More accurate trabecular bone imaging using UTE MRI at the resonance frequency of fat

Abstract

High-resolution magnetic resonance imaging (HR-MRI) has been increasingly used to assess the trabecular bone structure. High susceptibility at the marrow/bone interface may significantly reduce the marrow’s apparent transverse relaxation time (T2*), overestimating trabecular bone thickness. Ultrashort echo time MRI (UTE-MRI) can minimize the signal loss caused by susceptibility-induced T2* shortening. However, UTE-MRI is sensitive to chemical shift artifacts, which manifest as spatial blurring and ringing artifacts partially due to non-Cartesian sampling. In this study, we proposed UTE-MRI on the resonance frequency of fat to minimize marrow-related chemical shift artifacts and the overestimation of trabecular thickness. Cubes of trabecular bone from six donors (75±4 years old) were scanned using a 3T clinical scanner on the resonance frequencies of fat and water, respectively, using 3D UTE sequences with five TEs (0.032, 1.1, 2.2, 3.3, and 4.4 ms) and a clinical 3D gradient echo (GRE) sequence at 0.2×0.2×0.4 mm³ voxel size. Trabecular bone thickness was measured in 30 regions of interest (ROIs) per sample. MRI results were compared with thicknesses obtained from micro-computed tomography (μCT) at 50 μm³ voxel size. Linear regression models were used to calculate the coefficient of determination between MRI- and μCT-based trabecular thickness. All MRI-based trabecular thicknesses showed significant correlations with μCT measurements. The correlations were higher (examined with paired Student’s t-test, P<0.01) for 3D UTE images performed on the fat frequency (R²=0.59–0.74, P<0.01) than those on the water frequency (R²=0.18–0.52, P<0.01) and clinical GRE images (R²=0.39–0.47, P<0.01). Significantly reduced correlations were observed with longer TEs. This study highlighted the feasibility of UTE-MRI on the fat frequency for a more accurate assessment of trabecular bone thickness.

1. Introduction

Bone mineral density (BMD), as measured by dual-energy x-ray absorptiometry (DXA) at the spine or hip, is the standard clinical measure to diagnose osteoporosis and estimate bone fracture risk [1–4]. Despite the widespread use of BMD in clinics, a diagnosis of osteoporosis (based on DXA T-score ≤−2.5) often fails to predict fracture risk accurately [5–12]. Notably, because of the DXA-based BMD two-dimensional (2D) nature, its measurement cannot detect local changes in bone structure.

High-resolution magnetic resonance imaging (HR-MRI) has been demonstrated as a promising tool for in vivo trabecular bone imaging [13–23]. MRI is a noninvasive three-dimensional (3D) imaging modality free from ionizing radiation, which is a significant advantage over standard 3D bone imaging modalities. For example, computed tomography (CT) may expose subjects to a considerable dosage of ionizing radiation, depending on the imaging coverage and resolution. HR peripheral quantitative CT (HR-pQCT) has a much lower radiation dose but is still a research modality used only in peripheral sites. Moreover, MRI can acquire structural and compositional information from surrounding soft tissues such as bone marrow [24,25], tendon [26], cartilage [27], and muscle during the same scan session for more thorough pathological assessments.

HR MRI indirectly visualizes trabecular bone as dark regions surrounded by marrow with bright signals, because trabecular bone has much lower proton density and shorter apparent transverse relaxation time (T2*) than marrow[28]. Such indirect visualization has been used to track changes in trabecular bone microstructure in response to medical treatments [29–31] and for finite element analysis (FEA) based assessment [23,32]. With image post-processing, it is possible to render the 3D architecture and extract the corresponding structural parameters of trabecular bone [28,33–36]. Considering the average size of trabecular bone, the in-plane HR MRI pixel sizes are often selected to be near 0.2 mm [23,37].

