Effects of contrast on opportunistic computed tomography vertebral bone mineral density analysis using internal calibration

Abstract

Opportunistic CT, which repurposes clinical CT scans, has the potential to aid in osteoporosis screening and diagnosis. To convert CT scans to bone-equivalent densities without a calibration phantom, internal (or phantomless) calibration methods have been developed, using referent tissues of known density within the body. A key limitation is that intravenous iodinated contrast enhancement, which is routinely used in clinical settings, impacts quantitative measurements derived from these phantomless calibration techniques. Therefore, this cross-sectional study examined the influence of contrast enhancement on internal reference tissues and volumetric BMD (vBMD) measures derived using phantomless internal calibration. Hematuria protocol CT scans (N = 194) were retrospectively analyzed. This protocol included unenhanced and 2 enhanced CT scans from the nephrographic (~90 s delay) and urographic (~10 min delay) phases. The vertebrae were automatically identified and segmented with labels for the vertebral body and trabecular bone compartment in the third lumbar vertebra. Internal density calibration was performed using our previously established method with internal referent tissues (air, adipose, skeletal muscle, and cortical bone) to estimate the effective energy of each scan. A linear correction, derived based on training data from the nephrographic phase, was applied to test datasets from both the nephrographic and urographic phases. The nephrographic phase increased the vertebral body vBMD by 9.90 ± 4.28 mg/cm³ (5.52%, p < .0001) and the urographic phase by 2.33 ± 3.26 mg/cm³ (1.33%, p < .0001) compared to unenhanced scans. Applying the linear equation derived from the nephrographic training data reduced vBMD differences in the nephrographic test data but increased the variation in the urographic phase vBMD, demonstrating the need for phase-specific correction. This study demonstrates that vBMD measurements from contrast-enhanced CT are subject to significant error unless the interval between contrast administration and scan acquisition can be accounted for.

Introduction

Opportunistic CT leverages CT images acquired for a variety of clinical reasons to perform secondary analysis of skeletal health, including the assessment of volumetric BMD (vBMD; mg/cm³), which can help with earlier detection of osteoporosis.^1,2 This is particularly valuable given that the number of DXA screenings remains far lower than the frequency of CT scans being performed for various other diagnoses, creating an opportunity to assess bone health without additional imaging or radiation exposure.³

Traditionally, vBMD analysis requires using a density calibration phantom placed in the scan field of view (FOV) to convert the native scan Hounsfield Units (HUs) to bone equivalent densities. In clinical practice, however, phantoms are rarely included in scan FOV due to practical constraints. As a result, internal density calibration methods have been developed, which estimate vBMD using internal reference tissues of known densities, such as air, adipose, skeletal muscle, blood, and/or cortical bone. These methods either apply a scan-specific equation derived from pre-calibrated reference datasets^4–7 or estimate the scan’s effective energy derived from voxel-specific calibration.⁸ However, a major challenge for opportunistic CT analysis is that over 50% of all clinical CT scans are contrast-enhanced,^9,10 and this artificially alters the tissue densities required for accurate internal calibration.

The influence of contrast enhancement on density measures has been investigated at both the lumbar spine and proximal femur using phantom-based calibration.^11–15 Overall, vBMD estimates derived from contrast-enhanced CT scans are highly correlated with unenhanced CT scans and DXA T-scores; however, vertebral measurements are more substantially affected by contrast than those at the femur.^14,16 Additionally, the contrast phase (arterial vs venous) can significantly influence vBMD measures.^11,17 Whether these phases can be used interchangeably for vBMD estimation with internal density calibration remains unclear, and the effects of contrast on the reference tissues used to estimate scan effective energy in voxel-specific internal calibration are not well understood.

This study aimed to quantify the error in vBMD estimates obtained using our voxel-specific internal calibration method from contrast-enhanced CT scans, using a standardized hematuria protocol that includes an unenhanced scan followed by multiple intravenous iodinated contrast-enhanced scans within the same exam acquisition. By characterizing the impact of contrast enhancement on vBMD analysis, this work aims to further explore the potential of contrast-enhanced CT scans for opportunistic CT assessment.

