MRI-Derived Porosity Index is Associated with Whole-Bone Stiffness and Mineral Density in Human Cadaveric Femora

Abstract

Ultrashort echo time (UTE) magnetic resonance imaging (MRI) measures proton signals in cortical bone from two distinct water pools, bound water, or water that is tightly bound to bone matrix, and pore water, or water that is freely moving in the pore spaces in bone. By isolating the signal contribution from the pore water pool, UTE biomarkers can directly quantify cortical bone porosity in vivo. The Porosity Index (PI) is one non-invasive, clinically viable UTE-derived technique that has shown strong associations in the tibia with μCT porosity and other UTE measures of bone water. However, the efficacy of the PI biomarker has never been examined in the proximal femur, which is the site of the most catastrophic osteoporotic fractures. Additionally, the loads experienced during a sideways fall are complex and the femoral neck is difficult to image with UTE, so the usefulness of the PI in the femur was unknown. Therefore, the aim of this study was to examine the relationships between the PI measure in the proximal cortical shaft of human cadaveric femora specimens compared to (1) QCT-derived bone mineral density (BMD) and (2) whole bone stiffness obtained from mechanical testing mimicking a sideways fall. Fifteen fresh, frozen whole cadaveric femora specimens (age 72.1 ± 15.0 years old, 10 male, 5 female) were scanned on a clinical 3-T MRI using a dual-echo UTE sequence. Specimens were then scanned on a clinical CT scanner to measure volumetric BMD (vBMD) and then non-destructively mechanically tested in a sideways fall configuration. The PI in the cortical shaft demonstrated strong correlations with bone stiffness (r = −0.82, P = 0.0014), CT-derived vBMD (r = −0.64, P = 0.0149), and with average cortical thickness (r = −0.60, P = 0.0180). Furthermore, a hierarchical regression showed that PI was a strong predictor of bone stiffness which was independent of the other parameters. The findings from this study validates the MRI-derived porosity index as a useful measure of whole-bone mechanical integrity and stiffness.

Introduction:

Osteoporosis is a degenerative metabolic disease which impairs bone strength and often shows no symptoms until the first fracture [1, 2]. There are an estimated 200 million people in the world with osteoporosis, including 30% of all of the postmenopausal women in the United States and Europe [1]. Hip fractures are the second most common type of osteoporotic fracture, affect one in six postmenopausal women during their lifetime, and cause the greatest mortality and morbidity of all fractures, with a 20% mortality rate in the first year following a hip fracture [3, 4]. Clinical diagnosis of osteoporosis is based on radiographic evaluation of bone mineral density (BMD) obtained via Dual-energy X-ray absorptiometry (DXA), which is cheap, fast, and widely accessible [1, 2, 5]. However, DXA is commonly seen as a poor predictor of osteoporotic fracture and has a sensitivity of less than 50% [6–8]. Indeed, in a year-long study of 150,000 white postmenopausal women, of the 2,200 incident fractures recorded, 82% of them were in women above the DXA threshold for osteoporosis diagnosis and 77% did not meet treatment guidelines [9]. This is due in part to the 2D projections of bone acquired in DXA measurements being fraught with errors arising from complex 3D bone morphologies and heterogenous absorption from non-osseous tissue [7, 8]. Additionally, DXA measurements are incapable of differentiating between trabecular and cortical bone.

Cortical bone is subjected to roughly half of the loads in the femoral neck region during a sideways fall [10] and comprises 80% of the whole-body bony mass [11]. Furthermore, cortical bone undergoes extensive age-related changes characterized by endocortical erosion, periosteal expansion, and increase in intracortical pore volume [12–14] and it is a dominant factor in fractures above the age of 65 [15]. These changes are especially prominent following postmenopausal hormone changes and in diseases such as diabetes, hyperparathyroidism, and obesity, among others [16–20]. Cortical porosity is a major determinant of whole bone strength [21, 22] and is related to fractures independent of BMD measurements [23–27]. Existing osteoporotic medications, namely bisphosphonates and RANKL inhibitors, have been shown in several studies to improve cortical bone quality in the form of reduced cortical porosity, increased cortical bone mineral density, and increased cortical area [28–33]. As a result, there has been enormous interest in non-invasive in vivo methods to image and map cortical bone porosity.