The susceptibility difference between trabecular bone and bone marrow [38,39], the heterogeneity of complex hierarchical bone structure [40], and the presence of multiple fat resonance peaks [39], all contribute to the reduction in marrow T2*, resulting in marrow signal loss and thus an overestimation of trabecular bone thickness. Specifically, a portion of marrow at the bone boundary (reduced T2*) is likely to be considered as bone, falsely, by image thresholding and segmentation methods. Ultrashort echo time MRI (UTE-MRI) allows for acquiring signals from tissues with short T2* [41,42], such as bone and its neighboring marrow. UTE-MRI can minimize signal loss due to T2* shortening. Furthermore, the non-Cartesian sampling techniques such as radial or rosette/petal trajectories employed in UTE-MRI offer distinct advantages over Cartesian sampling employed in conventional MRI, providing improved k-space coverage and reduced susceptibility to motion artifacts, which is particularly beneficial for applications requiring motion robustness [43]. However, UTE-MRI is sensitive to chemical shift artifacts, which manifest as spatial blurring and ringing artifacts due to non-Cartesian sampling [44]. Specifically, in Cartesian sampling, the chemical shift artifact occurs in the frequency-encoding direction as a shift in the spatial location of fat voxels (no chemical shift artifact in the phase encoding direction). In non-Cartesian 3D UTE sampling, the frequency encoding is performed in three dimensions. As a result, spatial misregistration and ringing artifacts happen in all directions.

In this study, we proposed UTE-MRI on the resonance frequency of fat to minimize marrow-related chemical shift artifacts and signal loss due to susceptibility-induced T2* shortening, thereby reducing the overestimation of trabecular thickness. It is hypothesized that HR UTE-MRI can minimize marrow-related chemical shift artifacts if the center frequency is shifted to the fat peak. Bone is expected to be off-resonance in fat-centered imaging. However, bone signal, while detectable with UTE-MRI, is much lower than marrow signal due to its low proton density and ultrashort T2*. As a result, a negligible water-associated off-resonance artifact is anticipated. We will investigate the feasibility of HR UTE-MRI on the fat peak frequency for a more accurate depiction of trabecular bone structure in human distal tibial specimens at 3T. MRI-based trabecular bone thickness, the most common measure used in the literature, will be compared with the micro-computed tomography (μCT) results [21,45–53].

2. MATERIALS AND METHODS

2.1. Sample preparation

Fresh-frozen cadaveric specimens of the tibia from six donors (75±4 years old) were provided by the UC San Diego School of Medicine Medical Education/Anatomical Services. The axial section of the distal tibia, 30 to 50 mm above the medial malleolus, was cut into ~20 mm segments using a commercial band saw. One 20 mm³ cube was excised from the bone metaphysis region of each specimen using a low-speed diamond saw (Isomet 1000, Buehler, IL). Only trabecular bone was included in the final bone cubes. After being thawed, trabecular bone cubes were soaked in perfluoropolyether (Fomblin, Ausimont, Thorofare, NJ) and placed under a negative pressure vacuum for about 2 hours to reduce air bubbles. Next, specimens were placed in a rectangular plastic container (80×100×40 mm, approximately) filled with perfluoropolyether to minimize dehydration and susceptibility artifacts during the MRI scans.

2.2. UTE-MRI protocol

The UTE-MRI scans were performed on a 3T clinical scanner (GE Healthcare, Waukesha, WI) using an eight-channel transmit and receive knee coil. A 3D UTE Cones sequence with TR = 12.1 ms and five TEs (0.032, 1.1, 2.2, 3.3, and 4.4 ms) and a clinical 3D Cartesian gradient echo (GRE) sequence (TR = 9.6 ms, TE = 4.4 ms) were performed twice, first on the water peak frequency and then on the fat peak frequency. The fat peak frequency was selected by the MR operator after performing a manual pre-scan. The field-of-view (FOV), acquisition matrix, slice thickness, voxel size, and number of slices were 80 mm, 400×400, 0.4 mm, 0.2×0.2×0.4 mm³, and 160. The approximate scan time was 8.5 minutes for 3D UTE and 12.8 minutes for GRE sequences, respectively.

2.3. Micro-computed tomography (μCT)

Specimens were also scanned using a GE eXplore 120 Preclinical μCT scanner at 50 μm³ isotropic voxel size. Other scanning parameters were as follows: FOV = 100 mm, 60 kV voltage, 32 mA current, 0.5° rotation step, number of averages per frame = 2, and 80 minutes total scan time.

