Bone Mineral Density in Dual Energy X-Ray Absorptiometry and Hounsfield Units on Diagnostic Computed Tomography Scan: A Cross-Sectional Study

Abstract

Background

Previous studies have demonstrated a moderate correlation between bone mineral density (BMD) measured by dual energy X-ray absorptiometry (DXA) and Hounsfield units (HU) assessed on diagnostic computed tomography (CT), primarily in the lumbar spine. However, the correlation between HU values and DXA scores in other bones and at different anatomical landmarks remains underexplored. This study aimed to evaluate the relationship between HU values at various cortical and medullary regions and DXA-derived T-scores across multiple skeletal sites.

Methods

In this retrospective cross-sectional study, 214 patients who underwent both DXA and diagnostic CT imaging were included. The bones analyzed included the proximal humerus (N=96), distal tibia (N=56), distal femur (N=34), and proximal tibia (N=28). HU measurements were taken at five anatomic landmarks: the medullary center, anteromedial cortex, anterolateral cortex, posteromedial cortex, and posterolateral cortex. Correlations were analyzed between these HU values and the lowest T-score from the DXA scan.

Results

Significant moderate correlations were found between medullary HU values and the lowest T-scores in the proximal tibia (r=0.486, P=0.04) and distal tibia (r=0.458, P=0.01). In the proximal humerus, the anteromedial cortex HU also correlated with the lowest T-score (r=0.353, P=0.01). Mean HU values at several cortical and medullary regions significantly differed between osteoporotic and normal BMD groups.

Conclusions

There is a moderate correlation between HU values at specific anatomical landmarks and DXA T-scores. These findings suggest that diagnostic CT scans may serve as a supplementary tool in assessing bone quality, especially when DXA is not available.

GRAPHICAL ABSTRACT

INTRODUCTION

Osteoporosis is a systemic skeletal disorder characterized by low bone density and destruction of the bone microarchitecture, resulting in increased bone fragility.[ 1] The global prevalence of osteoporosis is reported to be 23.1% in women and 11.7% in men.[2] Along with the continuous increase in population aging, the number of patients living with osteoporosis also increased, so the age-standardized prevalence of osteoporosis in Korea is reported to be 62.7% in women and 24.6% in men.[3] A higher prevalence of osteoporosis is associated with a higher incidence of fragility fractures and a greater social, health, and economic burden.[4]

Currently, dual energy X-ray absorptiometry (DXA) is considered the gold standard method for determining bone mineral density (BMD).[5] However, the use of DXA in determining BMD is not without limitations. The accuracy of DXA may be impaired by various factors, such as osteoarthritis [6] or prior surgical interventions involving metallic implants, particularly bilateral hip arthroplasty, which can substantially alter attenuation values and reduce measurement accuracy. In addition, the BMD of obese osteoporotic patients may be classified as normal on DXA scan. [7] Therefore, finding alternative modalities to DXA has been an ongoing concern for researchers in this field.

In recent years, there has been a growing interest in the use of alternative imaging, such as computed tomography (CT) and magnetic resonance imaging (MRI), to assess BMD.[8] The use of CT for the measurement of bone density was first proposed in the mid-70s.[9] Despite its initial accuracy, a high dose of ionizing radiation is a main concern for the clinical implications of CT in routine BMD assessment.[ 9] Even so, a CT scan can still be used for the assessment of BMD in patients who have undergone CT scanning for other purposes, such as fracture diagnosis.

In this study, we evaluated the correlation between bone density obtained from DXA (T-score) and CT (Hounsfield units [HU]) in patients for whom a CT scan had already been obtained for diagnostic purposes. We hypothesized that if there is a significant correlation between bone density assessed by CT and DXA, available CT scans can be used to determine bone density, thereby reducing patient radiation exposure and the cost of medical services.

METHODS

This cross-sectional study was approved by the Institutional Review Board (P01-202201-01-026). As this was retrospective anonymized study, the requirement for informed consent was waived by the ethical committee of Busan Medical Center, and data collected between 2019 and 2023 were retrospectively reviewed. Inclusion criteria were available CT scan of the proximal humerus or distal femur or proximal tibia or distal tibia, available DXA scan for three regions of interest (spine, femoral neck, and distal radius), a maximum interval of one year between the date of DXA and CT performance median interval between DXA and CT as 127 days (interquartile range, 71–195 days). Patients were excluded if they had a fracture on the CT scan used for HU measurements. A total of 214 patients who met the study criteria were included in the analysis.

