T1-weighted spin-echo (SE) images are best to evaluate the cellular content in bone marrow because of high fat content interspersed with hematopoietic elements. The hydrophobic carbon-hydrogen groups in fat result in a short T1 relaxation time because of very efficient spin-lattice relaxation [1]; conversely, water has a long T1 relaxation time [2]. Yellow marrow has signal intensity (SI) comparable with subcutaneous fat, whereas red marrow has intermediate T1 relaxation with SI lower than subcutaneous fat but higher than disk or muscle [2] (Fig. 1). The exact SI of the marrow, however, is dependent on the proportion of red and yellow marrow. Cellular infiltration retains some intermixed fat, whereas replacement obliterates all fat within the bone marrow [3]. This replacement of the bone marrow characteristically appears hypointense relative to disk and muscle on T1-weighted images [3] and hypointense relative to normal marrow [4, 5]. Marrow T1-weighted hypointensity, however, is nonspecific. Generally, bone marrow signal that is hypointense to adjacent muscle or intervertebral disk at 1.5 T is abnormal with accuracy of 94% and 98%, respectively [6]. At 3-T field strength, Zhao et al. [7] showed a higher diagnostic accuracy comparing T1 marrow SI to muscle (89%) rather than disk (78%). For focal abnormalities, the bull’s-eye sign (i.e., a focus of high T1 signal intensity in the center of an osseous lesion) has been reported to be a specific indicator of normal hematopoietic marrow (sensitivity, 95%; specificity, 99.5%) [8] (Fig. 2). With diffuse abnormalities, extensive replacement of the vertebral bone marrow may initially create the impression of a normal study because of its homogeneity, and careful comparison of T1-weighted marrow signal to disk or muscle is required to make the diagnosis (Fig. 3).

Conventional T2-weighted sequences can also be helpful for evaluating bone marrow pathology. Fat protons have a relatively long T2 relaxation because of less-efficient spin-spin relaxation [1]. Yellow marrow shows intermediate and high SI on T2-weighted SE and T2-weighted fast SE (FSE) MRI, respectively. Fatty marrow appears higher in SI than muscle and equal to or slightly lower in SI than subcutaneous fat. Because water and fat are closer in SI, there is decreased contrast of the spinal marrow on this sequence. Moreover, FSE sequences do not fully suppress fat signal such that the contrast differences between red and yellow marrow can be difficult to delineate. Red marrow SI is slightly lower than that of yellow marrow [9]. Although metastatic lesions are usually brighter than normal bone marrow on T2-weighted MRI due to their high water content, they sometimes can be difficult to differentiate from normal marrow on T2-weighted sequences [10]. The halo sign, characterized by a rim of bright T2 signal, sometimes may be detected on T2-weighted MRI and indicates the presence of malignancy [8] (Figs. 4A and 4B). Schweitzer et al. [8] showed high specificity of both the halo sign and diffuse T2-weighted signal hyperintensity for metastatic disease (sensitivity, 75%; specificity, 99.5%) [8]. Both conventional SE and FSE T2-weighted sequences have been shown to detect the same number of bone marrow lesions [11]. The rapidity of the FSE T2-weighted sequence has made it the more common T2-weighted sequence used for clinical imaging.