Materials and methods

Image acquisition

This cross-sectional study used de-identified hematuria CT scans (N = 224) in patients over the age of 18 yr old that were retrospectively acquired for analysis through Alberta Health Services using the following health service codes corresponding with the hematuria protocol: plain abdomen, enhanced abdomen/pelvis, and delay. The CT scans were performed between January 1, 2018, to December 31, 2022, across 8 hospital sites on 3 CT scanner manufacturers, including GE Medical (Discovery CT750 HD, LightSpeed VCT, Revolution HD models), Toshiba (Aquilion One and Aquilion Prime models) and Siemens (Somatom Definition AS+, Somatom Definition Edge, and Somatom ForceT models). CT scans were excluded if there were imaging artifacts (eg, metal in the field of view, object outside the field of view) or inadequate referent tissue regions for internal calibration (eg, adipose infiltration into skeletal muscle, limited adipose). This study was approved by the Conjoint Health Research Ethics Board at the University of Calgary (REB22-0226).

The hematuria scan protocol included a non-contrast-enhanced scan immediately followed by dual-phased intravenous iodinated contrast-enhanced scans. The majority of examinations were performed using Iohexol (Omnipaque 350, GE Healthcare) as the contrast agent. In the presence of renal insufficiency (estimated glomerular filtration rate < 40 mL/min/1.73 m²), Iodixanol (Visipaque 320, GE Healthcare) may have been used at some sites. The standardized 2-phase contrast-enhanced scans include a nephrographic phase, approximately 90 s following the contrast injection, and a urographic phase, approximately a 10-min delay following the contrast injection.

Image processing

The third lumbar vertebra in each CT scan was automatically segmented using Anduin Bonescreen (https://anduin.bonescreen.de).¹⁸ This software utilizes a convolutional neural network-based framework trained on publicly available data to create 3 masks for each vertebra: (1) a full vertebral mask including the pedicles, (2) a vertebral body mask comprising the cortical shell and trabecular compartment (excluding the pedicles, referred to as vertebral body), and (3) a trabecular compartment mask. Because many participants exhibited very thin cortices, only the trabecular and vertebral body compartments were included in the final analysis of vBMD.

Internal calibration

Voxel-specific internal density calibration was performed for each CT scan using a previously established method by our group (code publicly available at: https://github.com/Bonelab/Ogo).⁸ In brief, internal ROIs for air, adipose, skeletal muscle, and blood were manually selected (Figure 1). The ROIs were placed on 3 consecutive slices at the L3 level to minimize the impact of image artifacts such as beam hardening and voltage modulation.¹⁹ To minimize partial volume effects when isolating cortical bone at L3 and to maintain consistency across scans with varying resolutions, the cortical bone ROI was extracted from the full vertebral mask (including the pedicles). Specifically, the highest density 40 mm³ voxels were sampled to define the cortical bone ROI. In cases where the cortical bone was especially thin, the cortical bone ROI volume was reduced to 20 mm³ to ensure accurate sampling of the available cortical bone. Previous work tested different combinations of internal referent tissues with this calibration method, demonstrating that a minimum of 4 tissues is recommended to reduce variability.¹⁹ As a result, blood was excluded from the effective energy estimation to prevent artificially inflated HU values due to contrast. The L3 vertebral masks (trabecular and vertebral body) were applied to the calibrated image, and vBMD was measured by summing the total voxels in the segmented ROI and dividing by the segmented ROI’s voxel count.

Figure 1.

Representative axial image for internal calibration regions of interest (ROI). ROIs were manually selected for air (yellow), adipose (red), skeletal muscle (teal), and blood (magenta) at the level of the L3 vertebra. The cortical bone ROI was automatically selected from the segmented L3 vertebra using volume thresholding to prevent partial volume artifacts.

Effects of contrast on vBMD

The effect of contrast enhancement on referent tissue (air, adipose, blood, cortical bone, and skeletal muscle) HUs and vBMD derived from internal calibration was evaluated across the entire dataset. Although blood was not included in the internal calibration, it was analyzed here to examine the influence of contrast enhancement. Linear correction factors were derived to evaluate the extent to which contrast-enhancement effects on vBMD could be reduced. For each contrast phase, scans were randomly divided into training (70%) and testing (30%) subsets. Linear regression models relating the unenhanced vBMD to the nephrographic and urographic phases were generated from the training sets and applied to the test sets. To assess the dependency of contrast-enhancement effects on phase, the linear equation derived from the nephrographic phase was applied to the urographic test data.

