During the last decade, diffusion-weighted imaging (DWI) has matured from an experimental tool to a clinically useful modality that has not only significantly impacted the diagnosis of (acute) cerebral stroke but has also shown utility in other abnormalities of the brain. Although DWI should be equally sensitive to changes in the spine, it has been used far less frequently in this region of the body. This is mainly because of the inhomogeneous magnetic environment, the small size of the spinal cord, and increased motion in and around the spine. However, once these limitations are overcome, a whole range of applications can be envisioned. Already now, DWI promises to be able to differentiate between benign and malignant vertebral compression fractures. As in the brain, the immediate reduction of diffusivity following ischemic damage in the spinal cord may provide an early identification of patients with infarction. The study of diffusion anisotropy may open new avenues for the detection and better understanding of damage to the long fiber tracts with important clinical implications for disorders like multiple sclerosis and amyotrophic lateral sclerosis. It may also be possible to address, in a more refined manner, mechanisms of damage such as occur with spondylotic myelopathy. To lay the basis for future research in these areas, we will discuss the most appropriate DWI methods for the spine. Following an overview of the basic principles of DWI and associated pitfalls, the most commonly used imaging methods are addressed. Finally, experimental and clinical applications in the spinal cord and the vertebral column and their clinical relevance thus far are reviewed.



Magnetic resonance imaging (MRI) is the method of choice for the detection and diagnosis of many disorders in the spine because of its inherent sensitivity to soft tissues and its capability of displaying long segments of the vertebral column in one examination. Above all it has revolutionized the assessment of patients who suffer from spinal cord symptoms because MRI can easily serve to rule out spinal cord compression or may indicate abnormalities of the spinal cord itself. On conventional MRI, however, intramedullary lesions can frequently appear rather nonspecific, ie, it may be difficult to separate between inflammatory, neoplastic, or even vascular changes, at least without longitudinal information. MRI is also successfully used to delineate disorders of the intervertebral discs and of the osseous spine. Again, a clear separation between degenerative changes of the vertebral bodies and inflammatory or neoplastic infiltration may sometimes be problematic on conventional MRI sequences, even following the application of contrast material.


Because of its unique contrast mechanism, diffusion-weighted MRI (DWI) promises to add to the diagnostic specificity and sensitivity of MRI in the spine. Based on the ability to depict microscopic motion of water protons, DWI can be used to sensitize image contrast to microstructural changes and thus can provide important information complimentary to regular MR sequences (1,2). During the last decade, DWI has experienced a rapid evolution from a mainly experimental to a promising, clinically useful imaging modality. It was both the dramatic improvement in scanner hardware and the invention of better MRI pulse sequences that has driven the quality of DWI methodology ever further. Early on, DWI has proven to be highly sensitive in detecting cerebral ischemic changes at a time when therapeutic intervention may still be successful (3,4). In addition to stroke, DWI also has demonstrated promising results in other pathologies, such as with neoplasms (5), multiple sclerosis (6), traumatic brain injury (7), abscesses (8), epilepsy (9), as well as in brain developmental disorders or degenerative diseases such as dyslexia (10), alcoholism (11), and Alzheimer’s disease (12). With the advent of novel acquisition and reconstruction techniques, the unique diffusion properties in white matter structures make it now even possible to explore neural fiber tracts and enable researchers to study neuronal connectivity noninvasively (13–15).


While there exist hundreds of reports about DWI of the brain, the application of DWI to the vertebral column and the spinal cord has been far less common, and pulse sequences specifically designed for the spine are not widely available. To a large extent, this is certainly due to the much greater technical difficulties that one encounters in and around the spine. First, the small size of the spinal cord and adjacent structures requires smaller voxel sizes and, thus, reduces the signal-to-noise ratio (SNR). Second, bulk physiologic motion within the thoracic cave or from swallowing, as well as motion from the spinal cord itself, can cause artifacts. Third, susceptibility gradients around the vertebrae and more pronounced global field variations adjacent to the cervicothoracic junction and the lungs may cause severe distortions that heavily impair overall image quality. Nevertheless, if all these problems are addressed carefully, DWI may turn out to be a useful adjunct in the diagnostic workup of spinal disorders.


To foster this development, we would like to discus the technical challenges and solutions for DWI in the spinal cord and the vertebrae following a general overview of the DWI method. Thereafter we will explore the potential areas of application of DWI in the spine based on existing work and clinical expectations.