High-resolution peripheral quantitative computed tomography (CT) (HR-pQCT) has been proven to reliably evaluate cortical bone porosity in the peripheral skeleton [10, 16, 25–29, 34–36]. Second generation scanners can image at 61 μm isotropic voxel size [34, 35] which is sufficient to image the larger pores that exist within the Haversian canals (40–100 μm) but cannot resolve smaller pores in the Haversian canals or the even smaller pores in the lacunae and canaliculi (10–30 μm and 0.1–1μm) [13, 37, 38]. On the other hand, μCT scans can image the whole cross-section of bone with voxel sizes of 9 μm which can resolve most of the pores in bone and is the accepted gold standard method for assessing bone porosity, but it can only be performed in cadaveric bone ex vivo or small animal in-vivo studies due to its exceedingly high ionizing radiation exposure and small scanning field-of-view [39]. HR-pQCT is also limited to the distal skeletal sites of the radius or tibia and cannot image deep structures such as the proximal femur, which is the most clinically relevant bone for osteoporotic fracture assessment. Although osteoporosis is a systemic disease, it is important to note that the microstructure in distal extremities are only weakly correlated to the those in deep central sites [36, 40, 41]. Recent advances in spectral CT scanners and photon counting detectors have enabled accurate, phantomless quantification of BMD [42, 43] and improvements in resolution of up to 150 μm [44], both of which could potentially improve in vivo imaging of femoral cortical bone, but they have not yet been validated for this application.

Although the signal in cortical bone is completely decayed by conventional clinical magnetic resonance imaging (MRI) echo times, ultrashort echo time (UTE) MRI is capable of directly imaging cortical bone signal in vivo. The MRI labile water protons in cortical bone are split into two pools, bound water and pore water. Pore water (PW) is water that is freely moving in larger bone pores and has a transverse relaxation time (T2*) above 1 msec, while bound water (BW) is water that is hydrogen bound to collagen which, based on its heavily restricted motion, has a much shorter T2* around 300 μsec [45]. Since pore water signal is proportional to the pore spaces within bone, isolating the pore water signal provides a direct measure of bone porosity. Various UTE-based bone water measures have been proposed including direct measurement with adiabatic radiofrequency (RF) pulses [46, 47], subtraction between total water (TW) content and adiabatic inversion RF pulse-isolated bound water content [48, 49], fitting the signal decay from multiple echoes with Bi-component [50, 51] or Tri-component models (where the third component is fat) [52, 53], measuring the magnetization transfer from macromolecules to water [54, 55], and computing the ratio of the two magnitude signals from a dual-echo UTE scan (i.e. Porosity Index (PI)) [13, 56]. While these methods differ greatly in their approach and complexity, they have all been validated independently and are related to cortical bone quality. More specifically, Manhard et. al quantified BW and PW from adiabatic inversion recovery and double adiabatic inversion recovery methods in 40 human cavaeric radii samples. PW showed a strong inverse relationship with bending strength and a positive relationship with μCT-derived porosity and BW was moderately related to bending strength [57]. Bae et. al demonstrated in tibial and femoral sections that pore water was strongly associated with μCT-derived porosity and bound water was moderately associated with ultimate stress [58]. Similarly, Jerban and colleagues showed in separate studies of femoral and tibial sections that macromolecular fraction from magnetization transfer modeling was correlated to cortical bone porosity, bone mineral density, ultimate stress, and histomorphometry [55, 59] and that tricomponent analysis obtained water content is related to cortical bone porosity and mechanical properties [52]. Finally, Porosity Index was shown to strongly correlate to μCT-derived porosity, pore size, pore water fraction [13], tibial stiffness, as well as near-infrared spectral imaging-derived bound water and collagen content [54].