As a part of the routine MR sequences, the combination of T1-weighted and either fat-saturation T2-weighted or STIR images is highly effective for the evaluation of bone marrow lesions [10]. Bone marrow contrast can be accentuated by using fat-suppressed sequences, either the chemically selective fat-saturation technique or STIR images (discussed later) [12]. Fat saturation can be applied to T2-weighted and gadolinium-enhanced T1-weighted images. At 1.5 T, a 210-Hz difference exists between the resonant frequency of fat and water [13]. For these sequences, a selective radiofrequency saturation pulse and dephasing gradient is applied based on the 210-Hz chemical shift, which suppresses the lipid signal but only minimally affects the signal coming from nonlipid tissues, such as water [13]. This narrow saturation band requires a very homogeneous magnetic field and is sensitive to magnetic field inhomogeneities from susceptibility differences, such as off-isocenter imaging and a large FOV. Poor fat saturation can be problematic, particularly when evaluating bony edema on T2-weighted and osseous lesions on contrast-enhanced T1-weighted images (Figs. 5A and 5B). Additionally, orthopedic implants may result in poor fat saturation because of local magnetic field inhomogeneity. This may be secondary to shifting of the resonant frequencies of water and fat, prevention of the frequency-specific saturation pulse from targeting fat, or unintended suppression of water signal [13] (Figs. 6A and 6B). Techniques such as increasing the bandwidth and orienting the frequency encoding direction parallel to the hardware can reduce related susceptibility artifact.

With the addition of fat-suppression techniques, bone marrow SI may be altered. On fat-saturated T2-weighted MRI, hematopoietic marrow shows intermediate SI similar to that of muscle, whereas fatty marrow shows SI lower than that of muscle. By comparison, most marrow pathology exhibits relatively high SI, greater than that of red and yellow marrow, on fat-suppressed images. Uchida et al. [14] studied 22 patients, comparing nonneoplastic and metastatic bone marrow using T1-weighted SE and chemical shift fat-saturated T1-weighted imaging. On fat-saturated T1-weighted images, they found that metastases show mixed to high SI, whereas nonneoplastic lesions have low SI.

The STIR sequence provides high tissue contrast, which is useful in evaluating bone marrow (Figs. 7A and 7B). The differences in longitudinal relaxation between fat and water protons result in differences in the T1 relaxation, which are the basis of the STIR technique [15]. An inversion time (TI) is chosen at the time to null the signal generated by fat. First, a nonselective 180° radiofrequency inversion pulse is applied. This is followed by a second 90° pulse, which is applied at the TI, which is the time the previously inverted longitudinal magnetization of fat crosses the null point [14, 15]. With this series of pulses, the signal from fat is suppressed and the signal from water is preserved in the selected slice.

The STIR sequence tends to produce more homogeneous fat suppression than T2-weighted FSE, which is performed with frequency-selective fat saturation. The main drawback of the STIR sequence is that it cancels every signal close to fat [15]. Troublesome examples of this include tissue that enhances with contrast administration (e.g., if STIR is performed after contrast administration) and blood products in a hematoma. Although the conspicuity of lesions is similar on STIR and fat-saturated T2-weighted imaging, the latter has several practical advantages, including acquisition of more slices per unit of time and improved tissue specificity [10]. Although the STIR sequence is time consuming, with only a limited number of slices acquired at one time, this can be overcome by using fast STIR sequences [16].

Gadolinium-enhanced T1-weighted images may be helpful in detecting some marrow lesions; however, enhancing lesions may be difficult to discern if they become isointense to normal bone marrow (Figs. 8A and 8B). Sequences that suppress the SI of normal fatty bone marrow allow better identification of the enhancing metastatic foci [17, 18]. However, Meyer et al. [19] showed, in their study of 91 patients with neuroblastoma being evaluated for bone marrow metastases, that although the conspicuity of lesions was increased with the heterogeneous pattern of enhancement on gadolinium-enhanced T1-weighted images with fat saturation, the sensitivity of lesion detection was not increased. On the other hand, contrast administration can be helpful in confirming malignancy or osteomyelitis and evaluating the extent of extramedullary spread [20]. In addition, contrast administration allows identification of lesions that may alter treatment, particularly lesions that spread to the epidural space and may result in compression of the spinal cord [21].

