MRI‐Based Quantitative Osteoporosis Imaging at the Spine and Femur

Osteoporosis is a systemic skeletal disease with a high prevalence worldwide, characterized by low bone mass and microarchitectural deterioration, predisposing an individual to fragility fractures. Dual‐energy X‐ray absorptiometry (DXA) has been the clinical reference standard for diagnosing osteoporosis and for assessing fracture risk for decades. However, other imaging modalities are of increasing importance to investigate the etiology, treatment, and fracture risk. The purpose of this work is to review the available literature on quantitative magnetic resonance imaging (MRI) methods and related findings in osteoporosis at the spine and proximal femur as the clinically most important fracture sites. Trabecular bone microstructure analysis at the proximal femur based on high‐resolution MRI allows for a better prediction of osteoporotic fracture risk than DXA‐based bone mineral density (BMD) alone. In the 1990s, T2* mapping was shown to correlate with the density and orientation of the trabecular bone. Recently, quantitative susceptibility mapping (QSM), which overcomes some of the limitations of T2* mapping, has been applied for trabecular bone quantifications at the spine, whereas ultrashort echo time (UTE) imaging provides valuable surrogate markers of cortical bone quantity and quality. Magnetic resonance spectroscopy (MRS) and chemical shift encoding‐based water–fat MRI (CSE‐MRI) enable the quantitative assessment of the nonmineralized bone compartment through extraction of the bone marrow fat fraction (BMFF). Furthermore, CSE‐MRI allows for the differentiation of osteoporotic vs. pathologic fractures, which is of high clinical relevance. Lastly, advanced postprocessing and image analysis tools, particularly considering statistical parametric mapping and region‐specific BMFF distributions, have high potential to further improve MRI‐based fracture risk assessments at the spine and hip.

