A Comprehensive Set of Ultrashort Echo Time Magnetic Resonance Imaging Biomarkers to Assess Cortical Bone Health: A Feasibility Study at Clinical Field Strength

Abstract

INTRODUCTION:

Conventional bone imaging methods primarily use X-ray techniques to assess bone mineral density (BMD), focusing exclusively on the mineral phase. This approach lacks information about the organic phase and bone water content, resulting in an incomplete evaluation of bone health. Recent research highlights the potential of ultrashort echo time magnetic resonance imaging (UTE MRI) to measure cortical porosity and estimate BMD based on signal intensity. UTE MRI also provides insights into bone water distribution and matrix organization, enabling a comprehensive bone assessment with a single imaging technique. Our study aimed to establish quantifiable UTE MRI-based biomarkers at clinical field strength to estimate BMD and microarchitecture while quantifying bound water content and matrix organization.

METHODS:

Femoral bones from 11 cadaveric specimens (n=4 males 67–92 yrs of age, n=7 females 70–95 yrs of age) underwent dual-echo UTE MRI (3.0 T, 0.45 mm resolution) with different echo times and high resolution peripheral quantitative computed tomography (HR-pQCT) imaging (60.7 μm voxel size). Following registration, a 4.5 mm HR-pQCT region of interest was divided into four quadrants and used across the multi-modal images. Statistical analysis involved Pearson correlation between UTE MRI porosity index and a signal-intensity technique used to estimate BMD with corresponding HR-pQCT measures. UTE MRI was used to calculate T₁ relaxation time and a novel bound water index (BWI), compared across subregions using repeated measures ANOVA.

RESULTS:

The UTE MRI-derived porosity index and signal-intensity-based estimated BMD correlated with the HR-pQCT variables (porosity: r=0.73, p=0.006; BMD: r=0.79, p=0.002). However, these correlations varied in strength when we examined each of the four quadrants (subregions, r=0.11–0.71). T₁ relaxometry and the BWI exhibited variations across the four subregions, though these differences were not statistically significant. Notably, we observed a strong negative correlation between T₁ relaxation time and the BWI (r=−0.87, p=0.0006).

CONCLUSION:

UTE MRI shows promise for being an innocuous method for estimating cortical porosity and BMD parameters while also giving insight into bone hydration and matrix organization. This method offers the potential to equip clinicians with a more comprehensive array of imaging biomarkers to assess bone health without the need for invasive or ionizing procedures.

1. Introduction

It is well established that bone is a composite biomaterial comprised of ~35–45% hydroxyapatite mineral crystals, ~40% organic matrix including type I collagen, and ~15–25% water by volume¹. These components combine to provide flexibility, toughness, and elasticity. Due to bone’s dynamic nature, these components can change independently or together due to aging and disease. Even so, the current gold standard approach to predict when a bone will fail relies on areal bone mineral density (aBMD in g/cm²) from clinical dual-energy X-ray absorptiometry (DXA) scans of the hip and spine. DXA has only moderate sensitivity^2,3 and misses many individuals at risk of fragility fracture when based on T-score thresholds. Importantly, this imaging technique assesses only the mineral, leaving the remaining ~55–65% of bone unaccounted for. The advent of high-resolution peripheral quantitative computed tomography (HR-pQCT) has permitted three-dimensional in vivo assessment of bone microstructure at peripheral skeletal sites (distal radius and distal tibia), potentially improving fracture risk prediction⁴. However, HR-pQCT is still X-ray-based, limited to imaging bone’s mineral phase, and lacks the capability to visualize prevalent fracture sites like the femoral neck or spine.

To enhance fracture prediction and the identification of those who would benefit most from treatment, there is a need for new clinically feasible techniques that assess the contribution of the bone matrix to bone health. Magnetic resonance imaging (MRI), a noninvasive and non-ionizing imaging technology, is an attractive imaging modality that has gained interest in bone research over the past decade due to several important technological advancements. Conventional MRI, while valuable for various tissues, is not well suited for characterizing bone due to its inherently long longitudinal (spin-lattice) relaxation time, T₁, and bone’s extremely short transverse (spin-spin) relaxation time, T₂, (primarily arising from bound or restricted water of bone tissue/collagen fibrils). In response to this limitation, ultrashort echo time (UTE) MRI has emerged as a promising solution capable of capturing signals from fast-relaxing tissues, including bone, thanks to TEs of less than one millisecond^5–7 without the use of ionizing radiation.

Rajapakse and colleagues⁸ introduced a method to compute the porosity index (PI) from dual-echo UTE acquisitions using the signal ratio from the two MRI images: one with a TE<0.05 and one TE long enough to ensure bound water signal has decayed to zero and the pore water and fat signals are in phase. Several papers have pointed to PI’s ability to determine porosity using UTE MRI in a manner that (1) is in good statistical agreement with standard high-resolution computed tomography-based approaches⁹, (2) can discriminate bone disease status^9,10, and (3) can indicate whole bone mechanical integrity and stiffness^11,12. Although porosity is a critical factor in comprehending bone quality and health, especially in diseases like chronic kidney disease (CKD) where porosity development is a hallmark of the cortical bone phenotype^10,13, augmenting ¹H UTE MRI assessment with a surrogate estimate for cortical BMD would further bolster its comprehensive evaluation of bone health.