Statistical analysis

Within-participant comparisons were performed between unenhanced and contrast-enhanced nephrographic and urographic CT scans. Mean HUs for adipose tissue, skeletal muscle, cortical bone, and blood, as well as vBMD values from the whole-body and trabecular compartments, were compared across scan phases using one-way ANOVA with Tukey post-hoc tests to assess the impact of contrast enhancement. Bland–Altman analyses were conducted to examine the agreement of BMD measurements across all CT phases (unenhanced, nephrographic, and urographic),²⁰ with limits of agreement (LoA) calculated as the mean difference ± 1.96 SDs. Linear regression analysis was used to derive the coefficients of determination (R²) to evaluate the relationship between unenhanced scans and both nephrographic and urographic phases. All statistical analyses were performed using R version 4.3.1 (2023-06-16) and RStudio (version 2023.06.1+524). Data are reported as mean ± SD, and the significance level was set to p < .05.

Results

Image acquisition

Of the 224 CT scans identified, 17 scans were excluded due to incomplete series, and 13 scans were excluded due to insufficient tissue for calibration or image artifacts, leaving a total of 194 de-identified hematuria CT scans for analysis. There were 102 males (mean age 60.6 14.7 yr) and 92 females (mean age 59.6 14.3 yr). Most scans were performed on GE Healthcare scanners (171 scans), with the remaining acquired on Siemens (16 scans) and Toshiba (7 scans) (Table S1).

Effects of contrast on internal referent tissues

As expected, the HU for the air ROI did not differ across the 3 CT phases (Table 1, Figure 2A). Adipose HU increased in both the nephrographic (+5.93 HU [5.62%], p < .001) and urographic (+7.04 HU [6.71%], p < .001) phases relative to the unenhanced CT scan, with differences between the nephrographic and urographic phases (+1.11 HU [1.09%], p < .001, Figure 2B). Similarly, skeletal muscle HU was greater in both the nephrographic (+13.31 HU [25.16%], p < .001) and urographic (+12.75 HU [24.24%], p < .001) phases compared to the unenhanced CT scan, with no significant difference between contrast phases (Figure 2D). Cortical bone HU decreased in both the nephrographic (−11.10 HU [−0.88%], p < .001) and urographic (−10.69 HU [−0.85%], p < .001) phases compared to unenhanced, with no significant differences between the 2 contrast phases (Figure 2E). Although the blood ROI was excluded from the internal calibration, the blood HU increased substantially in both the nephrographic (+118.12 HU [121.53%], p < .001) and urographic (+52.47 HU [81.51%], p < .001) phases compared to the unenhanced CT scan (Figure 2C), with a significant difference also observed between the 2 contrast phases (nephrographic +65.65 HU [−53.19%], p < .001).

Table 1.

Mean Hounsfield units (HU) for each tissue region of interest and corresponding mean difference and percent changes in HU from the unenhanced CT scan during the nephrographic and urographic phases.

Tissue	Unenhanced mean HU	Nephrographic  HU (%)	Urographic  HU (%)
Air	−997.74	−0.29 (−0.03)	−0.31 (−0.03)
Adipose	−108.44	5.93 (5.62)	7.04 (6.71)
Blood	38.13	118.12 (121.53)	52.47 (81.51) a
Cortical bone	1260.40	−11.10 (−0.88)	−10.69 (−0.85)
Skeletal muscle	46.23	13.31 (25.16)	12.75 (24.24)

Bolded values indicate statistical significance (p < .0001) from unenhanced.

^aIndicates statistical significance between contrast phases (p < .0001).

Figure 2.

Internal regions of interest Hounsfield units (HU). Mean HUs of air (A), adipose (B), blood (C), skeletal muscle (D), and cortical bone (E) across the unenhanced, nephrographic, and urographic phases. Statistical significance represented as ^***p < .001.

Volumetric BMD

Vertebral body vBMD assessed in both the nephrographic (R² = 0.99; Figure 3) and urographic (R² = 0.99) phases was strongly correlated with vBMD assessed in the unenhanced CT scan (Table 2). Compared to the unenhanced scan, vBMD in the nephrographic phase was significantly higher (9.90 ± 4.28 mg/cm³ [LoA: 1.51 to 18.23 mg/cm³], p < .0001). Despite the 10-min delay, the urographic phase remained significantly different from the unenhanced phase, though the magnitude of the difference was smaller (2.33 ± 3.26 mg/cm³ [LoA: −4.06 to 8.72 mg/cm³], p < .0001). Direct comparison of the 2 contrast phases revealed a significant difference of 7.57 ± 3.99 mg/cm³ (LoA: −0.25 to 15.39 mg/cm³, p < .0001), highlighting phase-specific effects on vBMD.

