Advances in Bone Marrow Imaging: Strengths and Limitations from a Clinical Perspective

Abstract

Conventional magnetic resonance imaging (MRI) remains the modality of choice to image bone marrow. However, the last few decades have witnessed the emergence and development of novel MRI techniques, such as chemical shift imaging, diffusion-weighted imaging, dynamic contrast-enhanced MRI, and whole-body MRI, as well as spectral computed tomography and nuclear medicine techniques. We summarize the technical bases behind these methods, in relation to the common physiologic and pathologic processes involving the bone marrow. We present the strengths and limitations of these imaging methods and consider their added value compared with conventional imaging in assessing non-neoplastic disorders like septic, rheumatologic, traumatic, and metabolic conditions. The potential usefulness of these methods to differentiate between benign and malignant bone marrow lesions is discussed. Finally, we consider the limitations hampering a more widespread use of these techniques in clinical practice.

Bone marrow disorders cover a wide range of conditions, such as neoplastic, septic, rheumatologic, traumatic, and metabolic disorders. Although conventional magnetic resonance imaging (MRI) remains the modality of choice for their assessment, the last few decades have seen the emergence of novel MRI techniques: chemical shift imaging (CSI), diffusion-weighted imaging (DWI), dynamic contrast-enhanced (DCE) MRI, and whole-body MRI, along with spectral computed tomography (CT) and nuclear medicine techniques.

These developments have been the focus of an increasing body of literature, with promising results covering both qualitative and quantitative analyses in a variety of conditions.

To understand how these techniques can be applied to various clinical scenarios, we provide an overview of their technical basics in relation to common physiologic and pathologic processes involving the bone marrow. Based on a review of the relevant literature, we discuss the strengths and limitations of these imaging methods to assess the most common non-neoplastic conditions and to differentiate them from neoplastic conditions, keeping the focus on clinical utility ( Table 1 ).

Table 1. Strengths and limitations of advanced imaging techniques in common clinical applications.

	Rationale for use	Normal marrow	Lesion detection and characterization	VCF	Infection	Inflammation	Metabolic	Other application	Pitfalls and limitations
CSI	Benign conditions contain residual fat	-SI drop on OP images -High but variable fat content	-Fat-only images have high CNR for detection of focal lesions -Marrow-replacing lesions: < 20% decreased SI on OP	> 20% decreased SI on OP images suggests a benign VCF		Detection of inflammatory and structural lesions in a single acquisition (T2-Dixon)	-Higher marrow adiposity in osteoporosis	-Protocol optimization by providing fat-sensitive and fluid-sensitive images on single T2-Dixon sequences (spine metastases, degenerative spine, spondyloarthritis)	-False positive in tuberculous spondylitis and sclerotic lesions
DWI	Increase in marrow cellularity and abscesses cause restricted Brownian motion	-Signal on DWI is variable (age, sex, Hb) -ADC of normal bone marrow: 0.2–0.6 ×10 −3 mm 2 /s	-Malignant and benign lesions have higher ADC than normal marrow	-Overlap of ADC values between benign and malignant VCF	-DWI may distinguish BME from osteomyelitis -DWI may detect abscesses if contrast injection contraindicated -DWI could differentiate between diabetic neuroarthropathy and osteomyelitis. Overlap exists	-Claw sign on lumbar MRI: Modic type 1 -DWI inferior to regular protocol of sacroiliitis	Limited value in osteonecrosis		-Variable protocols among different vendors and institutions -DWI is not a robust tool to differentiate infection from malignancy -DWI should be interpreted in conjunction with standard MRI sequences
DCE-MRI	-Tumors present neo-angiogenesis -Inflammatory conditions present increased vascularity	-Physiologic differences in marrow perfusion	-Fractional plasma volume lower in neoplastic lesions	-Pathologic VCF has higher perfusion parameters	K trans , K ep , and V e higher in osteomyelitis than in diabetic neuroarthropathy - K ep higher in malignant compared with tuberculous lesions -Tuberculous lesions do not show washout		-Marrow perfusion decreases in osteoporosis and around osteonecrotic lesions (ischemic penumbra)		-Frequent false positives and false negatives in VCF -Limited reproducibility of cut-off values between vendors and techniques
Spectral CT imaging	-Attenuation of tissues depends on atomic number Z; mineralized bone can be suppressed from images (VNCa images)	No standard HU values	-VNCa may improve detection of multiple myeloma and metastases	-VNCa images detect BME in acute VCF		-Detection of BME in sacroiliitis		-Detection of BME in trauma	-Validation of an established diagnostic threshold is difficult because of variety of acquisition and postprocessing techniques -Sensitivity of DECT in the diagnosis of acute but morphologically occult fractures remains to be addressed -False positives and false negatives are frequent
Nuclear medicine	-Bone marrow metabolism (FDG-PET) -Distribution in the hematologic system (WBC scintigraphy)				-High diagnostic accuracy in periprosthetic infection and in fracture-related infections (WBC scintigraphy) -Search for a focus of infection in FUO (FDG-PET)				-Intense uptake of red bone marrow hampers detection of bone marrow infection in the axial skeleton (WBC scintigraphy) -Focal increased uptake in benign marrow hyperplasia (FDG-PET)

Abbreviations: ADC, apparent diffusion coefficient; BME, bone marrow edema-like; CNR, contrast-to-noise ratio; CSI, chemical shift imaging; CT, computed tomography; DCE-MRI, dynamic contrast-enhanced MRI; DECT, dual-energy computed tomography; DWI, diffusion-weighted imaging; FDG, fluorodeoxyglucose; FUO, fever of unknown origin; Hb, hemoglobin; HU, Hounsfield Units; K _ep , rate constant; K ^trans , volume transfer constant; MRI, magnetic resonance imaging; OP, out-of-phase; PDFF, proton-density fat fraction; PET, positron emission tomography; SI, signal intensity; VCF, vertebral compression fracture; V _e , extra vascular space; VNCa, virtual non-calcium images; WBC, white blood cells.