osteoporosis in Western countries, resulting in one or more fragility fractures among $40% of these women in their remaining lifetime. 6,7 Among the male counterparts, still about 15% to 30% are expected to sustain at least one fragility fracture during life, making osteoporosis a public health concern. 7 Osteoporotic fragility fractures have been shown to considerably reduce health-related quality of life and to lead to premature mortality. [8][9][10][11][12] In this context, the presence of the initial fracture has been identified as a main risk factor for later additional fractures, with an increased risk of about 86% for a new fracture among people who have already sustained a fracture. 13 Vertebral fractures are estimated to be the most frequently occurring fractures in osteoporosis. [1][2][3] Affected subjects present a 2.3-fold increase in the risk of future hip fracture, a 1.6-fold increase in the risk of future distal forearm fracture, and a considerable 12.6-fold increase in the risk of future additional vertebral fractures. 14 Despite their frequent occurrence, osteoporotic vertebral fractures often remain hidden and can stay asymptomatic, with fewer than 10% of such fractures resulting in hospitalization, even in cases with symptoms. 15,16 Consequently, initiation of treatment can become drastically delayed, which is particularly problematic in the light of the increased risk for future additional fractures. 14 Following vertebral fractures, hip fractures represent the second most frequent osteoporotic fractures. [1][2][3] In contrast to osteoporotic vertebral fractures, hip fractures rarely stay silent, leading to high hospitalization rates including surgical treatment among affected people. Particularly after hip fractures, an increased risk of death can be observed in osteoporotic subjects, which is highest in the days and weeks following the fracture, remains elevated for months, and is causally related to 20-30% of deaths. [17][18][19] Moreover, only 40-60% of hip fracture patients can recover to their prefracture status of mobility, with 20-60% of patients who were independent in self-care activities before the fracture requiring assistance to perform these tasks after one year. 20,21 The management of osteoporosis largely aims at preventing fractures and/or at therapy of related complications. [1][2][3] In this regard, the objective of osteoporosis treatment is to lower the likelihood of fragility fractures by strengthening the skeleton, decreasing fall frequency, or both. 2 Besides recommendations of adaption of lifestyle factors (eg, regular physical activity and good nutrition), pharmacological interventions are appropriate for osteoporosis treatment, particularly in subjects with a high risk profile. 2 Several drugs have been shown to substantially decrease fracture risk by $30-70% for vertebral fractures and by as much as $50% for hip fractures. 2 Bisphosphonates are the most commonly used drugs in osteoporosis and, as antiresorptive agents, are capable of inhibiting osteoclasts to avoid or slow down further bone weakening, thus reducing fracture risks. 22,23 While bisphosphonates only have moderate effects on bone mineral density (BMD), newer drug options, such as denosumab or teriparatide, can facilitate a progressive BMD increase and provide the possibility that a specific BMD value may be defined to guide osteoporosis treatment decisions. 2,24 Furthermore, in subjects who have already sustained an osteoporotic fracture, surgery may be considered as a treatment option. While almost all hip fractures require timely surgery, conservative treatment could generally be appropriate for a considerable proportion of vertebral fractures except for cases with complications such as fracture-associated neurological deficits, high pain intensity and lasting immobility, significant instability, and/or progressing fracture sintering. [25][26][27] Areal BMD derived from dual-energy X-ray absorptiometry (DXA) of the spine and/or hip is the clinical standard for diagnosing osteoporosis and for assessing fracture risk. 1,28,29 However, it is well known that DXA has important limitations that can interfere with correct detection of osteoporosis, and that the DXA-based T-score-the diagnostic value for osteoporosis-is not appropriate as the sole factor to identify patients at high fracture risk. Specifically, it has been shown that DXA-derived BMD values of subjects with and without osteoporosis can overlap. 30,31 Thus, other techniques, including quantitative computed tomography (QCT) and magnetic resonance imaging (MRI), over the last years have experienced increasing attention for imaging-based studies of etiology, fracture risk assessment, and therapy monitoring. Recent reviews have highlighted the current and potential future role of QCT for osteoporosis diagnosis and monitoring. [32][33][34] Regarding quantitative MRI, studies dating back to the 1990s introduced T 2 * mapping and high-resolution trabecular bone imaging, which can enable the assessment of bone microstructure that is relevant in the context of osteoporosis. [35][36][37] In the 2000s, MRI studies started to demonstrate that bone marrow (BM), which fills the cavities of trabecular bone and mainly consists of adipocytes (yellow marrow regions) and hematopoietic red blood cells (red marrow regions), may play a key role in bone health and metabolism, with distinct alterations becoming increasingly evident in osteoporotic subjects. 38,39 MRI techniques to investigate the BM fat fraction (BMFF) include magnetic resonance spectroscopy (MRS) and chemical shift encodingbased water-fat MRI (CSE-MRI). However, a variety of other MRI methods, such as ultrashort echo time (UTE) imaging of cortical bone and quantitative susceptibility mapping (QSM) for imaging of trabecular bone, have recently found their way into the repertoire of MRI techniques for osteoporosis assessment. [40][41][42] While early MRI methods predominantly exploited the signal arising from BM to image the trabecular microstructure of bone, more recent approaches like UTE imaging and QSM enable more direct imaging of bone tissue and have only recently been applied to imaging of the spine.
The purpose of this article is to review the current literature on quantitative MRI methods and related findings in osteoporosis at the spine and proximal femur from a methodological and clinical perspective.

Background
The MRI signal from solid-state tissues such as bone is very low because of short T 2 * and, thus, trabecular bone appears dark in most clinical sequences. Hence, the ability to image the trabecular network is due to strong signal from the fatty BM, which surrounds the trabecular bone structure. Consequently, the goal of MRI acquisition is to maximize the BM signal and enhance the contrast to the cancellous bone. 36,43 Table 1 gives an overview of the technique's main characteristics in relation to other sequences.

