In Vivo Quantitative Imaging Biomarkers of Bone Quality and Mineral Density using Multi-Band-SWIFT Magnetic Resonance Imaging

Abstract

Bone is a composite biomaterial of mineral crystals, organic matrix, and water. Each contributes to bone quality and strength and may change independently, or together, with disease progression and treatment. Even so, there is a near ubiquitous reliance on ionizing x-ray-based approaches to measure bone mineral density (BMD) which is unable to fully characterize bone strength and may not adequately predict fracture risk. Characterization of treatment efficacy in bone diseases of altered remodeling is complicated by the lack of imaging modality able to safely monitor material-level and biochemical changes In vivo. To improve upon the current state of bone imaging, we tested the efficacy of Multi Band SWeep Imaging with Fourier Transformation (MB-SWIFT) magnetic resonance imaging (MRI) as a readout of bone derangement in an estrogen deficient ovariectomized (OVX) rat model during growth. MB-SWIFT MRI-derived BMD correlated significantly with BMD measured using micro-computed tomography (μCT). In this rodent model, growth appeared to overcome estrogen deficiency as bone mass continued to increase longitudinally over the duration of the study. Nonetheless, after 10 weeks of intervention, MB-SWIFT detected significant changes consistent with estrogen deficiency in cortical water, cortical matrix organization (T₁), and marrow fat. Findings point to MB-SWIFT’s ability to quantify BMD in good agreement with μCT while providing additive quantitative outcomes about bone quality in a manner consistent with estrogen deficiency. These results indicate MB-SWIFT as a non-ionizing imaging strategy with value for bone imaging and may be a promising technique to progress to the clinic for monitoring and clinical management of patients with bone diseases such as osteoporosis.

Introduction

In bone diseases of altered remodeling such as osteoporosis, current clinical practice primarily relies on ionizing x-ray-based approaches to characterize the mineral phase of bone [1–3]. While mineral is the focus of x-ray-based measures, it is understood that bone is a composite material composed of ~35-45% mineral crystals containing hydroxyapatite, ~40% organic matrix including type I collagen, proteoglycan and glycosaminoglycan, and ~15-25% water existing as free (in Haversian and Volkmann’s canals) and bound (hydrostatically to organic matrix constituents) by volume [4]. Each are critical to bone quality and fragility and may change independently, or together, with disease progression and in response to treatment [5, 6]. Even so, there is a near ubiquitous reliance on bone mineral density (BMD) to predict mechanical properties and ultimately serve as an In vivo surrogate for treatment efficacy while ignoring these other important contributors of bone strength [1]. It is widely accepted that BMD is an incomplete measure of fragility, accounting for a fraction of bone strength and may not adequately predict fracture risk [7, 8].

It has become increasingly evident that additional factors commonly referred to as markers of ‘bone quality’ play an important and unique role in governing the mechanical integrity of the bone and its overall health [4, 9–11]. Importantly, these factors are not accurately interpretable from traditional radiographic imaging approaches, but rather often rely on invasive or ex vivo techniques to quantify changes. Therefore, the ability to non-invasively measure biochemical processes and material-level changes In vivo as they relate to bone pathophysiology represents a powerful tool to guide therapeutic development and subsequent clinical trials [12]. The desire for non-destructive imaging biomarkers to characterize mineral in addition to material properties is immense and equally unmet.

There are a number of clinically available imaging modalities whose application in bone appear promising but have not yet been fully exploited [13, 14]. One such clinically available modality, magnetic resonance imaging (MRI), is gaining interest for its sensitivity to biochemical composition and its rich dynamic range. However, bone appears as a signal void in conventional MRI as it is unable to “capture” bone’s inherently ultra-short transverse relaxation time (T2) due to time domain constraints where the time delay between excitation and acquisition is too long to capture ultra-short T2s [15]. In bone, proton signal intensity is drawn from a limited hydrogen pool which includes water residing in microscopic Haversian canals and lacuno-canaliculi systems (free water, T2 > 1 ms), matrix water bound to collagen (bound water, T2 ⪡ 1 ms) and protons of the collagen backbone/sidechain (T2 < 0.1 ms) [4, 16]. In recent years, a number of remarkable advances have been made in the field of MRI to evaluate short T2 tissues. To date, there are at least three different and clinically attractive short T2 sensitive MRI methods: Ultrashort echo time (UTE) [17, 18], free induction decay (FID)-projection also known as zero TE (ZTE) technique [19–21], and SWeep Imaging with Fourier Transformation (SWIFT) [22, 23]. A well-known problem in imaging of short T2 species is image blurring due to off-resonance de-phasing and T2 decay. This is particularly problematic in bone where the cortices have the largest pool of strongly diamagnetic calcium salt in the body. Furthermore, cortical bone shares an interface with lesser diamagnetic tissues such as muscle, fat, marrow, and the periosteum which can result in magnetic field distortions [24]. A direct way to avoid image blurring is to increase the strength of encoding gradients and bandwidth (BW). However, all existing methods become less efficient at high BW (BW>100kHz): the UTE [17] sequence is restricted by gradient ramping time, and ZTE [19–21, 25, 26] is limited by maximum available RF amplitudes, while regular SWIFT [27] loses SNR due to decreased receiver duty cycle at higher BWs. Recently, ZTE imaging with frequency modulated (FM) pulses was proposed to alleviate this deficiency [28, 29]. The practical challenge of such approach is to obtain a flat excitation profile. An alternative approach to achieve significantly higher bands was a version of SWIFT using a multiband excitation and acquisition (MB-SWIFT) [30]. While MB-SWIFT allows the use of much smaller, practical RF amplitudes compared to ZTE, it achieves the RF power efficiency highly required for testing the objects with variable flip angles (VFA). Further, SWIFT and MB-SWIFT are an entirely different class of MRI, where the time domain signals are acquired in a time-shared manner during a swept RF excitation [22, 27]. This results in virtually no image acquisition delay (i.e., TE=0) allowing it to overcome the rapid signal loss due to low and restricted proton concentrations, and initial proof-of-concept studies applying SWIFT to bone, calcified cartilage or even dental tissue have been favorable [31–40]. MB-SWIFT can achieve high excitation BW and in theory mitigate susceptibility artifact at the cortical bone interfaces present during In vivo imaging.