Fat, Cellularity, and Vascularity: Bone Marrow Physiologic and Pathologic Changes Relevant to Marrow Imaging

To understand the contribution of different imaging techniques in investigating normal and pathologic conditions of bone marrow, following is a review of the basics of marrow pathophysiology.

First, normal bone marrow, whether yellow or red, contains a variable amount of fat. The proportion of water and fat content varies depending on the composition of bone marrow, related to several physiologic processes (including the transformation of red-to-yellow marrow with age, premenopausal status, etc.). In pathologic conditions the proportion of water and fat also changes with water content increasing relative to fat content. CSI and magnetic resonance spectroscopy (H1-MRS) can probe the fat content and quantify it to differentiate marrow-replacing lesions (where marrow fat is replaced, due to the presence of either malignant or benign lesions) from non–marrow-replacing lesions (where marrow fat is preserved, usually in relation to a benign lesion). Dual-energy computed tomography (DECT) and virtual non-calcium (VNCa) reconstructions can analyze bone marrow water and fat content by suppressing the attenuation component of mineralized bone, thanks to tissue characterization based on the atomic number Z and the photoelectric effect.

Second, many pathologic marrow conditions are characterized by increased water content, decreased fat content, increased vascularity, and destruction of the trabecular bone structure, all of which lead to increased diffusivity. DWI exploits the Brownian motion of water molecules and is sensitive to conditions where this motion is altered. The apparent diffusion coefficient (ADC) enables a quantitative evaluation of diffusion restriction.

Third, marrow pathologies show increased vascularity that may be assessed by DCE-MRI.

Finally, nuclear medicine and metabolic imaging examine the distribution of specific components of the hematopoietic and reticuloendothelial systems or the metabolic activity in bone marrow.

Chemical Shift Imaging

Technique

CSI takes advantage of the slight difference in resonance frequency that exists between water and fat protons to provide a series of four sets of images: in-phase (IP), out-of-phase (OP), water only (WO), and fat only (FO). Although Dixon described it in the 1980s, it has only recently become possible to associate this method with spin-echo–based sequences, the backbone of musculoskeletal MRI protocols, opening the door for multiple applications in this field.

CSI MRI can also be used to probe bone marrow fat content quantitatively, by measuring the signal drop between IP and OP images or by calculating the fat fraction (FF). This can be done both on gradient-echo and spin-echo sequences.

Normal Bone Marrow

The variation of signal intensity in normal bone marrow parallels the relative amount of water and fat protons, itself correlated to the relative amount of red and yellow bone marrow. Just as for T1-weighted sequences, red marrow presents lower signal intensity than yellow marrow on FO images due to its lower fat content. Furthermore, red marrow is lower in signal intensity on T2 IP images than red marrow.

At quantitative analysis, normal bone marrow shows a significant drop in signal intensity due to the presence of fat protons that at least partially cancel the signal of water, with a wide range of normal values due to variable marrow composition. It is reported that normal bone marrow should present a drop of at least 20% on OP compared with IP images. ^{1 2} The mean FF for normal vertebral marrow varies greatly, ranging between 13.7% and 83% (mean: 50.51 ± 14.69%). ³ FF of normal bone marrow varies with age, sex, and menstrual status; premenopausal women have higher marrow cellularity than men, whereas bone marrow fat increases with age in both sexes. ⁴

Clinical Applications

Robust Fat Suppression Technique

The Dixon method has mostly been used as a fat suppression technique that has proved to be more robust to magnetic field inhomogeneities than chemical shift selective suppression (CHESS) while providing a better signal-to-noise ratio than short tau inversion recovery (STIR) sequences. ^{5 6} Therefore, it is an appealing technique for large field-of-view (FOV) imaging of the bone marrow.

Optimization of Magnetic Resonance Imaging Protocols

The four sets of images generated by a Dixon sequence can be used to optimize MRI protocols, particularly when using a fast spin-echo (FSE) T2-weighted Dixon sequence. Through a single acquisition, both fat-suppressed and non–fat-suppressed fluid-sensitive images can be obtained. IP T2 images are equivalent to T2 FSE images, whereas WO images correspond to fat-suppressed T2 images. Furthermore, FO T2 images are not only sensitive but also specific to the signal of fat and can substitute T1-weighted images for the assessment of tissue fat content. ^{7 8 9 10 11} The potential of T2 FO images to replace T1-weighted images has been validated in a few studies addressing lesion detection in multiple myeloma, ⁸ metastases, ^{9 12} and characterization of vertebral compression fractures (VCFs). ¹³ For the common scenario of MRI in the context of nonspecific low back pain, a single sagittal T2-weighted Dixon sequence may replace the combination of T1-weighted, T2-weighted, and fat-suppressed T2-weighted sequences. ^{7 10 14} Of note, quantitative information on bone marrow fat content is also readily available when using Dixon sequences.

Lesion Characterization and Evaluation of Vertebral Compression Fractures

On standard MRI, the interpretation of decreased signal intensity on T1-weighted sequences is based on a comparison with an intrinsic reference (the signal intensity of muscles), and it helps discriminate between non–marrow-replacing lesions (T1 signal ≥ muscle/disk), which are usually benign, and marrow-replacing lesions (T1 signal < muscle/disk) ^{15 16 17} ( Fig. 1 ). This comes with a few caveats: the signal intensity of muscle and disks that serve as a reference can be altered, whereas some marrow-replacing lesions may be spontaneously high in signal intensity (e.g., due to the presence of melanin or high protein content), leading to false negatives. The interpretation of FO images, which are fat specific, is more straightforward. The absence of high-intensity pixels within a lesion is suggestive of a marrow-replacing lesion, whereas non–marrow-replacing lesions contain high-intensity pixels.

Fig. 1.