Technical Principles
In general, approaches to assess trabecular bone microarchitecture in vivo have been reported for MRI at 1.5T and 3T. High-resolution imaging to depict the trabecular bone microarchitecture of the proximal femur has been challenging, and at the spine is currently not feasible due to signal-tonoise ratio (SNR) constraints. The proximal femur is relatively deep in the human body and, hence, the signal response decays quickly. Furthermore, the proximal femur contains higher amounts of hematopoietic BM compared to the fattier BM in the extremities such as in the wrist or ankle. Hematopoietic BM has longer T 1 recovery times and therefore is usually darker in MRI. However, with the advent of higher field strengths, and advanced RF coils, the first in vivo high-resolution MRI of the proximal femur enabling trabecular bone analysis was published about 15 years ago. 44 Images featured a spatial resolution of 234 μm in-plane and 1500 μm slice thickness. 44 Since then, further progress has been made towards higher spatial resolution and better image quality. Current state-of-the-art in vivo MRI acquisition at the proximal femur can achieve a spatial resolution of 234 μm in-plane and 500 μm slice thickness in a clinically feasible scan time. 43,45 Initially, mainly gradient-echo-based pulse sequences were used for high-resolution imaging of the proximal femur. [43][44][45][46][47][48] The advantage of this type of pulse sequence is the high SNR efficiency and, thus, a shorter scan time. Some disadvantages include the susceptibility-induced broadening of the trabecular bone structure. 49 Furthermore, the prevalence of hematopoietic BM in some areas of the proximal femur limits the signal in these regions. For these reasons, a spin-echo-based pulse sequence (OVS-CUBE) has been proposed for trabecular bone MRI of the proximal femur. 50 OVS-CUBE combines a multi-spin-echo variable flip angle approach with outer volume suppression. The spin-echo implementation limits susceptibility-induced effects and optimizes signal from the hematopoietic BM by allowing longer repetition times (TRs) (Fig. 1).
Image postprocessing plays an essential role for assessment of trabecular bone microstructure. The first step of postprocessing includes correction of the coil intensity inhomogeneity. A simple method is based on low-pass filtering, whereas more advanced coil correction approaches include a fully-automated scheme based on a nonparametric, nonuniform intensity normalization approach that preserves image information while accurately correcting for coilinduced intensity inhomogeneities. [51][52][53] After correcting for the coil inhomogeneity and segmentation, the trabecular bone's microstructural information can be extracted. This commonly includes bone volume fraction (BVF) or bone volume to total volume fraction, trabecular thickness, trabecular spacing, and trabecular number. 54 All these parameters are usually titled "apparent" as they correlate well with true values from micro-CT, but absolute values may differ. 55 For analysis of the proximal femur, mainly fuzzy clustering with local bone enhancement at multiple scales has been used to obtain the BVF. 56,57 All other parameters are then obtained by thresholding the BVF map. 54 However, several other methods exist, including digital topological analysis, geodesic topological analysis, local inertial anisotropy, volumetric topological analysis, and tensor scale analysis. [58][59][60][61][62][63][64] Finite element analysis (FEA) has also been previously used to compute the elastic modulus and to measure bone strength. [65][66][67] In this regard, FEA applied to MRI data has shown to be a promising approach that can reliably provide fracture risk information and evaluation of bone strength with good reproducibility of FEA measurements, furthermore pointing at good agreement of the method against the gold standard of mechanical testing at the femur. 47,[68][69][70] Application and Main Findings In vivo microstructural analysis of high-resolution MRI of the proximal femur has been clinically applied to postmenopausal women with osteoporosis, 47,71 long-term glucocorticoid users, 48 as well as human immunodeficiency virus-infected (HIV) men. 45 Patients with fractures had lower elastic modulus than the control group in all proximal femur regions but no differences in T-scores. 47 Another study investigated 60 postmenopausal women (30 with and 30 without fragility fractures) who did not have osteoporotic BMD in the proximal femur; however, the fracture group was found to have significantly impaired bone microstructure. 71 One study in eight HIV-infected men found significantly lower bone quality compared to healthy controls, but BMD did not show significant differences between HIV-infected patients and healthy controls. 45 These findings suggest that assessment of trabecular bone microarchitecture might play an important role in the context of bone quality in addition to BMD measurements.

T 2 * and Quantitative Susceptibility Mapping
Background T 2 * mapping was the first quantitative MRI technique for assessment of bone as reported in the 1990s. [35][36][37] Bone is more diamagnetic than marrow. In the trabecular bone and BM interface, the two coexisting phases induce local magnetic field inhomogeneities. This can be measured as a shortening of the effective transverse relaxation time. It has been demonstrated that T 2 * correlates with the density and orientation of the trabecular bone. 36 A more direct measurement of the susceptibility difference between the vertebral trabecularized BM and surrounding tissues is based on QSM (Table 1).