MB-SWIFT could have a novel and specific application in bone research where capturing information beyond mineral alone is highly desired. Using MB-SWIFT, we sought to describe a comprehensive (although not exhaustive) set of MRI biomarkers to characterize aspects of bone quality that remain unavailable through conventional bone imaging approaches like clinical dual-energy X-ray absorptiometry (DXA) and pre-clinical micro-computed tomography (μCT). Specifically, we employed MB-SWIFT to longitudinally quantify cortical water, cortical matrix T1 relaxation times (a tissue-specific biomarker where measures are related to tissue organization), and marrow fat following ovariectomy in growing rats. Further, we established the efficacy of MB-SWIFT to directly measure changes in BMD in comparison to μCT. Findings contribute to the overall goal of establishing a suitable imaging strategy to simultaneously quantify material-level and biochemical alterations in bone safely, without the use of ionizing radiation.

Materials and Methods

Animal model and study design

All animal procedures were approved and conducted in accordance with the University’s Institutional Animal Care & Use Committee (IACUC) in compliance with the Animal Care & Use Office (ACUO) guidelines and complied with ARRIVE guidelines. Seven female Sprague Dawley rats, six-weeks old, were housed randomly (2-3 per cage) and fed a standard rat chow diet with access to tap water ad libitum. All rats were subject to bilateral ovariectomy (OVX) following baseline imaging (described in detail in the next section). Bilateral OVX was performed using a dorsal approach through 2 cm incisions from the second to fifth lumbar vertebrae ventral to the rector spinae muscles below the last rib. Sutures were placed around the cranial portion of the uterus and uterine vessels, uterine vessels were ligated, and the ovaries and oviducts were excised.

To track longitudinal changes in bone following estrogen deficiency in the growing rats, in vivo imaging of the right proximal tibia included μCT, a conventional MRI sequence with the lowest achievable TE, and the MB-SWIFT MRI approach. In vivo imaging occurred at baseline 48 hours prior to OVX procedure and at 2, 4, 10 and 12 weeks post-OVX surgery. All rats were euthanized following the 12-week imaging timepoint. Successful OVX was confirmed by assuring uterine horn atrophy during necropsy.

In vivo Magnetic Resonance Imaging (MRI)

In vivo high-field magnetic resonance imaging was performed with an Agilent 9.4T small animal imaging system (12 cm horizontal bore, Agilent Technologies Inc., Santa Clara, CA) running VnmrJ software (version 2.3 A) using a 33 mm diameter surface coil composed of materials (Teflon) not visible with ¹H-MRI. Rats were anesthetized and imaged under 2% isoflurane/air inhalation. Right tibiae were immobilized and secured so that the region of the proximal tibial growth plate was within the coil isocenter and aligned in the same direction for maintenance of a specific flip angle (FA) and to avoid radiofrequency (RF) non-uniformity.

A table of MRI scanning parameters for the study can be found in Table 1. MB-SWIFT images were acquired at three varying flip angles (FAs) in order to derive T1 relaxation times (using the variable FA method described in detail in the next section); one at the Ernst angle, one below and one above the Ernst angle. At each of the three FAs, three saturation schemes were acquired: fat saturation, water saturation, and no saturation. Fat saturation was obtained by applying a hyperbolic secant (HS4) saturation pulse of 1 kHz bandwidth centered at fat resonance frequency after every 16 views/projections (MB-SWIFT-FS). Water saturation was similarly applied at water resonance frequency (MB-SWIFT-WS). Images with no saturation were acquired with identical timing parameters (MB-SWIFT-NS). All MB-SWIFT acquisitions made use of 128 sidebands each of the 1395 Hz bandwidth (bw) in order to center-out k-space sampling. The terminus of the radial spokes is isotropically distributed on a sphere as a generalized spiral set [41]. To decrease motion artifacts, the spokes randomly regroup into sub-spirals with pseudo-isotropically distributed spokes. 800 dummy projections were applied prior to the first readout and 32 dummy projections were applied between sub-spiral to ensure that steady state achieves and holds during acquisition. Additional acquisition parameters specific to MB-SWIFT not found in Table 1 include: number of projections (np) = 2048; number of sub-spirals (nspirals) = 5; number of views in each sub-spiral (nv) = 2250. This highly under sampled acquisition protocol was chosen in a practical compromise between required resolution/snr of multi-sided study and minimization of the total time for the rats being under anesthesia. As a result each MB-SWIFT sequence took only 56 sec for a total in vivo scan time per animal of 16 min 57 sec. This time includes a conventional 3D gradient echo (3DGE) sequence (5 min 13 sec) that was acquired with no saturation scheme at the lowest achievable TE for comparison as a conventional imaging control.

Table 1.

MRI scan parameters.

Sequence	Saturation	Flip Angle (°)	TE (ms)	TR (ms)	Averages	Matrix	FOV (mm3)	Voxel Size (μm)	Time
3DGE	No Saturation	20	2.41	4.78	1	256x256x256	40x40x40	156x156x156	5 m 13 s
MB-SWIFT	No Saturation	2	~0	3.37	1	256x256x256	40x40x40	156x156x156	56 s
		4	~0	3.37	1	256x256x256	40x40x40	156x156x156	56 s
		6	~0	3.37	1	256x256x256	40x40x40	156x156x156	56 s
MB-SWIFT	Fat Saturation	2	~0	3.37	1	256x256x256	40x40x40	156x156x156	56 s
		4	~0	3.37	1	256x256x256	40x40x40	156x156x156	56 s
		6	~0	3.37	1	256x256x256	40x40x40	156x156x156	56 s
MB-SWIFT	Water Saturation	2	~0	3.37	1	256x256x256	40x40x40	156x156x156	56 s
		4	~0	3.37	1	256x256x256	40x40x40	156x156x156	56 s
		6	~0	3.37	1	256x256x256	40x40x40	156x156x156	56 s

TE= echo time; TR= repitition time; FOV= field of view; 3DGE= 3D gradient echo

MB-SWIFT images were reconstructed using gridding and a fast iterative thresholding algorithm (FISTA) available at CMRRpack (http://www.cmrr.umn.edu/swift/). Because of the 3D radial nature of MB-SWIFT, the sequence is highly insensitive to motion thus no motion correction was applied.

In vivo Micro-Computed Tomography (μCT)

In vivo μCT imaging (Bruker, Skyscan 1176) was acquired on the right proximal tibia with rats imaged supine and tibia secured to minimize motion. Images were obtained at an x-ray voltage of 70 kVp and current of 357 μA using a 1 mm aluminum filter to reduce beam hardening. Two frame averaging, 450 projections, and integration time of 85 msec with a total scan time of 6 min and 47 sec. Scans were reconstructed at 35 μm isotropic voxel size using vendor-supplied software (NRecon, Bruker, Version 1.7.1.6.).