Back to Top




Diffusion is the random translational motion of water molecules driven by their internal thermal energy. The mean-squared displacement of diffusing water protons that occurs within a certain observation interval (diffusion time) can be quantified by the apparent diffusion coefficient (ADC). Depending on 1) the type of the underlying tissue (cerebrospinal fluid [CSF], lipids, gray or white matter), 2) the orientation and composition of its microstructure (ie, direction of white matter tracts, presence of myelin), and 3) the characteristics of the intracellular and extracellular compartments (eg, normal tissue vs. cytotoxic edema), the ADC can vary over a large range.


In DWI the image contrast is achieved by the application of pairs of strong magnetic field gradients that are either bipolar or (more commonly) unipolar gradients bracket the 180° refocusing pulse (16). For the latter, the first gradient lobe causes the protons to dephase, while the second gradient (together with the refocusing pulse) inverts the dephasing process. Hence, only those spins that have changed their position along the direction of diffusion-weighting gradient will accrue a net phase change. Since diffusion causes each proton in the spin ensemble to move randomly, also the individual phase accruals are also random; the MR signal of the spins will interfere destructively and thus the overall signal will be attenuated to a degree which depends on the diffusion coefficient of the underlying tissue and the amount of diffusion-weighting. Here, the amount of diffusion-weighting (b-value) is determined by the duration ([delta]) and the temporal separation ([DELTA]) of both diffusion-encoding gradients and their gradient strength (G).


In tissues where the diffusion coefficient is high, the destructive interference between spins is also high; therefore, the diffusion-weighted signal is low. Conversely, tissues with smaller ADC values, such as lipids or ischemic neural tissue, exhibit little diffusion-weighting. On existing MR gradient hardware, diffusion coefficients of biologic tissues require echo times between 40 and 120 ms to achieve significant signal attenuation. As a consequence, the signal intensity of diffusion-weighted images does not purely reflect differences in diffusivity. Excessively long echo times will cause a noticeable amount of T2-weighting, and differences in T2-relaxation times can therefore obfuscate the image contrast of DWI scans (“T2-shine-through” effect). To provide a quantitative measure free from the aforementioned T2-effects, the ADC is usually calculated from at least two differently diffusion-weighted images (16); for one of these images, the diffusion-weighting gradients are frequently not played out or are negligibly small just to dephase flow (b0 ~0–50 s/mm2). Because a lower SNR also decreases the precision of ADC quantification, the effective echo time of DWI scans should be generally as short as possible for a given amount of diffusion attenuation. This can be accomplished by maximizing the diffusion gradients and using fully all the time that remains available between the RF pulses and the image readout to play out diffusion encoding gradients.


As mentioned earlier, diffusion in structured tissue, such as muscle fibers or white matter, is anisotropic. In other words, the ADC is different if measured along different directions in space; therefore, the image contrast in diffusion-weighted images varies depending on the direction of the diffusion-weighting gradient. The mathematical-physical construct to characterize anisotropic diffusion is the so-called diffusion tensor, a quantity that can also be determined by MRI (17). Basically, diffusion tensor imaging (DTI) is nothing other than a set of DWI acquisitions obtained from an object where the diffusion-encoding gradients are played out along (at least) six noncolinear directions (17). The latter will provide one with (at least) six independent linear equations from which the elements of the diffusion tensor (a symmetric 3 × 3 matrix) can be computed on a per pixel basis. Using DTI, scalar measures, such as the isotropic diffusion coefficient (<D>) or fractional anisotropy (FA), can be derived which allow one to characterize the properties of the diffusion tensor (17). In addition, the magnitude and direction of the principle diffusivities of the tensor, commonly known as the eigenvalues and eigenvectors of the diffusion tensor, can be calculated. Because of densely packed myelin sheaths around axons, it is widely assumed that protons diffuse more freely along white matter fibers than perpendicular to them. Thus, the orientation of the eigenvector that corresponds to the largest eigenvalue can be assumed to be oriented in parallel with the orientation of the white matter fibers and sets the basis for DTI-based tractography (13–15).