Technical Principles
The method of QSM can overcome the limitation of T 2 * relaxometry that para-and diamagnetic susceptibility sources similarly result in an increased relaxation rate (Fig. 2). In general, QSM has been proven to be capable of providing quantitative and reproducible data on magnetic susceptibility of different tissues of the body. 72 However, due to the fact that bone has a considerably shorter transverse relaxation time in combination with low signal at commonly used echo times (TEs) in gradient echo imaging, evaluation of osseous structures remains challenging. 42,72 It has been suggested that QSM of BM can be biased due to the low signal of cortical bone structures in the proximity of trabecularized BM. 42,73 Consequently, UTE might be important to correctly estimate susceptibility of musculoskeletal anatomies and, thus, could be relevant for derived BMD values in trabecularized BM. However, more research is required to assess the accuracy of UTE-based QSM for trabecularized BM imaging.

Application and Main Findings
Relevant studies on the BMFF quantification at the spine using T 2 * or QSM are shown in Table 2. In a recent study, T 2 * and a T 2 * ratio allowed for the discrimination between vertebral fractures of different origins, with a diagnostic accuracy of 73% and 89%, respectively, for distinguishing acute benign from malignant vertebral fractures. 76 While these values may not qualify the approach to be used independently, the T 2 * and T 2 * ratio may deliberately support diagnostics when combined with further dedicated anatomical sequences in the clinical setting. Regarding QSM, multiparametric trabecular bone R 2 * mapping and QSM based on multi-echo gradient-echo imaging were shown to be feasible, with good sensitivity of QSM for measuring trabecular bone density in yellow BM regions. 78 In this context, R 2 * represents the transverse, nonrecoverable relaxation rate, with R 2 * = 1/T 2 *. The decreases of calcium hydroxyapatite within vertebral bodies have been recently linked to increases in magnetic susceptibility values. 74,75,77 The first studies applying the technique showed significantly increased vertebral magnetic susceptibility in osteopenia and osteoporosis. 74,75,77 Furthermore, magnetic susceptibility was positively correlated with the proton density fat fraction (PDFF) as derived from CSE-MRI. 75 The sensitivity and specificity for differentiating osteoporosis from nonosteoporosis were 80.8% and 77.3%, respectively, and the area under the curve (AUC) for differentiating osteopenia from osteoporosis was significantly higher for the combination of magnetic susceptibility and PDFF than for PDFF alone. 74,75 Furthermore, QSM of the lumbar spine showed very good interobserver reliability and interscan reproducibility. 77 However, the studies on QSM referenced above employed methodologies that did not perform chemical shift encoding-based water-fat separation for generating the field map used in the magnetic susceptibility estimation and, therefore, their results should be primarily interpreted in a qualitative manner. 74