Image Analysis

All image analysis was performed using in-house algorithms developed in MATLAB (2018b, Mathworks, Natick, MA, USA). Full resolution μCT images (35 μm) were used to derive the trabecular and cortical segmentation masks for each animal tibia (n=7) at each imaging time point (baseline, 2, 4, 10, 12 wk) and used as the standard volume of interest (VOI) for comparing μCT and MRI outcomes (described in the next section) (Figure 1A,B). Specifically, trabecular and cortical masks were created 2.5 mm distal to the proximal tibial growth plate and spanned 3 mm in the z-direction. The cortical mask included only cortical bone while the trabecular mask contained both marrow and trabecular bone encompassing the entire non-cortical region inside the tibia. Masks were contoured using an automated segmentation algorithm based on a pre-defined threshold to isolate the cortical mask and were followed by a set of mathematical morphology operations to extract the trabecular mask.

Figure 1.

Image processing schematic. A. Representative coronal slices from the same region performed using In vivo Multi-Band SWeep Imaging with Fourier Transformation MRI (MB-SWIFT, 156 μm resolution) and μCT (35 μm resolution) of the right proximal tibia acquired at baseline and post ovariectomy (OVX). B. Trabecular and cortical segmentation masks were derived using μCT images at each time point. Masks began 2.5 mm distal from the growth plate spanning 3 mm distally and created the volume of interest (VOI) for analysis. Cortical masks included only cortical bone and trabecular masks contained both marrow and trabecular bone encompassing the entire non-cortical region inside the tibia. C. μCT image was registered to the MB-SWIFT MRI image assuming a rigid body transformation, where the checkerboard images of the resultant registration were used as quality control for accuracy. D. Registered μCT image (156 μm) was used for further analysis. E. The transform from each registration for each animal (n=7 animals at 6 time points for a total of 42 registrations) was applied to corresponding trabecular and cortical VOIs. F. Transformed VOIs were used across imaging modalities to guide image analysis (G). Total registration time took ^~ 1 min and 3 sec and VOI transformation took ^~47 sec.

Image Registration

To quantify longitudinal changes in the cortical and tibial VOIs following OVX, μCT images were registered to the MB-SWIFT MR images for each rat at each time point using Elastix (version 4.8), an open-source registration algorithm with mutual information as an objective function and simplex as an optimizer. The registration was automatically performed and assumed rigid-body geometry (only rotation and translation) of the right tibia (Figure 1C,D). The resulting transforms derived from each registration [each animal (n=7 at each time point for a total of 42 registrations)] were applied to the μCT-trabecular and cortical VOIs which created identical VOIs between μCT and MB-SWIFT MRI to guide quantitative imaging analysis (Figure 1E–G).

Signal to Noise Ratio (SNR)

Signal to noise ratio (SNR) was used to compare signal efficiency between the conventional 3DGE MRI and the MB-SWIFT-NS and MB-SWIFT-FS across baseline in vivo acquisitions using the right proximal tibia. Mean SNR was calculated in manually segmented cortical regions beginning 3 mm distal from the growth plate and spanning 5 slices distally according to the following equation:

$$\text{SNR}=\text{SImean}/\text{SDnoise}$$

where SImean is the mean signal in the region and SDnoise is the standard deviation of the background noise in the image.

Bone Mineral Density (BMD)

Trabecular and cortical volumetric BMD (vBMD) was calculated in the original μCT images reconstructed at 35 μm resolution and registered μCT where resolution was down sampled to match that of the MRI scans (156 μm) and referred to herein as down sampled μCT (DS μCT). Scans were acquired with a known calcium hydroxyapatite density calibration phantom (CHA), a water phantom, and an air region. The mean attenuation coefficient of air (μA) and water (μW) were determined and used to convert the image attenuation coefficient (μ_ijk) to Hounsfield units (HU_ijk) at a given voxel location, ijk, by applying a linear transformation to the data as

$${\text{HU}}_{\text{ijk}}={\mu }_{\text{ijk}}\ast \text{M}+\text{B}$$

where

$$\text{M}\left(\text{the}\text{slope}\right)=1000/\left(\mu \text{W}-\mu \text{A}\right)$$

$$\text{B}\left(\text{the}\text{intercept}\right)=\left(-{\text{M}}^{\ast }\mu \text{W}\right)$$

Once in HU’s, the mean HU of the CHA phantom (HU_CHA) VOI was calculated. The mean HU of the trabecular (HU_TRAB) and cortical (HU_CORT) VOIs was determined, and the vBMD for each was calculated as

$${\text{vBMD}}_{\text{TRAB}\text{or}\text{CORT}}=\left\{\frac{\left[H{U}_{TRABorCORT}\times {\rho }_{CHA}\right]}{H{U}_{CHA}}\right\}$$

where, ρ_CHA is the known concentration of the CHA standard (1.64 mg/mm³).

Using MB-SWIFT MRI image with the lowest FA (closest to pure proton density in MB-SWIFT), signal intensity can be used to estimate BMD (SI MRI BMD) using the signal model adapted from Ho et al. [42]. We concurrently imaged a CHA standard and water phantom to derive the following

$$\text{SI}\text{MRI}{\text{BMD}}_{\text{TRAB}\text{or}\text{CORT}}=\left\{1-\left[\frac{S_{TRABorCORT}-S_{CHA}}{S_{WATER}-S_{CHA}}\right]\right\}\times {\rho }_{CHA}$$

where SI MRI BMD is the estimated BMD based on signal intensity, S_{TRAB or CORT} is the mean signal of the VOI of interest of the trabecular or the cortical region, and S_WATER is the average signal intensity for water. In the original equation described by Ho et al., S_WATER can be any standardized region in which there is no mineral and there is a maximum signal in the region [42]. In the original description, the authors used a region of subcutaneous fat; however, in our model of estrogen deficiency during growth, we refrained from using an internal reference to ensure no unintended effect of estrogen deficiency in the fat region and instead chose a standardized water phantom. A depiction of the acquired CHA standard and water phantom using MB-SWIFT-NS can be found in Figure 2.

Figure 2.

Example image of the calcium hydroxyapatite (CHA) phantom in water acquired using MB-SWIFT MRI with no saturation scheme. Calibration phantom and a separate glass tube filled with water (water phantom) were acquired under the same experimental parameters. Regions of water, air, and CHA phantom were used to convert MR images to BMD values based on signal intensity proposed by Ho et al. MB-SWIFT= Multi-Band SWeep Imaging with Fourier Transformation.