The T1 FLAIR sequence is proving to be a useful sequence in spinal imaging (Fig. 9). FLAIR techniques typically use long TE readout (increased T2-weighting) to null CSF signal and thereby improve the conspicuity of tissue abnormalities without the diagnostically problematic effects of CSF pulsation artifacts and volume averaging [22–25]. Rapid acquisition with relaxation enhancement (RARE) techniques, which are equivalent to FSE, can be incorporated with basic FLAIR sequences to improve the efficiency [26]. With this imaging technique, a fast T1-weighted inversion recovery sequence with different TI pairs to null CSF and vary T1 effects on non-CSF tissues is coupled to a RARE readout [27]. This sequence is also particularly useful with 3-T MRI because there is decreased fluid contrast that is associated with the lengthened T1 relaxation times seen with higher magnetic strength [25]. In addition to evaluating spinal cord pathology, Melhem et al. [28] have shown that the T1 FLAIR sequence optimizes the tissue contrast between fatty marrow and abnormal tissue. Thus, there is improved conspicuity of edema and metastatic lesions in the bone marrow. Recent literature has also shown that the T1 FLAIR sequence is sensitive for detecting bone marrow pathology [29].

DWI evaluates the tissue-specific molecular diffusion of protons, which may be useful in differentiating bone marrow pathology. For example, in tissues with high cell densities (such as with neoplasm), a decreased apparent diffusion coefficient (ADC) is expected because the exaggerated amount of intra- and intercellular membranes results in restricted diffusion. However, the utility of DWI on differentiating benign from metastatic spinal lesions is controversial. One study using DWI found all benign vertebral compression fractures to be hypoto isointense compared with adjacent normal vertebral bone marrow, whereas pathologic compression fractures were hyperintense [5]. This study found a statistically significant difference (p < 0.001) between the calculated contrast ratios comparing benign and pathologic vertebral fractures [5]. In contrast, Castillo et al. [30] evaluated 15 patients with metastases and showed that DWI of the spine offered no advantage in the detection and characterization of vertebral metastases compared with unenhanced T1-weighted imaging. Nevertheless, they did conclude that DWI was superior to T2-weighted imaging in detecting malignant lesions [30]. Others have shown that, rather than qualitative assessment, the quantitative evaluation of the ADC in vertebral bodies may be useful for differentiating malignant from benign vertebral tissue [31]. In their small study of eight patients with epidural metastatic disease, Plank et al. [32] concluded that DWI may be a useful technique for the evaluation of epidural lesions causing spinal cord compression. They showed decreased diffusivity in neoplastic lesions with higher cellularity [32].

In addition to tumor evaluation, DWI may be useful for differentiation of degenerative and infectious endplate abnormalities. In a recent study of 16 patients with endplate abnormalities, DWI showed hyperintensity in five patients with infection and no hyperintensity in 11 patients with degenerative endplate changes, seven of which were classified as Modic type 1 [33]. ADC values of infectious endplates were significantly higher than normal and degenerative endplates [33].

DWI also may prove useful in distinguishing progressive changes in bone marrow among treatment sequelae, fracture, or tumor, which can be difficult on conventional MRI. In one study, Hanna et al. [34] compared MRI scans using T1-weighted, T2-weighted, STIR, and contrast-enhanced T1-weighted sequences with histologic specimens at 21 sites, 7 of which contained tumor and 14 of which were tumor-free. For all of the tumor-positive sites, the MRI scans revealed abnormalities. For the sites without tumor, there was a significant (more than 50%, depending on the pulse sequence) false-positive rate, presumably because tumor could not be distinguished from the effects of treatment [34]. However, DWI sequences may show decreased SI of metastatic disease of the vertebral marrow with successful treatment [35]. In a study of 24 patients with metastatic disease of the spine, Byun et al. [35] showed decreased DWI signal of the spine greater than 1 month after therapy in patients with clinical improvement, whereas persistently abnormal hyperintense signal was found when there was no clinical improvement.

