Differences in Skeletal Kinetics Between Vertebral and Humeral Bone Measured by 18F-Fluoride Positron Emission Tomography in Postmenopausal Women


  • Dr. Gary J. R. Cook,

    Corresponding author
    1. Department of Radiological Sciences and Medical Engineering, Guys, Kings and St Thomas' School of Medicine, Kings College, London, U.K.
    • Department of Nuclear Medicine Guys Hospital London, SE1 9RT, U.K.
    Search for more papers by this author
  • Martin A. Lodge,

    1. Department of Radiological Sciences and Medical Engineering, Guys, Kings and St Thomas' School of Medicine, Kings College, London, U.K.
    Search for more papers by this author
  • Glen M. Blake,

    1. Department of Radiological Sciences and Medical Engineering, Guys, Kings and St Thomas' School of Medicine, Kings College, London, U.K.
    Search for more papers by this author
  • Paul K. Marsden,

    1. Department of Radiological Sciences and Medical Engineering, Guys, Kings and St Thomas' School of Medicine, Kings College, London, U.K.
    Search for more papers by this author
  • Ignac Fogelman

    1. Department of Radiological Sciences and Medical Engineering, Guys, Kings and St Thomas' School of Medicine, Kings College, London, U.K.
    Search for more papers by this author


We have sought to investigate regional differences in skeletal kinetics between lumbar vertebrae and the humerus of postmenopausal women with 18F-fluoride positron emission tomography (PET). Twenty-six women, mean age 62 years, had dynamic PET scans of the lumbar spine and lower humerus after the injection of 180 MBq 18F-fluoride ion. Plasma arterial input functions (IFs) were calculated from a mean IF measured arterially from 10 women and scaled according to late individual venous activity. Vertebral and humeral time activity curves were measured by placing regions of interest (ROI) over lumbar vertebrae and the humeral shaft. Using a three-compartmental model and nonlinear regression analysis the macroconstant Ki, representing plasma clearance of fluoride to bone mineral, and the individual rate constants K1 (related to regional skeletal blood flow) and k2 to k4 describing transport between plasma, an extracellular fluid compartment and a bone mineral compartment, were measured. Mean vertebral Ki (3.47 × 10−2 ml · min−1 · ml−1) and K1 (1.08 × 10−1 ml · min−1 · ml−1) were found to be significantly greater than humeral Ki (1.64 × 10 2 ml min−1 ml−1; P < 0.0001) and K1 (3.90 × 10−2 ml · min−1 · ml−1; P < 0.0001) but no significant differences were found in k2, k3, and k4. These findings confirm differences in regional skeletal kinetics between lumbar vertebrae and the lower humerus. These observations may help increase our understanding of the regional differences in pathophysiology and response to treatment that have been observed in sites consisting predominantly of either trabecular or cortical bone. 18F-fluoride PET may prove to be a valuable technique in the noninvasive measurement of regional skeletal metabolism.


Differences in regional skeletal metabolism are important in the understanding of the pathophysiology of metabolic bone diseases and the differential responses to treatment between cortical and trabecular bone.(1) Global measurements of skeletal function, for example, biochemical markers or radionuclide clearance and tracer retention techniques are unable to infer the contribution from or relative changes between sites rich in either trabecular or cortical bone.(2–5) The greatest changes in mass after successful treatment of osteoporosis are seen at sites rich in trabecular bone.(6) However, cortical bone comprises approximately 80% of the skeleton and its importance in the development of osteoporosis, not only at the femoral neck but also in vertebrae, may be underestimated.(1,7–9)

Although it is possible to measure regional bone mass or structure with, for example, dual-energy X-ray absorptiometry (DXA), quantitative computed tomography, and quantitative magnetic resonance imaging, the measurement of regional metabolism is more difficult in man.(10) Studies on regional turnover or regional blood flow are possible in animal models but in man histomorphometry is used to infer information on regional metabolism.(11–13) However, this is a relatively invasive technique that generally is limited to the iliac crest and is not always practical in a clinical setting.