Sensitivity to water loss in cortical bone subject to sequential drying

MB-SWIFT sensitivity to changes in cortical water was evaluated under progressive dehydration using bone from three four-month-old female rats who had not undergone OVX. Immediately following sacrifice, bilateral tibiae were removed and dissected free of soft tissue including removal of the periosteum. The proximal and distal ends were removed and marrow was flushed completely from the cortices. Using the left tibia (n=3), prepared bone was pat dry and weighed in ambient air with an electronic balance (Mettler-Toledo AE50, Columbus OFI) at baseline. Dehydration was performed in an oven at 110°C while maintaining a constant vacuum. Weight was acquired every 5 minutes for the first 100 minutes and every 30 minutes thereafter until full dehydration was reached. Full dehydration was marked by a plateau in weight for at least three consecutive observations 30 minutes apart. Weight and time were plotted, and dehydration time was calculated for 1/4 dehydration, ½ dehydration, and full dehydration for each of the three tibiae as:

$$\%\text{Water}\text{loss}\text{by}\text{weight}=\left({\text{W}}_{\text{initial}}-\text{W}\right)/{\text{W}}_{\text{initial}}$$

where, W_initial denotes the initial bone wet weight, and W is the weight taken throughout the dehydration protocol.

The contralateral tibias (n=3, right) were imaged ex vivo at baseline and following dehydration under oven/vacuum condition to reach 1/4 dehydration, ½ dehydration and full dehydration using the imaging protocol described in detail above. Tibiae entered the 110°C oven under vacuum for time each contralateral tibia required to reach 1/4 dehydration. Tibia were immediately imaged, returned to oven and the process was repeated until ν dehydration, and full dehydration had been imaged. To mitigate any effects of transmit and receive gain changes over image acquisition at multiple time-points, a water phantom was imaged in addition to the experimental tibia as an internal control for any MR field homogeneity alterations (i.e. B1 field changes). Weights were taken before and after imaging to ensure bone did not alter weight due to further dehydration in air by more than 0.05%, which has been used in previous reports. Water fraction maps were calculated (described in the next section) in the cortical bone VOI to evaluate MB-SWIFT’S sensitivity to cortical water content loss during sequential drying.

Water Fraction and Fat Fraction MRI Maps

Signal fat fraction (FF) maps and water fraction (WF) maps were derived from consecutive MB-SWIFT images acquired with fat saturation (MB-SWIFT-FS), water saturation (MB-SWIFT-WS), and no saturation scheme (MB-SWIFT-NS) each using a minimal FA in order to achieve near proton-density image and minimize T1 bias. The FF maps which distinguish signal arising from fat protons in the image, were calculated as:

$$\text{FF}=1-\left({\text{S}}_{\text{waterFat}}/{\text{S}}_{\text{water}}\right)=\left({\text{S}}_{\text{water}}-{\text{S}}_{\text{waterFat}}\right)/{\text{S}}_{\text{water}}$$

where S_waterFat is the signal derived from the water saturated image, and S_water is the signal from image with no saturation from fat.

And the WF maps were calculated as:

$$\text{WF}=1-\left({\text{S}}_{\text{fat}}/{\text{S}}_{\text{fatFree}}\right)=\left({\text{S}}_{\text{fatFree}}-{\text{S}}_{\text{fat}}\right)/{\text{S}}_{\text{fatFree}}$$

where S_fat is the signal from the fat saturated image, and S_fatFree is the signal from the fat image where there was no saturation from water.

MB-SWIFT T1 Relaxation using the Variable Flip Angle (T1-VFA) Method

T1 relaxation maps of the MB-SWIFT-NS MRI using the variable flip angle method (T1-VFA) were fitted using a voxel-by-voxel linear fitting relative to equation E1 [40], defined below, based on theoretical signal intensity described by Treier et al. [43]. The signal intensity is given by

$$S=M_0\frac{\left[\text{sin}\alpha \times \left(1-E1\right)\times E2\right]}{\left[1-E1\times \text{cos}\alpha \right]}$$

Where M₀is the equilibrium longitudinal magnetization, α is the FA, and

$$E1=-e^{\left(TR/T1\right)}$$

$$E2=-e^{\left(TE/T2\right)}$$

where, T1 & T2 are the relaxation times, TE is the echo time, and TR is the repetition time. The above equation can be represented in the linear form as

$$S/\text{sin}\alpha =E1\times S/\text{tan}\alpha +M_0\times \left(1-E1\right)\times E2$$

However, in the above equation E2 is set to 1 since the effective TE for MB-SWIFT-NS acquisitions is 0. Since TR is kept constant and there is no input from TE, the measured signal changes from each FA is fit to a line characterized by the slope α = E1 and the intercept b= M0*(1-E1) and T1 values are calculated in the cortical and trabecular VOI.

Histology

To support MRI outcomes, tibia from one additional rat euthanized at six weeks of age (no OVX, representing our baseline time point) and one rat from the 12 week post-OVX group were stained for marrow fat deposition using an Oil Red O- isopropanol method described by Lillie et al. [44]. In short, tibia was removed free of soft tissue, fixed for 24 hours in 10% neutral buffered formalin, rinsed in tap water and embedded in optimal cutting temperature compound (OCT, Tissue-Tek; Sakura Finetek USA) in preparation for cryosectioning following standard laboratory procedure. Embedded non-decalcified tibia were sectioned longitudinally at 5 μm thickness on a cryostat (Leica CM30505, Nussloch, Germany), where sections were adhered using the Kawamoto tape method [45] followed by the staining protocol. Oil Red O stained sections were imaged (bright-field) at 20x magnification using a Nikon Eclipse Ni-U microscope (Nikon Instruments Inc., Melville, NY).

