Prediction of low DEXA T‐scores by routine computed tomography body scans at different kilovoltage peaks

Previous studies have demonstrated positive correlations between computed tomography (CT) attenuation of lumbar spine vertebrae and their bone mineral density (BMD) measured by dual‐energy X‐ray absorptiometry (DEXA). However, these studies were performed using a standard 120 kilovoltage peak (kVp) setting. As radiation attenuation in mineralised tissues varies by the tube voltage applied, we determined the diagnostic accuracy of CT attenuation at identifying individuals with low BMD at different kVp settings.


Introduction
Osteoporosis is a prevalent disease marked by reduced bone mineral density (BMD), micro-architectural deterioration and elevated fracture risk.Dual-energy X-ray absorptiometry (DEXA) is commonly used to estimate areal BMD at the hip and lumbar spine.While the test is useful in the detection of poor bone health and increased fracture risk, it is generally underutilised even in populations that are eligible for subsidised screening. 1Schreiber et al. 2 was the first to suggest that CT scans may be used opportunistically to measure vertebral bone density.Thus, CT scans performed for various indications could be retrieved retrospectively and without the use of additional tools or radiation exposure to screen for reduced bone density.][5][6][7][8][9][10][11] However, virtually all of these studies used single CT scanning systems with X-ray tube voltages fixed at 120 kVp.Yet, in clinical practice, patients are imaged on different CT scanning systems and at different X-ray tube voltages (kVp) depending on the clinical indication, patient size, age and renal function.Importantly, CT bone density measurements are affected by variations in kVp as ionising radiation is attenuated differently in diverse materials and at altered energy levels. 12e herein report the diagnostic accuracy of CT lumbar spine attenuation at different kVp settings in identifying individuals with low DEXA T-scores.

Methods
The Sydney Local Health District Human Research Ethics Committee (SLHD HREC) approved this study (2022/ ETH01370).For inclusion into the study, patients 18 years or older had to have both CT body imaging incorporating the L1-4 lumbar vertebrae and a DEXA scan of the lumbar spine and hip within 6 months interval of each other.All scans were done at the same institution between 1 July 2019 and 30 June 2022.Exclusion criteria were significant degenerative spine disease and/ or scoliosis, primary or metastatic malignancies affecting the lumbar spine or hip, or prior spinal surgery.Patients who had DEXA scan within 7 days of CT contrast agents were excluded to avoid compromising precision of the DEXA measurement. 13Patients with lumbar spine fractures were included but the specific fractured segment was excluded from the CT and DEXA analysis.
Dual-energy X-ray absorptiometry was performed using standard techniques on a Lunar Prodigy (GE Healthcare, Chicago, IL, USA) or Hologic Horizon densitometer (Hologic Inc).A spine phantom was used daily for calibration.BMD of the lumbar spine (L1-4) and left hip were obtained.We avoided defining osteoporosis by DEXA as many patients with fragility fractures do not have osteoporotic DEXA T-scores. 14As such, patients were categorised into three groups based on the lowest T-score measured at either the lumbar spine, femoral neck or total hip: T-score À2.5 or less, T-score À1.0 to À2.49 or T-score greater than À1.0.
Suitable CT scans covering the lumbar vertebrae L1-4 were included in the analysis.The scans were performed at various X-ray tube voltages and intravenous contrast protocols, reflecting the wide range of imaging indications.Scanners used include Revolution CT (GE Healthcare) or CT Definition AS+ (Siemens Healthcare, Erlangen, Germany).Quality control was maintained at or above the manufacturer's recommendations, including regular calibration to ensure attenuation values were correct.Images were viewed on a standard radiology picture archiving and communication system workstation (PACS) (GE Centricity Workstation, GE Healthcare).Non-contrast series were analysed where possible including when a multiphase study was performed.For dual-energy studies, the Revolution CT produced a monochromatic 70 keV series while the CT Definition AS+ produced mixed series.
Trabecular density of the lumbar vertebral body was measured by a senior radiologist blinded to the DEXA results.To evaluate CT attenuation in Hounsfield units, a single oval region of interest (ROI) was placed over an axial section of mid-vertebral body trabecular bone (Fig. 1).The ROIs were drawn as large as possible while avoiding the vertebral cortex, venous plexus, focal lesions or dense bone islands.Lumbar vertebrae from L1 through L4 were assessed.Any fractures of the hip, pelvis or vertebrae (thoracic, lumbar or sacral) were identified.Vertebral fractures were assessed by the Genant semi-quantitative method. 15To remove any doubt regarding borderline or mild compression deformities, only moderate (grade 2, 25%-40% loss of height) and severe (grade 3, >40% loss of height) compression fractures were included.
Statistical analysis was performed using SPSS software (version 25, IBM Corp, Armonk, NY, USA).Reference values including the mean, standard deviation and 95% confidence intervals were calculated for the cohort.Oneway analysis of variance (ANOVA) was used to compare continuous variables.Correlation between HU values and T-scores were calculated using the Pearson correlation coefficient (weak r = <0.30,moderate r = 0.3-0.7 and strong r = >0.7). 16Area under the ROC curve (AUC) was calculated by receiver operating characteristic (ROC) analysis.Sensitivity and specificity were used to determine thresholds that would yield balanced diagnostic performance.P < 0.05 were considered statistically significant.