Noninvasive, functional imaging techniques therefore are attractive both as research and as clinical tools in the measurement of regional skeletal metabolism. For example, by using single-photon emission computed tomography (SPECT) and the bone tracer 99m-technetium methylene diphosphonate (99mTc MDP), Israel et al. have previously shown differences in turnover between sites consisting predominantly of either cortical or trabecular bone in normal and osteoporotic women.(14)

18F-fluoride positron emission tomography (PET) allows the measurement of regional skeletal kinetics using compartmental analysis and has previously been described in the evaluation of the axial skeleton in normal volunteers and in a small number of subjects with metabolic bone disorders.(15,16) The advantages of PET over SPECT include superior spatial resolution and more accurate quantification so that parameters may be expressed in absolute units, for example, milliliters per minute per milliliter of bone.

18F-fluoride is a positron emitting radionuclide that was first described as a bone imaging agent nearly 40 years ago.(17) After diffusion through bone capillaries into bone extracellular fluid, fluoride ion exchanges with hydroxyl groups in the hydroxyapatite crystal to form fluoroapatite.(18) Fluoride is preferentially deposited at the surfaces of bone where remodeling and turnover is greatest.(19,20) Employing plasma clearance methods, this tracer has been used to measure total skeletal blood flow, which has been shown to correlate with osteoblast work rate in iliac trabeculae and skeletal influx rate of calcium in osteoporotic patients.(21)

The aim of the present study was to measure and compare regional skeletal kinetics at sites consisting of either predominantly trabecular or cortical bone, using 18F-fluoride PET in nonosteoporotic postmenopausal women.


Twenty-six postmenopausal women (mean age 62 years) who did not have osteoporosis by World Health Organization (WHO) criteria for bone densitometry measured by DXA were included. Information on age, relationship to the menopause, and bone density is detailed in Table 1.(22) No subject was known to suffer from a metabolic bone disorder or was receiving drugs known to affect skeletal metabolism. The study was approved by the local ethical committee and informed, written and witnessed consent was obtained.


18F-fluoride was produced by the 18O (p,n) 18F nuclear reaction in a cyclotron by bombarding highly enriched H218O with 11 MeV protons. After purification, the solution was passed through a 0.22-μm filter for sterilization and injected in aqueous solution.

Table Table 1.. Age, Years Since Menopause, Lumbar Spine, and Femoral Neck DXA Bone Densitometry T Scores in Study Group of 26 Females
 Mean (SD)Range
  1. T score = (measured BMD – young adult mean BMD)/young adult standard deviation. Lumbar spine using manufacturers normal database and femoral neck using data from NHANES III.(23)

Age (years)62 (8.8)46–77
Years since menopause10.7 (9.4)1–34
Lumbar spine T score−0.63 (0.65)−2.49–1.67
Femoral neck T score−1.39 (0.67)−2.35–−0.28

Blood sampling

In all subjects venous blood samples were taken at 40, 50, and 60 minutes after the injection of 18F-fluoride. A portion of each sample was centrifuged and the activity concentration in plasma and whole blood were measured using a well-counter that had previously been cross-calibrated with the PET scanner. In addition, arterial samples were taken from the first 10 subjects via a radial artery cannula. For the first 30 minutes of the study a continuous measurement of the activity in blood from the radial artery was made using an on-line crystal scintillation counter.(24) These continuous whole blood data were used in conjunction with a series of discrete arterial samples from the same radial artery cannula, taken at 5-minute intervals, with plasma separated and counted in the same way as the discrete venous samples, to correct for plasma and whole blood differences and to derive an arterial plasma IF.