Ultrashort Echo Time Imaging
Background In high-resolution imaging of trabecular bone, primarily applicable to the proximal femur, the signal voids are identified as bone due to the bone's low water content and short T 2 relaxation (Table 1). T 2 * and QSM utilize the BM signal and the perturbance of such signal due to the susceptibility differences between trabecular bone and BM. 42,72,[79][80][81] Technical Principles Another approach besides T 2 * and QSM is the direct measurement of the bone matrix signal. The direct detection of the bone tissue signal itself is a challenging task due to its extremely rapid signal decay, which precludes the depiction of bone structures with conventional MRI sequences with TEs of 2-10 msec. Several MRI sequences were introduced with radial k-space sampling that allow the sampling at an UTE of 30-200 μs, which can enable the direct visualization of trabecular and cortical bone. [82][83][84] Cortical bone consists of $40% mineral, 35% collagen, and 25% water by volume. 36 The short T 2 component detected by UTE imaging is predominantly from bound water and is in the range of 0.42-0.50 msec. 84,85 The visualization of trabecular bone is limited by the presence of high signals from long-T 2 water and fat and due to the much lower proton density of trabecular bone. The suppression of long-T 2 tissues is critical for trabecular bone imaging and to achieve a desirable high contrast. Several methods for long-T 2 suppression have been proposed: dual-echo subtraction methods, [86][87][88] water-fat saturation techniques, 89,90 and long-T 2 suppression techniques based on inversion recovery (IR). [91][92][93] All of the presented approaches are suitable to generate contrast between long-T 2 and short-T 2 tissues. However, IR approaches can be designed to be time-efficient by combining a short TR and a turbo field echo readout. 91,94 Therefore, the inversion pulses are applied with a TR < T 1 , leading to a quasi-steady-state of longitudinal magnetization. The z-magnetization of water in the steady-state is low due to its relatively long T 1 , as compared to bone, yielding a contrast between water and bone signal. To eliminate the signal from fat, the UTE acquisition starts at the null point of the fat signal. A reduction of scan time is achieved when several readouts near the null point are acquired. Signal suppression is still possible, since excited signals before the null point and after the null point are of opposite polarity and since the data acquisition starts in the center of k-space for each readout. During the gridding process the signals from fat with opposite polarity, acquired before and after the null point of fat, annul each other. 91

Application and Main Findings
In principal, UTE imaging in combination with long-T 2 suppression is a promising novel approach to visualize bone and to measure trabecular bone density directly (Fig. 3). It was recently shown that adiabatic IR-prepared UTE imaging in combination with a calibration phantom yields the trabecular bone proton density. 93 Specifically, using a hip agarose bone phantom and measurements in healthy subjects, it was demonstrated that high-contrast imaging of trabecular bone can be achieved ex vivo and in vivo (with fitted T 2 * values of 0.3-0.45 msec and proton densities of 5-9 mol/L), suggesting the applicability of both selective imaging and quantitative assessment of short-T 2 water components in trabecular bone. 93 However, IR-prepared UTE sequences often have long scan times and the clinical value of this technique needs to be investigated further for application at the spine or femur. However, UTE imaging has been used ex vivo for  valid cortical bone porosity assessment of the tibia, as well as in healthy subjects to investigate pore water concentration mapping of tibial cortical bone, for instance, thus pointing at potential clinical utility. 95,96 Magnetic Resonance Spectroscopy

Background
To date, single-voxel proton MRS has been the most frequently applied technique for quantification of BM (Fig. 4). This technique acquires data from a prescribed localized volume of interest (VOI) with high spectral resolution. Thus, it enables the quantification of individual chemical components based on their resonance frequency, namely water and fat in the case of BM (Table 1).