Cohort characteristics
Baseline characteristics are shown in Table 1.Of the 268 adults, 63% were female.The mean age was 66.9 years  The most common type of CT scan was abdomen and pelvis +/À chest (73%) followed by kidney, ureter and bladder (11%) and lumbar spine (6%).Scans were done for various clinical indications, most commonly for suspected malignancy or malignancy surveillance.Approximately, one third of scans were unenhanced while the remainder were obtained after administration of intravenous contrast.CT X-ray tube voltage settings in the study include 100 kVp (43%), 120 kVp (20%) and dual energy (80 kVp/140 kVp) (37%).There was no difference in the mean L1 Hounsfield unit between the Siemens Healthcare and GE Healthcare CT scanners (156 vs. 162 HU, P = 0.469).

Lumbar spine CT attenuation and X-ray tube voltages
Mean CT attenuation values (HU) consistently decreased as the tube voltage increased.As a result, mean HU densities at L1 and L1-4 varied significantly depending on the tube voltage used (Table 2).The largest difference was observed between scans performed at 100 and 120 kVp, where the L1 and L1-4 attenuation values differed by 24.9 and 25.3 HU respectively.Patient numbers were distributed proportionally across the spectrum of DEXA T-scores with about half of patients having Tscores between À1.0 and À2.49.In contrast, there was a higher proportion of patients with T-scores of À2.5 or below in the group scanned with dual energy.HU values correlated positively with DEXA-derived T-scores independent of whether the mean of L1-4 or L1 measurements were used.Correlation coefficients ranged between 0.349 (dual energy/L1) and 0.550 (100 kVp/ L1) (P < 0.01 for all).

CT threshold values for the prediction of low T-scores on DEXA
Lumbar vertebral attenuation was significantly lower in patients with DEXA T-scores of À2.5 or less, compared to patients with DEXA T-scores greater than À2.5 (Table 2).Using attenuation values for L1 and mean L1-4, ROC curves were constructed to predict DEXA T-scores of À2.5 or less (Fig. 2).Threshold values for high sensitivity, high specificity and balanced sensitivity/specificity were determined (Tables 3 and 4).Of note, these thresholds differed depending on the CT X-ray tube voltage used.As the use of CT contrast may affect attenuation, we analysed scans with and without prior contrast separately.The AUC for the ROC curves for predicting a DEXA T-score of À2.5 or less was 0.878 (95% CI 0.771-0.984)for the unenhanced series and 0.789 (95% CI 0.686-0.892)for the enhanced CT series.The difference between the ROC performance for contrast and non-contrast series was not statistically significant (P = 0.206).

