained from computed tomography examinations performed for trauma patients who were stratified for age and sex. In order to further assess the correlation between material density, the HU value, and mechanical strength, compression tests and computed tomography were performed on polyurethane foam of various densities.
Subject Cohorts
A database of subjects with either spinal trauma or vertebral compression fractures who had been managed in an academic medical center was reviewed after we had obtained approval and a waiver of patient informed consent from our institutional review board. Subjects who had undergone either an abdominal or lumbar computed tomography scan as well as a lumbar spine dual x-ray absorptiometry scan within twelve months of each other were candidates for inclusion in the study. The exclusion criteria included lumbar spine instrumentation, previous spinal fusion, previous vertebroplasty, or invalid dual x-ray absorptiometry findings due to spinal degeneration or deformity as judged by the reading radiologist. One thousand five hundred charts were reviewed, and twenty-five subjects were identified.
Control subjects (N = 80), utilized to determine normative computed tomography data, were identified from a list of trauma patients who underwent computed tomography because of a possible lumbar spine injury. Patients with prior lumbar spinal instrumentation were excluded. The eighty subjects were equally divided according to sex in each decade of life between ten and ninety years, with five male and five female subjects in each group.
Imaging
All dual x-ray absorptiometry scans were performed with use of the Lunar Prodigy Advance densitometer (General Electric, Milwaukee, Wisconsin). Information from the dual x-ray absorptiometry scans, including T-scores and bone mineral density (measured in g/cm2), were obtained for the first through fourth lumbar vertebrae.
For quantitative computed tomography, a helical sixty-four-channel computed tomography scanner (LS VCT 64; General Electric) was utilized for all subjects. Computed tomography parameters included a slice thickness of 1.25 mm with a 0.625-mm interval, a tube voltage of 120 kVp, a tube current of 300 mA (Smart mA/Auto mA range, 150 to 750), and a bone reconstruction algorithm (window width/window level, —3000/300). Two-dimensional reconstructions were obtained in the coronal and sagittal planes. Quantitative computed tomography was performed without the use of phantoms.
A McKesson Picture Archiving and Communication System (McKesson, San Francisco, California) was used to calculate an average HU value by placing an elliptical region of interest that was confined to the medullary space of the vertebral body to reduce the potential for beam hardening and volume averaging from the adjacent cortical bone. Regions of interest were measured on the axial images at L1 through L4 at three separate locations: immediately inferior to the superior end plate, in the middle of the vertebral body, and superior to the inferior end plate (Fig. 1). For each measurement, the largest possible elliptical region of interest was drawn, excluding the cortical margins to prevent volume averaging. In pilot testing, no significant difference was present when using mean values of multiple small regions of interest compared with a single maximally sized ellipse. On the basis of these findings, the latter was utilized for the present study. The HU values from the three axial slices were averaged to give a mean HU value for each lumbar vertebral body. Measurements were performed by two independent observers who were blinded to the subjects’ dual x-ray absorptiometry scores and were averaged.
Computed tomography scans illustrating the method of determining the HU value with use of an elliptical region of interest function. The top image shows the axial planes of interest on a sagittal slice of a computed tomography scan of the second lumbar vertebral body. Slice A was taken just inferior to the superior end plate, slice B was chosen at the middle of the body of L2, and slice C was taken just superior to the inferior end plate. The bottom images show the HU values generated by the imaging software program.
Biomechanical Methods
The mechanical properties of solid rigid polyurethane foam blocks were assessed with use of a mechanical testing machine. Foams of four different densities (10, 15, 20, and 50 lb/ft3) were obtained from Sawbones (Vashon Island, Washington) for the study. Each type of foam was tested five times with five different 4 × 4 × 4-cm foam blocks. Purely vertical compression was performed with an MTS Sintech 10/GL mechanical testing machine (MTS, Eden Prairie, Minnesota), with a displacement rate of 5 mm/min, until the foam block failed. A 22.24-N preload was applied to the specimens.
Displacements were measured from the crosshead of the MTS machine. For each compression test, engineering stress was calculated by dividing the load recorded at each data point by the original cross-sectional area of the block and engineering strain was calculated by dividing the crosshead displacement by the original height of the block. The elastic modulus was estimated from the slope of the linear portion of the stress-strain curve.
All foam blocks were scanned with a Lightspeed 16 computed tomography scanner (General Electric) with use of a tube voltage of 120 kVp, a tube current of 30 mA, a slice thickness of 2.5 mm, and a slice spacing of 2.5 mm. Four calibration phantoms (100, 400, 1000, and 1750 g/cm2) were included in the scan. To estimate computed tomography densities of the foam blocks, three-dimensional computed tomography-based voxel models of the blocks were constructed with use of Mimics software (Materialise, Plymouth, Michigan). Only the central portions of the foam blocks (including the air pockets) were segmented to avoid the influence of partial volume effects. The HU densities reported in this study are the average HU values calculated with the software on the basis of the segmented region.