Image acquisition

Scans were performed on a Siemens ECAT 951R PET scanner with a 10.8-cm axial field of view. Subjects were positioned so that the midlumbar spine (L2–L4) and adjacent lower humeral shaft were within the field of view. Although measurement of the femoral neck as an area rich in cortical bone would be of interest, simultaneous dynamic acquisition with the lumbar spine would not be possible because of the size of the field of view. Therefore, the humerus at the same axial level and in the same field of view as the lumbar spine was used as an area representing relatively cortical-rich bone. A 15-minute transmission scan was performed using 68Ge/68Ga external rods for subsequent attenuation correction. One hundred eighty mega-becquerels of 18F-fluoride was injected intravenously and flushed with 10 ml of saline over 10 s. The injection method was controlled in this fashion to minimize differences in the time to peak plasma activity between subjects in order to simplify the subsequent application of a population IF technique.

A dynamic acquisition was performed over 60 minutes with time frames of 12 × 10 s, 4 × 30 s, and 14 × 240 s. An attenuation map produced from the transmission scan was used to correct images for the effects of attenuation and images of the lumbar spine and humeral shaft were reconstructed using a Hann 0.5 filter resulting in an image spatial resolution of 8.5 mm (full width at half maximum). The 31 × 3.48 mm slices were reconstructed for each frame with 2-mm pixels. Frames 23–30 were summed to produce static images for placement of regions of interests (ROIs).

Plasma arterial IF

For the purposes of kinetic analysis by nonlinear regression, the arterial plasma concentration of 18F-fluoride with relation to time (IF) is required. Because on-line arterial sampling can be uncomfortable for the subject, occasionally associated with morbidity and technically complex, we sought to simplify measurement of the IF by deriving a mean population IF (IFp), which could then be scaled individually for each subject according to venous samples in a manner similar to that previously described for other PET tracers.(25) The IFp was derived from the mean of arterial plasma IFs from the first 10 subjects after the arterial IFs were temporally shifted such that the peak activities coincided.

In the 10 subjects who had both arterial and venous sampling it was noted that arterial to venous plasma activity ratios did not change after 2400 s and were not significantly different from 1 (mean 1.015 ± 0.03).

In each of the 26 subjects the mean activity in the three venous plasma samples was calculated and multiplied by 1.015 to estimate arterial activity (Vc). The mean activity (Ac) from the same three time points was then taken from the IFp. The IFp was then scaled by the factor Vc/Ac and an individual time offset was introduced, based on the delay observed in a time-activity curve derived from an ROI over the aorta. This process therefore produced an individual IF by scaling a mean population arterial plasma IF according to late venous plasma activity.

Mathematical model

A three-compartment model was used as described by Hawkins et al. (Fig. 1).(15) The three compartments consist of plasma, a bone extracellular fluid (ECF) compartment, and a bone mineral compartment. Rate constants describing the transport of fluoride ion between compartments include K1, unidirectional clearance from plasma to total bone tissue; k2 describing the reverse transport from the bone ECF compartment back to plasma; and k3 and k4 describing entry and release from the bone mineral compartment, respectively (Fig. 1). The K1 is related to blood flow by the equation

equation image(1)

where E is the unidirectional extraction fraction, Q is regional blood flow, and PCV is the packed cell volume.

Figure FIG. 1..

A three-compartment model describing fluoride ion kinetics as proposed by Hawkins et al.(15) The K1 represents plasma clearance to the bone ECF compartment, while k2 is the rate constant for return of fluoride to plasma; k3 and k4 are the rate constants describing movement of fluoride into and out of the bound bone (bone mineral) compartment. The PET ROI is made up of the compartments surrounded by the dashed line but also will include a contribution that is caused by a small fraction of the ROI that is comprised of blood.

The plasma clearance of fluoride to bone mineral (Ki) can be calculated from

equation image(2)

and like K1 has units of milliliters per minute per milliliter whereas k2–k4 have units of min−1. A fifth parameter, fractional blood volume, also was included in the model to account for the plasma and red cell 18F-fluoride activity in the tissue region.