2.4. MRI and μCT image analysis

MRI images were analyzed using 10 slices in the middle of each specimen, covering 4 mm in the Z direction. Three ROIs were selected at each slice by an experienced image analyst while avoiding regions affected by infiltrated air bubbles. 30 ROIs per specimen (180 ROIs for six specimens) were selected in total. Corresponding μCT images were selected manually (8 consecutive μCT slices for each MRI slice). A 2D semiautomatic registration algorithm was used to map the selected ROIs onto the μCT images. Registration was performed after selecting the matching corners of each specimen in MRI (moving in the registration) and μCT (fixed in the registration) images. Notably, employing a 3D automatic registration algorithm was not applicable due to the artifacts in MRI images caused by the trapped air in the marrow space. Also, selecting ROIs on registered MR images would make it difficult to detect and avoid air bubbles, as enlarged registered MR images looked noisier and did not cover the entire specimen.

A local adaptive gray level thresholding algorithm was used to segment bone pixels from marrow pixels within each selected polygon ROI with a size of ~1.5×1.5 cm² on the MRI and μCT images. Local thickness was calculated at each pixel using the distance transform performed on the segmented images. Specifically, the local trabecular thickness in a 2D fashion equals the diameter of the largest covering circle (Figure 3 H).

Figure 3.

A representative trabecular bone specimen (excised cube from metaphysis in distal tibial) in the axial plane using 3D UTE sequence (0.2×0.2×0.4 mm³ voxel size) acquired on the fat peak frequency, TE=0.032 ms (A). Signal inversion of 3D UTE image highlights trabecular bone structure (B). μCT provides high isotropic (0.05×0.05×0.05 mm³ voxel size) imaging of trabeculae (C). The corresponding zoomed regions indicated with the red boxes are shown in the second row (D-F). Schematics of the trabecular bone boundaries (G) and the largest covering circles to calculate the local bone thickness (H). The thin regions of bone are indicated with small covering circles in red, while the thick regions are indicated with larger covering circles in yellow and white.

For each ROI in MRI images, the contrast-to-noise ratio (CNR) was calculated between the bone and marrow regions ($CNR=({Signal}_{marrow}-{Signal}_{bone})/{Noise}_{background}$). CNR is expected to always be a positive variable as marrow demonstrates a higher signal than bone.

2.5. Statistical analysis

A simple linear regression model of MRI-based trabecular bone thickness as a function of μCT-based trabecular bone thickness was defined (${MRI}_{Th}=A\times {\mu CT}_{Th}+B$, where A and B are constant values) to calculate the coefficient of determination, R². The statistical significance was determined using the Student’s t-test. To ensure that intra-specimen dependency did not affect the results, all correlation studies were repeated using one average measure per sample instead of multiple ROIs. All image processing steps and statistical analyses were performed using in-house developed programs in MATLAB (version 2021, The Mathworks Inc., Natick, MA, USA). P values less than 0.05 are considered as significant.

3. Results

Figure 1 shows HR MR images of a representative distal tibia specimen acquired at the fat peak frequency and water peak frequency, respectively, using 3D UTE acquisition with a TE of 0.03 ms. Significant chemical shift artifacts were observed in UTE images acquired at the water peak frequency (arrows in Figure 1A), blurring trabecular bone structure (Figure 1A and zoomed Figure 1C). In contrast, UTE images acquired at the fat peak frequency showed no noticeable chemical shift artifacts (Figure 1B and zoomed Figure 1D), providing a superior depiction of trabecular bone structure. The contrast between bone and marrow is noticeably higher in UTE images acquired at the fat peak frequency than at the water peak frequency.

Figure 1.

A representative trabecular bone specimen from the distal tibial was imaged in the axial plane using the 3D UTE sequence on the water peak frequency (A) and the fat peak frequency at TE=0.03 ms (B). The corresponding zoomed regions indicated with the red dashed-line boxes are shown in the second row (C-D). The voxel size in all images is 0.2×0.2×0.4 mm³. UTE images acquired at the water peak frequency show significant chemical shift artifacts, manifesting as blurred trabecular bone structure and ringing artifacts compared to the fat peak. The contrast between bone and marrow is noticeably higher in UTE images acquired at the fat peak frequency than at the water peak frequency.