Statistical Analysis
Interobserver and intraobserver reliability calculations were performed with use of the interclass correlation coefficient, reported as a score between 0 and 1 (with 0 representing no agreement and 1 representing perfect agreement). A score of >0.8 is considered to indicate excellent agreement7. Correlations between HU values and bone mineral density, T-scores, age, and elastic modulus were all calculated with use of the Pearson correlation coefficient. The sensitivity, specificity, and positive and negative predictive values of the HU value were calculated on the basis of the dual x-ray absorptiometry standard. For this model, bone density was dichotomized as normal or abnormal on the basis of whether the HU value was less than the confidence intervals for subjects with T-scores of -1.0 or more.
Source of Funding
There was no external funding for this study.
Reliability
Measurement of the HU value was reliable, with excellent intraobserver and interobserver reliability of 0.964 and 0.975, respectively.
Normative Data
The HU values obtained from computed tomography for the eighty consecutively presenting trauma subjects were stratified by sex and decade of life (Table I) (see Appendix). The HU value decreased relatively linearly by decade, ranging from a mean of 255.1 in the second decade of life to 78.7 in the ninth decade of life. The correlation between age and HU was significant (r = 0.83, p < 0.05).
There were no significant differences in the HU value at L1 through L4 in either males or females (Table II). The HU values averaged 164.9, 164.0, and 176.6 at the cranial aspect of the body, the middle of the body, and caudal aspect of the body, respectively. These differences were not significant.
Correlation of Dual X-Ray Absorptiometry and Hounsfield Units
The enrolled subjects consisted of eighteen women and seven men with a mean age of 71.3 years (range, thirty-three to eighty-seven years). The HU values for the twenty-five subjects ranged from 31.1 to 277.3 (mean [and standard deviation], 107.1 ± 36.8). Dual x-ray absorptiometry T-scores ranged from —5.1 to 2.8 (mean, —1.42 ± 1.51), and bone mineral density ranged from 0.593 to 1.509 g/cm2 (mean, 1.104 ± 0.186 g/cm2). The correlations between the HU value and both bone mineral density (r2 = 0.44) and T-score (r2 = 0.48) were significant (p < 0.0001) (Figs. 2 and 3).
Scatter plot showing the correlation between Hounsfield units obtained from computed tomography and bone mineral density scores obtained from dual x-ray absorptiometry scans of the first through fourth lumbar vertebral bodies of twenty-five subjects. A significant correlation was found (r2 = 0.44, p < 0.0001).
Scatter plot showing the correlation between Hounsfield units obtained from computed tomgraphy and T-scores obtained from dual x-ray absorptiometry scans of the first through fourth lumbar vertebral bodies of twenty-five subjects. A significant correlation was found (r2 = 0.48, p < 0.0001).
On the basis of World Health Organization (WHO) criteria, subjects were stratified according to T-scores as normal (-1.0 or greater), osteopenic (less than -1.0 or greater than -2.5), or osteoporotic (-2.5 or less)8,9. The mean HU values for normal, osteopenic, and osteoporotic subjects were 133.0 (95% confidence interval, 118.4 to 147.5), 100.8 (95% confidence interval, 93.1 to 108.8), and 78.5 (95% confidence interval 61.9 to 95.1), respectively (Table III). The differences between groups in terms of the HU values were significant (p < 0.00001).
Each subject was determined to have normal or abnormal bone on the basis of being above or below the lower limit of the 95% confidence interval (118.4) as determined above. With use of the T-score as the standard, the sensitivity, specificity, positive predictive value, and negative predictive values were 0.73, 0.71, 0.85, and 0.556, respectively.
Foam Compression Data
In the polyurethane foam blocks, there was a perfect linear correlation between material density and the HU value (r2 = 1.0, p < 0.0001) (see Appendix). Similarly, there was a significant correlation between the HU value and the elastic modulus (r2 = 0.998, p < 0.0001) (see Appendix). Thus, in the polyurethane foam, HU could very accurately predict material density and elastic modulus.
The results of the present study demonstrate that regional cancellous bone mineral density can be approximated from Hounsfield units (HU) measured on computed tomography examinations utilizing automatic exposure control. It was shown that this information not only is reliable but also follows a predictable pattern with increasing age. In our biomechanical study involving polyurethane foam, an increase in the HU value correlated linearly with increasing material density and increasing compressive strength. Given the correlation that was demonstrated between the HU value and bone mineral density with use of dual x-ray absorptiometry, we believe that computed tomography (when performed for other purposes) may be used to identify patients with diminished bone density and those at risk for fracture.