Figure 3.

Pearson correlations and Bland–Altman for third lumbar vertebral body volumetric BMD (vBMD). Linear regression analyses comparing vertebral body vBMD between the unenhanced and nephrographic phases (A) and between the unenhanced and urographic phases (B), along with corresponding Bland–Altman plots illustrating differences between unenhanced vertebral body vBMD and the nephrographic (C) and urographic (D) phases.

Table 2.

Agreement and differences in volumetric BMD (vBMD) across contrast phases.

Compartment	Comparison	R 2	Mean difference    SD (mg/cm3)	Limits of agreement (mg/cm 3 )	p-Value
Vertebral body	Unenhanced vs nephrographic	0.99	9.90 ± 4.28	1.51 to 18.23	<.0001
	Unenhanced vs urographic	0.99	2.33 ± 3.26	−4.06 to 8.72	<.0001
	Nephrographic vs urographic	–	7.57 ± 3.99	−0.25 to 15.39	<.0001
Trabecular bone	Unenhanced vs nephrographic	0.98	10.68 ± 4.58	1.70 to 19.66	<.0001
	Unenhanced vs urographic	0.99	2.54 ± 3.44	−4.19 to 9.27	<.0001
	Nephrographic vs urographic	–	8.14 ± 4.25	−0.20 to 16.47	<.0001

Values are reported as mean difference SD. Limits of agreement represent that 95% Bland–Altman limits. R² values reflect linear correlation with the unenhanced CT reference.

Similarly, trabecular vBMD also demonstrated a strong correlation between both the nephrographic (R² = 0.98, Figure 4) and urographic (R² = 0.99) phases. Isolating trabecular bone resulted in a more pronounced effect. The nephrographic phase demonstrated a significantly higher trabecular vBMD compared to unenhanced CT (10.68 ± 4.58 mg/cm³, [LoA: 1.70 to 19.66 mg/cm³], p < .0001), while the urographic phase showed a smaller but still significant difference (2.54 ± 3.44 mg/cm³ [LoA: −4.19 to 9.27 mg/cm³], p < .0001). In trabecular bone, the difference between the nephrographic and urographic phases was larger than that observed for vertebral body vBMD (8.14 ± 4.25 mg/cm³ [LoA: −0.20 to 16.47 mg/cm³], p < .0001), indicating a greater phase-dependent effect when isolating trabecular bone.

Figure 4.

Pearson correlations and Bland–Altman for the third lumbar vertebra trabecular region volumetric BMD (vBMD). Linear regression analyses comparing trabecular vBMD between the unenhanced and nephrographic phases (A) and between the unenhanced and urographic phases (B), along with corresponding Bland–Altman plots illustrating differences between unenhanced trabecular vBMD and the nephrographic (C) and urographic (D) phases.

Correcting for contrast enhancement

The linear regression equations, derived separately for each vertebral region (vertebral body and trabecular compartment) from the nephrographic and urographic phases in the training dataset, were subsequently applied to the corresponding nephrographic and urographic test datasets. The application of the nephrographic linear equation reduced the mean error in vBMD in the vertebral body (0.60 ± 4.32 mg/cm³ [LoA: −7.86 to 9.06 mg/cm³], p = .55, Figure 5). Similarly, the application of the urographic linear equation reduced mean error in vBMD in the vertebral body (−0.30 ± 3.18 mg/cm³ [LoA: −6.53 to 5.92 mg/cm³], p = .75). The trabecular compartment demonstrated similar results in both the nephrographic (0.55 ± 4.79 mg/cm³ [LoA: −8.84 to 9.94 mg/cm³], p = .66, Figure 6) and the urographic phase (−0.19 ± 3.32 mg/cm³ [LoA: −6.70 to 6.32 mg/cm³], p = .90, Figure 6). In contrast, applying the linear equation derived from the nephrographic phase to the urographic test data increased the variation in both vertebral body vBMD (−7.44 ± 3.10 mg/cm³ [LoA: −13.52 to −1.35 mg/cm³], p < .0001) and trabecular vBMD (−7.76 ± 3.27 mg/cm³ [LoA: −14.18 to −1.35 mg/cm³], p < .0001) from the unenhanced CT scan.

Figure 5.