A 74-year-old woman with ovarian adenocarcinoma. ( a ) Sagittal reformat of fused fluorodeoxyglucose positron emission tomography/computed tomography images showing focal uptake in L3 vertebral body (arrow; maximum standardized update value  = 6). Sagittal ( b ) T1-weighted, ( c ) fat-only, ( d ) water-only, ( e, g ) in-phase, and ( f, h ) out-of-phase T2-weighted Dixon images. The lesion (arrows) contains fat with a drop in signal intensity of 36%, suggestive of a non–marrow-replacing lesion. ( g ) Note the low signal intensity on the non–fat-suppressed image that concurs with benign nodular marrow hyperplasia. The lesion was unchanged on subsequent follow-up studies.

As discussed earlier, gradient-echo and FSE Dixon images have been used to assess the fat content of bone marrow quantitatively. Most reports found a threshold of 20% to be effective in differentiating bone-marrow-replacing lesions (signal drop ≤ 20%) from non–bone-marrow-replacing, (signal drop > 20%), which are usually benign. ^{1 2} The need for biopsy could be eliminated in > 60% of patients with benign disease as demonstrated in a monocentric study. ¹⁸ In addition, CSI is useful to confirm the diagnosis of focal hematopoietic bone marrow islands and avoid unnecessary follow-up. ¹⁹ However, the measurement may depend on the types of sequences used, and other thresholds have occasionally been reported. ²⁰

A quantitative evaluation of intralesional fat through the calculation of FF maps, either with gradient-echo or spin-echo–based sequences, has also shown potential to differentiate benign from malignant lesions accurately. ^{3 21 22}

Axial Spondyloarthropathy

A single T2-weighted Dixon sequence was shown to provide all the information to assess inflammatory and structural lesions of spondylarthritis in the spine and sacroiliac joints, and it may replace standard T1 and fat-suppressed fluid-sensitive sequences ^{23 24} ( Fig. 2 ). The robust fat suppression and reduced examination time are additional assets for large FOV acquisitions.

Fig. 2.

A 34-year-old man with axial spondyloarthropathy. Coronal oblique ( a ) water-only and ( b ) fat-only images of a single T2-weighted Dixon show bone marrow edema-like signal intensity (a, white asterisks), subchondral erosions (a, arrows), and fat metaplasia (b, arrows). The T2-weighted Dixon technique enables accurate depiction of signs of disease activity and structural bone changes in a single acquisition.

Miscellaneous Marrow Conditions

Quantitative studies of marrow adiposity in osteoporosis showed increased marrow fat fraction with age and with decreased bone mineral density. ^{25 26 27} In clinical research, FF could be used as a biomarker for osteoporosis, knowing there is an overlap with healthy subjects. ²⁷ In addition, some authors used OP images to detect ankle and foot fractures, ²⁸ measure tumor size, ²⁹ and assess femoral head osteonecrosis. ³⁰ However, it is important to mention that the hypointense lines referred to as “India ink artifacts” seen on the OP images correspond to areas with an equal amount of water and fat protons and should not be mistakenly interpreted as fracture lines.

Pitfalls

Dixon images, especially two-point techniques, are prone to fat-water swapping artifact that could be easily characterized as such by a side-to-side comparison of FO and WO images.

Areas that contain a disproportionate amount of water protons relative to fat protons (such as in cysts, abscesses, and Schmorl's nodes) intrinsically have low or no signal drop on OP images, potentially leading to false-positive results ³¹ ( Fig. 3 ). Sclerotic metastases have also been shown to be a potential source of false-positive results, whose cause might be multifactorial ³¹ ( Fig. 3 ).

Fig. 3.

An 82-year-old man with nontraumatic back pain. Sagittal ( a ) T1-weighted, ( b ) in-phase, ( c ) water-only, and ( d ) fat-onlyT2-weighted Dixon images show a recent vertebral compression fracture (VCF) of T12 with an intravertebral horizontal fluid-filled cleft (b, c, arrow) in favor of a benign VCF. However, the vertebral body did not show any residual fat (d), and there was no signal drop on out-of-phase images ( e ), suggestive of a marrow-replacing lesion. In view of the discordance, a vertebral biopsy was performed (not shown) that was negative for malignancy. ( g ) Sagittal reformat of computed tomography (CT) images show sclerotic changes in the posterior third of the vertebral body (arrow) and gas-containing cleft (arrowhead) as sources of false-positive findings on quantitative analysis of Dixon imaging. Follow-up fluorodeoxyglucose positron emission tomography/CT was negative for malignancy (not shown). Of importance, regions of interest should exclude fluid-filled regions to avoid false-positive results ( f ).

Another pitfall is hypocellular neoplasia, such as in multiple myeloma, where the presence of substantial remaining fat may lead to false-negative FF findings. ^{3 32}

Finally, FF measurements depend on the acquisition method used and their relative sensitivity to T1, T2, and T2* decay, ^{4 33} in addition to the lack of standardization and variable thresholds that have been used across studies. ^{3 21 22} These limitations apply to most currently available quantitative applications. As a rule, overreliance on quantitative assessments should be avoided and quantitative information should rather be used as an adjunct to, rather than a substitute for qualitative assessment, when necessary.

Diffusion-weighted Imaging

Technique

DWI assesses the self-diffusion, that is, the Brownian motion of water molecules, influenced by the microscopic structure and organization of biological tissues. The DWI signal is influenced by the choice of the sequence and the strength of the diffusion weighting given by the b-value. If two or more images with different b-values are acquired, a quantitative measurement of diffusion may be obtained: the ADC. ³⁴

Most clinical DWI examinations are performed with diffusion-weighted single-shot spin-echo echo planar imaging (EPI) sequences, which have the advantage of being fast, and therefore relatively robust against motion artifacts. ³⁵ However, EPI sequences have limited spatial resolution (i.e., 128 × 128 pixels), increased susceptibility to magnetic field inhomogeneities, and eddy currents. ³⁵ They are prone to significant image distortion and susceptibility artifacts due to the presence of the cortical bone–fat interface, and the vicinity to the lungs and great vessels. ³⁶ Single-shot reduced FOV EPI can reduce geometric distortion effects but reduces signal-to-noise ratio and therefore requires longer scan times. ³⁶ Multishot readout-segmented EPI sequences were more recently proposed to decrease susceptibility and motion artifacts compared with single-shot EPI, with reasonable scanning times. ³⁷

Of importance, the presence of fat reduces ADC values, and the use of fat suppression techniques is usually required. ^{34 35}

Normal Bone Marrow

The ADC values of normal vertebral bone marrow range between 0.2 and 0.6 × 10 ⁻³ mm ² /s. ^{35 38} These variations are partly explained by technical parameters, such as differences in hardware, pulse sequences, diffusion weightings, and postprocessing.