Technical Principles
Both point-resolved spectroscopy (PRESS) and stimulated echo acquisition mode (STEAM) sequences were predominantly used. However, STEAM has relevant advantages over PRESS. It enables shorter minimum TEs and, thus, features higher signal for the short-T 2 water component of the BM spectrum despite its intrinsic signal loss of about 50% compared to PRESS. MRI-based quantification of fat of BM can be expressed as the BMFF, but most notably as the PDFF, which can be regarded as a fundamental tissue property and is defined as the ratio of density of mobile protons from fat (triglycerides) and the total density of protons from mobile triglycerides and mobile water. 97 Confounding relaxation effects have to be corrected for in order to quantify the PDFF. Measurements at multiple TEs have to be acquired at suitable TRs to allow for the correction of T 2 -weighting and T 1 -weighting effects, respectively. Regarding the placement of the VOI, it is important to take into account chemical shift displacement effects due to finite RF pulse bandwidths of the MRS sequence. At 3T, this can typically result in an offset of up to 15% of the voxel size in each spatial dimension between the water and fat localization. 98 To reduce this effect and subsequent quantification errors, the VOI should be placed with caution within the vertebral body or proximal femur to avoid severe inhomogeneities, degenerative changes including subchondral sclerosis, or vascular structures.    using MRS at 1.5T in 82 men and 103 women at the level of the lumbar spine. 39,101 In detail, the vertebral BMFF was significantly increased in both elderly men and women with osteoporosis (males: 58.2 ± 7.8%, females: 67.8 ± 8.5%) as compared to subjects with normal BMD values (males: 50.5 ± 8.7%, females: 59.2 ± 10.0%). 39,101 Analogously,    Griffith et al used MRS at 1.5T with VOIs placed in the femoral head, neck, and subtrochanteric region of the femoral shaft, reporting on significant BMFF increases with decreased BMD for these locations among a cohort of 120 elderly women. 111 At the femoral neck, the BMFF amounted to 80.8 ± 9.4% in normal subjects, whereas it was 86.2 ± 6.5% in osteopenic and 88.4 ± 4.8% in osteoporotic subjects, respectively. 111 At the spine, increased BMFF and decreased BMD have also been related to biomechanical properties: an in vitro study harvested vertebrae from human cadavers and demonstrated a significant negative correlation between the BMFF and the biomechanical failure load. 107 Other studies reported on the BMFF being positively associated with prevalent vertebral fractures in men and on fat unsaturation levels being negatively associated with the prevalence of fragility fractures. 104,105 Generally, fat unsaturation levels were reported to be decreased in osteoporotic subjects, as well as in diabetic subjects, and were proposed as complementary imaging biomarkers to assess fracture risk, especially in subjects with increased fracture risk despite regular BMD, such as in type 2 diabetes mellitus. 100,103,104 For the femoral neck, a combined approach using MRS and diffusion tensor imaging (DTI) was described to be sensitive and specific in identifying osteoporotic subjects, with a positive correlation being reported for fractional anisotropy (FA) and BMFF in osteopenic and/or osteoporotic subjects. 112,113 Regarding the femoral neck of elderly women, methylene, glycerol, and total lipid resonances were significantly lower in healthy as compared to osteoporotic subjects, and significant correlations were revealed between total lipids and the T-score, respectively. 114 Vertebral BMFF was reported to be dependent on age and gender, and it shows anatomical variations. 99,102,103,109 Subcutaneous adipose tissue (SAT) volume and total adipose tissue volume correlated significantly with the BMFF in females with and without diabetes, while significant   and HbA1c, the most important blood marker for diabetes, were observed only in diabetic females. 103 Vertebral BMFF was higher in men as compared to women; however, this has been observed among healthy subjects and not explicitly in subjects with osteoporosis. 99 Furthermore, the BMFF shows an increase from the L1 to L4 vertebral bodies in subjects with low BMD, indicating anatomical variations at the spine. 102 Lastly, BMFF measured at the L3 vertebral body was significantly associated with BM parameters on iliac crest biopsies, which was not the case for BMFF measured at the intertrochanteric region. 106 The reproducibility of the MRSbased BMFF measurements at the spine, reported in terms of the coefficient of variation, was determined to be equal to 1.7%. 102

Chemical Shift Enconding-Based Water-Fat Imaging
Background Assessment by CSE-MRI allows spatially resolved quantification of the BMFF (Fig. 5). Sequences typically used at 3T include bipolar gradient readout sequences acquiring all echoes (usually six echoes) within a single TR and monopolar time-interleaved gradient echo sequences where all echoes (usually six echoes) are acquired in multiple TRs (Table 1). At 1.5T, nontime-interleaved monopolar readouts acquiring all echoes in a single TR are also used. 115 Technical Principles For fat quantification, complex-or magnitude-based parameter estimations were conducted. [115][116][117] While complex-based techniques have shown superiority to magnitude-based approaches due to lower sensitivity to fat signal modeling errors and better noise performance, they are nevertheless prone to phase errors. In addition, T 1 bias and T 2 * decay effects need to be considered in order to extract the PDFF. 115,116,[118][119][120] These effects can introduce considerable bias during BMFF measurements, thus requiring an appropriate selection of experimental parameters; for example, small flip angles to reduce T 1 bias, and d1istinct correction of T 2 * decay effects during the postprocessing stage. 115,[121][122][123] Compared to applications of the PDFF in other organs, T 2 * decay effects have to be considered in particular when measuring the PDFF in the presence of trabecular bone, which shows was found between total lipids and the T-score, and a significant positive correlation was found between water and the T-score.