Correcting volumetric BMD (vBMD) for contrast enhancement in the third lumbar vertebra vertebral body using the linear equation derived from nephrographic data. Linear regression analyses comparing both corrected and uncorrected vertebral body vBMD between the unenhanced and nephrographic phases (A) and between the unenhanced and urographic phases (B), with corresponding Bland–Altman plots illustrating the differences between unenhanced vBMD and the corrected and uncorrected nephrographic (C) and urographic (D) phase measurements.

Figure 6.

Correcting volumetric BMD (vBMD) for contrast enhancement in the third lumbar vertebra trabecular region using the nephrographic linear equation. Linear regression analyses comparing both uncorrected and corrected trabecular vBMD between the unenhanced and nephrographic phases (A) and between the unenhanced and urographic phases (B), with corresponding Bland–Altman plots illustrating the differences between unenhanced trabecular vBMD and the uncorrected and corrected nephrographic (C) and urographic (D) phase measurements.

Discussion

This study establishes how contrast administration and the CT acquisition phase influence voxel-specific internal calibration, demonstrating their impact on reference tissues and resulting vBMD measurements. During the nephrographic phase (~90 s post-injection), contrast enhancement led to increased vBMD values across skeletal regions of the L3 vertebra (both trabecular and vertebral body compartments), as well as in referent tissues including adipose, skeletal muscle, and blood. In contrast, urographic phase (~10 min post-injection) mean vBMD values closely matched those from unenhanced scans. The effects observed in both contrast-enhanced phases were reduced with phase-specific linear adjustments; however, applying a uniform correction to urographic data increased variability, and even corrected contrast-enhanced measurements may still result in substantial errors in an individual’s vBMD.

Our findings are consistent with prior work in the field using alternative calibration methods, confirming that contrast leads to overestimation of vBMD and that scan timing relative to contrast administration is critical.^11–16,21 In their systematic review, Kutleša and colleagues²² reported spinal vBMD increases of 10%–15% in the venous phase following contrast administration. These results are higher but generally in line with the increases we observed at the L3 vertebra during the nephrographic phase, both in the trabecular compartment (10.68 mg/cm³ [8.27%]) and in the whole vertebral body (9.90 mg/cm³ [5.52%]). Comparisons with studies using other internal calibration approaches show that contrast effects on lumbar spine vBMD were generally larger than those observed here. In our analysis, the effect was generally smaller than reported by Bauer and colleagues (30.3%)¹⁵ and Engelke and colleagues (13.1%),²¹ both based on scans acquired 60-90 s after contrast administration, and comparable to that reported by Kaesmacher and colleagues (9%),¹³ which was measured during the portal venous phase (approximately 70 s post-contrast). The influence of contrast on internal calibration tissues was also consistent with prior literature, with slight increases in both skeletal muscle and adipose tissue, and markedly larger increases observed in blood.²¹ Interestingly, cortical bone demonstrated a slight decrease in HU after contrast enhancement. This observed decrease may be due to beam hardening from the high-density contrast enhancement (Figure S1, Table S2).²³ Despite these tissue-level changes, estimated effective energy remained largely stable, as the internal calibration is anchored by the relatively unaffected reference tissues of air and cortical bone (Table S3). Taken together, these findings demonstrate that contrast enhancement can lead to substantial alterations in vBMD measurements when using voxel-based internal calibration.

Although all contrast phases differed significantly from the unenhanced phase, the magnitude of these differences may not be clinically meaningful in every context. For example, the urographic phase was only 2.33 mg/cm³ greater than the unenhanced phase—a difference sufficiently small (1.33%) that such scans could still be suitable for opportunistic osteoporosis screening using retrospective clinical CT data. According to the American College of Radiology (ACR) guidelines for CT-based vBMD, values >120 mg/cm³ are considered normal, 80 and 120 mg/cm³ indicate low bone mass (formerly osteopenia), and vBMD <80 mg/cm³ is defined as osteoporosis.²⁴ With these diagnostic thresholds, a small upward shift in vBMD observed in the urographic phase is unlikely to alter diagnostic classification in most cases, suggesting that the urographic phase scans remain appropriate for screening or diagnostic use. The magnitude of change in the urographic phase is similar to the variability introduced by imaging protocol choices, such as the application of a smoothing reconstruction kernel, which is generally regarded as acceptable in opportunistic CT analysis.²⁵ Even the larger difference observed in the nephrographic phase (9.90 mg/cm³ [5.52%]) remains well within the 40 mg/cm³ thresholds that separate diagnostic categories, supporting potential use in certain clinical scenarios. For applications requiring precise quantification, such as monitoring longitudinal changes or evaluating therapeutic response in pharmaceutical trials, these differences could be more consequential. In such settings, even modest shifts in vBMD caused by contrast may obscure true biological change, and careful interpretation or correction would be necessary.