Ho et al. described a mathematical relationship between signal intensities from proton-density-weighted in-phase images generated by multi-fat-peak T2*-IDEAL MRI and BMD by imaging a set of calibration standards with various concentrations of hydroxyapatite in water. The authors reported a strong linear correlation between hydroxyapatite concentration and MRI signal intensity. When their signal model was carried out in volunteers imaged using the T2* IDEAL sequence, the estimated BMD from MRI signal intensity correlated significantly with BMD measured using quantitative computed tomography¹⁴. Building upon this work, we adopted a similar approach in a preclinical model of post-menopausal osteoporosis, utilizing the SWIFT sequence^15,16, which allows for near-zero TE acquisition¹⁷. This work employed direct imaging of a calcium hydroxyapatite phantom with a known density value and incorporated this information into an equation, enabling us to estimate BMD based on the signal intensity of the resulting MR image. Those studies not only successfully estimated BMD and detected changes consistent with estrogen deficiency, but it also exhibited strong agreement with BMD values quantified using microCT. However, the performance of BMD estimation by UTE MRI at clinical field strength remains an unexplored area in the field.

While estimating cortical porosity and BMD are important to understanding bone health, we still require quantitative information about bone water and matrix organization and how these aspects of bone quality change in response to aging, disease, or treatment. Multiple lines of evidence indicate that bone water, occupying a quarter of bone by volume, greatly influences mechanical properties essential for fracture resistance¹⁸. Remarkably, recent developments allow distinguishing between bound and free water via various analysis techniques^19–22, facilitating the quantification of bone’s distinct water compartments in vivo from UTE images. Further, groups have begun to derive quantitative indices to measure parameters associated with cortical bone’s collagen fracture and bone matrix organizing^23–25.

To fully exploit the capabilities of UTE MRI in bone-related applications, the current study aimed to establish a rigorous set of quantitative MRI biomarkers, not only to characterize aspects of bone that go beyond what X-ray-based methods offer but also to estimate cortical porosity and matrix density. We conducted UTE MRI and HR-pQCT scans on human femoral bone specimens for this pilot, proof-of-concept study. Our primary objective was to validate the effectiveness of UTE MRI in estimating BMD and PI by comparing them to HR-pQCT results. Furthermore, we expanded the utility of UTE MRI by introducing a bound water index based on two distinct echo times and examining T₁ relaxation times in different subregions within the cortical bone. We hypothesize that UTE MRI will be in good agreement with HR-pQCT and that UTE MRI will also be able to detect regional differences in bound water and T₁ relaxation times throughout the cortex.

2. Materials and Methods

2.1. Bone Specimens

The right femur was harvested from eleven consecutively available non-embalmed human donors (n=4 males 67–92 yrs of age, n=7 females 70–95 yrs of age) from the Indiana University School of Medicine Anatomical Education Program. No protected health information was collected so institutional review approval was not required. Femora were cleaned of muscle, the proximal and distal ends removed using a Stryker Autopsy Saw (Stryker, Kalamazoo, MI), and the shaft imaged using HR-pQCT (described below). A 5 cm long section of the most proximal end of the shaft was sectioned using an IsoMet Low Speed Saw (Buehler, Lake Bluff, IL). The prepared 5 cm long bone specimen was wrapped in phosphate buffered saline (PBS)-soaked gauze and immediately stored at −20°C until further imaging.

2.2. High-Resolution Peripheral Quantitative Computed Tomography (HR-pQCT)

Femora were imaged using a XtremeCT II HR-pQCT scanner (SCANCO Medical AG, Brüttisellen, Switzerland) to acquire three-dimensional cortical geometry, microarchitecture, and volumetric BMD. Specimens were wrapped in PBS-soaked gauze and sealed in a plastic bag to maintain hydration during scanning. Specimens were positioned within a cylindrical, carbon-fiber molded sample holder provided by the manufacturer and propped up with towels to position the specimen in the bore’s isocenter. The scanner operated at an effective energy of 68 kVp, 1470 μA intensity, and 43 ms integration time. A multi-stack acquisition²⁶ of about 22 consecutive 168-slice stacks was executed to acquire ~3,370 slices (~20.5 cm of bone length prior to creating the 5 cm section) with a voxel size of 60.7 μm. The average scan time was ~40 minutes. Phantoms containing known density and volume inserts were scanned daily to ensure scanner stability, following the manufacturer’s instructions. Bone scans were reconstructed according to the manufacturer’s algorithm and exported in DICOM image file format for further image processing and analysis.

2.3. Ultrashort Echo Time Magnetic Resonance Imaging (UTE MRI)

The 5 cm bone specimen was removed from −20°C storage and thawed at room temperature for two hours while wrapped in PBS-soaked gauze to prevent dehydration. Thawed specimens were fully immersed in perfluoropolyether (Fomblin, Thorofare, NJ), a viscous fluoroinert solution, to minimize susceptibility artifact during UTE MRI. Importantly, Fomblin is insoluble in water and, therefore, forms a “plug” to contain the water in the bone sample thus minimizing dehydration during imaging.