1. DXA

BMD of the spine (L1–L4), femoral neck, and distal radius was evaluated using a DXA device (Hologic Inc., Bedford, MA, USA). BMD was expressed in grams per square centimeter (g/cm²), and the T-score, a dimensionless standardized index, was also recorded. A T-score greater than −1 was considered normal bone density, a T-score between −1 and −2.5 was osteopenia, and a T-score less than 2.5 was considered osteoporosis.[10] The lowest T-score was used in the analysis.

2. Quantitative CT scanning

All CT scans were obtained by a 128-channel helical CT scanner (Ingenuity; Phillips Healthcare, Best, The Netherlands). The bone density was evaluated in five different points, including the anteromedial corner, anterolateral corner, posteromedial corner, posterolateral corner, and center of medulla. The intersection of the mediolateral and anteroposterior lines was considered the center of the medulla. At each anatomic point, HU was measured three times, and the average value was used for analysis. The picture archiving and communication system (GE Digital Healthcare Inc., Chicago, IL, USA) was used to calculate the HU values. To this aim, an elliptical region with a diameter of 1 mm was placed over the area of interest.[11]

All CT scans were obtained without the use of intravenous contrast agents to avoid HU alterations due to enhancement. Image reconstruction was performed using a standard bone algorithm kernel with routine scanner calibration. Beam-hardening correction was not applied. For HU measurement, an elliptical region of interest (ROI) with a diameter of approximately 2 mm was placed over each anatomical landmark. This ROI size was selected to minimize image noise while ensuring precision in small cortical areas. To reduce the heterogeneity, some evaluation landmarks were defined for each anatomic location. For the proximal humerus, we went from the proximal to the distal sections in the axial CT scan, and the first axial cut in which the humerus head was no longer visible was used for the evaluation of bone density. For the distal femur, we went from the distal to proximal sections in the axial CT scans, and the first axial cut in which the femoral condyle was not observable was used for the evaluation of bone density (Fig. 1). For the proximal tibia, we went from the distal to proximal sections in the axial CT scans, and the first axial cut in which the tibial tubercle was not identified was used for the bone density evaluation. For the distal tibia, the axial CT section 2.5 cm above the distal tibial joint was used for the evaluation of bone density.[11]

Fig. 1.

(A) Identification of the center of medulla in the axial computed tomography (CT) scan of the distal femur. (B) Calculation of Hounsfield unit of the center of medulla in the axial CT scan of the distal femur. (C) Calculation of Hounsfield unit of the cortex in the axial CT scan of the distal femur. Std Dev, standard deviation.

3. Sample size and statistical analysis

Choi et al. [12] reported a correlation of 0.447 between the T-score and HU values of the total spine. Based on this correlation, effect size=0.07, power of 95%, and type 1 error 0.05, a number of 20 patients was sufficient to conduct this correlation study using G*Power software. To increase the power of the study, we included all patients referred to our center during the study period.

Statistical analysis of the data was conducted using SPSS for Windows, version 16 (SPSS Inc., Chicago, IL, USA). Descriptive statistics were demonstrated by mean±standard deviation for quantitative variables or numbers with percentages for qualitative variables. A one-way ANOVA test or its nonparametric counterpart (Kruskal-Wallis test) was used to compare mean values among groups. Pearson or Spearman correlation coefficient were used to evaluate potential correlations between CT and DXA measurements. A P-value of less than 0.05 was considered statistically significant.

RESULTS

The study population included 56 males and 158 females with a mean age of 63.5±10 years (range, 51–90). The mean lowest T-score of the patients was −3.1±1.3. Accordingly, the majority of included patients were osteoporotic (136 of 214). The most frequent location of CT was the proximal humerus (96 or 214). Characteristic features of the patients are demonstrated in Table 1.

Table 1.

Characteristic features of included patients

Variable	Value (N=214)
Age (yr)	63.5±10.0
Sex
Male	56 (26.2)
Female	158 (73.8)
The lowest T-score of the three region	−3.1±1.3
Bone status according to T-score
Normal	6 (2.8)
Osteopenia	72 (33.6)
Osteoporosis	136 (63.6)
CT location
Proximal humerus	96 (44.9)
Distal femur	34 (15.9)
Proximal tibia	28 (13.1)
Distal tibia	56 (26.2)

The data is presented as mean±standard deviation or N (%).

CT, computed tomography.