Statistical analysis

For comparison of In vivo signal efficiency, one-way analysis of variance (ANOVA) with Bonferonni corrections were used to compare SNR between conventional 3DGE MRI with the lowest achievable TE, MB-SWIFT-NS and MB-SWIFT-FS for cortical, marrow, and muscle VOIs. Pearson correlation coefficients were used to test the linear association between μCT BMD measures at 35 μm resolution, the adjusted resolution DS μCT BMD measures at the nominal resolution of the MRI (156 μm), and the MB-SWIFT SI MRI BMD. Simple linear regressions were used to determine the predictive value of SI MRI BMD (156 μm) against μCT BMD at 35 μm resolution, our gold standard approach. To determine MB-SWIFT’s sensitivity to changes in water during sequential drying (dehydration), a simple linear regression model was used to determine association between WF and gravimetric weight in the cortical bone. Sensitivity of In vivo imaging outcomes to longitudinal changes in cortical and trabecular VOIs following OVX from baseline was determined using repeated measures ANOVA (RM ANOVA) followed by Holm-Sidak’s multiple comparison test. Specifically, we tested the sensitivity of μCT BMD, SI MRI BMD, WF, and T1 relaxation to detect changes following OVX in the cortical VOI. For the trabecular VOI, analysis was carried out using the following outcomes: μCT BMD, SI MRI BMD and FF. GraphPad Prism Version 8.0.2. was used for data analysis. In all cases, p ≤ 0.05 was considered statistically significant.

Results

MB-SWIFT achieves significantly greater signal to noise ratio (SNR) in the tibia compared to 3DGE MRI

Representative consecutive in vivo MRI acquisitions of the rat proximal tibia using 3DGE, MB-SWIFT-FS, and MB-SWIFT-NS were acquired for the study (Figure 3A,B). For the cortical bone VOI, MB-SWIFT-NS and MB-SWIFT-FS achieved significantly greater SNR compared to 3DGE acquired with the lowest achievable TE (Figure 3C). Note, the sensitivity enhancement in bone regions in MB-SWIFT relative to 3DGE was observed despite a 5 times faster acquisition time.

Figure 3.

In vivo MRI signal to noise ratio (SNR) measures. A. We compared the signal efficiency in terms of SNR between conventional 3D gradient echo (3DGE, left, contrast-adjusted for visualization) with the lowest achievable TE, Multi-Band SWeep Imaging with Fourier Transformation with fat suppression (MB-SWIFT-FS, middle) and MB-SWIFT with no saturation scheme (MB-SWIFT-NS, right). B. SNR measures were taken in the right tibia beginning 3 mm distal from the growth plate and spanned 5 consecutive slices distally. C. SNR mean ± standard deviation was reported using manually drawn volumes of interest (VOIs) in the cortical bone. MB-SWIFT achieved significantly greater SNR in the cortical bone. There were no significant differences in SNR between MB-SWIFT-NS and MB-SWIFT-FS. ****p<0.0001. Significance was set at p ≤ 0.05.

MRI-derived BMD significantly correlates with μCT BMD measures

We observed a significant positive linear correlation between μCT BMD and MB-SWIFT-derived SI MRI BMD in both the cortical and trabecular VOIs (Figure 4A,B). Specifically, cortical BMD derived from the full resolution μCT, our gold standard, demonstrated a strong linear correlation with SI MRI BMD (r=0.65) (Figure 4A). Correlations were also statistically significant between trabecular μCT and SI MRI BMD (r=0.62) (Figure 4B). BMD derived between μCT methods (full resolution μCT vs. DS μCT) were highly correlated (cortical: r=0.97; trabecular: r=0.90, data not shown). Linear regressions between μCT BMD and SI MRI BMD were significant in all cases and y-intercepts for trabecular BMD were significantly different (p<0.0001).

Figure 4.

Relationship between MRI-derived bone mineral density (BMD) and μCT BMD (gold standard). BMD derived from MB-SWIFT using the signal intensity method described by Ho et al. (SI MRI BMD) is plotted against BMD calculated from the full resolution μCT. The relationship between the μCT and MRI methods to derive BMD were statistically significant in both the cortical (A) and trabecular (B) VOIs in all cases (denoted by bolded p-values next to the respective regression equation). The y-intercept between BMD methods in the trabecular VOI were statistically different. For each BMD method, regression lines (solid) and coefficients (dotted with shaded fill) are shown with the 95% confidence intervals with r-values of r=0.65 for cortical (A) and r=0.62 for trabecular (B) VOIs. Significance was set at p ≤ 0.05. MB-SWIFT = Multi-Band SWeep Imaging with Fourier Transformation where the no suppression acquisition was used for calculation.

Both MRI and μCT detected significant longitudinal increases in BMD

We evaluated longitudinal changes in cortical and trabecular BMD measured using full resolution μCT and MB-SWIFT-derived SI MRI BMD using RM ANOVAs (Figure 5A,B). For cortical BMD, RM ANOVA results for each modality was significant (μCT: p<0.0001; MRI: p<0.0001). Cortical BMD increased from baseline following OVX and follow-up Holm-Sidak’s multiple comparison test determined this was significant by 2 weeks which was observed in both imaging modalities. For trabecular BMD, RM ANOVA results were significant for SI MRI BMD (p<0.0001) but not for μCT (p=0.06). Trabecular BMD measured with MB-SWIFT increased from baseline following OVX and was significantly different by 2 weeks post-OVX. Trabecular BMD measured using μCT was significantly increased from baseline by 10 weeks post-OVX.

Figure 5.

Longitudinal in vivo measurement of cortical and trabecular bone mineral density (BMD) from full resolution μCT images (A, top row) and MB-SWIFT-derived BMD based on signal intensity (SI MRI BMD) (B, bottom row) presented as mean BMD (g/cm³) ± standard deviation for each time point. Repeated measures ANOVA results were significant for cortical μCT BMD and cortical and trabecular SI MRI BMD (p<0.0001, p<0.0001, p<0.0001, respectively) but not for trabecular μCT BMD (p=0.06). Results from follow-up post-hoc analysis using Holm-Sidak’s multiple comparison test to detect differences in BMD compared to baseline (pre-OVX) are indicated by stars when significance was reached at P≤0.05.

MB-SWIFT is sensitive to small alterations in bone hydration and detected a significant decrease in cortical water by 10 weeks post OVX

Correlations between ex vivo cortical MB-SWIFT WF and water loss (measured as a % loss by volume) during sequential drying under oven/vacuum were performed and results can be seen in Figure 6A. As WF decreased, percent water loss by volume increased; this strong negative correlation was significant between the two measures (r=−0.98; p=0.01). Results indicate MB-SWIFT WF is highly sensitive to cortical water loss. While BMD measures significantly increased in our model (likely as a result of growth overcoming estrogen deficiency changes), we evaluated the utility of WF to detect additional bone quality changes in the cortical bone consistent with OVX. The longitudinal effects of OVX on cortical WF significantly decreased longitudinally from baseline following OVX (RM ANOVA WF: p<0.0001) and was significantly decreased from baseline by 10 weeks post-OVX (Figure 6B and C). Further, cortical WF negatively correlated with cortical μCT BMD (r=−0.6; p=0.0002) (data not shown).