Application and Main Findings
Relevant studies on the BMFF quantification using CSE-MRI are shown in Tables 5 and 6. As for MRS, the majority of studies investigated the spine, with fewer studies performing CSE-MRI at the proximal femur. Importantly, good agreement has been reported between MRS-based BMFF quantification as the gold-standard method and CSE-MRIbased BMFF quantification for the spine as well as the proximal femur. [131][132][133][134] At the spine, the BMFF derived from CSE-MRI showed increased values in subjects with osteoporosis and inverse correlations with BMD and Tscores. 81,130,132 The reproducibility of BMFF measurements, expressed as an absolute precision error for the PDFF, has been reported in the range of 1.70% (C3 to L5 vertebral bodies) and 1.45% (L1 to L4 vertebral bodies). 124,132 Along the spinal axis, the BMFF showed an increase in the craniocaudal direction. 124,129 Specifically, as derived from measurements from C3 to L5, segment-specific differences in measurements between males and females have been reported, with the BMFF averaged over C3-C7, T1-T6, T7-T12, and L1-L5 vertebral bodies in young, healthy men and women (mean age: 26 years) amounting to 31.7 ± 7.9% and 23.0 ± 7.8%, 33.8 ± 6.8% and 24.6 ± 8.8%, 33.8 ± 6.4% and 26.1 ± 6.4%, and 38.8 ± 7.6% and 31.5 ± 12.4%, respectively. 124 In children, cervical, thoracic, and lumbar BMFF measurements were significantly associated with the natural logarithm of age, with differences in the BMFF between boys and girls being not yet significant. 133 At the proximal femur, subregion-specific significant differences in fat composition in postmenopausal compared to premenopausal women were observed, with postmenopausal women demonstrating higher saturation (+14.7% to +43.3%), but lower mono-(−11.4% to −33%) and polyunsaturation (−52% to −83%). 135 Specifically, within red marrow adipose tissue, postmenopausal women showed lower fat content (−16% to −24%) and decreased polyunsaturation (−80% to −120%) in the femoral neck, greater trochanter, and Ward's triangle. 135 An accelerated fatty conversion of BM was observed in women as compared to men with increasing age, particularly evident after the menopause. 125 Of note, significant correlations between fat fractions of paraspinal and vertebral BM compartments were observed recently in postmenopausal women, thus potentially providing a hint for direct associations between vertebrae and other body fat compartments. 128 Furthermore, in postmenopausal women BM heterogeneity as evaluated by means of texture analysis based on CSE-MRI was significantly increased at the spine. 129 Among the parameters estimated using gray-level cooccurrence matrix analysis, contrast and dissimilarity performed best in differentiating pre-and postmenopausal women. 129 The PDFF derived from CSE-MRI has recently been shown to facilitate discrimination between benign osteoporotic and malignant vertebral fractures, which can be challenging in the acute clinical setting. 126,127 The PDFF of vertebral BM and a PDFF ratio (fracture PDFF divided by normal vertebrae PDFF) were significantly lower in malignant vertebral compression fractures when compared to acute osteoporotic fractures (PDFF: 3.48 ± 3.30% vs. 23.99 ± 11.86%; PDFF ratio: 0.09 ± 0.09 vs. 0.49 ± 0.24). 126 Moreover, the diagnostic accuracy was 96% regarding the differentiation of benign lesions and acute vertebral fractures from malignancy. 127 Acute vertebral fractures showed an average PDFF that was clearly higher than those of malignant tumors like vertebral metastases, multiple myeloma, and chordoma, while being lower than the PDFF of hemangioma or degenerative changes of vertebral endplates. 127