Back to Top




In vivo imaging of the spine with diffusion techniques has various problems. The small size of the spinal cord requires the use of smaller voxels to achieve a reasonable spatial resolution. As a consequence, the maximum obtainable MR signal is usually very low. In addition, partial volume effects from the CSF, with its long T2-component, can impair lesion conspicuity, while bulk motion (ie, swallowing, breathing, and CSF pulsation) in the strong gradient field may cause large ghosting artifacts. Also, depending on the voxel size, the rapid transition between the myelon and the surrounding CSF space can cause significant Gibbs-Ringing artifacts, which can impair conspicuity. Bulk physiologic motion, however, is the major challenge for clinical DWI/DTI; any macroscopic movement in the presence of the strong diffusion-encoding gradients can lead to a significant phase accrual and intravoxel dephasing. This additional phase term is usually not known and is competing with the regular phase modulation that is created by the phase-encoding gradients during image formation. Therefore, if bulk physiologic motion (eg, brain pulsation, cord jiggling) is present, DWI (if uncorrected) is heavily distorted. In an attempt to diminish such imaging artifacts, widely known as ghosting artifacts, several DWI methods have been introduced over the past decade (16). Basically, these methods can be categorized as: 1) techniques that avoid conventional phase encoding, 2) techniques that measure the additional motion-induced phase terms and correct for these phase errors either in real time or retrospectively, and 3) single-shot techniques. Each of these three categories has certain advantages but also drawbacks, and there is still no consensus about what is the optimal diffusion-weighted sequence.


Back to Top


Echo-Planar Imaging


The most frequently used technique for DWI is still single-shot echo-planar imaging (EPI) (18). The EPI sequence is capable of acquiring an entire image within approximately 100 ms. Thus, diffusion-weighted images can be acquired very rapidly, which is very important to complete time-demanding DTI experiments within a reasonable amount of time. However, the most important property of EPI is that it acquires the entire k-space at once. Phase perturbations caused by bulk motion during the time when diffusion-encoding gradients are active are therefore equal for each line in k-space and will have no consequences if only magnitude images are used for further processing. Alternative single-shot readout strategies (19,20) have been proposed, but EPI is by far the most frequently used scheme. Even if other methods provide better image quality, the colossal number of images that are sometimes required in DTI precludes some acquisition techniques from being used within a reasonable amount of scan time. Despite EPI’s fast image acquisition, the interval between subsequent k-space profiles is long enough for off-resonant spins (eg, from chemical shift, B0-inhomogeneities, susceptibility gradients, eddy currents) to develop significant phase errors. To that end, the strong chemical shift artifacts in EPI can be avoided by fat suppression or spectrally selective excitation. Susceptibility gradients, however, can lead to signal pile-up or intra-voxel dephasing causing signal loss. Therefore, the harsh magnetic environment in and around the spine makes it extremely difficult to produce EPI images of sufficient quality. Mainly because of that reason, only a few investigators have used single-shot EPI for DWI of the spine, thus far. Furthermore, EPI’s acquisition matrices are usually limited to 128 × 128 (or at most to 128 × 256) and the T2*-decay during the whole EPI-train acts as a low pass filter on the images along the phase-encode direction. Both facts are unfortunate since they affect image resolution and can cause severely blurred images.


Back to Top


Interleaved Echo-Planar Imaging


Recently, it has been shown that interleaved echo-planar imaging (IEPI) (21,22) is a promising alternative to single-shot EPI for DWI (Fig. 1). By means of proper k-space segmentation, IEPI allows a tradeoff between fast image acquisition and the creation of artifacts from off-resonant spins. In contrast to regular EPI, IEPI uses multiple interleaves to fill the entire k-space. In the case of n interleaves, the spacing between adjacent lines in k-space within one interleaf is n-times larger than it is for a single-shot EPI scan. Therefore, the k-space trajectory of one such interleaf traverses through k-space n-times faster and, thus, reduces the phase errors due to off-resonant spins by the same amount. However, by leaving the single-shot regimen, random motion-induced phase-fluctuations between interleaves can perturb the phase-encoding and lead to considerable image ghosts. Therefore, diffusion-weighted IEPI usually requires phase-navigation (23). Phase-navigation, sometimes referred to as “Navigator Echo(es),” measures the contribution of phase errors due to physiologic motion, which then can be corrected either during the data acquisition (24) or retrospectively during image reconstruction (21–23).