In addition, the diffusion properties of normal bone marrow depend on physiologic parameters. The relative amount of fat and water influence the ADC values, with the ADC of fat close to zero. ^{34 35} Therefore, DWI of the normal bone marrow may vary depending on factors influencing the FF, including sex, age, premenopausal status, as well as anatomical location (axial versus peripheral skeleton, different vertebral levels, pelvic bones, etc.). ^{4 39 40} Any condition influencing the red versus yellow marrow distribution and the proportion of mineralized bone (osteoporosis, anemia) may influence DWI ^{35 40} ( Fig. 4 ).

Fig. 4.

A 65-year-old man with newly diagnosed rectal cancer referred for staging by magnetic resonance imaging (MRI) of the abdomen and pelvis. Transverse ( a ) diffusion-weighted image (DWI) (b = 600) and ( b ) corresponding apparent diffusion coefficient (ADC) map and ( c ) T1-weighted images show a diffuse increased signal of the marrow on DWI (a, asterisk) with restricted diffusion (ADC value: 0.62 ×10 ^–3 mm ² /s) and normal T1 (c, asterisk). Known aneurysms of the iliac arteries treated by endovascular treatment (arrows). Transverse T1-weighted gradient-echo ( d ) in-phase and ( e ) out-of-phase images performed in the routine protocol of the abdomen show a 68% signal drop, indicating residual fat in keeping with marrow hyperplasia. The patient had severe anemia (Hb: 6.1 mg/dL).

Clinical Applications

Lesion Characterization and Evaluation of Vertebral Compression Fractures

Bone marrow lesions, whether benign or malignant, have higher ADCs than normal bone marrow. On one hand, this finding can be explained because normal bone marrow has a low signal on DWI and a low ADC value, mostly attributed to the high fat content and the presence of bone trabeculae. On the other hand, bone marrow lesions exhibit high tumoral cellularity, decreased FF, increased water content and vascularity, and disrupted trabeculae that likely contribute to the higher signal on DWI and higher ADC values. Thanks to the difference between normal marrow and malignant lesions, DWI has been established as a method of reference to assess bone marrow involvement in the oncologic setting. ^{41 42 43} DWI has also been used to differentiate between benign and malignant bone marrow lesions.

In a meta-analysis, differentiation between benign and malignant vertebral lesions based on quantitative analysis of ADC values had pooled pooled sensitivity and specificity of 89% and 87%, respectively. ⁴⁴ Note, however, that the group of benign lesions included a wide range of pathologies including infectious lesions, nodular hyperplastic bone marrow, and other primary bone lesions lesions. Due to their different tissue characteristics, these lesions should present different ADC values (e.g., hyperplastic bone marrow has low ADC value due to the preserved bone and bone marrow structures), making the analysis difficult. For the differentiation of benign and malignant VCFs, ADC values yielded pooled sensitivity and specificity of 92% and 91%, respectively. ⁴⁴ In practice, morphological analysis of VCFs most of the time is sufficient for the diagnosis, ¹³ and the added value of DWI in cases where morphological analysis is inconclusive remains to be determined.

Diabetic Foot

In diabetic patients, the distinction between pedal osteomyelitis and diabetic neuroarthropathy (Charcot's arthropathy) is challenging, and both conditions may coexist. Few studies have investigated the use of DWI in diabetic feet with discordant results. In two studies, ADC values were higher in diabetic neuroarthropathy than in osteomyelitis, suggesting a cut-off value of 0.98 × 10 ⁻³ mm ² /s to differentiate the two conditions, ^{45 46} whereas there was a significant overlap between the ADC values in a larger study that used a normalized signal intensity for ADC. ⁴⁷ In a recently published study, a readout-segmented multishot echo planar DWI showed promising results to distinguish osteomyelitis from bone marrow edema (BME) related to other conditions, although there was again some overlap between the two groups, with a 95% accuracy achieved in only 73% of cases. ³⁷

Spinal Infections

In vertebral end endplates with BME, on visual inspection of DWI, the presence of the “claw sign,” defined as well-marginated linear regions of high signal located within the adjacent vertebral bodies at the interface of normal-abnormal marrow, was shown to be highly suggestive of Modic type 1 degenerative changes, with its absence suggesting infection. ⁴⁸ However, in practice, morphological assessment on fat- and fluid-sensitive sequences, in addition to contrast-enhanced sequences if needed, is sufficient in most cases, and the added value of the claw sign in equivocal cases remains to be determined.

ADC values of infectious bone marrow are significantly higher than normal and degenerative bone marrow, and DWI has been proposed as an adjunct to conventional MRI to detect spinal infections especially if intravenous contrast injection is contraindicated ^{49 50 51} ( Fig. 5 ). The applicability of thresholds, however, is difficult in practice. The usefulness of ADC values in differentiating malignancy from infection was not consistently demonstrated across studies. ^{50 52 53}

Fig. 5.

A 29-year-old man with pyogenic spondylodiskitis. ( a ) Sagittal fat-suppressed T2-weighted image shows T8–T9 spondylodiskitis with a spread of the infection to the prevertebral soft tissues (arrow). Sagittal left paramedian ( b ) fat-suppressed T2-weighted diffusion-weighted image (DWI) (b = 700) and apparent diffusion coefficient map show a phlegmon involving the paravertebral soft tissues (asterisk) with areas of restricted diffusion ( c, d, arrows) corresponding to (micro)abscesses. DWI can be useful to detect soft tissue abscesses if intravenous contrast injection is contraindicated.