CT attenuation and DEXA in patients with prevalent fractures
Prevalent fractures were diagnosed in 79 patients on the CT.Of these, five patients had pelvic fractures and 74 had vertebral fractures (45 at multiple levels).Among the fracture group, only 29 (37%) had DEXA T-scores of À2.5 or less while 39 (49%) had T-scores between À1.0 and À 2.49, and 10 (14%) had T-scores above À1.0.When comparing the fracture and non-fracture groups, there was no statistical difference in the mean DEXA BMD (0.787 g/cm 2 vs. 0.800 g/cm 2 , P = 0.498) or T-score (À1.99 vs. À1.79,P = 0.214).At 120.9 HU, the pooled mean CT attenuation at L1-4 was significantly lower in the fracture group compared to 151.5 HU in the non-fracture group (P < 0.001).Attenuation values were also lower in the fracture group at all three individual kVp settings.

Discussion
CT scans frequently cover the lumbar spine, as such offering the opportunity to evaluate a patient's bone density status without adding expense or radiation exposure.Previous research has demonstrated significant correlations between CT attenuation values at lumbar vertebrae and BMD as measured by DEXA.However, all of these studies [2][3][4][5][6][7][8][9][10][11] have utilised the same CT kilovoltage setting of 120 kVp, when in clinical practice, patients are imaged on different CT scanning systems and at different X-ray tube voltages (kVp).In addition, there is a trend to perform CT body scans at lower kVp to reduce ionising radiation and contrast agent volumes.As the attenuation in most tissues decreases within increasing kVp, we aimed at providing voltage-specific, probability-optimised thresholds for the identification of persons likely to have low BMD on DEXA scanning.In a first step, we confirmed an inverse relationship between CT X-ray tube voltage and lumbar vertebral attenuation.L1 and mean L1-4 trabecular bone decreased by 24.9 and 25.3 HU, respectively, when the kilovoltage peak increased from 100 to 120 kVp.Similarly, Garner et al. found the linear attenuation of L1 fell by 45.5 HU when Xray tube voltage increased from 100 to 140 kVp, and by 76.4 HU when changed from 80 to 140 kVp.Importantly, attenuation of bone is known to be more affected by tube voltage than soft tissues or fluids. 12Consequently, previously published attenuation thresholds for estimating BMD can only be applied to CT scans performed at 120 kVp.
In a second step, we then established the optimal HU thresholds at L1 and mean L1-4 for identifying patients likely to have DEXA T-scores of À2.5 or less at different tube voltages.Of note, while L1 threshold values were relatively similar for CT scans performed at 100 kVp and dual energy (80 kVp/140 kVp), they were significantly different for scans performed at 120 kVp.Based on these voltage-specific thresholds, we found that the AUCs for identifying patients with a DEXA T-score of À2.5 or less to range between 0.743 and 0.925.Thresholds were again different when using the mean L1-4 attenuation, but AUC values were similar (range: 0.707-0.933).
Depending on demographic and screening goals, different attenuation thresholds can be employed.For instance, with 100 kVp scans choosing a lower L1 HU threshold such as <145 HU (approximating to 90% specificity) can help to rapidly identify at risk individuals where further assessment is warranted given this threshold is fairly specific for predicting low T-score at DEXA.Conversely, a higher L1 threshold such as >180 HU (approximating to 90% sensitivity) would help to rule out the majority of low BMD cases and avoid an unnecessary DEXA.
When it comes to measuring CT attenuation, L1 is often preferred over mean value of L1-4 for several reasons.These include ease and convenience for the radiologist to determine the attenuation value, less susceptibility to degenerative changes and inclusion in both chest and/or abdominal CT scans. 17We also found threshold values and AUCs for L1 and mean L1-4 did not differ significantly.More recent studies have focused solely on L1 for these reasons.Pickhardt et al. screened over 1,800 subjects in one the largest studies so far focusing on the L1 vertebral body within abdominal CT scans performed at 120 kVp.The authors reported that a threshold of <160 HU was 90% sensitive for diagnosing a DEXA T-score of À2.5 or less, while a threshold of <110 HU was 90% specific for the same outcome.A HU value of <135 achieved balanced sensitivity and specificity. 3A subsequent study by Li et al. (n = 109) using CT settings similar to Pickhardt et al. found a near identical L1 threshold value of <136 HU optimising the ROC. 11We found our optimal L1 threshold was comparable at <128 HU for the subgroup scanned at 120 kVp.