1. Association between the lowest T-score and HU of the proximal tibia

In the proximal tibia, the HU of the center of the medulla was significantly correlated with the lowest T-score of the patients (P=0.04). The HUs of the other landmarks were not significantly correlated with the T-score (Table 2). The mean HU of the medullary center was significantly lower in patients with osteoporosis (P=0.036, Table 3). Also, a ROC curve analysis was performed to determine a potential cutoff for HU in the proximal tibia medullary region. The Youden index yielded a threshold of 45.3 HU (area under the curve=0.732, sensitivity=0.71, specificity= 0.68).

Table 2.

Correlation between the lowest T-score and Hounsfield units of different anatomic landmarks in the proximal tibia

Anatomic landmark	N	Minimum	Maximum	Mean±SD	R	P-value
Anteromedial cortex	28	266	1,466	595.9±296.0	−0.072	0.80
Anterolateral cortex	28	326	1,002	698.3±240.0	0.025	0.93
Posterolateral cortex	28	222	1,368	784.0±342.0	0.263	0.36
Posteromedial cortex	28	315	1,237	822.0±348.0	0.192	0.51
Center of medulla	28	−70	136	30.7±47.4	0.486	0.04

SD, standard deviation; R, correlation coefficient.

Table 3.

Association between Hounsfield units of different anatomic landmarks in the proximal tibia and bone status based on T-score

Anatomic landmark	Bone status	Mean±SD	P-value
Anteromedial cortex	Osteopenia	445.6±59.0	0.16
	Osteoporosis	679.4±344.0
Anterolateral cortex	Osteopenia	695.2±274.0	0.97
	Osteoporosis	700.1±237.0
Posterolateral cortex	Osteopenia	863.8±272.0	0.53
	Osteoporosis	739.8±383.0
Posteromedial cortex	Osteopenia	893.6±344.0	0.58
	Osteoporosis	782.2±364.5
Center of medulla	Osteopenia	52.4±49.0	0.036
	Osteoporosis	18.8±44.6

SD, standard deviation.

2. Association between the lowest T-score and HU of the distal femur

In the distal femur, the HU at none of the anatomic landmarks was significantly correlated with the lowest T-score (Table 4). The mean HU of the center of the medulla was not significantly different between the osteopenic and osteoporotic patients (Table 5).

Table 4.

Correlation between the lowest T-score and Hounsfield units of different anatomic landmarks in the distal femur

Anatomic landmark	N	Minimum	Maximum	Mean±SD	R	P-value
Anteromedial cortex	34	173	949	476.1±240.0	0.002	0.99
Anterolateral cortex	34	212	1,029	471.6±245.0	0.195	0.84
Posterolateral cortex	34	410	1,369	822.0±272.0	−0.053	0.84
Posteromedial cortex	34	322	1,285	659.4±280.0	0.149	0.56
Center of medulla	34	−29	86	25.4±37.0	−0.068	0.74

SD, standard deviation; R, correlation coefficient.

Table 5.

Association between Hounsfield units of different anatomic landmarks in the distal femur and bone status based on T-score

Anatomic landmark	Bone status	Mean±SD	P-value
Anteromedial cortex	Osteopenia	440.8±181.0	0.62
	Osteoporosis	500.7±281.0
Anterolateral cortex	Osteopenia	496.2±329.0	0.74
	Osteoporosis	454.3±184.0
Posterolateral cortex	Osteopenia	876.2±250.0	0.51
	Osteoporosis	784.0±294.0
Posteromedial cortex	Osteopenia	651.4±317.0	0.92
	Osteoporosis	665.0±268.0
Center of medulla	Osteopenia	15.7±48.0	0.28
	Osteoporosis	32.3±27.5

SD, standard deviation.

3. Association between the lowest T-score HU of the proximal humerus

In the proximal humerus, the HU of the anteromedial cortex was significantly correlated with the lowest T-score of the patients (P=0.01). The HUs of the other landmarks were not significantly correlated with the T-score (Table 6). The mean HU of the posterolateral cortex was significantly smaller in the osteoporotic patients (P=0.003; Table 7).

Table 6.

Correlation between the lowest T-score and Hounsfield units of different anatomic landmarks in the proximal humerus

Anatomic landmark	N	Minimum	Maximum	Mean±SD	R	P-value
Anteromedial cortex	96	342	1,311	875.5±220.0	0.353	0.01
Anterolateral cortex	96	290	1,720	982.5±279.0	0.212	0.14
Posterolateral cortex	96	645	1,693	1,081.2±272.0	0.180	0.22
Posteromedial cortex	96	394	1,622	961.1±280.0	0.113	0.44
Center of medulla	96	−79	161	−0.3±47.0	0.254	0.08

SD, standard deviation; R, correlation coefficient.