Multidimensional Investigation and Parametric Mapping
Background Regarding region-specific interactions, there is a lack of knowledge concerning the 3D spatial interrelationships between BMFF and BMD. However, investigations of such spatial associations could improve our understanding of bone fragility in critical anatomical regions like the proximal femur. Statistical parametric mapping (SPM) is an advanced image analysis technique with a framework that enables this type of spatial assessment. 136 Technical Principles SPM is a computational anatomy technique where voxelbased (volumetric) or vertex-based (surface) maps of tissue  parameters in a population (eg, maps of BMFF) are spatially normalized to a common template using affine and deformable registration, effectively establishing local anatomical correspondence across structures. After spatial normalization, smoothing of the parametric maps is commonly performed to compensate for inaccuracies of the registration step. Spatial normalization and smoothing steps then enable meaningful comparisons between groups of spatially varying data on a voxel-by-voxel or vertex-by-vertex basis. Comparisons are usually performed in the form of general linear models providing the opportunity to incorporate covariates, with the result of these local comparisons being in the form of a statistical map. The identification of specific subregions where the tissue feature under investigation is significantly associated with the variable of interest becomes possible with this approach. The main advantage of SPM is that by removing the anatomical variability in the population study, there is no longer the need of individualized regions of interest for the analyses. Thus, SPM provides the capability of performing analyses that are not biased to one particular region, and it gives an evenhanded and comprehensive assessment of feature statistics throughout the different tissues of interest across an entire study population.

Application and Main Findings
Using the SPM framework, Carballido-Gamio et al investigated the local interrelationships of the BMFF measured by MRI and volumetric BMD measured by QCT in the proximal femur of a small group of postmenopausal women. 137 In this pilot study, the authors included 15 postmenopausal women, six without fracture (control group, mean age: 62.4 ± 8 years), and nine with fragility fractures (mean age: 62.8 ± 9 years). Volumetric BMD maps were derived from QCT scans obtained with a calibration phantom, and BMFF maps were derived from CSE-MRI acquisitions at 3T applying six echoes and a multi-frequency model of fat (Figs. 6 and 7). 138 Within-group voxelwise correlations of BMFF and volume BMD were assessed, revealing mostly negative correlations between the BMFF and volumetric BMD across the whole proximal femur. 138 However, the strength of these correlations was spatially heterogeneous, and their spatial distribution was visually distinct between cases and controls (Fig. 7). 138 Thus, SPM-based analysis of CSE-MRI to characterize spatially resolved BMFF may become a promising option for evaluating osteoporotic bone. Beyond investigating the interrelationships of BMFF and volumetric BMD, CSE-MRI measurements of the  BMFF were recently used to increase the accuracy of QCT measurements of the spine. 139 Furthermore, it was reported that subjects with at least one prevalent vertebral fracture showed decreased MRI stiffness (up to 17.9%) and QCT density (up to 34.2%) at the distal extremities compared to a nonfracture group. 140 High BMFF can lead to false low measurements in single-energy QCT. These errors can be reduced by a factor of five using dual-energy CT or corrections by CSE-MRI measurements of the BMFF and should be considered to investigate the true relationship between BMD and bone marrow adipose tissue in humans. 141 However, it remains to be investigated if such corrections also lead to a better prediction of fracture risk in osteoporotic patients.

Conclusion and Perspectives
Different quantitative MRI methods have been introduced for imaging of osteoporosis at the spine and proximal femur over the last decades, initially based on high-resolution trabecular bone imaging and T 2 * mapping. QSM, which can overcome some of the limitations of T 2 * mapping, shows also potential for trabecular bone quantifications at the spine, whereas UTE imaging provides surrogate markers of cortical bone quantity and quality. Moreover, MRS and CSE-MRI enable the evaluation of the nonmineralized bone compartment, represented by BM, and extraction of the PDFF. Novel postprocessing and analysis approaches, such as FEA and SPM, have shown potential to further improve the assessment of fracture risk in osteoporotic patients. Thus, MRI is a viable option for radiation-free, quantitative assessment of osteoporotic bone and is fast developing, given the variety of sequences and techniques at hand to date. However, it has to be acknowledged that the methods reviewed in this article have not yet accomplished the ultimate transition to utility in broad clinical routine, which is partially related to limited evidence of clearly improved fracture prediction beyond BMD. Anticipated technical developments and investigations in large representative cohorts, alongside with broader distribution of MRI systems capable of applying these methods, may facilitate quantitative MRI application for osteoporosis imaging in the foreseeable future.