Isotropic 3D radial UTE imaging was performed with a research application sequence on a 3T whole-body MRI scanner (MAGNETOM Prisma Fit, Siemens Healthineers AG, Erlangen, Germany) equipped with an 18-channel body coil (Body 18, Siemens Healthineers AG, Erlangen, Germany). UTE sequence parameters were selected for subsequent bone-based calculations described in detail in the following sections. One dual echo acquisition was acquired with an ultrashort echo time (TE1 = 0.04 ms) to capture the signal from fast (bound water) and slower (free water) relaxing protons in the bone cortex, while the second, longer echo time (TE2 = 2.8 ms) captured signal solely from slower relaxing protons and no signal contribution from the fast-relaxing protons at 3.0T clinical field strength²⁷. Additional scan parameters were as follows: repetition time (TR): 12 ms, flip angle (FA): 12°, averages: 2, number of radial views: 50,000, matrix size: 352, field of view (FOV): 160 mm, isotropic voxel size: 0.45 mm, radiofrequency (RF) pulse duration: 40 μs, dead time between RF and ADC: 20 μs, with an acquisition time of 20:02 [min:sec]. An additional dual-echo acquisition was acquired with identical parameters except for FA: 18°, which was used for subsequent variable flip angle (VFA) T₁ calculation²⁸.

In each UTE MRI acquisition, specimens were scanned with a series of water phantoms with ascending concentrations of deuterium oxide (heavy water, D₂O) doped with 27 mmol/L MnCL₂ and a calcium hydroxyapatite phantom (0.75 g/cm³ CaHA) to evaluate the sensitivity to fast-relaxing T₂ components and support converting signal intensity to estimate BMD, respectively (Figure 1). The equation to achieve the signal-intensity-based estimate of BMD using the first echo of the UTE MRI acquisition (Figure 1A) can be found in Section 2.6.1.

Figure 1. Dual-echo UTE MRI images of bone specimen and phantom.

A 5 cm long bone specimen was taken from the proximal portion of the femoral shaft beginning about 2 cm distal from the pectineal line from eleven donors post-mortem. A–B) Representative UTE image from one specimen and the co-imaged phantoms. The top panel is TE1=0.04 ms (A), and the bottom is TE2=2.8 ms (FA=12°, B). The five water (H₂O) phantoms doped with deuterium oxide (D₂O) + 27 mmol of MnCL₂ are circled in blue, the calcium hydroxyapatite (CaHA) phantom is circled in yellow, and lines indicating the cortical bone and marrow are in orange. Substantial signal intensity is lost in the TE2=2.8 ms image where four of the D₂O-doped phantoms, the CaHA phantom, and the cortical bone appear as a signal void. Specimens also underwent HR-pQCT imaging (not pictured).

2.4. Registration, Segmentation, and Subregional Division

The HR-pQCT image was cropped to contain just the proximal femoral shaft region (~5.5 cm). The HR-pQCT (moving) was automatically registered to its corresponding UTE MRI scan (fixed, using TE2) using Analyze (V.14, AnalyzeDirect, Inc., KS, USA) and assuming rigid-body geometry (only rotation and translation). During this process, the HR-pQCT image resolution was down-sampled to match the resolution of the UTE MRI (0.455 mm isotropic); thus, the transformation was applied to the original HR-pQCT scan (60.7 μm nominal isotropic) for use in further analysis.

The remaining image analysis was carried out using in-house algorithms developed in MATLAB (2021b, Mathworks, Natick, MA, USA). Using the HR-pQCT image, a 75-slice region of interest (ROI) spanning 4.55 mm in the z-direction (along the length of the shaft) was chosen from the central part of the 5 cm bone section. Masks were then contoured from the 75-slice HR-pQCT images using an automatic segmentation algorithm based on a predefined threshold. The masks were further segmented into anterior, posterior, medial, and lateral subregions of the cortical bone by locating the centroid of the cross-sectional image to facilitate the division. This mask created for HR-pQCT was then applied to the UTE MRI images. Finally, HR-pQCT and UTE MRI outcomes were quantified in the total cortical ROI and the four subregions, as detailed below. A schematic of the image processing scheme can be found in Figure 2.

Figure 2: Image Processing Scheme.

A) Representative femoral cross-sectional images from the same region using ultrashort echo time MRI (UTE MRI, 0.45 mm) and high-resolution peripheral quantitative computed tomography (HR-pQCT, 0.061 mm voxel size). B) Rigid-body registration was applied with the UTE MRI (fixed) and HR-pQCT (moving) thus down-sampling the HR-pQCT image resolution to the resolution of the UTE MRI. The resultant transform was applied to the full-resolution HR-pQCT image (0.061 mm) which was used for the remaining image processing steps and analysis. C) Two separate cortical segmentation masks were derived from the HR-pQCT image. The thresholded mask (top) isolates the cortical bone, with pores depicted as void spaces, while the thresholded+filled mask (bottom) encompasses the cortical bone along with the pores. D) Both cortical masks were divided into four subregions (anterior, posterior, lateral, and medial) based on a user-selected centroid. E) The segmentation masks were then applied to the UTE MRI and HR-pQCT images to derive cortical bone outcomes.