Spondyloarthropathy

Few studies have assessed DWI to evaluate sacroiliitis, mostly showing inferiority to STIR images, limited added value to diagnose sacroiliitis, limited specificity, and poor interobserver agreement. ^{54 55 56} In one study, ADC values of edematous changes of vertebral endplate of rheumatologic origin were higher than those in Modic type 1 changes. ⁵⁷

Osteonecrosis

The T2-blackout effect on DWI (low DWI, low ADC) was described as a sign of early bone infarct and sequestration in a patient with sickle-cell disease. ⁵⁸ However, DWI has limited value in the staging of femoral head osteonecrosis in clinical practice. ^{30 59}

Pitfalls

DWI encompasses a spectrum of sequences heavily influenced by technical parameters and patient-related factors. First, substantial technical variability exists among different vendors and different institutions (e.g., EPI or FSE technique, field strength, fat suppression, choice of b-values). ^{34 35} Second, DWI is influenced by patient-related factors, whether physiologic (e.g., marrow composition, age, sex) or pathologic (e.g., anemia, osteoporosis) ⁶⁰ ( Fig. 4 ). Third, the most widely used DWI techniques are EPI based that are prone to susceptibility artifacts at tissue boundaries (bone, lungs), more pronounced at higher field imaging, and a source of pitfalls in image interpretation. ⁶¹ For example, sclerotic metastases may alter image contrast on DWI, and blood products and metal debris may give erroneous ADC values in a postoperative setting. ⁵¹ Finally, due to the technical factors and lack of standardization, quantitative measurements have limited reproducibility, and there is frequently substantial overlap between cut-off values, hindering their universal applicability ( Figs. 6 and 7 ). Finally, the added value of DWI in comparison with conventional morphological imaging is yet to be determined.

Fig. 6.

A 12-year-old girl with painless progressive kyphosis. Sagittal ( a ) diffusion-weighted image (DWI) (b = 500) and ( b ) corresponding apparent diffusion coefficient (ADC) map, ( c ) transverse contrast-enhanced fat-suppressed T1-weighted images, and ( d ) coronal T2-weighted images show abnormal restricted diffusion of T12 and L1 vertebrae (asterisks) with epidural extension (b, arrow) and paravertebral peripherally enhancing collections with low T2 rim (c, d, arrows). The patient was found to have tuberculous spondylitis. The usefulness of ADC values in differentiating malignancy from infection is not consistent (courtesy of Joseph El-Khalil, MD, Beyrouth, Lebanon).

Fig. 7.

A 53-year-old man with fever and back pain and suspicion of spondylodiskitis on computed tomography (CT) (not shown). Sagittal ( a ) T1-weighted, ( b ) fat-suppressed T2-weighted, ( c ) contrast-enhanced T1-weighted, ( d ) diffusion-weighted image (DWI) (b = 500), and ( e ) corresponding apparent diffusion coefficient (ADC) map show marrow replacement of T2 and L1 vertebral bodies, restricted DWI (asterisks) with an ADC value of 0.58 ×10 ⁻³ mm ² /s, a peripherally enhancing prevertebral collection with restricted DWI, causing scalloping of the anterior vertebral wall (thin arrows) and elevation without disruption of the anterior longitudinal ligament (a, arrowhead). The differential diagnosis included tuberculous spondylitis and lymphoma. The patient was diagnosed with Hodgkin's lymphoma on biopsy and positron emission tomography/computed tomography (not shown). There is an overlap of ADC values between malignant and infectious lesions, making the differentiation difficult based on quantitative imaging alone.

Marrow Perfusion: Dynamic Contrast-enhanced MRI

Technique

DCE-MRI consists of assessing tissue perfusion through serial acquisitions of images before and after a bolus of intravenous contrast injection and the assessment of the variation of MR signal intensity of the tissues of interest, both qualitatively and quantitatively.

The nonmodeled quantitative and visual assessments in DCE-MRI are easily implemented in clinical settings and not computationally demanding. Images can be visually analyzed and regions of interest drawn on an area of interest to obtain DCE time-intensity curves. Qualitative assessment of contrast uptake, wash-in, and washout rates is the most widespread method in clinical routine.

The quantitative analysis is based on the Tofts bicompartmental model. ^{62 63} This quantitative analysis involves the conversion of signal intensity to gadolinium concentration and fitting of the data into a tissue model, and yields four main parameters (volume transfer constant [K ^trans ], rate constant [K _ep ], fractional plasma volume [V _p, ], and extravascular space [V _e ]) that reflect the gadolinium distribution between the intravascular and the extravascular-extracellular compartment. ⁴² Quantitative DCE-MRI has been mainly used in a research setting to gain knowledge of the pathophysiology of various diseases such as osteoporosis, osteonecrosis, and osteoarthritis. ^{25 64 65}

Normal Bone Marrow

Normal bone marrow shows variable enhancement after intravenous contrast injection (mean: 20%; range: 3–59%). ⁶⁶ However, vascularity depends on marrow composition, and red marrow is highly vascularized compared with yellow marrow. ^{16 67 68} Therefore, bone marrow perfusion on MRI may vary depending on marrow composition, age, and sex, with higher levels in women and decreasing levels with age. ^{66 69 70 71} The interpretation of DCE-MRI studies should be performed considering these physiologic differences.

Clinical Applications

Lesion Characterization and Evaluation of Vertebral Compression Fractures

Among the potential applications of DCE-MRI, it was suggested that the V _p , a quantitative parameter extracted from DCE-MRI, could differentiate benign from malignant VCFs with a sensitivity of 93% but a specificity of 78%. ⁷² Pathologic VCFs were shown to have higher perfusion parameters (V _p , K ^trans , wash-in slope, peak enhancement, and area under the curve) compared with benign fractures. ⁷³ However, the added value of DCE-MRI in cases where morphological analysis is inconclusive remains to be determined.