Figure 6.

A. MB-SWIFT water fraction (WF) is sensitive to cortical water loss during sequential drying. As MB-SWIFT WF decreased, percent water loss by volume increased; this strong negative correlation was significant (r=−0.98; p=0.01). B. Cortical WF decreased longitudinally following OVX (reaching significance from baseline by 10 weeks post-OVX). Results from follow-up post-hoc analysis using Holm-Sidak’s multiple comparison test to detect differences in WF compared to baseline (pre-OVX) are indicated by stars when significance was reached at p≤0.05. **p=0.004; ****p<0.0001. C. In vivo coronal water fraction (WF) images with cortical VOI overlay at baseline through 12 weeks post-OVX. MB-SWIFT= multi-band SWeep Imaging with Fourier Transformation.

Marrow fat fraction significantly increased by 10 weeks post OVX

Marrow fat measured as the FF in the trabecular VOI increased longitudinally following OVX (Figure 7A). RM ANOVA was significant (p<0.0001) and Flolm-Sidak’s post hoc determined that FF significantly increased from baseline by 10 weeks post-OVX. Undecalcified cryosections stained with Oil Red O support the finding where increased marrow fat deposition is observed in 12 week post-OVX sections (Figure 7B). Longitudinal changes in marrow fat from baseline through 12 weeks post-OVX for one rat can be observed in Figure 7C.

Figure 7.

A. Trabecular fat fraction (FF) increased longitudinally following OVX (reaching significance from baseline by 10 weeks). Significance was set at p ≤ 0.05. ****p<0.0001. B. Bright field images of tibial marrow region stained using Oil Red O at baseline (top, no OVX) and following 12-weeks post OVX (bottom) depict increases in lipid deposit (red) following OVX. C. In vivo coronal FF images with trabecular VOI overlay at baseline through 12 weeks post-OVX. FF increase is denoted by the increasing signal (white) within the trabecular region over time. Representative longitudinal FF images are from one rat. VOI=volume of interest.

Cortical matrix T1 relaxation time decreased significantly by 10 weeks post-OVX

Mean T1 relaxation measured using VFA MB-SWIFT increased two weeks following OVX and then began to decrease over time through the 12 week post-OVX timepoint (Figure 8A–B). RM ANOVA results comparing T1 relaxation times longitudinally were significant (p=0.0005) and Flolm-Sidek post hoc test for multiple comparisons determined that T1 relaxation was significantly decreased from baseline by 10 weeks post-OVX. There was an observed negative relationship between T1 relaxation times and BMD but this association was not significant (p=0.08, data not shown). T1 map overlay demonstrate a slight increase in T1 values at 2 weeks post-OVX followed by a gradual decrease in T1 values in the cortical bone through 12 weeks (Figure 8B).

Figure 8.

Cortical matrix T1 relaxation times decrease significantly following OVX in growing rats. A. Mean ± standard deviation (SD) of T1 relaxation times in the cortical VOI measured using the variable flip angle method in MB-SWIFT images at baseline through 12 weeks post-OVX (n=7). RM ANOVA with a Holm-Sidak post-hoc test determined T1 relaxation times significantly decreased from baseline by 10 weeks post-OVX (brackets and stars). 10wks *p=0.03; 12wks *p=0.02. B. Representative axial tibia MB-SWIFT images with cortical T1 relaxation map overlay for one rat at baseline through 12 weeks post-OVX. MB-SWIFT= Multi-Band SWeep Imaging with Fourier Transformation MRI (used to acquire volumetric T1 relaxation time measures); VOI= volume of interest; OVX= ovariectomy.

Discussion

Because the high atomic mass number associated with calcium (Z=20) gives rise to a higher photoelectric absorption compared to that of soft tissue, there is a near ubiquitous reliance on ionizing x-ray based techniques to visualize the mineral phase of bone [46]. As a result, characterization of disease progression and treatment efficacy for metabolic bone diseases is complicated by the lack of imaging modality able to safely monitor material-level and biochemical changes In vivo. To improve upon x-ray-based BMD, we tested the efficacy of a novel 3D-MRI approach, MB-SWIFT, in an estrogen deficient (OVX) model of osteoporosis during growth. This provided a rapidly changing system to evaluate our proposed MRI bone biomarkers of cortical WF, cortical matrix T1 relaxation times, and marrow FF in addition to an MRI-derived measure of BMD. BMD measured by MB-SWIFT correlated significantly with BMD measured using the pre-clinical gold standard, μCT, in the trabecular and cortical regions that significantly increased longitudinally over the duration of the study. Growth appeared to overcome estrogen-deficient changes in bone mass in our rat model yet MB-SWIFT MRI detected significant changes consistent with estrogen deficiency by 10 weeks in cortical WF, cortical matrix organization (T1 relaxation times), and marrow FF. Further, MB-SWIFT-derived cortical WF was strongly correlated to water loss during sequential drying indicating the technique is highly sensitive to even small changes in an already limited cortical water pool. Together, findings point to MB-SWIFT MRI’s ability to quantify BMD in good agreement with the gold standard μCT, while demonstrating the additive benefit of detecting biochemical and material-level alterations consistent with disease independent of the mineral phase suggesting its value for bone imaging.

Not surprisingly, MB-SWIFT MRI was able to achieve significantly greater SNR in bone compared to conventional 3DGE using the lowest achievable TE. This increase highlights the immense improvement in signal efficiency achievable in the extremely short T2 of the cortical bone In vivo. SNR itself has been used as a biological outcome measure in studies applying regular SWIFT, acquired ex vivo, in cortical bone [31, 32]. While we did not use SNR as a biological outcome measure in the present MB-SWIFT study, it should be noted that Sukenari and Minami et al. each determined SNR was positively correlated with new bone area measured using histomorphometry in rats with diabetes and following OVX and that SNR weakly correlated with increasing BMD following OVX [31, 32]. Further, Minami et al. observed that SWIFT-derived SNR significantly detected differences between diabetic rats and controls by 2 weeks where changes in BMD were not detectable until week 8 [31].