Taken together, these studies suggest the immense potential of UTE bone water measures for assessing cortical bone quality in vivo. However, the aforementioned studies were performed in machined cortical sections of the tibia or femur and in the distal radius. While machined bone sections allow for more controlled inspection of the relationship between MRI measures and bone properties, to our knowledge the relationship between UTE bone water measures and femoral whole-bone integrity has never been investigated. Additionally, the complex loads experienced by the proximal femur during a catastrophic sideways fall typically result in fractures at the femoral neck [60]. However, the cortical bone in the femoral neck is an extremely thin and deep structure which has lower contrast from its surrounding trabecularized regions and lower signal-to-noise than other parts of the femur on UTE scans [61]. We therefore hypothesize that UTE-derived porosity measures in the compact cortical bone of the proximal femoral shaft, which has higher contrast and is easier to reliably contour on UTE images than the femoral neck, can provide a useful surrogate for evaluating whole bone strength and cortical bone quality. As such, the primary aim of this study was to examine the relationship between UTE-MRI derived cortical Porosity Index and whole bone experimental stiffness obtained by the sideways loading configuration in whole, unfixed human cadaveric femora. Sideways fall mechanical test are designed to recapitulate the complex in vivo loading conditions experienced during catastrophic femoral fractures. Porosity index was chosen since it is the simplest of the pore water mapping methods and does not require the use of external calibration phantoms or specialized adiabatic inversion pulses and it can be performed in clinically viable acquisition times. Secondary aims of this study were to investigate the relationship between Porosity Index, CT-derived BMD, and cortical thickness, as well as assess the feasibility of porosity mapping of the proximal femur in human subjects.

Methods:

Bone Specimens

Fifteen fresh frozen whole human cadaveric femora (age 72.13 ± 14.99, range 44–93) were harvested from ten male (age 73.10 ± 17.60, range 44–93) and five female donors (age 70.2 ± 7.0, range 65–81) which were obtained from a local biobank (National Disease Research Interchange, Philadelphia, PA). Bones were stored frozen at −30° C and were thawed for a minimum of 12 hours prior to UTE-MRI data acquisition which, based on previous studies, produces signal to noise ratio that best recapitulates in vivo conditions without degrading the samples [62, 63]. Following UTE scanning, femora were refrozen and at a later time were rethawed and scanned with a clinical CT scanner. The specimens were refrozen and then thawed one final time for an average of 24 hours prior to mechanical testing. A thin layer of soft tissue was left intact on the femora for UTE and CT imaging and was subsequently removed prior to mechanical testing to ensure the proper contact and consistent testing conditions were maintained.

UTE MR Imaging

MR imaging was performed at the University of Pennsylvania on a 3-T whole body clinical MRI scanner (Siemens Prisma, Erlangen, Germany) with an 18-element flexible RF coil (Siemens Body 18, Erlangen, Germany). Each whole femora was loaded onto the scanner bed in a feet first supine position to mimic standard clinical hip imaging protocols and were imaged using clinically available hardware. The femora were imaged with a custom-designed 3D ultrashort echo time sequence (Figure 1) that was written in Sequence Tree [64]. Briefly, in UTE acquisition the signal reception for Echo 1 at an echo time 1 (TE₁) begins immediately after a hardware-constrained dead time in a ramp-up sampling manner and maps the signals in a spherical k-space with a center-out trajectory. This study used a dual-echo UTE sequence where an additional gradient-recalled echo (Echo 2) is acquired at a later echo time (TE₂) with the same readout gradients as used for the first echo. Sequence parameters include flip angle 15°, isotropic FOV (180 mm)³, 50k radial spoke projections, TEs of 50 and 2150 μsec, volumetric rectangular RF pulse of duration 40 μsec, repetition time (TR) 12 ms, readout bandwidth = 250 kHz, readout ramp-up duration 240 μsec, reconstructed matrix 256×256×256, isotropic voxel size 0.7mm, scan time 10 minutes. Image reconstruction was performed using custom Matlab R2020a (Mathworks, Natick, MA, USA) scripts with the following steps: (A) Correcting for the k-space sampling positions using a pre-determined, actual trajectory information [65, 66], (B) Applying non-uniform Fourier transform [67] individually to all acquired raw datasets, and (C) Taking a square root of sum-of-squares across receive channel images.