GraphicFigure 1



Despite the fact that IEPI can significantly reduce distortions from off-resonant spins, water-fat shift artifacts can still be considerable and fat suppression is therefore required. Yet field inhomogeneities at the cervicothoracic junction (Fig. 2) can cause incomplete (frequency-selective) fat saturation and residual chemical shift artifacts (Fig. 3). In addition, substrates that contain both water- and lipid-bound protons can cause errors in ADC quantification if fat suppression is incomplete (both have significantly different diffusion coefficients). Alternatively, spectral-spatial excitation pulses (ie, water-only excitation), which are less sensitive to field inhomogeneities, can be used. Since these pulses do not require a spectral inversion time, more slices per TR can be acquired, although the duration of these RF pulses is longer than that of regular slice-selective RF pulses. Generally, one disadvantage of spectral fat suppression methods is, however, that lipid resonances also occur at the water resonance peak and, as mentioned above, should be considered for exact quantification of the ADC. Such shortcomings might be ameliorated by instead using short-tau inversion recovery (STIR) fat suppression techniques. However, the latter usually suffers from a strongly reduced SNR. Even if the signal from lipid-bound protons is suppressed, the presence of these molecules can significantly increase the spin tortuousity and, thus, reduce the apparent diffusion coefficient.

GraphicFigure 2

GraphicFigure 3


Diffusion imaging also has the potential to play an important role in traumatic spinal cord injuries. However, in such injuries a hemorrhagic component is likely to exist, causing additional susceptibility gradients. If and to what extent IEPI can then be used for such injuries is not yet fully clear, since IEPI sequences demonstrate some degree of T2*-weighting. Likewise, small susceptibility gradients in certain areas of the spine could impair lesion conspicuity. In contrast to single-shot EPI, the authors’ experience with IEPI is that with a sufficient number of interleaves (ie, to increase the bandwidth per pixel along the phase-encoded direction) severe susceptibility artifacts occur only seldom and usually do not impair the delineation of the spinal cord. This finding can be explained by the relatively small susceptibility difference between bone and tissue (-8.86 vs. -9.05 ppm cm-3) as compared with air (0.0 ppm cm-3). Nevertheless, slight distortions in IEPI series can sometimes be observed, such as in cases of osseous spinal stenosis.


Notice that more recently developed parallel imaging techniques (25), together with EPI, allow one to traverse k-space much faster, similar to IEPI, but without the need for phase navigation (26,27).


Back to Top


Fast Spin Echo


Diffusion-weighted fast spin echo (FSE) may constitute another alternative to image areas located in an inhomogeneous magnetic environment. Due to its train of RF refocusing pulses, FSE is less susceptible to field inhomogeneities and chemical shift artifacts. For this reason, FSE does not necessarily require extra fat suppression, although it might potentially be useful in imaging the vertebrae. Conversely, it is known that artifacts may occur in FSE if the Carr-Purcell-Meiboom-Gill (CPMG) condition is not met, ie, when the net transverse magnetization is not perfectly in phase with subsequent RF pulses in the FSE-sequence (28). This is a situation that is most likely to happen in DWI due to bulk physiologic motion. Regardless of whether a single- or multi-shot readout is chosen, the violation of the CPMG condition causes destructive interference between spin and stimulated echoes in the FSE train and, hence, leads to unstable echoes and phase fluctuations (29,20). The same problem would also apply to combined spin- and gradient echo acquisitions. In this context, just recently a few methodological improvements were proposed to overcome some of these problems (31,32). In a recent study (33), the authors found that navigated diffusion-weighted IEPI yielded better results in spinal cord imaging than navigated diffusion-weighted multi-shot FSE (Fig. 4 and Table 1).

GraphicTable 1

GraphicFigure 4


One very elegant method has recently been published that solves the signal instability problems of diffusion-weighted FSE methods by varying the phase of the refocusing pulses and combining it with advances in radial scanning (34) (Fig. 5). This work is based on the PROPELLER (Periodically Rotated Overlapping ParallEL Lines with Enhanced Reconstruction) method: by sampling rotating strips (blades) that cover the center of k-space, this method inherently includes in each FSE train some two-dimensional navigator information. For each FSE train, the collected echoes are phase-encoded to form a sufficiently sampled blade, which goes through the center of k-space. The width of the blade in k-space is equal to the echo train length divided by the field of view. The basic idea of PROPELLER is that the oversampled region in the center of k-space can be compared between blades to correct for inconsistencies prior to combining the data (ie, phase navigation). For diffusion applications, the primary inconsistency will be image-space phase differences. After data correction, the blades are weighted to correct for sampling density variation as well as uncorrected errors, then gridded onto a Cartesian array (35), and Fourier-transformed to form the image. Generally, PROPELLER-FSE-DWI uses FSE data collection, which provides far greater immunity to geometric distortion than that obtained with EPI sequences. Note that, as with other radial techniques, off-resonant spins cause radial blurring while geometric distortions are generally more benign.