Figure 2 shows HR MR images of the same distal tibia specimen (Figure 1) acquired at the fat peak frequency using 3D UTE acquisition with increasing TEs from 0.03 ms to 4.4 ms (Figure 2A–E and zoomed Figure 2G–K). These images are visually compared with 3D GRE (Figure 2F and zoomed Figure 2J) acquisition at the fat peak frequency. The darker pixels within the specimens represent the trabecular bone, which seems larger and overestimated with increasing TEs (from the left column to the right column subfigures). Interestingly, the darker regions seem more pronounced at TE 1.1 ms and 3.3 ms in some locations (indicated with thin arrows and arrowheads in the lower row) due to the water and fat

Figure 2.

A representative trabecular bone specimen from distal tibial imaged in the axial plane using 3D UTE sequence at the fat peak frequency at TE=0.03 ms (A), TE=1.1 ms (B), TE=2.2 ms (C), TE=3.3 ms (D), and TE=4.4 ms (E), and clinical GRE (3D-Cartesian) sequence at TE=4.4 ms (F). The corresponding zoomed regions indicated with the red dashed-line boxes are shown in the second row (G-L). The voxel size in all images is 0.2×0.2×0.4 mm³. The darker region associated with the trabecular bone increased with TE, leading to an overestimation of trabecular thickness.

Figure 3 shows a representative trabecular bone section used for MRI- and μCT-based assessment of the trabecular bone thickness. UTE-MRI on the fat peak frequency shows excellent bone marrow structure in the axial plane at 0.2×0.2×0.4 mm³ voxel size (Figures 3A, D). Trabecular bone shows as signal void due to its much lower proton density and shorter T2* relaxation time. Simple signal inversion reversed the image contrast with a high signal from trabecular bone and a low signal from marrow (Figures 3B, E). Figures 3C and F show the corresponding μCT images acquired at 0.05×0.05×0.05 mm³ voxel size. Figure 3G shows the schematics of the trabecular bone boundaries in a small zoomed-in section. The schematic largest covering circles used to calculate the local bone thickness are depicted in Figure 3H.

Figure 4 shows the local trabecular thickness map within a representative ROI selected manually (on the fat peak frequency UTE image) to avoid artifacts caused by air trapped in the bone of the same representative trabecular bone section in Figure 3. The local trabecular thicknesses were calculated based on the MR images on the fat peak frequency (3D UTE at TEs=0.03, 1.1, 2.2, 3.3, and 4.4 ms and GRE at TE=4.4 ms), and based on the ground truth μCT images, respectively. The estimated local trabecular thickness was higher for MR images acquired at longer TEs. MRI-based trabecular thicknesses were noticeably higher than μCT-based results.

Figure 4:

A representative ROI selected on the fat peak frequency UTE image (A), and the corresponding local trabecular bone thickness map generated for TE=0.03 ms (B), TE=1.1 ms (C), TE=2.2 ms (D), TE=3.3 ms (E), and TE=4.4 ms (F). The trabecular thickness maps for the matched representative ROI on the clinical GRE (G) and μCT images (H) are also shown. The estimated local trabecular thickness was higher for MR images acquired at higher TEs. MRI-based trabecular thicknesses were obviously higher than μCT-based results.

Figure 1S in the supplemental materials demonstrates the local trabecular thickness maps using the corresponding MR images centered on the water peak frequency.

Figure 5 demonstrates the scatter plots and the linear regressions of the MRI-based trabecular thicknesses on the μCT-based results, including 30 ROIs per specimen (180 ROIs for six specimens, degrees of freedom = 178). All MRI-based trabecular thicknesses showed significant correlations with the μCT-based results. The correlations were higher for 3D UTE MR images performed on the fat peak frequency (R²=0.59–0.74, P<0.01, F-value =198–502) than those on the water peak frequency (R²=0.18–0.52, P<0.01, F-value = 40–189). UTE and TE =1.1 ms images on the fat peak frequency had higher correlations (R² = 0.74 and 0.73, F-value =502 and 494, respectively, P<0.01) with μCT-based results than UTE images acquired at longer TEs (R² = 0.59–0.67, P<0.01, F-value =198–358), or UTE images acquired on the water peak (R² = 0.18–0.52, P<0.01, F-value = 40–189), or the clinical GRE images (R² = 0.47 and 0.53, P<0.01, F-value = 158 and198).