Although both nephrographic and urographic phases showed strong correlations with unenhanced vBMD, correction was phase dependent. The derived nephrographic linear equation reduced mean vBMD differences when applied to nephrographic data but was less accurate when applied to urographic data, underscoring the importance of interpreting individual-level data and also contrast timing. Prior work has shown that early post-contrast phases exhibit rapid and unstable vBMD changes, whereas later phases demonstrate greater stability, allowing correction formulas to be applied more reliably.¹¹ In our study, the nephrographic phase required correction, while the urographic phase differed only slightly from unenhanced vBMD, suggesting that correction may be unnecessary at this later time phase. However, despite small mean differences at the group level, the relatively wide LoA indicate substantial variability at the individual level, which could lead to misclassification of bone health status. These findings emphasize that a single equation cannot be generalized across phases, and that contrast dose timing plays a critical role in vBMD estimation.^11,17 A key limitation of this approach is that deriving phase-specific correction equations requires paired contrast-enhanced and unenhanced scans, which are rarely available in retrospective clinical datasets. While density-based conversion equations have previously been published to correct for contrast enhancement^15,26 for vBMD measurement, extrapolating these equations to other data sets results in inaccurate BMD measures.¹⁷ This limits the broad applicability of the method and highlights the need for alternative strategies to account for contrast effects. Automated algorithms for phase detection may help address this issue in future studies,^27–29 but further validation in the context of vBMD estimation is needed. Until a robust and generalizable solution is established, nephrographic phase CT data should be used with caution for opportunistic vBMD analysis unless a validated, phase-specific correction factor is available.

Several limitations of this study should be acknowledged. Correcting for contrast enhancement remains challenging given the strong influence of timing and dosage. However, the use of a standardized hematuria CT protocol provided a consistent framework for comparing unenhanced, nephrographic, and urographic phases, minimizing variability in contrast timing and enabling more reliable comparisons of vBMD measurements. The analysis was limited to the vertebrae, as the proximal femur was outside the scan field of view, restricting assessment of other clinically relevant skeletal sites. Additionally, although scans were acquired across multiple scanner platforms, their distribution was highly imbalanced across manufacturers and models (with 88% of scans being acquired on GE Healthcare scanners), limiting the ability to perform meaningful scanner-specific comparisons.

While opportunistic CT provides a valuable opportunity to diagnose osteoporosis and reduce the screening gap, contrast-enhanced scans typically lead to overestimation of vBMD and must be interpreted with caution. Although a reliable and universal correction method could overcome this limitation, such an approach has not yet been established. Until then, the effect of contrast remains a key barrier to the broader clinical adoption of opportunistic CT for vBMD assessment.

Author contributions

Bryn E. Matheson and Kirsten N. Bott contributed equally to this work.

Bryn E. Matheson (Formal analysis, Methodology, Software, Visualization, Writing—review & editing, Investigation), Kirsten N. Bott (Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Software, Validation, Visualization, Writing—original draft, Writing—review & editing, Visualization), Andrew G. Kiraly (Investigation, Methodology, Writing—review & editing), Richard E.A. Walker (Conceptualization, Data curation, Writing—review & editing, Methodology), Steven K. Boyd (Conceptualization, Data curation, Formal analysis, Resources, Software, Supervision, Writing—review & editing, Methodology), and Sarah L. Manske (Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Supervision, Writing—original draft, Writing—review & editing)

Funding

B.E.M. received graduate fellowship support from Alberta Innovates and from the Natural Sciences and Engineering Research Council (NSERC). K.N.B. received postdoctoral funding from the Canadian Institute of Health Research (CIHR), Health System Impact Fellowship and the University of Calgary T Chen Fong Postdoctoral Fellowship in Medical Imaging Science.

Conflicts of interest

S.L.M. holds the position of Guest Editor for JBMR Plus and has been recused from reviewing or making decisions for the manuscript.

Data availability

Analysis code is available at https://github.com/Bonelab/Ogo. Due to data acquisition agreements with Alberta Health Services, we are unable to make image data available.