2.5. Cortical Porosity

2.5.1. UTE MRI Porosity Index

To calculate the UTE MRI metric for cortical porosity, known as the porosity index (PI), the intensity values within the long echo (TE2=2.8 ms) and short echo (TE1=0.04 ms) images were divided to create a resultant image. The PI map was generated using the following equation as described elsewhere²⁹:

$$\mathit{\text{Porosity}}\mathit{\text{Index}}\mathit{\text{Map}}=\frac{\mathit{\text{TE}}2\mathit{\text{Intensity}}}{\mathit{\text{TE}}1\mathit{\text{Intensity}}}$$

The segmentation routine stored the PI map values in a unique vector for each ROI (total bone, anterior, posterior, lateral, and medial). The mean value of each vector was calculated and multiplied by 100 to get a representative PI value.

2.5.2. HR-pQCT Cortical Porosity Percent and Pore Volume

To calculate cortical porosity percent from the HR-pQCT images, the thresholded+filled mask was used to calculate the total number of pixels representing the bone and pores (total area) and compared pixel-by-pixel to the thresholded mask. The number of pixels only containing pores was summed, divided by the total cortical area and multiplied by 100 to get porosity as a percentage.

2.6. Bone Mineral Density (BMD)

2.6.1. UTE MRI Signal-Intensity-Based estimate of Bone Mineral Density

An estimation of BMD based on signal intensity from the UTE MRI was utilized using the equation below described by our group¹⁷ which was adapted from Ho et al.¹⁴.

$$\mathit{\text{SI}}\mathit{\text{UTE}}\mathit{\text{MRI}}\mathit{\text{BMD}}=\left\{1-\left[\frac{S_{\mathit{\text{CORT}}}-S_{\mathit{\text{CaHA }}}}{S_{\mathit{\text{MARROW}}}-S_{\mathit{\text{CaHA}}}}\right]\right\}\times {\rho }_{\mathit{\text{CaHA }}}$$

In this equation, SI UTE MRI BMD is the estimated BMD based on the signal intensity of the nominal echo time image, TE1. S_CORT represents the average signal intensity of the cortical ROI, S_CaHA is the average signal intensity in the CaHA phantom, S_MARROW is the mean signal intensity of the bone marrow, and ρ_CaHA is the known density value of the CaHA phantom (0.75 g/cm³). A depiction of the CaHA phantom and one cortical bone specimen using the UTE MRI at TE1 can be found in Figure 1A. As a note, only ¹H UTE MRI was conducted. ³¹P UTE MRI was not conducted in this study.

2.6.2. HR-pQCT Bone Mineral Density

We computed the volumetric BMD (mgHA·cm⁻³) from the HR-pQCT images within the 4.55 mm ROI using the SCANCO standard clinical evaluation protocol, as detailed elsewhere by our group^30,31. The precise location of the ROI within the original HR-pQCT was determined, and with identified slice numbers, the periosteal and endosteal surfaces were contoured. Images were then filtered using a low-pass Gaussian filter (sigma 0.8, support 1.0 voxel), and fixed thresholds were applied to extract the cortical bone (450 mgHA·cm⁻³). The volumetric BMD served as the reference standard for comparison with the estimated UTE MRI BMD.

2.7. UTE MRI Cortical Matrix T₁ Relaxation Time Analysis

Estimated matrix organization could be observed using estimated T₁ relaxometry mapping. The map was generated voxel-by-voxel by performing the linear fitting described in the equation below.

$$T_1\mathit{\text{map}}(x,y,z)=\frac{-TR}{\ln \left(\frac{\alpha }{\beta }\right)}$$

Where:

$$\alpha =\frac{I_{FA1(x,y,z)}}{\sin \left(F{A}_1\right)}-\frac{I_{FA2(x,y,z)}}{\sin \left(F{A}_2\right)}\beta =\frac{I_{FA1(x,y,z)}}{\tan \left(F{A}_1\right)}-\frac{I_{FA2(x,y,z)}}{\tan \left(F{A}_2\right)}$$

In the equation, TR represents repetition time, FA₁ represents a flip angle of 18°, FA₂ represents a flip angle of 3°, I_FA1(x,y,z) represents the signal intensity of the voxel of interest at an FA of 18°, and I_FA2(x,y,z) represents the signal intensity of the voxel of interest at an FA of 3°. The average T₁ relaxation time was quantified for the total cortical ROI and the individual cortical subregions. Radiofrequency (RF) field (B1) correction maps were not recorded due to technical difficulties at the time of image acquisition, and therefore this correction did not occur. As a result, our interpretation of the data was done within this study only (no comparisons made to previously reported T₁ relaxation time in cortical bone).

2.8. UTE MRI Bound Water Index

The bound water index (BWI) introduces a novel and straightforward metric utilizing two distinct echo times to quantify the relative content of bound water, specifically the extremely fast relaxing population within the cortical bone. The BWI is calculated by subtracting TE2 (2.8 ms) from the shortest echo time (TE1, 0.04 ms) image and dividing the result by TE1, as expressed in the following equation:

$$\mathit{\text{BWI}}=\frac{TE1\mathit{\text{ intensity }}-TE2\mathit{\text{ intensity }}}{TE1\mathit{\text{ intensity }}}$$

TE1 should be minimized to capture the proton signal from nearly all cortical water, including both bound and free water components. TE2 must be sufficiently long to nullify the contribution of bound water while still capturing the signal from the free/pore water pool. Following the creation of BWI images using the provided equation, the average intensity values for the total cortical ROI and its four subregions were quantified.