BMD is used clinically to classify the onset and extent of osteoporosis [1, 47] and as a pre-clinical endpoint in osteoporosis related research [48]. Because MB-SWIFT can achieve ^~TE=0, we hypothesized that it would be sensitive to the mineral phase of bone which has a short T2 ^~ 10 μs [15], and therefore could quantify BMD without the use of harmful ionizing radiation. We imaged a CFIA standard and water along the rat tibia using MB-SWIFT and quantified trabecular and cortical BMD using the signal intensity approach described by Flo et al. [42]. Compared to the high resolution μCT BMD, MB-SWIFT SI MRI BMD significantly correlated to both trabecular and cortical BMD demonstrating its promise to resolve changes in BMD in good agreement with an established technique.

MB-SWIFT BMD was able to significantly detect longitudinal increases in cortical bone comparable to BMD measured by μCT. Flowever, for the trabecular region, MB-SWIFT BMD detected a significant increase by 2 weeks while μCT was not significantly elevated from baseline until week 10. In aged models of estrogen deficiency (OVX) bone loss, bone loss is greater and occurs more rapidly in trabecular bone compared to cortical bone [49, 50]. In the present study, the estrogen-related changes in the trabecular region may have been stronger than in the cortical bone, therefore growth-induced increases in BMD may not have overcome estrogen deficiency-induced bone loss until 10 weeks post-OVX when measured using μCT. While MB-SWIFT demonstrated significant correlations with μCT BMD in the trabecular region, the correlations were not 1:1 due to the method in which signal is captured where MB-SWIFT MRI excites protons vs. attenuation of the tissue (μCT). Inspection of the regression plots in the trabecular VOI show that the MRI-derived approach appears to overestimate BMD compared to μCT. The contribution of signal from the marrow (fat and water) may have impeded the BMD values in the region. In contrast, μCT trabecular VOI shows tissue attenuation for the trabeculae and little signal contributions from the marrow and water. As such, the increased BMD values for the trabecular region in MRI may have been confounded by the increased dynamic range of the MRI image compared to that achieved using μCT. Even so, we believe the statistical correlations in both VOIs coupled with the demonstrated sensitivity to longitudinal change, especially in the cortical VOI, support the utility of MB-SWIFT to characterize the mineral phase of bone, in vivo, without the use of ionizing radiation.

In human subjects, Li et al. demonstrated the utility of UTE MRI-derived bulk water (total) measures at the mid diaphysis of the tibia where positive correlations were observed with age and negative correlations with BMD [51]. Using MB-SWIFT to quantify water fraction in the cortical region of the rat proximal tibia, we also observed a significant negative correlation between cortical water and BMD measured using μCT. While both bound and free water pools contribute to the signal that are captured in the MB-SWIFT water fraction maps, the bone has a larger portion of water existing in the bound form located in matrix [52] found either loosely bound to collagen [15] or more tightly to the mineral [53]. Further, Du et al. reported that UTE MRI signal in human cortical bone was greater than 77.6% attributable to the bound water pool [54] therefore we hypothesize that the greater contribution of our measured water signal was from bound water although it should be re-stated that we did not directly measure contribution separately from each pool (but instead from all pools). Using a solid-state NMR spectroscopy technique, Cao et al. suggested that protons bound to the matrix more closely relate to the matrix composition and can infer mineral density of bone [55]. The authors used a solid-state approach to achieve zero-TE where the most molecularly immobile components could be imaged including signal in the solid bone matrix (where authors used suppression of fluid signals, like our study, to exploit this). With MB-SWIFT MRI, time-domain signals are acquired during a swept RF excitation in a time-shared manner thus achieving a nearly zero TE and therefore the technique can capture all protons in the bone including matrix bound water [22]. Bound water has shown to decrease with age in rats [56] and in human cadaveric femurs [16] and decrease in rats with high bone turnover rates in a model of chronic kidney disease [57]. Bone turnover rate is increased both during aging and with osteoporosis and can lead to deterioration of bone microarchitecture affecting bone quality [58] which may be observed independent of BMD. In the present study, we observed a significant decrease in cortical water by 10 weeks post-OVX, consistent with increased bone turnover due to estrogen deficiency, even while BMD continued to increase.

Free pore water concentration on the other hand may be a surrogate to measure cortical porosity [16]. Li et al. applied a saturation recovery scheme to selectively quantify pore water in ex vivo cadaveric tibia specimens and demonstrated that UTE MRI demonstrated strong positive correlations with both age and μCT porosity [59]. It has been previously shown that porosity increases as a result of age and this change accounts for over 70% of reduction of strength in bone [60, 61]. Using 1H NMR, Horch et al. observed that bound water positively correlated with peak stress while free water (pore) negatively correlated using human cortical bone specimens [62]. The ability to separate these water pools using in vivo acquisitions is desirable and work has been done using bi-component analysis of cortical water pools acquired by UTE both in preclinical setting and in humans at clinical field strengths [52, 63]. In a future study, we would adopt MB-SWIFT to T2 sensitive experiments using magnetization preparation schemes [64] as well as using MB-SWIFT with saturation schemes at the pore and bound peak to derive water contributions from each pool.

Traditional x-ray-based approaches also fall short in characterizing bone marrow fat. There is a strong relationship between increased marrow fat observed in patient biopsies with osteoporosis [65–67] suggesting its promise as an imaging target for the disease. Conventional MRI has demonstrated utility in quantifying marrow fat in vivo [68–70], but conventional MRI is unable to provide additive information regarding the surrounding mineral and other material-level properties in bone. In the present study using MB-SWIFT, we detected an increase in marrow FF longitudinally following OVX. This increase was statistically significant from baseline by 10 weeks supportive of clinical experience of increased bone marrow fat in osteoporotic bone samples [65–67]. Prior studies have suggested that fatty marrow is associated with reduced trabecular bone mass. Studies have demonstrated a negative relationship between lower bone density measured by DXA and higher fat content measured by conventional MRI [69–72], where it has been suggested that low BMD may result from increased differentiation of mesenchymal stem cells to adipocytes instead of osteoblasts [73, 74]. However, in our study we did not observe a statistical relationship between increasing FF and μCT BMD in the growing rats following OVX (r=0.27; p=0.21). It has been reported that both DXA and CT are vulnerable to incorrect reading of BMD in the presence of large increases in marrow fat [75, 76]. Even so, we reason that the lack of association was likely more attributable to the continued growth of the animal influencing the increased BMD in our study. We believe the detectable increase in FF observed independent of increasing BMD in our growing animals following OVX highlights the discriminating power of MB-SWIFT-measured FF. Supported by histology, these data suggest its additive utility as an imaging biomarker of increased bone turnover due to disease and estrogen deficiency.