Figure 5:

Scatterplots and linear regressions of MRI-based trabecular bone thicknesses on the μCT-based results, including 30 ROIs per specimen (n=180).

Figure 6 demonstrates similar scatter plots and the linear regressions when the bone thicknesses were averaged per specimen (n=6, degrees of freedom = 4) to ensure that intra-specimen dependency does not affect the results. All trabecular thicknesses from fat peak frequency MRI showed significant correlations with the μCT-based results (R²=0.88–0.99, P≤0.01, F-value =20–264). For trabecular bone thicknesses from water peak frequency MRI, correlations were statistically significant only for images at TE=1.1 (R²=0.81, P=0.01, F-value =17) and TE= 4.4 ms (R²=0.78 and 0.82, P=0.02 and 0.01, F-value =14 and 18, respectively). UTE and TE=1.1ms images on the fat peak frequency had the highest correlations with μCT-based results (R²=0.95 and 0.99, F-value =73 and 264, respectively, P<0.01). In summary, using the paired Student’s t-test, the MRI-μCT correlations were significantly higher (P<0.01) when 3D UTE MR images were performed on the fat peak frequency than those on the water peak frequency.

Figure 6:

Scatterplots and linear regressions of MRI-based trabecular bone thicknesses on the μCT-based thickness. All thicknesses were averaged per specimen (n=6).

Figure 2S in the supplemental materials demonstrates the average CNR between the trabecular bone and the marrow in the studied specimens using different 3D UTE MRI sequences. Fat-centered 3D UTE images showed higher CNRs than water-centered images. The UTE image on the fat peak frequency showed the highest CNR compared with other acquisitions.

4. Discussion

This study was the first to investigate the feasibility of using UTE-MRI at the fat peak frequency to improve trabecular bone thickness assessment. The μCT-based trabecular bone thickness strongly correlated with MRI-based results when the fat peak frequency UTE or TE=1.1 ms images were used. Correlations were lower on average when higher TEs were utilized. The μCT-MRI correlations were higher for MR images performed on the fat peak frequency than those on the water peak frequency, on average. UTE MR imaging centered on the fat peak frequency significantly minimized marrow-related chemical shift artifacts. Specifically, bone trabeculae are expected to be off-resonance in fat-centered imaging, and due to the much lower signal in bone than in marrow, the water-associated off-resonance artifact is negligible. Meanwhile, UTE data acquisition minimizes signal loss due to susceptibility-induced T2* shortening. As a result, UTE MRI at the fat peak frequency improves indirect imaging of trabecular bone by resolving chemical shift artifacts while minimizing susceptibility effects. The superior performance of fat-centered UTE-MRI is likely to be more pronounced when a higher fat fraction in the marrow is expected, such as in older osteoporotic patients, particularly in their lower extremities [54–56]. Nevertheless, such hypotheses are to be examined in future investigations.

Trabecular bone thickness has been the most used metric in the literature for trabecular evaluation [21,45–53]. Thus, an accurate MRI-based trabecular thickness is of great interest to the bone research community. Moreover, accurate rendering of trabecular bone structure can be coupled with micro-FEA to assess bone mechanical competence [23]. Predicting the mechanical properties of bone, particularly in trabecular bone sites, is challenging because of bone’s heterogeneous and anisotropic nature, such as the dependencies on loading direction, anatomical location, and sample dimension [57–59]. In micro-FEA, the displacements, forces, and stress and strain tensors can be calculated throughout the meshed volume of trabecular bone to predict its mechanical competence. Developing micro-FEA based on fat-centered UTE imaging of trabecular bone is likely to improve mechanical competence prediction, which has not been investigated yet [23].