Diabetic Foot

DCE-MRI could be a useful adjunct to standard MRI protocols to help differentiate acute neuroarthropathy from pedal osteomyelitis. In one study, the K ^trans , K ^ep , and V _e values of bones with osteomyelitis were higher than those of acute neuropathic arthropathy. ⁷⁴ In another study, the K ^trans allowed a reliable differentiation between both entities but was inferior to the visual assessment of fluorodeoxyglucose positron emission tomography (FDG-PET)/CT. ⁴⁷ DCE-MRI has also been used to predict and monitor treatment response in acute Charcot's foot. ⁷⁵

Spinal Infections

Quantitative analysis of DCE-MRI was shown to contribute to the early diagnosis of brucella spondylitis ⁷⁶ and the differential diagnosis between spinal metastatic tumor, brucella spondylitis, and spinal tuberculosis. ^{76 77} Tuberculous vertebral lesions, especially in the early phases, may mimic malignant lesions on conventional MRI sequences but also on DWI ( Figs. 6 and 7 ). DCE-MRI has been used to assess spinal tuberculosis. On quantitative analysis, the presence of washout or a K _ep ≥ 1.17 min ⁻¹ was shown to be highly predictive of malignancy. ^{78 79}

Pitfalls

Significant variations in imaging protocols, scanner types, and postprocessing methods hamper the universal applicability and reproducibility of quantitative DCE-MRI parameters. For example, quantitative parameters may vary depending on the acquisition protocols (amount of contrast agent, temporal resolution, and scan duration). ⁸⁰

Whole-body MR Imaging for Non-neoplastic Marrow Conditions: Applications and Challenges

Whole-body MRI has emerged as a useful tool in oncologic imaging (e.g., for staging, assessing disease burden, and treatment response). MRI acquisition protocols vary, but experts agree it should include T1-weighted, STIR, and DWI sequences, with recent trends toward using T2-weighted Dixon images to replace T1-weighted and STIR. ^{8 9 41 81}

Non-oncologic applications of whole-body MRI include the investigation of fever of unknown origin (FUO) in children, ^{82 83} the diagnosis and assessment of disease activity and treatment response in chronic recurrent multifocal osteomyelitis, ^{84 85 86 87 88 89} and synovitis, acne, pustulosis, hyperostosis, osteitis (SAPHO). ⁹⁰ Whole-body MRI has also been used to detect clinically occult inflammatory lesions and to provide a simultaneous evaluation of the axial and appendicular skeleton in inflammatory arthritis. ^{91 92 93 94 95 96 97 98} Other applications include the evaluation of the disease burden in Gaucher's disease ⁹⁹ and multifocal osteonecrosis. ¹⁰⁰

However, the incorporation of whole-body MRI into the routine clinical workflow remains challenging because of examination duration, the need for an experienced reader, and, in many countries, potential billing difficulties. ⁴¹ The use of gradient-echo techniques, FSE T2-weighted Dixon sequences, and acceleration techniques, such as simultaneous multislice sequences and potentially reconstruction algorithms based on artificial intelligence, may contribute to decreasing acquisition time and allow wider use. ^{101 102}

Other Techniques of Marrow Imaging Using MRI

¹ H-MRS spectroscopy provides concentrations of specific metabolites, and its main application in marrow imaging is the quantification of fat. ^{36 103} In clinical research, it has been investigated in a few applications related to osteoporosis, fracture risk assessment, and response to treatment in patients with metabolic diseases like anorexia nervosa, ^{104 105} Gaucher's disease, ¹⁰⁶ rheumatoid arthritis, ¹⁰⁷ multiple myeloma, ¹⁰⁸ or Charcot's neuroarthropathy. ¹⁰⁹ However, in clinical practice, its use has significantly decreased in favor of CSI. ¹¹⁰

Intravoxel incoherent motion MRI uses low b-value DWI to acquire perfusion maps without the need for contrast injection. This technique has been applied in neuroradiology and oncology, and it may be used to assess muscle perfusion. However, its usefulness in marrow imaging in clinical practice remains to be validated by further studies. ^{111 112}

Spectral Computed Tomography Imaging

Technique

Spectral CT is based on the principle that the attenuation of tissues depends not only on their density but also on their atomic number Z, as well as on the energy of the photon beam. ¹¹³ Using these properties, spectral CT may be used to characterize and quantify certain tissue components. As an example, mineralized tissues of bones can be subtracted from the image and VNCa images may be obtained, allowing the detection of bone marrow lesions. ^{114 115}

The most commonly available subset of spectral CT is DECT, in which two X-ray energy spectra are used. More recently, photon counting detectors were introduced. This new technology allows multi-energy imaging, a direct count of individual incoming photons, and a measure of their energy level. ^{116 117} Photon counting detectors can provide higher spatial resolution, dose reduction, and better material differentiation, and they are less prone to beam-hardening artifacts. ^{116 117 118} The following sections summarize some applications of DECT for bone marrow imaging, keeping in mind that photon counting CT has the potential to improve the diagnostic performance of DECT for these applications, although this is yet to be validated.

Normal Bone Marrow

DECT may be used to assess bone marrow composition. ¹¹⁹ However, to the best of our knowledge, no consensus has been reached on the definition of normal bone marrow on DECT. In fact, image analysis is based on a comparison of areas of interest with the presumed normal bone marrow within the FOV and is influenced by postprocessing algorithms and thresholding parameters.

Clinical Applications

Detection of Bone Marrow Edema and Marrow-replacing Lesions

The performance of conventional CT to detect bone marrow lesions that do not alter the mineralized bone is poor. VNCa reconstructions have the potential to improve the detection of BME-like lesions, as well as bone marrow–replacing lesions. For sake of simplicity, we use “BME” to refer to BME-like lesions in the rest of this section.