2.9. Statistical Analysis

Statistical analysis was conducted using GraphPad Prism v9.5.1. Pearson correlation coefficients assessed the linear association between HR-pQCT and UTE MRI estimates of porosity and BMD. A simple linear regression model was used to evaluate the predictive value of UTE MRI against HR-pQCT. Repeated measures ANOVA (RM ANOVA) was employed for T₁ relaxation times and BWI to identify spatial differences in UTE MRI measures across the four cortical subregions. This was followed by a Tukey multiple comparison test when the RM ANOVA term was significant. In all instances, a significance level of p ≤ 0.05 was utilized.

3. Results

3.1. UTE MRI Porosity Index vs. HR-pQCT Percent Porosity

There was a strong positive linear correlation (r=0.73, p<0.01) and significant overall regression (R²=0.53, F [df regression, df residual] = 10 [1, 9], p=0.01) between HR-pQCT percent cortical porosity and PI in the total cortical bone ROI (Fig. 3A). A strong linear correlation (r=0.71, p<0.01) and significant regression (R²=0.50, F [df regression, df residual] = 9.1 [1,9], p=0.01) was also observed in the medial subregion (Fig. 3F). The posterior subregion had a significant, moderate linear correlation (r=0.54, p=0.045), but the slope of the regression model did not reach significance (R²=0.29, F [df regression, df residual] = 3.6 [1, 9], p=0.08) (Fig. 3C).

Figure 3: Relationships between UTE MRI porosity index and HR-pQCT percent cortical porosity.

We identified a significant positive linear correlation in the A) total cortical ROI, C) posterior, and F) medial subregions and B) depicts the color-coded subregions corresponding to the plots. However, no statistically significant correlation was observed in the Pearson correlation analysis between the D) lateral and E) anterior subregions, and the linear regression analyses yielded non-significant results. Each plot displays the regression lines (solid), coefficients (dotted with shaded fill), and their respective 95% confidence intervals. An asterisk in front of the linear regression formula indicates significance.

There were no relationships between PI and HR-pQCT cortical porosity in the anterior (r=0.34, p=0.16, Fig. 3E) or lateral (r=0.11, p=0.38, Fig. 3D) subregions and non-significant linear regressions (p=0.31 and p=0.75 for anterior and lateral subregions, respectively).

3.2. Signal-intensity-based UTE MRI estimated BMD vs. HR-pQCT BMD

A significant positive linear correlation was observed between HR-pQCT BMD and UTE MRI estimated BMD based on signal intensity in the total cortical bone ROI, anterior, and medial subregions. Specifically, there was a strong linear correlation between the total ROI (r=0.79, p=0.002, Fig.4A) which also had a significant overall regression (R2=0.62, F[df regression, df residual]= 14.49 [1, 9], p=0.004). The anterior (r=0.71, p=0.007, Fig.4E), and medial (r=0.70, p=0.008, Fig.4F) subregions also had strong linear correlations between UTE MRI and HR-pQCT measures and each regression model was significant (anterior: R2= 0.50, F[df regression, df residual]= 9.13 [1,9], p=0.02; medial: R2= 0.49, F[df regression, df residual]= 8.75 [1, 9], p=0.02).

Figure 4: Relationships between ¹H UTE MRI estimated BMD based on signal intensity and HR-pQCT BMD.

There was a significant, strong positive linear correlation in BMD from HR-pQCT and estimated BMD based on the signal intensity model first described by Ho et al. in the A) total cortical ROI, E) anterior, and F) medial subregions and B) depicts the color-coded subregions corresponding to each plot. There was no statistical association between imaging modalities in the C) posterior and D) lateral regions, although r values were moderate. The regression lines (solid) and coefficients (dotted with shaded fill) are shown with 95% confidence intervals for each plot. An asterisk in front of the linear regression formula indicates significance. Significance was defined as p ≤ 0.05.

The results for the Pearson correlation analysis in the posterior (r=0.48, p=0.067, Fig.4C) and lateral (r=0.49, p=0.065, Fig.4D) subregion showed a moderate positive and nearly significant linear correlation. Neither the posterior (p=0.12) nor the lateral (p=0.13) linear regression model was significant.

3.3. Cortical T₁ Relaxation Time: UTE MRI

Ten of the 11 femoral bone specimens were analyzed for T₁ relaxometry. Images from one specimen were excluded, as the average signal intensity for both recorded FA differed by a factor of four, which was not representative of the expected Gaussian distribution of signal intensity per flip angle (with 12° being the maximum FA, which should have had no more than double the signal intensity of the lower flip angle).