Chemical shift, or in- and out-of-phase, imaging may be useful in detecting pathology because it takes advantage of the difference in resonance frequency between fat and water protons, which are abundant in bone marrow. Water and fat protons are in phase with one another at a TE of 4.6 milliseconds and 180° opposed at a TE of 2.4 milliseconds at 1.5 T. When a given voxel contains both fat and water, there will be some SI loss on images that are obtained when the protons are in their opposed phase (TE 2.4 milliseconds). More SI suppression occurs when the volume of fat and water is roughly equal. The presence of both fat and water in normal marrow results in suppression of SI on opposed-phase images [36, 37]. An early study by Wismer et al., [38] showed that hematopoietic marrow yields low SI on out-of-phase images and pathologic conditions in red marrow typically result in increased SI because of higher proportions of lipid or water. On the other hand, because of its high fat content, yellow marrow normally yields high SI on out-of-phase images. Pathologic conditions in yellow marrow exhibit decreased SI because of increased tissue water due to the combination of accumulated tissue water and the normally high fat content in these sites [38]. Red marrow reconversion will display normal low signal on out-of-phase imaging because both fat and water elements are present [39].

In- and out-of-phase imaging can be used to determine whether a signal abnormality in bone marrow is due to a neoplastic process [40]. Neoplastic conditions replacing normal marrow will not have signal dropout on out-of-phase images (Figs. 10A and 10B). In their study of 30 patients, Disler et al. [40] calculated relative SI ratios to predict whether lesions were neoplastic or nonneoplastic, and using a 0.81 cutoff value, they achieved sensitivity and specificity of 95%. This technique has also been used to differentiate acute benign versus malignant vertebral fractures [41]. In benign compression fractures, although conventional SE sequences reveal abnormal SI, no marrow-replacing process has occurred. This normal marrow fat results in suppression of SI on opposed-phase images. In pathologic fractures, normal fat-containing marrow is replaced with tumor, which results in lack of suppression on the opposed-phase images. Studies have found a difference in the SI ratio between benign and pathologic compression fractures (sensitivity, 0.95; specificity, 1–0.8) [41, 42].

There are pitfalls to consider in chemical shift imaging, Radiation therapy may normalize the signal-to-noise ratio in treated lesions [41], whereas marrow fibrosis may result in a false-positive interpretation. Susceptibility artifact associated with sclerotic metastases and fracture-related hematoma may result in a false-positive result [39]. On the other hand, metastases containing fat, such as renal cell carcinoma and infiltrative multiple myeloma, may result in a false-negative interpretation [39]. In such cases, histopathologic confirmation may be needed.

Studies have shown an increase in SI within the bone marrow of healthy individuals after gadolinium administration because of bone vascularity [43, 44]. Montazel et al. [45] showed that normal spinal marrow DCE-MRI patterns are dependent on age and fat content [45]. Furthermore, younger subjects may exhibit a percentage of SI increase within the range typical of diffuse malignant marrow infiltration [44]. Higher perfusion parameters are noted in hematopoietic marrow, which is characterized by numerous fenestrated vessels, large vascular pools and channels, and small amounts of poorly vascularized fat [2]. Bone marrow enhancement decreases markedly with increasing age and conversion to fat, although there is significant variability among subjects [45]. This difference in fat composition of bone marrow may be responsible for the interindividual differences in SI after contrast enhancement [44]. Arteriosclerosis associated with aging may alter marrow perfusion and result in ischemic change [46]. Chen et al. [46] showed decreased perfusion in both sexes with increasing age, more marked in women. Studies have also shown a reduction in perfusion indexes within the vertebral body and paraspinal tissues supplied by the same artery as the bone mineral density decreases in both genders [47, 48].

DCE-MRI of bone marrow has been used in patients with lymphoproliferative disease and diffuse marrow infiltration as well as in the assessment of response to chemotherapy [49–51]. The percentage of SI increase has been seen in intermediate-grade and high-grade diffuse marrow infiltration compared with healthy subjects [44]. This MR technique has also been applied in predicting the progression of collapse of an osteoporotic vertebral fracture. Kanchiku et al. [52] found an increased tendency of progression of vertebral collapse with increased unenhanced areas on DCE-MRI.