Figure 1:

Sequence diagram for the dual-echo UTE sequence used in this study. Signal reception for TE₁ in the form of a free-induction-decay starts immediately following RF excitation and a hardware-constrained dead time. Spins were rewound following the first echo acquisition and a second echo acquisition was subsequently acquired at TE₂. A spoiler was applied at the end of each TR to squash the residual transverse magnetization. The RF pulse used had a short volumetric rectangular shape of 40 μsec duration.

Porosity Index

The cortical Porosity Index map was computed using the equation: [13]

$$\mathit{\text{Porosity}}\mathit{\text{Index}}(\text{\%})=\frac{{\mathit{\text{Echo}}}_2\mathit{\text{intensity}}}{{\mathit{\text{Echo}}}_1\mathit{\text{intensity}}}*100\text{\%}$$

Here Echo₁ corresponds to the first echo image of 50 μs and Echo₂ corresponds to the second echo image at 2150 μs. The first echo time should ideally be as short as possible to maximize SNR but 50 μs was the shortest achievable with our hardware. The second echo time of 2150 μs was chosen because it shown previously to provide the best combination of SNR and the dynamic range of Porosity Index.

The cortical bone region was manually segmented from the first echo of each scan using the Mimics 17.0 software. Since the cortical bone in the femoral neck has low contrast with the surrounding trabecularized region and soft tissue on UTE scans and it is difficult to reliably segment, we instead analyzed the nearby compact, cortical region of the proximal femora shaft that is just inferior to the lesser trochanter. This region allowed for a consistent and reproducible segmentation across all specimens. An analysis region of 1 cm (15 slices * 0.7 mm voxel size) was chosen and an additional step of a one voxel morphological erosion was applied to the outer periosteal boundary of the cortical segmentation to ensure that all voxels containing soft tissue were appropriately removed (Figure 2C). The average cortical porosity index within the analysis region was computed.

Figure 2:

Flow chart illustrating workflow for this study. (A) and (B) show analogous sagittal slices obtained from UTE and CT scans within the same cadaveric specimen. (C) highlights the image analysis methods employed, with a representative Echo 1 image and the corresponding manually segmented region of interest in the proximal femoral shaft. The analysis region highlighted in red was chosen as a continuous 1 cm region just inferior to the lesser trochanter. This region has high contrast with surrounding tissue and is consistent across different femoral morphologies and modalities which makes it easy to reliably segment. In panel (D), an overview of the mechanical testing methodology is shown. A representative picture of a cadaveric femora sample within the mechanical testing from is shown at the top. A cartoon of a typical mechanical testing load-deformation curve is shown in the bottom left. The stiffness is computed as the slow of the linear part of the load-deformation curve, which spans from the origin to the red dot on the curve. Finally, in the bottom right of (D), a cartoon of the loading conditions are overlaid on top of the UTE scan.

Additionally, the average cortical bone thickness (CbTh) for each cadaver was computed from the manual segmentations. Briefly, the cortical thickness of each slice was computed by modeling the periosteal and endosteal masks as uniform circles. The difference of the average periosteal and endosteal radii was defined as the cortical thickness for one slice. The reported cortical thickness was an average of the thickness over all of the slices within the analysis region.