The generated T₁ maps were used to assess matrix organization. The total bone ROI had an average T₁ value of 0.49 seconds, with a standard deviation of 0.16 seconds (Fig. 5A). One-way RM ANOVA testing did not reveal significant differences (p=0.58) between the averages for the posterior (mean ± standard deviation = 0.48 ± 0.15 seconds), lateral (0.54 ± 0.25 seconds), anterior (0.47 ± 0.21 seconds), and medial (0.46 ± 0.20 seconds) subregions (Fig. 5B). The lateral region had a trend for the highest T₁ relaxation time.

Figure 5: UTE MRI T₁ relaxation time analysis using the variable flip angle approach.

A) Representative T₁ maps from each specimen. Data presented as mean T₁ relaxation time in seconds (s). Anatomical orientation is indicated on the cross-section in the top left corner. B) The RM ANOVA test did not reveal significant differences between the subregions.

3.4. Bound Water Index: UTE MRI

Water phantoms with increasing concentrations of pure water were imaged with the specimens and used to identify the optimal pair of echo times for assessing the bound water index (BWI). Our analysis established that TE1 at 0.04 ms and TE2 at 2.80 ms offered an accurate representation of the water behavior, achieving an R2 value of 0.995, as illustrated in Supplemental Figure 1. The curve fitting was performed using the exponential plateau fit in GraphPad Prism.

The average BWI calculated for the total cortical bone was 0.59 ± 0.19. The RM ANOVA testing revealed no significant differences between the mean BWI values for any subregions (p=0.11, Fig.6A). We employed Pearson correlation coefficient analysis and observed that BWI had a strong, negative relationship to the measured T₁ relaxation time (r=−0.87, p=0.0006, Fig.6B) where bones with the highest BWI had the lowest T₁ relaxation time. The simple linear regression model was significant (p=0.001) with an R2 of 0.75 and an F statistic of 24 [1,8].

Figure 6. Cortical bound water index and relationship to T₁ relaxation time.

A) An RM ANOVA analysis revealed no significant difference between the bound water index (BWI) across cortical subregions. B) The BWI and T₁ relaxation times (s) are highly and negatively correlated (r = −0.87, p = 0.0006). The regression lines (solid) and coefficients (dotted with shaded fill) are shown with 95% confidence intervals. The asterisk in front of the regression formula indicates significance.

4. Discussion

In this study, we employed 3D radial UTE MRI at clinical field strength (3.0T) to image cortical bone specimens from the shaft of 11 human femora. We demonstrate a strong positive linear relationship between the signal-intensity-based estimated BMD and the PI, both derived from UTE MRI, with HR-pQCT derived BMD and percent porosity, respectively. Additionally, UTE MRI provided valuable quantitative information about the cortical bone’s limited bound water pool using the newly described BWI and matrix organization, as measured using a T₁ relaxation technique. The results underscore that UTE MRI has an added advantage in identifying biochemical and material-level changes independently of the mineral phase. Collectively, these results highlight the potential significance of UTE MRI in bone imaging for a more comprehensive picture of bone health.

DXA has been widely used as a fracture prediction and screening tool over the past 50 years for its user-friendly features, extensive reference databases, cost-effectiveness, and accessibility. However, our current capability for early detection across various diseases and aging stages which largely relies on DXA, continues to fall short of optimal standards, leading to underdiagnosis and under-treatment of osteoporosis and associated fragility fracture^32,33. The introduction of HR-pQCT has made it feasible to evaluate essential factors contributing to bone strength, including compartment-specific vBMD, geometry, and microarchitecture including cortical porosity³⁴. As a result, HR-pQCT is gaining momentum in the research setting, and its applications continue to broaden^35,36. In this study, we utilized HR-pQCT as our gold standard modality to quantify cortical porosity (due to its high achievable resolution of 60.7 μm,) and vBMD for our cortical bone specimens derived from the cortical shaft.

Early work by Ho et al., described a theoretical signal model to estimate BMD using ¹H MRI-imaged hydroxyapatite phantoms in good agreement with a quantitative CT measure¹⁷. The authors exploited the idea that bone minerals would have an inverse relationship to signal intensity. The HA phantom (representing a region of a known density of mineral) and the bone marrow (representing a region with no mineral) could be used to calibrate the cortical bone signal intensities to a quantitative, surrogate bone mineral density map. Previously, our group modified this technique to be applied using a near-zero echo time MRI approach, SWIFT, at preclinical field strength (9.4T) to image a cohort of rats with progressing estrogen deficiency. Here we captured longitudinal changes in the signal-intensity-based estimated BMD consistent with microCT BMD, and we also demonstrated a strong correlation between both modalities in the cortical region¹⁷. In the current study, we used the same theoretical signal model to derive an estimate of BMD based on the UTE MRI signal intensity with the help of a concurrently scanned CaHA phantom and demonstrated a significant correlation with vBMD measured with HR-pQCT in the total cortical region. However, when the analysis was performed within the four cortical subregions, the correlations varied, with significant correlations in the anterior and medial regions. The non-significant correlations in the posterior and lateral regions may be attributed to the small sample size in this proof-of-concept study. Future studies with larger cohorts, spanning age, sex, and ethnicity, are needed to further investigate these associations.