The DECT depiction of BME most commonly uses two material decomposition algorithms (calcium/water) based on different attenuation profiles at different energies. VNCa images can hereby be generated, allowing the assessment of bone marrow attenuation. VNCa images are often interpreted as color maps coding the attenuation of bone marrow and may be fused with native bone images for better anatomical correlations. ¹²⁰

Bone marrow neoplastic lesions, unless lytic or sclerotic, are difficult to detect on conventional CT. VNCa images were shown potentially useful for the detection of bone marrow lesions in multiple myeloma ¹²¹ and metastases of solid tumors. ¹²²

Moreover, studies have used DECT to differentiate malignant from nonmalignant tumors, ¹²³ osteoblastic metastases from bone islands, ¹²⁴ osteolytic metastases from Schmorl's nodes, ¹²⁵ and infections. ¹²⁶ However, MRI so far remains the keystone morphological modality for imaging bone marrow, and more validation studies of DECT are required.

Vertebral Compression Fractures

DECT allows the detection of BME and the differentiation of acute from chronic VCFs. ^{120 127 128 129 130 131} In a recent systematic review and meta-analysis, DECT had 89% sensitivity and 96% specificity to diagnose BME related to a recent VCF, possibly obviating the need for a confirmatory MRI in the emergency setting. ¹³² However, CT with a single-source technique had poorer specificity (78%) compared with those with a dual-source technique (98%), the diagnostic accuracy for the detection of BME depended on the reader's experience, and the absence of BME did not confidently rule out the diagnosis of acute VCF if the clinical suspicion was high. ¹³² Of importance, the sensitivity of DECT in the diagnosis of acute but morphologically occult VCFs remains to be addressed. ¹³²

Trauma of the Appendicular Skeleton

In clinical practice, CT is used in traumatic contexts with noncontributory radiographs and clinical suspicion of fracture. However, the lack of cortical discontinuity or trabecular displacement can make the diagnosis challenging. MRI is considered the gold standard for detecting BME in occult fractures; however, it is not widely available in the emergency context.

VNCa images may improve visualization of occult fractures and improve the reliability and diagnostic confidence among less experienced readers. ^{120 133} In a recent meta-analysis, DECT had a pooled sensitivity and specificity of 86% and 93%, respectively. ¹³⁴ DECT reduces the reading time while analyzing lower extremity CTs for fractures when the radiologist is presented with BME maps, and time reductions were more evident for unskilled readers. ¹³⁵ DECT could potentially function as a one-stop shop and obviate the need for confirmatory MRI.

Low-energy Trauma of Older Adults

DECT increases the sensitivity to detect radiographically occult fractures of the pelvic girdle and proximal femurs in older adults, whether related to low-energy trauma or bone insufficiency. ^{136 137 138 139 140} Some authors have suggested that a quantitative analysis of VNCa images might help differentiate pathologic from nonpathologic fractures. ¹⁴¹ However, this finding should be confirmed in larger studies.

Rheumatologic Disorders

DECT can detect inflammatory lesions in sacroiliitis with relatively high sensitivities (between 81% and 93%) and specificities (91–94%). ^{142 143} The incidental detection of periarticular BME may help in the early diagnosis of rheumatologic disorders. ¹²⁰

Miscellaneous Marrow Conditions

DECT can detect nontraumatic BME of the hip and knee with a sensitivity of 88.4% and specificity of 96.1%. ¹⁴⁴ If MRI is contraindicated, DECT could help depict BME associated with cortical erosions confirming osteomyelitis. This possibility should be validated in further studies. ¹²⁰

Pitfalls and Limitations

Validation of established diagnostic thresholds for bone marrow alterations is difficult because of the variety of acquisition and postprocessing methods ^{132 145 146} ( Fig. 8 ). The ability of the material decomposition in DECT increases with lower spectral overlap. ¹⁴⁷ Therefore, a wider separation between the energies of the two tubes could improve DECT sensitivity to detect BME. ¹⁴⁸

Fig. 8.

A 46-year-old man with left knee pain. ( a ) Coronal intermediate-weighted fat-suppressed image shows edema-like signal intensity of the lateral femoral condyle and the tibial eminence (asterisks). Two types of coronal color-coded dual-energy computed tomography reconstructions are displayed: ( b ) virtual non-calcium and ( c ) virtual non-hydroxyapatite, showing the influence of postprocessing. There are false positives (arrows) and false negatives (arrowhead) (b).

As a general recommendation, VNCa maps should not be assessed in isolation, and the reader must be aware of possible pitfalls. Any process that locally increases the attenuation of the bone marrow to a density higher than fat could show false-positive results. Red marrow hyperplasia can be misinterpreted as BME (i.e., in the proximal femurs and flat bones as vertebrae). Hence careful comparison with the contralateral side and adjacent structures is recommended when looking for an occult fracture.

Bone sclerosis can cause both false-positive findings due to locally increased Hounsfield unit levels, and false-negative findings as an extensive subtraction process can hide BME detection. In the presence of a splint or cast, BME may not show even in displaced fractures, probably due to a reduction in the attenuation differences caused by the dense material around the limb. ¹²⁰ In general, VNCa maps should always be assessed together with conventional images to avoid pitfalls.

Additional limitations related to technical challenges, standardization of reconstruction, and decomposition algorithms, as well as window level and width settings, remain to be addressed before the widespread implementation of DECT in marrow imaging.

Nuclear Medicine and Molecular Imaging

Technique

Nuclear imaging relies on the intravenous injection of radiopharmaceuticals to assess the distribution of hematopoietic or reticuloendothelial cells. Imaging of the hematopoietic component can be achieved by white blood cell (WBC) scintigraphy, by injecting the patient's WBCs after they have been radiolabeled in vitro with technetium (Tc)-99m (Tc) or indium-111, or by injecting Tc-99m-labeled mouse anti-granulocytes monoclonal antibodies or antibody fragments. The reticuloendothelial cells may be imaged thanks to radiolabeled colloids (Tc-99m-sulfur colloid and Tc-99m-nanocolloid) that are phagocytosed by the reticuloendothelial system in bone marrow, spleen, and liver. Bone scintigraphy with intravenous injection of Tc-99m-labeled diphosphonates allows imaging of bone osteoblastic activity.