Other parameters of physiological interest that can be derived from those above are the ratio of K1/k2, which describes the fraction of the ROI occupied by the bone ECF compartment; the ratio of Ki/K1, which describes the unidirectional extraction efficiency to bone mineral; and k4 × k2/(k2 + k3), which describes the rate constant for efflux of fluoride from bone mineral back to plasma.

Data analysis

ROIs were derived from the summed static image and were placed over the vertebral bodies and humeral shaft with an automated method using 50% of maximum bone activity within the image set as a threshold (Fig. 2). Posterior vertebral elements and disc spaces were excluded from the analysis. Either two or three complete vertebrae were available within the field of view for analysis in each subject. The ROIs were then applied to the dynamic data set to produce time-activity curves for individual vertebrae or the humerus (Fig. 3).

Individual kinetic parameters were estimated from bone and arterial plasma curves by standard nonlinear regression/ least squares iterative analysis. A mean value of each parameter was derived from the whole vertebra or lower humeral region scanned to give regional kinetic parameters for each subject.

A skewness and kurtosis test for normality was made on the distribution of the measured parameters and no significant difference was found from a normal distribution. Statistical comparison therefore was made between vertebral and humeral indices using the paired t-test.

Figure FIG. 2..

(Left) Transaxial 18F-fluoride PET image through the lumbar vertebra showing the ROI around the vertebral body. (Right) Transaxial scan through the humerus with ROI.

Figure FIG. 3..

A representative study showing arterial plasma IF (continuous line), vertebral (triangles), and humeral (circles) ROI time-activity curves. All are corrected for radioactive decay. The vertebral and humeral values have been multiplied by 5 for easier comparison with the IF.


Comparison of results obtained using derived IF with results using individually measured IFs

In the first 10 subjects, when comparing the parameters Ki and K1 to k4, calculated using the individual IFp derived IFs, with those calculated using the directly measured arterial plasma IFs, the mean error for each of the parameters in the group of 10 subjects was not significantly different from zero, suggesting no systematic error.

The root mean square errors and correlation coefficients for each of the parameters using the scaled IFp compared with the individually measured IFs are displayed in Table 2.

Regional skeletal kinetics

The mean values for the measured parameters are detailed in Table 3. Vertebral values of plasma clearance of fluoride to bone mineral (Ki) and unidirectional clearance of fluoride from plasma to total bone tissue (K1) were statistically significantly higher than those for the humerus. No significant difference was found between vertebral and humeral values of k2, k3, and k4, the rate constants describing exchange of fluoride from bone ECF back to plasma, from bone ECF to bone mineral, and from bone mineral to bone ECF, respectively. Calculation of the ratio Ki/K1, reflecting the unidirectional extraction efficiency to bone mineral, revealed significantly higher results in the humerus whereas K1/k2, describing the fraction of the ROI occupied by the bone ECF compartment was higher in vertebrae. No difference was found in the rate constant describing efflux of fluoride from bone mineral back to plasma (k4 × k2/(k2 + k3)) between the vertebral and humeral sites.

Table Table 2.. RMS Error, Correlation Coefficient (r-Value), and Significance of Correlation (P-Value) When Comparing Results Obtained Using Derived IF with Results Using Individually Measured IFs
ParameterRMS errorr-ValueP-Value
  1. RMS, root mean square.

Ki5.6%0.973P < 0.0001
k4 × k2/(k2 + k3)19.2%0.840.004


With the use of 18F-fluoride PET we have observed differences in regional skeletal kinetics, including plasma clearance to bone mineral and parameters related to skeletal blood flow, in postmenopausal women. We also have validated a noninvasive technique using a population IF approach to measure individual IFs without direct arterial sampling in healthy postmenopausal women.(25)