The HU value was found to correlate moderately with T-scores, which are used in the WHO guidelines to diagnose osteoporosis8,9. Thus, the HU value obtained from computed tomography of the lumbar spine may alert the physician to metabolic bone diseases such as osteoporosis. In this manner, the HU value may be used as a guide to perform further investigations, such as dual x-ray absorptiometry, for metabolic bone disease. Clinically, increases in bone mineral density result in decreased fracture risk3. The spine is a particularly important region as approximately half of all osteoporosis-induced fractures are vertebral compression fractures10. An advantage of using the HU value over using dual x-ray absorptiometry is that the former method can readily be used in the cervical and thoracic spine, for which dual x-ray absorptiometry standards are not available. Normative data to allow for the calculation of T-scores based on HU values in these areas are needed. In the present study, subjects with normal bone density had a mean lumbar HU value of 133.0, those with osteopenia had a mean lumbar HU value of 100.8, and those with osteoporosis had a mean lumbar HU value of 78.5.
The biomechanical experiments involving polyurethane foam demonstrated the proof of concept that the HU value can be used to estimate regional bone strength. We found a linear correlation between the HU value and compressive strength for all ranges of density, which would include cancellous bone11. Further experiments using cadaver bone and other failure modes such as shear and torque are planned.
The effect of regional spinal bone mineral density on the success of various implant devices has been investigated, and important correlations have been demonstrated12-14. Of particular interest is the use of the HU value for the planning of screw fixation. Measuring the HU value in the planned screw trajectory may predict the strength of the bone-screw interface before screw insertion. Implant success based on computed tomography with automatic exposure has been tested in dentistry but not in orthopaedic surgery. Hounsfield units have been used to approximate regional oral bone mineral density and have been shown to strongly correlate with insertion torque and the stability of metal implants both in vitro and in vivo15-17. While HU normative values are available for specific regions of the mandible and maxilla18,19, this information was not previously available for the lumbar spine. As in dentistry, the HU value may prove to be an important prognosticator of implant stability, not only for the lumbar spine but for any osseous region.
The present study had several limitations. First, the data obtained for correlations between the HU value and dual x-ray absorptiometry were from patients presenting with either spinal trauma or osteoporosis. In general, these patients were older (mean age, 71.3 years) and had decreased bone mineral density (mean T-score, -1.42) compared with the population as a whole. For this reason, these data may not accurately extrapolate to the general population. Furthermore, the time between computed tomography and dual x-ray absorptiometry (minimum, one year) may have influenced the results, although changes are likely to be small in the absence of the institution of osteoporosis therapy or the onset of a severe endocrine disorder. Second, with the advent of technologies such as quantitative computed tomography and finite element methods, one may claim that an HU analysis is an overly simplistic approach to bone mineral density. We do not propose replacing these technologies, but merely supplementing them with widely available and easily interpretable data. Additionally, with automatic exposure control, HU measurements may be just as accurate without necessitating the use of phantoms. Third, while polyurethane foam is widely used as a model for cancellous bone, the low-density foams are not universally accepted as a model for osteoporotic bone20. This is particularly important given the patient population studied. The computed tomography protocols were different between subjects and foam, but we do not believe that, with the use of the modern scanning protocols, these differences would account for any important differences in the results. Fourth, cancellous bone is heterogeneous and so three axial sections may not accurately summarize bone quality21. The data that were used to determine age-related changes in the HU value were only from a small sample size in comparison with the normative data for thousands of subjects that are used to determine standards for dual x-ray absorptiometry. The control group was selected from a population of patients with blunt trauma, which may not reflect the true normal population. Finally, the correlations in the present study are based on imaging modalities that evaluated different components of bone.
While dual x-ray absorptiometry scans assess both cancellous and cortical bone, our analysis of the HU value involved the evaluation of only cancellous bone, which may account for some of the differences observed between the dual x-ray absorptiometry findings and the HU values. Although cancellous bone has been shown to be more important than cortical bone for vertebral load sharing and fracture risk, we may be ignoring another important component22. While dual x-ray absorptiometry is thought to be the gold standard to determine bone density, it has limitations. The reference standards may not apply to all populations or patients, and the width of bone is not taken into account, so that smaller patients in general have lower bone densities when evaluated with dual x-ray absorptiometry.
In conclusion, the present study introduces a novel application of data that are widely available yet seldom used. Obtaining the HU value from a region of interest designation on a computed tomography scan made with automatic exposure control is straightforward and can be done in an accurate and reliable manner with minimal time or training requirements. It does not require dedicated spine computed tomography as it can be done on computed tomography scans windowed for examinations of the chest, abdomen, or pelvis. There was moderate correlation both between the HU value and bone mineral density of cancellous bone and between the HU value and compressive strength in our osseous model. We propose that the HU value may prove to have several clinical applications, including the ability to predict the stability of orthopaedic implants and to assist in surgical decision-making. In addition, it may serve as a diagnostic tool for screening of regional bone mineral density when computed tomography is performed for other reasons, thereby alerting the physician to initiate further work-up in individuals in whom metabolic bone disease would not otherwise be suspected. We do not suggest that computed tomography scans be used instead of dual x-ray absorptiometry for screening or the assessment of bone density, although we believe that HU values may provide valuable additional information for those individuals who undergo computed tomography scans for other reasons.