Mechanical Testing

Specimens were nondestructively mechanically tested in a sideways fall configuration as previously reported (see Figure 2D) [63]. The head of the femur was positioned in a polymethylmethacrylate (PMMA, Trim Plus PMMA, Patterson Companies, Mendota Heights, Minnesota) cup designed to imitate the acetabulum and placed under the actuator of the mechanical test frame (Model 8874 servohydraulic testing machine, Instron Inc., Norwood, MA, 25kN load cell). The greater trochanter of each femora was aligned in a rigid base of polymethylmethacrylate above a metal base allowing for unrestricted translation in the X-Y plane. A 3-degree of freedom vise firmly held the femoral shaft roughly 10° above the horizontal to model the contact orientation of a sideways fall on one’s hip, similar to Keyak et al. [68, 69]. The PMMA cup was lowered till contact was established with the acetabulum. Contact was defined when a nominal amount of compressive load (<10 N) was observed on the load cell. All femora were subjected to 3 triangular waveforms of non-destructive loading between −50N and −1200N at a loading rate of 1.67 mm/s. The lower load limit for non-destructive tests was determined by destructively loading one sacrificial femur sample to failure. The chosen value of −1200N was 70% of the maximum failure load observed for this sample. This ensured that all samples would be loaded well into the linear region and would not encroach the plastic region. Time (sec), displacement (mm), and force (N) were recorded at 100 Hz throughout the test. Mechanical stiffness was computed as the slope of the linear region of the load-displacement curve from the last cycle.

CT Imaging

Bone mineral density was obtained using a clinical CT scanner (Revolution HD, GE Medical Systems, Chicago, Illinois) and commercially available calcium rod calibration phantoms of 50, 100, and 300 g/cc (Gammex Multi-Energy CT Phantom, Sun Nuclear Corporation, Melbourne, Florida) which were placed next to the cadaveric specimen within the FOV during the scans. Data were obtained for all but one specimen which was not available at the time of imaging. Scans were performed using an existing clinical axial bone imaging protocol on the scanner with the following parameters: 120 kVp, tube current 150 mAs, rotation time 0.8 seconds, Bone convolution kernel, 0.3 to 0.4 mm in-plane voxel size, 1.25 mm slice thickness and 0.5 mm slice increment, FOV 205 × 205 × 500 mm, matrix size 512 × 512 × 1000.

The calibration curve between Hounsfield Units (HU) and volumetric BMD (vBMD) was determined by taking 7 equally spaced manual ROIs covering the entire length of each calibration phantom and performing a 3-point fit to the known concentrations of the calcium rods. The fit was found to be stable between all scans (R² > 0.99) so a single calibration equation was used for all scans:

$$\mathit{\text{vBMD}}(mg/cc)=0.3286*HU-14.95$$

The endosteal and periosteal boundaries for each slice were manually segmented for each scan using the Mimics 17.0 software. An analysis region, matching that used for porosity index calculation, of 1 cm (21 slices * 0.5 mm slice thickness) was chosen just inferior to the lesser trochanter (see Figure 2C). The average vBMD of the cortical bone within this region was computed.

Statistical Analysis

Statistical analyses were conducted using JMP Pro Discovery Software (JMP 14.0 SAS Institute, Inc., Cary, NC, USA) and R version 3.5.1 (R Foundation for Statistical Computing, Vienna, Austria) with statistical significance set at α = .05. All tests were two-tailed. PI, BMD, and CbTh were correlated with each other using Pearson’s r. A nested regression was performed to assess if the three imaging parameters provide complementary information about whole bone stiffness. ANOVAs were used to compare nested models, as well as to check for interactions. Each permutation of PI, BMD, and CbTh was evaluated; additionally, the predictive power of the combination of PI and CbTh was compared to that of BMD alone using an ANOVA. The model determined to best predict whole bone stiffness was validated using a stepwise regression. For the purposes of comparing linear models, row-wise deletion was used to exclude the one specimen for which BMD values were could not obtained.