Skeletal metabolic heterogeneities are of interest in the study of the pathophysiology of metabolic bone disorders including osteoporosis and also may help us to understand the different treatment responses observed with regard to bone mass and fracture risk at different skeletal sites.(6,9,11,14) Whyte et al. found that it was not possible to predict regional heterogeneity in postmenopausal osteoporotic skeletal dynamics without histomorphometric analysis but predicted that radionuclide methods might allow these measurements in the future.(26) Unlike histomorphometry, which essentially is limited to the iliac crest, 18F-fluoride PET has the potential to measure kinetics at any skeletal site although to date there is only limited experience in the spine.(15,16,27,28)

Table Table 3.. Parameter Results (Mean and SD) for Vertebral and Humeral Kinetics in 26 Postmenopausal Women
ParameterVertebral mean (and SD)Humeral mean (and SD)Statistical significance of comparison (P-value)
Ki (ml · min−1 · ml−1)3.47 (0.82) × 10−21.64 (0.74) × 10−2<0.0001
K1 (ml · min−1 · ml−1)1.08 (0.29) × 10−13.9 (2.4) × 10−2<0.0001
k2 (min−1)2.6 (1.3) × 10−12.08 (1.65) × 10−10.11
k3 (min−1)1.16 (0.28) × 10−11.3 (1.2) × 10−10.28
k4 (min−1)9.05 (2.1) × 10−31.1 (0.99) × 10−20.32
K1/k20.49 (0.19)0.34 (0.34)0.04
Ki/K10.33 (0.08)0.43 (0.22)0.03
k4 × k2/(k2 + k3) (min−1)6.05 (1.6) × 10−35.7 (6.8) × 10−30.41

Fluoride clearance to bone mineral (Ki) represents fluoride mass influx to the bone mineral compartment and has been shown to be correlated with histomorphometric indices including adjusted mineral apposition rate and bone formation rate as well as serum alkaline phosphatase and parathyroid hormone levels in renal osteodystrophy.(27). Schiepers, using 18F-fluoride PET, has shown differences in axial skeletal kinetic parameters in a variety of metabolic bone diseases and found that Ki best differentiated the disorders.(16)

In addition, the 18F-fluoride ion has previously been used to estimate global skeletal blood flow and has been shown to correlate with the work rate of osteoblasts and the rate of influx of calcium into the exchangeable pools of bone in osteoporotic subjects.(3,21) In animals and human pathological specimens fluoride has been shown to be preferentially deposited at bone surfaces where remodeling is greatest.(19,20)

Plasma clearance of fluoride to bone mineral (Ki) and unidirectional plasma clearance to total bone tissue (K1) are significantly greater in the lumbar spine, which is a predominantly trabecular site compared with the humeral shaft, which is predominantly cortical. We have found Ki to be on average 2.1 times greater in the lumbar spine. This is in keeping with greater turnover and bone formation in trabecular bone and reflects higher mineralization rates. This is likely to be related to the greater available surface area and mass of newly mineralized bone at trabecular sites.

In tomographic imaging the measured activity concentration within small volumes is underestimated because of the limits of the resolution of the imaging system, small volume activities being averaged over a larger volume.(29) Although this partial volume effect, which starts to become significant at sizes less than twice the resolution of the system, may contribute to the differences observed between large vertebral ROIs and relatively smaller humeral ROIs in this study, with tight ROIs around structures of greater than 1.7 cm, we feel that the effect will be small.

The unidirectional plasma clearance of fluoride to total bone tissue (K1) is related to blood flow and the unidirectional extraction fraction of fluoride [see Eq. (1)]. After studies on rabbit hind leg bones it has previously been assumed that the unidirectional extraction of fluoride is close to 100% and K1 therefore equates to regional blood flow.(13) Although this may be true in cortical bone in which blood flow is relatively low, at higher blood flows such as found at trabecular-rich sites, for example, vertebrae, the extraction fraction is likely to fall and so K1 may underestimate true regional flow in these situations.(28,30) The differences we have observed in vertebral and humeral values of K1 in this study therefore should be regarded as minimum differences if regional total bone tissue plasma clearance is assumed to reflect regional blood flow.