While we used ¹H UTE MRI to estimate density, it should be discussed that solid-state ³¹P zero echo time MRI can directly quantify bone mineral density as shown in several previous studies including the most recent work by Jones et al. and colleagues³⁷. The authors demonstrate that ³¹P density³⁸ was positively strongly associated with DXA aBMD (r=0.66, p<0.001) and moderately with quantitative CT vBMD (r=0.44, p=0.01). Their conclusion from this prospective study of post-menopausal women, with and without porosity, suggests that solid state ³¹P MRI of the cortical bone has the potential to reflect mineralization in vivo. One current limitation hindering the widespread implementation of ³¹P imaging is that it requires specialized hardware, including custom RF coils.

To understand cortical microstructure via UTE MRI through estimation based on signal, a singular dual-echo 3D UTE acquisition can be used to calculate PI based on the signal of free water predominately residing in the cortical pores which others have demonstrated correlations to percent porosity measured using CT-based approaches⁸. In our work, we observed a significant correlation with PI and percent porosity measured via HR-pQCT in the total cortical ROI which again varied when the same analysis was carried out by subregion (with posterior and medial regions remaining correlated). The sample size again may have come into play with our subregional analysis. The recently described suppression ratio may increase the precision of a porosity based measurement based on free water at clinical field strength¹¹ as well as UTE magnetization transfer analysis which is a fast approach to gain information regarding pore water³⁹. It is imperative to mention that fat suppression was not employed in the MRI acquisition of our study. While there is only a nominal amount of lipid present in the cortical bone itself, the lipid signal was still present in the UTE acquisition, particularly for TE2, where both free water and fat signal are in-phase at 3T and thus could have skewed the results, particularly if there was lipid present in the pores. Marrow contains nearly ~70% fat by volume⁴⁰; thus, while we used the HR-pQCT to segment the cortical bone precisely and then applied the segmentation to the UTE MRI for analysis, an erroneous segmentation that included marrow voxels would substantially increase the measure of porosity. While there are instances of using the PI without fat suppression in the literature¹², PI with fat suppression, including through inversion recovery methods, would give a more precise measure of the free water residing in the critical pores, thus a more precise measure of the PI.

MRI shines is in its ability to resolve water content and its relationship to the local macromolecular structures. With the advent of UTE and zero echo time MRI approaches with negligible time delays between RF excitation and signal acquisitions, quantifying bone’s bound and free water¹⁸ located within the relatively dense material is now feasible⁷. Techniques to achieve bound water quantification from UTE include using bi- and tri-exponential fitting of multiple acquired echo times as well as the magnetization transfer technique, each of which suffer from long acquisitions times⁷. The inversion recovery-based UTE techniques are far more rapid⁴¹ but require an adiabatic preparatory pulse. Hence, we find UTE MRI fractional indices intriguing due to their practicality, as they only require two distinct echo times for computation, resulting in faster acquisition times and straightforward analysis compared to schemes necessitating multiple echoes. Building upon the PI fractional index, we introduce the BWI, which employs ultrashort TE and a sufficiently long TE to nullify the signal from bound water, thus isolating the free water signal. To validate our measure, we employed water phantoms doped with varying deuterium oxide concentrations to change the T₂ relaxation times, resulting in a robust R2 value of 0.995. While our BWI evaluation across subregions did not reveal significant differences, the lateral and medial regions exhibited the highest BWI. Further research is required to assess BWI variations in healthy and diseased bone tissues and its changes throughout disease progression. Regardless of the method used to quantify bound water, the ability to do so via MRI is a powerful tool, supporting the development of therapeutics targeting bound water^42–45 and providing a non-invasive, in vivo metric to evaluate their efficacy.

Although mechanical testing was not conducted in our study, it is pertinent to acknowledge the body of literature that underscores the significance of bound water in the context of bone fracture in preclinical studies. Loosely bound water found at the collagen and mineral interfaces allows sliding between collagen and mineral, thus playing an essential role in transferring loads. The removal of loosely bound water by thermal dehydration results in a bone that is less tough illustrating the role of bound water in governing post-yield mechanical behavior^{46 47}. Moreover, tightly bound water, localized within the collagen triple helices, has been found to exhibit a positive association with bone toughness and a negative correlation with stiffness⁴⁶. Depletion of tightly bound water has been observed to induce structural alterations in collagen type I, leading to its shortening^48,49. Collectively, data supports the importance of understanding bound water dynamics and its relationship to bone mechanical properties in vivo. Excitingly, the current bone research landscape is marked by a growing interest in UTE MRI, which is capable of quantifying bound and free water in vivo at clinical field strength (3.0T).