T1 relaxation time holds promise as a quantitative imaging biomarker based on tissue-specific relaxation properties because T1 reflects the energy flow between spins and their local microenvironment. T1 relaxation times have shown to be sensitive to small variations in healthy tissue [77, 78] and more importantly, is able to distinguish changes due to disease [77, 79–81]. Until the advent of UTE and zero-echo time (ZTE) MRI, T1 relaxation times of short T2 species such as bone has been unattainable. Since the description of SWIFT, Wang et al. described its utility of conventional SWIFT using the variable flip angle (VFA) method to measure T1 relaxation times in aqueous suspensions of iron oxide nanoparticle in excellent agreement with spectroscopic measures [82] and others have applied ex vivo VFA SWIFT to quantify T1 relaxation times in osteochondral specimens [40]. We applied the VFA method using MB-SWIFT to reduce susceptibility artifact at the cortical borders where we calculated volumetric T1 relaxation times in vivo. We observed the cortical matrix T1 relaxation times increased immediately after baseline and then decreased longitudinally and were statistically different from baseline by 10 weeks. While cortical water also decreased longitudinally in the study, there was no statistical association between decreasing T1 and decreasing cortical WF. Interestingly, while not statistically significant, T1 relaxation time decreased with increasing BMD. Nissi et al. observed that T1 relaxation times decreased closer to the more mineralized region in the cartilage bone interface reflecting the higher sensitivity to short T2 spins located at the cartilage/bone interface [40]. Using 3D solid state phosphorus-31 NMR projection imaging ,Wu et al. reported that subtle molecular or crystalline structural differences in the mineral were reflected as large changes in T1 highlighting the discriminating power of the parameter to its local environment [83]. We hypothesize that the significant decrease in VFA MB-SWIFT T1 reflects its ability to directly image the changing bone matrix composition. This included an increase in mineral density and changes in the collagen environment including cross link concentration and collagen morphology both of which are observed in osteoporosis [11, 84, 85] and would affect the mobility of protons measured during longitudinal relaxation (T1) resulting in shorter T1 relaxation times. The fast view-shared data acquisition, saturation recovery Look-Locker scheme for SWIFT could be used in future studies to shorten the time needed to generate 3D T1 maps [86].

Limitations

We were limited by the age of our animals at the time of OVX (six weeks old) where the rats were reproductively mature yet still growing [87]. The continued growth appeared to compete with estrogen deficiency-induced changes in the mineral phase where BMD significantly increased longitudinally over the course of the study. The impact of growth on OVX-related changes has been highlighted in other studies. Turner et al. observed that rats ovariectomized at seven weeks of age demonstrated a significantly increased bone formation rate at the tibial diaphysis by four weeks compared to sham controls [88]. The authors did not observe osteopenic changes in the OVX animals and reasoned that increasing radial growth overshadowed increases in endosteal resorption. In another study, rats ovariectomized at 10 weeks of age significantly increased bone formation in the tibia (measured at 5 weeks post-OVX) and the authors postulated estrogenic antagonism may have been overshadowed by the stimulatory effects of growth hormone [89]. It should be restated that our use of OVX in growing rats satisfied the goal of the present study where we sought to compare imaging modalities, including the sensitivity of their outcomes, longitudinally in a system of changing bone mass and material composition. While BMD increased over time in our growing rats following OVX, MB-SWIFT was able to characterize BMD in a manner which significantly correlated to μCT and MRI provided additive quantitative outcomes about bone quality in a manner consistent with estrogen deficiency. The present study was not cross sectional and we did not use a SHAM operated control group for comparison. Instead, we focused on the sensitivity of MRI to detect longitudinal In vivo changes following OVX compared to measures acquired at baseline. A SHAM group, however, may have permitted OVX-related changes in BMD to be detected over time even during growth. Lietner et al. demonstrated that SHAM-operated 3 month old female rats longitudinally increased bone growth (significantly) by 10% at four weeks [90]. In the OVX rats, the authors determined that this growth rate “dampened” BMD results measured longitudinally from baseline. When the authors compared the OVX rats to the SHAM operated group, a significant decrease in OVX BMD compared to SHAM was observed. Lastly, we used μCT to derive our BMD measures in order to characterize the skeletal changes with the highest resolution possible in our lab. DXA remains the clinical gold standard approach to quantify BMD in patients including those with post-menopausal osteoporosis. Therefore, it would be advantageous for future studies to compare DXA-based BMD with BMD derived using the MB-SWIFT MRI approach described in this paper to fully compare to the current clinical standard.

There is increasing interest in characterizing microarchitecture, in addition to BMD, as it relates to bone health and biomechanics [91–93] which is almost ubiquitously measured using high-resolution μCT pre-clinically and high-resolution peripheral quantitative computed tomography (HR-pQCT) in clinical studies [94]. Compared to μCT, proton-based MRI is limited by the In vivo spatial resolution it can achieve while maintaining a sufficient SNR and limited its ability to resolve microarchitecture of the trabeculae, for example. Using our current resolution In vivo MB-SWIFT MR images, we were able to achieve significant correlation in BMD measures between high resolution μCT (r = 0.67 for cortical region and r = 0.62 for trabecular region). Future work should focus on increasing resolution to determine if resolving cortical thickness and eventually trabecular microarchitecture of the rat bone is attainable while maintaining enough SNR to obtain meaningful biomarker measures in the bone.

Conclusions

There are currently no accepted approaches to measure bone quality, non-invasively, in the clinical setting. We show MB-SWIFT’s pre-clinical potential by overcoming MRI’s inability to directly image the mineral phase of bone In vivo while simultaneously providing quantitative bone quality measures. We believe our results support the efforts towards eventual clinical implementation of ultra-short and zero-echo time MR techniques to image bone. Specifically, MB-SWIFT significantly detected longitudinal increases in BMD and correlated to BMD acquired using the gold standard approach. Results support the utility of MB-SWIFT to characterize the mineral phase of bone, longitudinally, without ionizing radiation. While BMD increased following OVX, likely as a result of growth, the proposed MB-SWIFT MRI biomarkers targeting additional measures of bone quality detected significant changes in the bone at 10 weeks that were consistent with estrogen deficiency following OVX without the use of harmful ionizing radiation. Results support the promise of MB-SWIFT to non-invasively image bone In vivo.

Highlights:

• MB-SWIFT MRI-derived BMD correlated significantly with BMD measured using μCT

• MB-SWIFT MRI provided additive quantitative outcomes about bone quality

• Changes in cortical water, matrix organization and marrow fat were detected by MRI