In describing skeletal fluoride kinetics, a three-compartment model has previously been shown to be preferable to a two-compartment one.(15) However, even this is likely to be an oversimplification of physiological processes because of the heterogeneity of tissue within PET ROIs. For example, these models do not take into account the bone marrow space within the ROI of the PET scan. Bone marrow ECF is likely to show uptake of the small fluoride molecule as well as cellular uptake within the marrow leading to limited access to the bone mineral surface.(31)

Unless kinetics of the central ECF compartments are measured by a separate technique, the addition of further compartments and rate constants to account for a marrow compartment would make it impossible to calculate any of the parameters with sufficient accuracy by nonlinear regression. We feel that K1 therefore is more appropriately described as plasma clearance to total bone tissue, which is related to total bone tissue blood flow [see Eq. (1)].

The presence of a bone marrow compartment is supported by our observations on the ratio of K1/k2 (Table 3), which reflects the proportion of the ROI occupied by the central ECF compartment. The vertebral ECF (bone and bone marrow) space is larger than that in the humerus as would be expected if trabecular-rich bone contained more marrow. The proportion of total ECF space in vertebral (49%) and humeral (34%) bone regions is in concordance with previously reported estimations in canine tibias, which are 10% for bone ECF space alone and 24% for bone marrow ECF.(32)

In addition, because of the possible limitations of the model, it is felt that the physiological significance of parameters k2 and k3 is not meaningful in relation to the mineralized skeleton and that the ratio of Ki/K1 [= k3/k2 + k3, in Eq. (2)], which reflects the fractional forward movement of fluoride from the central ECF compartment to bone mineral, is preferable, although still limited by uncertainties related to the amount of regional bone marrow. Our results have shown a higher ratio of Ki/K1 in the humerus (Table 3) but this probably reflects the greater access of fluoride to the bone mineral surface from within the humeral ECF compartment because there is proportionally less bone marrow compared with vertebrae, which contain more trabecular bone.

Similarly, the calculation of k4 is affected by inaccuracies in the model and hence consideration of the rate constant describing return of tracer from the bone mineral compartment to plasma is preferable. This is equal to k4 × k2/(k2 + k3). We have not observed differences in k4 or this parameter between vertebral and humeral sites (Table 3).

One of the problems with compartmental model analysis by nonlinear regression is that parameter estimation becomes less stable and more inaccurate the larger the number of parameter estimates required. If it can subsequently be shown that k4 varies little between subjects and skeletal regions, it would then be possible to fix k4 at a population average, reducing the errors in the estimation of the other parameters of interest. Indeed, it has been suggested that k4 can be ignored because it is small or it is assumed to be zero as the calculation of Ki by Patlak graphical analysis (which assumes k4 = 0) produces a straight line plot and correlates with Ki estimated by nonlinear regression.(15)

Although the calculation of the individual rate constants is not independent and therefore errors in one will propagate to others, these errors cancel out to some extent in the calculation of Ki, which is the most stable and precise parameter. This parameter has best differentiated a number of metabolic bone disorders, but further studies are required to see if the computationally simpler methods, such as measurement of the standardized uptake value (SUV), which reflects the proportion of injected dose per milliliter of tissue taken up at a given time postinjection and is normalized to body weight, Ki, derived from Patlak graphical analysis, or model-independent methods such as deconvolution analysis, are equally robust.(16,33)


18F-fluoride PET is a potentially powerful technique to measure regional skeletal metabolism in man in a manner that has not previously been available. It allows estimation of a number of kinetic parameters of interest related to regional mineralization and blood flow and could be developed with wider applications to study focal bone diseases or the regional treatment effects in metabolic bone diseases.

By using this technique we have observed differences in vertebral and humeral bone kinetics in postmenopausal women. In the future this technique and these observations may help in our understanding of the regional differences in pathophysiology and response to treatment that have been observed between cortical and trabecular sites.


This study was funded by a research grant from the Arthritis Research Campaign.