Furthermore, the metabolic activity of bone marrow can be probed by ¹⁸ F-Fluoro-deoxyglucose positron emission tomography ( ¹⁸ FDG PET).

Osteomyelitis

WBC scintigraphy has high accuracy in evaluating infections involving the appendicular skeleton. However, its diagnostic performance for the detection of infection in the axial skeleton is hampered by intense physiologic bone marrow uptake.

¹⁸ FDG PET/CT may be used in patients in whom MRI is contraindicated to detect sites of infection in the axial skeleton (i.e., spondylodiscitis). Focal FDG uptake in the diabetic foot may be a useful sign to differentiate osteomyelitis from Charcot's neuro-osteoarthropathy when MRI is inconclusive. ⁴⁷ In a 2017 meta-analysis, WBC scintigraphy and FDG-PET had a comparable sensitivity to MRI but a higher specificity to diagnose pedal osteomyelitis. ¹⁴⁹ In addition, WBC scintigraphy has high diagnostic accuracy to evaluate fracture-related infections ¹⁵⁰ ( Fig. 9 ).

Fig. 9.

A 54-year-old man with a history of previous right tibial fracture and suspicion of osteomyelitis. Anterior planar images were acquired at ( a ) 4 hours and ( b ) 24 hours after intravenous injection of mouse monoclonal 99m-technetium-antigranulocyte antibodies showing focal accumulation of leukocytes in the right tibial diaphysis (arrow). ( c ) Coronal fused single-photon emission computed tomography/computed tomography images confirmed that the uptake corresponds to the site of the previous tibial fracture (arrow).

WBC imaging is also a valuable tool to assess periprosthetic infections. Bone scintigraphy is highly sensitive but not specific for septic loosening, whereas WBC scintigraphy has high sensitivity and specificity for infection. ^{151 152 153} In practice, the absence of periprosthetic uptake on bone scan practically rules out infection, whereas an increased periprosthetic uptake should be complemented by WBC scintigraphy, with or without a colloid scan. ¹⁵³

In cases of FUO, FDGPET has higher diagnostic accuracy than WBC scintigraphy to detect the infectious site, although the latter is more specific. ^{152 154 155} A new technique consisting of radiolabeling WBC with FDG might replace Tc-99m WBC in this setting because it is more specific than FDG-PET/CT and more sensitive than WBC scintigraphy. ¹⁵⁶

Vertebral Compression Fractures and Occult Fractures

Bone scintigraphy can depict VCFs because it is a very sensitive method for detecting active bone remodeling. ¹⁵⁷ In patients with MRI contraindications and multiple vertebral fractures, it may help guide therapy by pointing out the most recent fracture. Bone scintigraphy is also an excellent imaging alternative to MRI in patients with suspicion of occult fractures. ¹⁵⁸

Focal Marrow Hypermetabolism on FDG-PET/CT

Normal bone marrow metabolism is variable and declines with age. ¹⁵⁹ Diffuse increase in bone marrow metabolism can occur in hematologic malignancies but is frequently observed in non-neoplastic conditions resulting in marrow stimulation such as sepsis, rebound hematopoiesis after chemotherapy, or following the administration of hematopoietic growth factors. ^{160 161} A focal FDG uptake may be due to infection or focal medullary hyperplasia that may mimic metastatic lesions in an oncologic setting. ¹⁹ In cases where the characterization of a single uptake is crucial for staging (i.e., no other metastases), MRI with CSI may be useful ( Fig. 1 ).

Miscellaneous Applications

WBC and, to a lower degree, colloid scintigraphy can be used for the diagnosis of extramedullary hematopoiesis. ¹⁶² In addition, WBC scintigraphy can also be used to map bone marrow in cancer patients with impaired bone marrow function to help predict the effect of therapies such as external radiotherapy or metabolic radiotherapy on hematopoiesis ¹⁶³ ( Fig. 10 ).

Fig. 10.

( a ) Anterior planar image of the whole body acquired 4 hours after intravenous injection of mouse monoclonal 99m-technetium (Tc)-antigranulocyte antibodies showing a normal distribution pattern of the radiotracer in a 37-year-old patient. Note the distribution of the radiotracer in the axial skeleton and the absence of uptake in the appendicular skeleton, paralleling the normal red marrow distribution in adults. ( b ) Anterior planar image of the whole body acquired 4 hours after intravenous injection of mouse monoclonal 99m-Tc-antigranulocyte antibodies in a 75-year-old man with metastatic prostate cancer, showing an intense abnormal uptake in the appendicular skeleton (arrows), together with the absence of uptake in the axial skeleton as compared with the normal pattern in (a). Bone marrow hyperplasia in the peripheral skeleton has occurred as a consequence of diffuse marrow replacement by metastases in the axial skeleton.

Limitations and Pitfalls

Nuclear imaging techniques present some general limitations including the need for a specific technical platform (WBC scintigraphy), the radiation dose, and cost-related issues.

Pitfalls exist and vary according to the radiotracer used. An extensive review of these pitfalls is beyond the scope of this article, but a good understanding of the physiology and distribution of each radiotracer is essential to avoid several of these pitfalls, including false-positive findings.

Conclusion

Over the past decades, there has been a growing interst in novel imaging techniques for the assessment of bone marrow, both qualitatively and quantitatively. Although many of these methods have been successfully used in the research setting, their incorporation in clinical practice has been limited. Many challenges remain to be addressed, including availability, cross-vendor and cross-institutional reproducibility, issues related to reimbursement, as well as examination and processing time. Once these challenges are overcome and techniques are standardized, further studies will be needed to assess the added value of these methods in relation to conventional protocols.