In addition to HR structural MRI, other MRI-based techniques for trabecular bone assessment can be categorized into two groups: bone-marrow relaxometry (T1, T2*, and T2’ analyses) in relatively low-resolution images [33,60–65], and direct bone imaging via UTE-MRI with marrow signal suppression [66–69]. Most low-resolution MRI-based analyses of trabecular bone have been focused on marrow relaxometry or magnetic susceptibility measurements [33,60–65]. These techniques can indirectly quantify trabecular bone density and structure [41,53,54,63]. The strong susceptibility between trabeculae and marrow interface leads to greatly reduced relaxation times for bone marrow, depending on bone volume and bone-specific surface [33,62]. Bone marrow relaxation times are correlated with BMD in different studies [28,33,48,62,63,70], however as determined by the low image resolution, they are not capable of trabecular bone thickness assessment achievable by the proposed fat-centered UTE MRI technique. Direct trabecular bone imaging has been reported using water- and fat-suppressed MRI techniques [66–69]. To selectively image trabecular bone, it is critical to suppress signals from long T2 tissues, particularly marrow fat, because of its much higher signal than bone. The fat-suppressed UTE sequence detects signals from water bound to the organic matrix [71], thereby providing an indirect measurement of organic matrix density in trabecular bone [69]. However, this technique is subject to an inherently low signal-to-noise ratio (SNR) efficiency due to the low water concentration in porous trabecular bone. It is difficult or impossible to directly image the trabecular bone networks, therefore fat-suppressed UTE sequences are able to map organic matrix density but unable to assess the trabecular bone thickness, which is achievable with the fat-centered UTE-MRI technique.

The limitations of this study can be described in the following five aspects. First, this study was performed on bone specimens cut from the bone metaphysis region, in which cortical bone and the surrounding muscles were removed. The presence of muscles, other soft tissues, and subject motion will contribute to the reduced performance of all HR MR imaging techniques in vivo compared with ex vivo studies. However, it is possible to apply the fat-centered UTE-MRI technique to assess trabecular bone structure in vivo, as demonstrated in the supplemental Figure 3S. The calcaneal trabecular bone of a 43-year-old male volunteer was depicted with higher contrast in images acquired at the fat peak frequency than the water peak frequency and greater marrow signal loss was observed in images acquired at longer TEs. Future in vivo investigations are necessary to demonstrate the superiority of fat-centered UTE imaging for better depiction of trabecular bone microstructure. Second, FEA based on MRI data would provide interesting comparisons between the examined sequences. In future studies, the FEA of trabecular bone structure based on fat-centered UTE-MRI is expected to provide more accurate bone mechanical characteristics. Third, the HR MR imaging was performed using an anisotropic voxel size, which was appropriate based on the authors’ experience. Future investigations should be performed to seek the practical anisotropic and isotropic voxel sizes for optimal human trabecular bone depiction with sufficient signal-to-noise ratio while avoiding long scan times. Fourth, the trabecular number is an important outcome but was not investigated. Fourth, Fourth, the trabecular number is an important outcome but was not investigated. This study only focused on the trabecular bone thickness. The performance of the proposed UTE-MRI sequence in detecting the number of trabeculae should be investigated in future studies. Fifth, this study was performed using UTE sequences. While all the major MR vendors have UTE techniques or variants available on their systems, these sequences are still in the research stage, which limits their availability to other researchers and clinicians.

5. Conclusions

The feasibility of UTE-MRI at the fat peak frequency for more accurate trabecular bone assessment was investigated for the first time in the literature. The correlations between μCT- and MRI-based trabecular thicknesses were higher when MR was performed on the fat peak frequency instead of the water peak frequency. UTE MRI at the fat peak frequency minimizes the marrow-related chemical shift artifacts and susceptibility-induced T2* shortening simultaneously. The μCT-based trabecular bone thickness strongly correlated with MRI-based results when the fat peak frequency UTE or TE=1.1 ms images were used. Correlations were lower on average when higher TEs were utilized. This study highlighted the feasibility of UTE on the fat peak frequency for a more accurate trabecular bone thickness assessment.