In addition to DWI, proton MRS [53] has been used in assessing bone marrow changes with varying bone mineral densities. Proton MRS allows the noninvasive quantitative assessment of bone constituents, particularly fat, at the molecular level [54, 55]. In this technique, the global MR signal from bone is divided into the two major segments of water and lipid separated by 3.1 ppm. The spectra of proton MRS have a water peak, originating mostly from red marrow, and a lipid peak, arising from yellow marrow [2, 56]. The lipid signal is composed of at least eight fractions [57], the largest of which is from the methylene group at 1.6 ppm. Signal peaks can vary among subjects and are influenced not only by the water-lipid proton quantity and the tissue environment but also by the surface coil and the distance between the voxel and surface coil [58]. Because of these variables, instead of measuring the absolute signal peak for signal quantification, a lipid-water ratio is often measured [58]. The fat fraction value determined from MRS has been adopted as a predictor of bone marrow fat content in most studies [47, 58].

MRS has been used to illustrate the increase in fat content with age [59]. Demmler et al. [60] showed that a decrease in cancellous bone was accompanied by a corresponding increase in fat cells and decrease in arterial capillaries and sinuses in bone marrow. Not only has it been shown that vertebral marrow fat increases as bone marrow density decreases [47, 48], but also inferences of osseous quality may be made. In a study of 22 patients, Schellinger et al. [58] used proton MRS to show that the percentage fat fraction was higher in lumbar vertebral bodies of subjects with weakened bone compared with the control group, suggesting it could serve as a measure of bone quality. Because the density and spatial orientation of the trabeculae influence the microscopic homogeneity of the magnetic field inside the marrow [55], the line width may be used to reflect bone density. Increased bone density will cause greater magnetic field inhomogeneity and wider spectral peaks as opposed to narrow peaks due to decreased bone mineral density [58].

As stated previously, MR diffusion techniques depend primarily on the interstitial fluid flow, cellularity, and extracellular water volume and, to a lesser extent, on microvascular perfusion [61]. Increased interstitial fluid flow along bone canaliculi from repeated mechanical deformation has been proposed as an explanation for the anabolic adaptive responses to mechanical stress occurring within bone. Although Griffith et al. [47] found no relationship between bone marrow ADC and bone density, Liu et al. [62] showed a positive correlation between ADC and bone mineral density. Griffth et al. did find a weak negative association between marrow ADC and marrow fat content, indicating a reduction in molecular diffusion as marrow fat content increased. Similarly, ADC has been shown to correlate negatively to fat fraction values of vertebral bodies [62], which is due to the increased fat packing of bone marrow taking place in the dilated intertrabecular space because of osteoporosis.

Higher-magnetic-field-strength MRI has the advantage of increased signal-to-noise ratio, which can be used to improve contrast and spatial resolution [63]. Another advantage is that the more rapid scanning time is associated with reduced patient motion artifacts. However, higher magnetic field strengths are associated with particular challenges, and the sequence parameters have to be adjusted. The increased T1 relaxation time with increased magnetic field strength can result in reduced contrast resolution on standard T1-weighted SE images, although there may be improved background suppression on MR angiography. Increased susceptibility effects from trabeculae result in lower SI of bone marrow [64] (Figs. 11A and 11B). The transverse relaxation may have to be increased on FSE sequences. Other modifications for higher field strength include maximizing echo-train length and decreasing slice thickness and time to echo. The latter is necessary to achieve contrast resolution on T2-weighted imaging because of the reduced T2/T2^* relaxation times [24].