Results:

The porosity index maps demonstrated good spatial agreement with CT images (Figure 3). A summary of all correlation plots is shown in Figure 4. The porosity index was strongly related to whole bone stiffness (r = −0.82, P = 0.0014), QCT-derived vBMD (r = −0.64, P = 0.0149), and CbTh (r = −0.60, P = 0.0180). CbTh and BMD were also significantly related to whole bone stiffness (r = 0.75, P = 0.0021 and r = 0.58, P = 0.0374, respectively). CbTh and BMD were also significantly related (r = 0.53, P = 0.05).

Figure 3:

Coronal and transverse views comparing same slice location between UTE and CT scans. It is important to note that, while there is little contrast in the trabecularized region on the axial Echo 1 or Echo 2 images, the Porosity Index map shows excellent spatial agreement with the CT scan in this region. The Porosity Index values are percentages which are displayed based on the colorbar shown, where the highest level of porosity is shown in dark red and corresponds to pure fat, and the lowest level of porosity is shown in dark blue and indicates pure compact cortical bone.

Figure 4:

Correlation plots between imaging parameters and whole bone stiffness. Error clouds indicate 95% Confidence Intervals.

Multivariate Regressions of Whole Bone Stiffness

The results of the nested regressions for all three imaging parameters, PI, BMD, and CbTh, are shown in Table 2. Regardless of the order of the parameters added into the models, the addition of PI to any combination of the other parameters significantly improved the model in all cases (ΔR² = 0.21, 0.33, 0.20 and P = 0.02, 0.01, and 0.02). Similarly, cortical bone thickness improved the relationship to stiffness when added to BMD (ΔR² = 0.23, P = 0.01), but not when added to any model already containing PI (P = 0.06, 0.09). The addition of BMD to any of the other permutations did not improve the model in any case (P = 0.39, 0.47, 0.93). The model that best predicted whole bone stiffness was PI plus CbTh (ΔR² = 0.78, P = 0.001). A stepwise regression validated this finding. No interactions were found.

Table 2.

Analysis of Nested Models Predicting Stiffness (kN/mm)

Model	R2	Pa,b	ΔR2	Pa,c
BMD	0.34	.04
BMD + CbTh	0.57	.02	0.23	.01
BMD + CbTh + PI	0.78	.003	0.21	.02
BMD	0.34	.04
BMD + PI	0.67	.004	0.33	.01
BMD + PI + CbTh	0.78	.003	0.11	.06
CbTh	0.56	.003
CbTh + PI	0.76	.001	0.20	.02
CbTh + PI + BMD	0.78	.003	0.02	.39
CbTh	0.56	.003
CbTh + BMD	0.57	.02	0.01	.47
CbTh + BMD + PI	0.78	.003	0.21	.02
PI	0.67	.001
PI + CbTh	0.76	.001	0.09	.09
PI + CbTh + BMD	0.78	.003	0.02	.39
PI	0.67	.001
PI + BMD	0.67	.004	0	.93
PI + BMD + CbTh	0.78	.003	0.11	.06

Abbreviations: BMD, volumetric Bone Mineral Density (mg/cc); PI, Porosity Index (%); CbTh, cortical thickness (mm).

^aBold text indicates statistical significance at α = .05.

^bP-value of the linear model.

^cP-value of the difference between each linear model and the reduced model in the previous row.

Discussion:

This study was the first to examine the relationship between whole bone stiffness and UTE-MRI biomarkers of cortical bone. The cortical Porosity Index biomarker in the proximal femoral shaft was shown to be strongly associated with both whole bone stiffness and bone mineral density. Importantly, the hierarchical regression demonstrated that, while all three imaging measures were related to bone strength, the Porosity Index was a strong predictor of stiffness and the information it provided about bone health was independent of the other imaging parameters (Table 2). This confirms the notion that cortical porosity is related to bone strength independent of BMD [23–27].