CT is superior to DEXA in identifying morphological spine fractures which, if sustained after a low energy trauma, make the diagnosis of osteoporosis likely regardless of the DEXA result.Our study showed that the majority of subjects (63%) with fractures had DEXA Tscores above À2.5 and there was no difference in the mean DEXA BMD or T-scores between the fracture and non-fracture group.This highlights the limitation of DEXA particularly with BMD values becoming falsely elevated due to degenerative changes.In fact, many patients with T-scores above À2.5 go on to sustain minimal trauma fractures. 14It is well established that BMD, measured by any technique, is not the sole determinant of fracture risk and that any technical measurement needs to be interpreted in the context of clinical risk factors.In addition, fracture risk calculators such as the Fracture risk assessment tool or Garvan risk calculator may be useful in assessing absolute fracture risk. 18e found mean L1-4 attenuation was significantly lower in the fracture group compared to the non-fracture group (120.9HU vs. 151.5HU).This supports clinical experience where in certain cases, CT lumbar spine attenuation may provide a more accurate representation of bone density than DEXA.However, CT lumbar attenuation values have not been validated in terms of future fracture risk and cannot be used with fracture risk calculators.Therefore, the American College of Radiology Appropriateness Criteria 2022 recommends that measurement of lumbar attenuation by conventional CT scans should not be considered a substitute for the evaluation of BMD by DEXA. 19ur study has several limitations.Firstly, the nature of the single centre retrospective study may have limited the  generalisability of our results; larger cohorts and external validation is required.Secondly, both contrast and noncontrast scans were included in our study which may have introduced some heterogeneity into our results.However, we found AUC values for predicting DEXA T-score of À2.5 or less were similar for contrast and non-contrast series.4][5] Thirdly, we included several densitometers and CT scanners, although no difference in the mean T-score or L1 Hounsfield unit was found; and the machines were calibrated regularly using industry and institutional standards.Nonetheless, phantom-based cross-calibration between machines is warranted. 20Fourthly, the two CT scanners used had different methods for producing and displaying dual-energy images, which could impact on the threshold scores.Finally, a clinical review could not be performed to determine whether the fractures found were truly minimal trauma fractures which would reflect bone health independent of the measured BMD.
Our study has significant strengths.To the best of our knowledge, this is the first study to report L1 and mean L1-4 HU threshold values for identifying patients likely to have DEXA T-scores of À2.5 or less at kilovolt settings outside of 120 kVp.We are also the first to publish threshold values in an Australian local population.We limited our time interval between CT and DEXA scans to 6 months to minimise possible changes in bone health or density.Finally, our study provides opportunity for prospective studies to investigate the cost-effectiveness of referring patients who meet the screening threshold on CT scan for additional DEXA scanning.

Fig. 1 .
Fig. 1.HU measurement on CT abdomen and pelvis performed at 100 kVp.Oval region of interest (ROI) was placed over axial section of L1 midbody showing attenuation value of 88 HU.

Table 1 .
Baseline characteristics of patients included in the study.Data reported as number (%) and standard deviation unless otherwise stated

Table 2 .
Lumbar spine CT attenuation and X-ray tube voltages by DEXA T-score © 2023 The Authors.Journal of Medical Imaging and Radiation Oncology published by John Wiley & Sons Australia, Ltd on behalf of Royal Australian and New Zealand College of Radiologists.

Table 3 .
Threshold HU values at L1 for patients with a DEXA T-score of À2.5 or less

Table 4 .
Threshold HU values at L1-4 for patients with a DEXA T-score of À2.5 or less