Table 7.

Association between Hounsfield units of different anatomic landmarks in the proximal humerus and bone status based on T-score

Anatomic landmark	Bone status	Mean±SD	P-value
Anteromedial cortex	Osteopenia	952.4±176.0	0.23
	Osteoporosis	835.8±236.0
Anterolateral cortex	Osteopenia	1,043.9±222.0	0.31
	Osteoporosis	961.3±303.0
Posterolateral cortex	Osteopenia	1,266.4±234.0	0.003
	Osteoporosis	995.7±285.0
Posteromedial cortex	Osteopenia	1,009.3±272.0	0.39
	Osteoporosis	946.6±285.0
Center of medulla	Osteopenia	6.1±34.0	0.18
	Osteoporosis	−13.0±51.0

SD, standard deviation.

4. Association between the lowest T-score and the HU of the distal tibia

In the distal tibia, the HU of the center of the medulla was significantly correlated with the lowest T-score of the patients (P=0.01). The HUs of the other landmarks were not significantly correlated with the T-score (Table 8). The mean HU of the posteromedial cortex and center of the medulla were significantly larger in patients with normal BMD (Table 9).

Table 8.

Correlation between the lowest T-score and Hounsfield units of different anatomic landmarks in the distal tibia

Anatomic landmark	N	Minimum	Maximum	Mean±SD	R	P-value
Anteromedial cortex	56	178	1,218	621.1±251.0	0.172	0.38
Anterolateral cortex	56	348	1,524	826.3±315.0	0.065	0.74
Posterolateral cortex	56	269	1,368	759.2±260.0	0.178	0.36
Posteromedial cortex	56	217	1,232	682.1±304.0	0.306	0.11
Center of medulla	56	−13	301	132.0±89.0	0.458	0.01

SD, standard deviation; R, correlation coefficient.

Table 9.

Association between Hounsfield units of different anatomic landmarks in the distal tibia and bone status based on T-score

Anatomic landmark	Bone status	Mean±SD	P-value
Anteromedial cortex	Osteopenia	624.7±231.0	0.22
	Osteoporosis	587.2±236.0
	Normal	912.0±432.0
Anterolateral cortex	Osteopenia	854.1±378.0	0.74
	Osteoporosis	798.0±295.0
	Normal	970.5±378.0
Posterolateral cortex	Osteopenia	680.3±247.0	0.58
	Osteoporosis	785.0±272.0
	Normal	843.5±253.0
Posteromedial cortex	Osteopenia	567.1±184.0	0.04
	Osteoporosis	679.8±317.0
	Normal	1,161.0±35.0
Center of medulla	Osteopenia	147.5±70.0	0.04
	Osteoporosis	110.8±88.0
	Normal	260.5±57.0

SD, standard deviation.

DISCUSSION

In this study, we evaluated the correlation between BMD obtained from DXA (T-score) and HU values measured from diagnostic CT images. HU values were calculated at five anatomical landmarks (four cortical and one medullary site) in four bones: distal tibia, proximal humerus, distal femur, and proximal tibia. The HU of the medullary center in the proximal and distal tibia showed significant positive correlations with the lowest T-score. In the proximal humerus, the HU of the anteromedial cortex also showed a significant correlation with T-score. Additionally, osteoporotic patients had significantly lower HU values at the posterolateral cortex of the proximal humerus, the medullary center of the proximal tibia, and both the posteromedial cortex and medullary center of the distal tibia compared to patients with normal bone density.

Several prior studies have investigated the correlation between CT-derived HU and DXA-based BMD, primarily in the spine. Schreiber et al. [13] first introduced the use of HU values on conventional CT scans as a proxy for BMD. In their study of lumbar vertebrae, HU values in the medullary space were moderately correlated with BMD and T-score (r²=0.44–0.48), and were significantly higher in patients with normal bone density compared to those with osteoporosis.[13] Our findings extend this concept to peripheral skeletal sites, showing a comparable pattern of stronger HU–T-score correlations in trabecular-rich regions such as the proximal tibia.

While many studies have focused on vertebral HU, few have addressed peripheral bones. Our findings provide novel evidence that diagnostic CT scans of peripheral sites—often acquired for other clinical reasons—may provide valuable information on bone health. Given the practical implications for cost and radiation reduction, further prospective validation is warranted.