The Porosity Index biomarker was first proposed by Rajapakse et. al in 2015 and was shown in cadaveric tibiae to strongly correlate to the gold standard μCT measurements of porosity and pore size [13]. Subsequent tibiae studies have shown the PI is inversely related to MRI-obtained bound water and phosphorus content [48] and is also inversely related to near infrared spectroscopy-obtained collagen and bound water content [56]. While these suggest Porosity Index as an effective metric for quantifying tibial health, its efficacy in the proximal femur, the site of the most devastating osteoporotic fractures, was unknown. Chen et. al investigated the porosity index in the femoral neck in a cohort of 68 people and compared it to age, BMI, and tibial PI [61]. Interestingly, the femoral neck Porosity Index was related to age only in men and there was no linear relationship found between femoral neck Porosity Index and tibial PI or BMI. However, Chen et. al reported low contrast and signal-to-noise ratio in the cortical bone of the femoral neck due to it being a small and deep structure. Furthermore, their study did not include elderly subjects who would likely exhibit the highest cortical porosity values. While Chen’s study demonstrated the feasibility of PI measurement in the proximal femur, there were no comparisons made to gold standards, osteoporotic status, or incident fractures, so the clinical usefulness of the PI measurement in the proximal femur was still known. Based on their results in the femoral neck, we elected to focus on the more compact cortical region in the proximal femoral shaft because it is larger and has higher contrast between the cortical bone and the surrounding soft tissue. Notably, the relationship between the cortical PI in this region and the stiffness of the femur in a sideways fall (r = −0.82) was strikingly similar to a previous study which compared PI to axial stiffness in machined cadaveric tibiae samples (r = −0.79) [56]. The results presented validate the Porosity Index as a useful measure of whole-bone mechanical integrity and stiffness.

Cortical porosity is a major determinant of bone strength that is related to fracture risk independent of BMD and other risk factors including sex, weight, height, smoking status, and the presence of an osteoporotic comorbidity [24, 70]. In a case-control study of 211 patients with nonvertebral bone fractures, cortical porosity identified an additional 26% of patients with fracture compared to DXA alone [24]. Osteoporotic medications have been shown to reduce cortical porosity in the proximal femur and improve bone strength, but they are under prescribed [71]. Clinical evaluation of cortical porosity could identify patients who are at high risk of bone fracture and who would most benefit from pharmaceutical intervention. MRI quantification of cortical porosity is advantageous over CT-based methodologies because it does not produce ionizing radiation. This is particularly useful in clinical studies evaluating the efficacies of pharmacological interventions on reducing cortical porosity because the lack of radiation makes short-term longitudinal scanning feasible and safe.

Cortical bone thickness significantly improved the association to whole-bone integrity when added to BMD, but when added to porosity index it was close albeit not significant (P = 0.06 and 0.09, respectively). This is likely a result of a relatively small sample size and the fact that cortical thickness itself was moderately associated with porosity index (r = −0.60). Since cortical porosity and cortical thickness are both measures of bone quality and strength that are related to other factors including age [72–77], medication [78–80], physical activity [16, 81, 82], surgery [83], and presence of disease [84, 85], it is not surprising that these parameters were related, especially since several of the specimens were over 65 years old when cortical bone undergoes the most drastic changes [75]. Indeed, past studies have reported weak to moderate correlations between cortical porosity and cortical thickness [13, 86]. Regardless, since stiffness is related to both the size and the quality of the bone, it is likely that cortical thickness does provide at least marginal additional information about bone health that is not captured by porosity. For example, in one large study of 211 postmenopausal women with nonvertebral fractures and 232 controls, both cortical measures remained significantly associated with incident fracture even after adjustment for BMD and other clinical risk factors, although porosity was the better predictor of incident fracture [87]. However, further research is needed to verify the combined utility of cortical thickness and porosity for in vivo bone assessment.