While it has been noted for some time that bone toughness declines more significantly than bone strength during aging (with strength being related to bone mass)⁵⁰, current clinical assessment for osteoporosis lacks a direct measure of the underlying integrity of bone’s organic matrix. The absence of methods to monitor alterations in the interconnected network of collagen I, water, mineral, and non-collagenous proteins using imaging prevents effective evaluation of osteoporosis medications targeting the bone’s extracellular matrix, necessitating expensive clinical trials with fractures as the primary outcome. T₁ relaxation time, derived from MRI acquisitions, offers a valuable means to explore tissue-specific relaxation properties⁵¹. It quantifies the energy dynamics of spins influenced by their local microenvironment and is highly sensitive to subtle variations in healthy and diseased tissues^52,53. In preclinical studies, when combined with UTE or ZTE, T₁ relaxation times tend to decrease as tissue becomes more mineralized⁵⁴. This phenomenon was also observed in our previous work when we used T₁ to monitor cortical bone changes following ovariectomy in a rat model¹⁷. Although this study didn’t allow us to track T₁ changes, we employed the VFA approach at clinical field strength to assess T₁ values across subregions. We didn’t find significant differences in T₁ values but did note an interesting inverse relationship with cortical porosity: regions with lower porosity exhibited higher T₁ values, and vice versa. We postulate that the bones with high porosity had higher free/pore water and likely a concurrent decrease in bound water indicating less water residing in the collagen triple helices and at the collagen mineral interfaces (in part due to less matrix and in part because a more porous matrix is likely more disorganized even within the matrix that remains intact. Notably, we observed a significant negative correlation between T₁ relaxation time and the BWI. In other words, regions with lower BWI concentration displayed longer T₁ relaxation times. Further validation in the field is warranted, involving sequential denaturation of collagen, removal of collagen crosslinks, and/or demineralization of the mineral content. This will help assess the specificity of T₁ in characterizing the local macromolecular content and organization within bone.

Limitations

This study had several limitations which require discussion. A drawback of some MRI approaches lies in the prolonged acquisition time. In the present study employing UTE, the acquisition time for the dual echo acquisition of our ex vivo bone specimens was 20:02 [min:sec]. To conduct the T₁ relaxation analysis using the VFA approach, an additional dual echo UTE acquisition was obtained, extending the total time to 40:04 [min:sec]. We recognize that lowering the TR in our acquisition would improve the overall time cost. While the increasing need for fragility fracture screening may pose a time-related limitation, noteworthy progress in leveraging deep learning and compressed sensing approaches is underway to significantly reduce acquisition times particularly using UTE. This includes utilizing 3D UTE imaging with a spiral cones trajectory⁵⁵ instead of regular cartesian trajectories. Integrating such acquisition with compressed sensing reconstruction enables acceleration up to 2.5 – 5× times the regular acquisition speed. Isotropic radial 3D trajectories are also used in conjunction with compressed sensing for fast acquisition of musculoskeletal imaging⁵⁶. Additionally, supervised⁵⁷ and self-supervised⁵⁸ deep learning techniques are applied to expedite data acquisition and correct motion artifacts in UTE pulmonary MRI. Apart from UTE sequences, deep learning has been frequently employed to overcome the artifacts caused by under sampling for the brain and knee MRI, demonstrating its notable effectiveness in reducing acquisition time^59–62.

Bone images were acquired ex vivo in the current pilot feasibility study. The study itself had a small sample size making it a pilot endeavor but future work will transition the approach in vivo using a larger cohort. Bones were prepared by carefully removing musculature and adipose tissue. Future research should assess the performance of the described biomarkers in in vivo acquisitions, considering the influence of neighboring soft tissues and determine how the results compare to bone that was imaged with no soft tissues submerged in Fomblin. This evaluation should span various age, sex, and ethnicity groups to comprehensively understand the efficacy across populations. Additionally, further investigations are warranted to precisely delineate the association between UTE MRI porosity and an estimated BMD based on signal intensity of the MRI and their counterparts in HR-pQCT. While these measures exhibit correlations, it’s crucial to acknowledge that ¹H MRI cannot directly measure BMD, ³¹P MRI would be necessitated to directly calculate BMD. Nonetheless, these correlations hold promise as indicators of declining bone quality. While we pose UTE as an MRI pulse sequence able to more comprehensively characterize bone health, we still utilized HR-pQCT images to create our segmentation masks. Manual image segmentation to create bone-selective MR masks directly from the MR images is time and labor-intensive (requiring expert rater), posing a significant obstacle to the widespread adoption of the UTE technique in clinical practice. Recent research by Khandelwal et al. outlines a multi-atlas pipeline for UTE segmentation of the cranium. This approach is not only rapid but also demonstrates promising agreement with expert-labeled segmentation derived from ground truth CT images, suggesting its potential to support clinical implementation of UTE⁶³. Finally, while we chose parameters to minimize the dead time, or the time where no RF pulses are applied and no data is acquired, to be 20 μs in order to capture short T₂ and increase SNR, it is plausible that we may have suffered from coil ring down artifact which could have impacted TE1 in our study.

Conclusions

Currently, established non-invasive methods for measuring bone quality in clinical settings are lacking. Our study underscores the potential of UTE MRI to offer a more comprehensive set of outcomes that provide important information about bone health. Further research plans include assessing the performance of our proposed biomarkers in vivo, in a larger cohort, including across different populations (bone disease, response to bone treatment, age, sex) to determine sensitivity to subtle changes in bone health.

Highlights.

• UTE MRI porosity index correlated with porosity measured by HRpQCT

• Estimated BMD based on UTE MRI signal intensity correlated with HRpQCT BMD

• UTE MRI could additionally quantify bound water

• T₁ relaxation time by UTE MRI varied across cortical bone specimens

• UTE MRI could provide a comprehensive imaging evaluation of bone health