The correlation between the bone densitometry assessed by DXA and CT scan has been investigated in several earlier studies. In 2011, Schreiber et al. [13], for the first time, introduced the HU as a tool for determining bone density on conventional CT scans without exposing patients to the high radiation dose used in Quantitative CT scans. Twenty-five patients undergoing lumbar spine DXA and CT were included in this study. ROI was confined to the medullary space of the vertebral body to minimize the beam hardening and volume averaging by the adjacent cortical bone. The HU values were significantly correlated with both BMD (r²=0.44) and T-score (r²=0.48). The mean HU was significantly greater in the normal population and smaller in the osteoporotic population.[13] In the present study, we evaluated HU values in the cortexes, in addition to the medulla. The most significant correlation was observed between the medullary HU of the proximal tibia and the lowest T-score (r=0.486). Similarly, we observed significantly higher HU values in the normal population and lower HU values in the osteoporotic population. One possible explanation for the strong correlation observed between the medullary HU of the proximal tibia and the lowest T-score lies in the anatomical and biomechanical characteristics of this region. The proximal tibia contains a large volume of trabecular (cancellous) bone, which is metabolically more active and sensitive to systemic changes in bone turnover compared to cortical bone. Moreover, as a major load-bearing site in the lower extremity, the proximal tibia is subjected to substantial mechanical stress during weight-bearing activities, which may contribute to earlier or more pronounced bone loss in osteoporotic conditions. These structural and mechanical factors may enhance the sensitivity of HU measurements in this region as a surrogate marker of systemic BMD.

Lee et al. [5] evaluated the correlation between BMD measured by DXA and HU measured by diagnostic CT in the lumbar spine. The HU was assessed in the middle of the vertebral body, not in the cortex. The HU values were significantly smaller at higher ages (P<0.001). The HU values showed a significant positive correlation with L1–4 T-score (r ranged between 0.673 and 0.794). The HU of the L1–4 vertebrae was considerably higher in the normal population and remarkably smaller in the osteoporotic population.[5] Similar to the study of Lee et al. [5], the mean HU of the medulla (in the distal tibia) was significantly higher in the normal population of the present study. Likewise, we found a significant positive correlation between the T-score and Medulla HU of the proximal and distal tibia. However, the observed correlation was weaker in the present study, which may be attributed to differences in the selected bones.

The correlation between bone density assessed by DXA and HU measured on diagnostic CT has also been evaluated in several studies. Ahmad et al. [14] performed an updated systematic review of correlations between BMD from CT and DXA. A total of 15 CT studies were included in this review. The pooled correlation coefficient of vertebral HU versus vertebral DXA, vertebral HU versus hip DXA and vertebral HU versus lowest T-score were 0.600, 0.500, and 0.600, respectively. The correlations were worse in the degenerative spine population. They concluded that there is only a moderate correlation between the DXA scores and HU measured on diagnostic CT, and at this time, it is not clear which modality is a better representative of bone quality. They attributed this uncertainty to inconsistencies in cohorts, imaging timing, and measurement protocols, which may contribute to the observed heterogeneity.[14] In the present study, for the first time, we evaluated the correlation of T-score with HU values of different bones to determine which bone shows the most significant correlation with the lowest T-score, which turned out to be the proximal tibia. Similarly, we found a moderate correlation between the T-score with HU values.

The present study was not without limitations. Since we included patients referred to the Fracture Liaison Service, the majority of the patients were in the osteopenia or osteoporosis group, and the number of patients with normal BMD was very small. In addition, all the CT and DXA images in the present study were obtained by the same scanner. Therefore, the results might not be reproducible in other CT and DXA scanners. One of the study limitations is the relatively long interval allowed between DXA and CT scans (up to one year), which could introduce temporal variability in BMD, especially in patients undergoing treatment for osteoporosis. Finally, HU values were calculated separately in trabeculated bone (medulla) and cortical bone, whereas DXA values include both cortical and trabeculated bone. This difference may weaken the correlation strength between CT and DXA and should be addressed in future studies.

A significant moderate correlation was found between HU values of different bones at various anatomic landmarks and the lowest T-score of the patients. The strongest correlation was found between the medullary HU of the proximal tibia and the lowest T-score. The mean HU at some landmarks was associated with the bone status evaluated by DXA, so the medullary HU was significantly smaller in the osteoporotic patients and larger in normal population. These findings highlight the value of HU in determining bone quality. However, more standardizations are required before the clinical implications of HU for determining bone quality.