The increased effects of magnetic susceptibility can result in signal loss in regions of greater signal dephasing on T2-weighted MRI, such as at tissue-bone and tissue-air interfaces. Increasing the spatial resolution with thinner slices or using parallel imaging may help minimize these magnetic susceptibility effects [24]. With higher magnetic field strength, there is a proportional increase in chemical shift artifact. This spatial misregistration of fat tissue relative to water tissue in the frequency-encoding direction degrades standard SE images [24]. Modifications such as increasing bandwidth can reduce chemical shift artifact but result in decreased signal-to-noise ratio.

Another important consideration is the potentially increased energy deposition in the patient and hardware with higher field strengths. The rate of radiofrequency-induced heat deposited in the patient is increased. This increased specific absorption can be managed by reducing the acquisition flip angle and decreasing the number of phase encoding steps.

To identify abnormal bone marrow on imaging, an understanding of the normal pattern of bone marrow maturation that occurs with age is important. The developmental maturation occurs through the replacement of active hematopoietic marrow, which is actively producing mature blood cells from progenitors, by primarily fatty marrow that no longer produces hematopoietic cells [2]. Humans are born with nearly their entire skeleton composed of red marrow, which converts to fatty marrow over time in an organized predictable fashion. This proceeds in a distal to proximal manner until the age of 25 years when the adult pattern of marrow is in place, with the axial and proximal appendicular skeleton containing the remaining red marrow [2].

MRI of bone marrow has a variable appearance because of the age-related distribution of cellular and fatty marrow. These patterns have large interindividual variations among healthy subjects of a given age. In contrast, marrow distribution and SI patterns show little variation among each vertebral body of the same subject [65]. In addition, the normal distribution patterns of cellular and fatty marrow can be difficult to distinguish from focal or diffuse marrow infiltration [3, 66, 67]. For example, the SI of hematopoietic marrow in neonates may be slightly lower than that of muscle on T1-weighted MRI, reflecting the larger percentage of cellular marrow. Subsequent to the neonate period, however, the SI of bone marrow increases progressively on T1-weighted MRI, reflecting the progressive increase in fat content. Therefore, marrow SI lower than normal muscle almost always indicates pathology.

Spinal bone marrow is characterized by the presence of red marrow throughout life, but the proportions of red and yellow marrow in the axial skeleton vary by age and environmental factors. Women have a larger amount of hematopoietic marrow in early adulthood, with a decline in the water-fat fraction after 60 years compared with men, in whom the decline is in the first 25 years [36]. Red marrow tends to be diffuse in the axial bone marrow under the age of 40 years with only small areas of yellow marrow around the basivertebral plexus [68]. Within individual bones, red marrow is often symmetric [4, 9]. With advancing age, generally over age 40 years, the vertebral bone marrow becomes increasingly replaced with fatty marrow. This may occur in one of three patterns: a bandlike pattern of fatty replacement along the endplates, small foci of fatty marrow replacement, or larger globular areas of fat replacement [68] (Figs. 12A, 12B, 12C, and 12D). Some elderly patients and patients with malnutrition or osteoporosis may have near-complete replacement of vertebral marrow by fat [47, 48].

The bone marrow is a dynamic organ that undergoes changes both during development and, due to the extensive vascularity of bone marrow, in relatively rapid response to changes in environmental factors, including dietary changes, anemia, chronic hypoxia, chemotherapy, and other medications, through various cytokines [69, 70]. Chronic smoking may result in tissue hypoxia because of elevated carboxyhemoglobin levels [71], and long-distance running can lead to anemia from chronic mechanical hemolysis [72]. In pregnancy, there may be folate deficiency related to rapid cell turnover with development of megaloblastic anemia [73] (Fig. 13). The bone marrow of such patients may show benign hematopoietic hyperplasia with patchy areas of T1 hypointensity. Patients with sickle cell disease or other chronic diseases that stress the bone marrow will tend to reconvert the yellow marrow to red marrow or may never convert the red marrow to yellow marrow. This adaptive nature of bone marrow must be kept in mind when reviewing spinal MRI.