UTE MRI is especially well suited for measuring bone porosity. At ultrashort or zero echo times, signal is detected from all water protons in bone, including those that exist in microscopic pores which are beyond the detection limit of μCT. While UTE MRI biomarkers have been validated compared to μCT-derived porosity in numerous studies [13, 57, 58], one remarkable study demonstrated moderate to strong correlations between UTE measures and porosity evaluated from histological slides of 0.2 μm resolution, which is well beyond the limit of μCT [59]. This indicates that UTE MRI is capable of indirectly probing the microarchitecture of cortical bone. Additionally, Porosity Index in cadaveric tibiae specimens demonstrated strong correlations with Biexponential Analysis-derived pore water fraction and pore water T2*, confirming that different UTE MRI measures provide similar information about bone microstructure. However, further research is needed to compare the relative efficacy of different UTE MRI biomarkers at quantifying bone health.

This entire study was performed using commercially available hardware. The 3-Tesla scanner and the flexible body matrix RF coil used are routinely used for clinical hip imaging. Indeed, 3-Tesla clinical scanners are ideal for solid state imaging of cortical pore water since higher field strengths cause susceptibility effects between water and bone which reduce the T2* of pore water, thus diminishing the separation between bound and pore water T2* and making it more difficult to separate the signals [88]. Furthermore, although a custom dual-echo UTE sequence was used to acquire the data, similar sequence variants are sold by all major MR vendors and are becoming increasingly more prevalent. Another important feature of this study is that, unlike other UTE MRI biomarkers, the cortical Porosity Index does not require the use of specialized adiabatic inversion pulses or external calibration phantoms. Further, since the PI is computed as the naïve ratio of the two echo images, the computation is easily performed with any free image analysis software and does not require customized scripts or programmatic knowledge.

This study and protocol are not without limitations. Due to difficulties in acquiring specimens, this study was conducted with a relatively small sample size. Additionally, the segmentation of the cortical bone is crucial for accurate PI quantification. Erroneous segmentation can often lead to inadvertently including bone marrow voxels of near-100% porosity into the volume of interest which can drastically increase the measured porosity. As such, incorporating fat or long-T2 suppression into the sequence could likely improve the results and make them less sensitive to accurate segmentation. Furthermore, the reproducibility of the cortical Porosity Index in the proximal femur must be evaluated before it can be used in future in vivo studies. Another limitation of this study was the time length of the pulse sequence. Although 10 minutes is not prohibitively long, clinical sequences tend to be roughly 5 minutes so further improvements in hardware and software are needed before this protocol can be truly clinically viable. We note, however, is that UTE sequences are inherently radial and oversample the center of k-space by design, thus making them ideal candidates for sparse or compressed sensing sampling schema designed to reduce scan time [89, 90]. Finally, although this study took the first steps in validating PI in the proximal femur, its true usefulness as a non-invasive biomarker of fracture risk can only be properly evaluated in a large prospective study in comparison to the existing clinical standards of DXA or QCT.

Conclusion:

The efficacy of the cortical Porosity Index for assessing femoral bone health was investigated as compared to both whole bone stiffness obtained in the non-destructive sideways fall loading configuration and as compared to QCT-derived vBMD. Cortical Porosity Index in the proximal femoral shaft showed strong correlations with whole-bone stiffness, QCT-derived vBMD, and with cortical thickness. Notably, the information Porosity Index provided about bone stiffness was independent of other imaging parameters. This study further demonstrated that UTE methods are useful for directly assessing cortical bone porosity and that cortical porosity is an important determinant of bone strength.

Table 1:

Mean, standard deviation, and range for the cadaveric specimens.

Age	Porosity Index (%)	Cortical Thickness (mm)	vBMD (mg/cc)	Stiffness (kN/mm)
72.1 ± 15.0 [44–93]	34.0 + 4.8 [24.9–41.4]	4.1 ± 0.9 [2.3–58]	506.2 ± 27.2 [458.2–547.6]	1.3 ± 0.3 [0.4–1.9]