GraphicFigure 5



Back to Top


Diffusion-Weighted Line Scan Imaging (LSDI)


Line scan imaging (36) is especially attractive for DWI because no phase encoding steps are necessary to form an image. Hence, for magnitude-only reconstruction, additional phase terms due to bulk motion are of no consequence. Moreover, LSDI is not as demanding in terms of hardware performance as other methods, such as EPI.


In LSDI, each column is formed by the intersection of two planes selected by two-slice selective RF pulses at different angles relative to the desired image plane; only those spins that have seen both RF pulses will refocus. A standard frequency-encoding readout and Fourier transform are then performed along the selected column to obtain the spatial information along this line. Shifting the resultant cross-sectional area (defined by the intersecting pulses) within the image plane allows one to scan the image line by line. Assembling adjacent lines ultimately yields the final image. The width of the selected column determines the in-plane resolution, while the height of the cross section defines the slice thickness of the image.


LSDI is a very reliable method and has been demonstrated to be very useful for imaging the spinal cord (37,38) and the spine (39) (Fig. 6). However, LSDI is partly limited by its longer scan times and lower SNR. The latter is due to the fact that only a single line contributes to each spin echo as opposed to all the spins from the whole slice, as in regular 2D spin warp imaging.

Figure 6



Back to Top


Steady-State Free Precession (SSFP) Sequences


A condition, called steady-state free precession (SSFP), will develop if a train of equidistant RF pulses with flip angle [alpha] and TR < T2 are played out by an MR pulse sequence. To that end, rapid image formation is possible with SSFP because of the very short TR time. SSFP imaging has long been known for its high sensitivity to flow and diffusion in the presence of magnetic field gradients (40,41). However, the signal formation in SSFP is a complex interplay between numerous spin and stimulated echoes which may be formed through a multitude of coherence pathways (29,30,42) and are limited only by the natural decay times T1 and T2. The complex signal formation therefore renders the quantification of diffusion very difficult. In particular, the diffusion attenuation (b-factor) may vary from tissue to tissue because the b-factor in SSFP sequences is also determined by parameters like relaxation times and B1-uniformity. Hence, one therefore has to deal not only with “T2-shine-trough” effects but also with the fact that b-values are weighted by the underlying relaxation times and other confounding factors.


Back to Top




Because of its high vascularity, the spine is a common target site for metastatic spread (43). Hence, vertebral metastases are frequently observed in patients with cancer (30%-70%) and, besides pain, may lead to various secondary problems, ranging from fractures to compression of the dural sac by epidural masses. MRI has evolved into one of the preferred methods for detecting vertebral metastasis, mainly because MRI rapidly detects when normal bone marrow is replaced by more hypercellular tissue or by other pathologies that cause a higher water content (44,45). However, with conventional imaging, such as T1- and T2-weighted (fast) spin echo, or short tau inversion recovery sequences, it is often difficult to discriminate whether the cause of an acute vertebral compression fracture is osteoporosis or metastasis (46,47). Morphologic signs, such as complete replacement of vertebral marrow, involvement of the posterior elements, and epidural or paraspinal masses, can be used to improve the diagnostic accuracy but may be equivocal.


In this context, results from recent studies raise hope that DWI might be able to differentiate benign from malignant acute vertebral fractures (48–52). It has been reasoned that proton diffusivity is elevated in osteoporotic fractures because of bone marrow edema (48). Conversely, metastatic lesions might change diffusivity only moderately or even decrease it. Zhou et al. (52) postulated that a high cellularity of metastatic lesions, especially of actively growing tumors, would increase their intracellular volume fraction relative to the interstitial space. Because the water diffusion coefficient is approximately 10 times lower in the intracellular space than that in the extracellular space, this should lower the ADC values of metastatic vertebral infiltration.


Along these lines, Baur et al. (48) reported that osteoporotic fractures in 22 patients generally showed hypointense or isointense signal on diffusion-weighted SSFP sequences relative to adjacent unaffected bone marrow. On the other hand, they found that metastatic fractures in 17 patients appeared hyperintense on SSFP DWI. In a following study, the same group found signal hyperintensities on DWI scans in all 25 malignant vertebral fractures, while 35 of 38 benign fractures appeared hypointense or isointense (49). Three of the 38 benign lesions, however, appeared false positive, showing hyperintense signal on DWI scans. Similar specificities of DWI regarding the separation of benign and pathologic vertebral fractures were reported by Tasaly et al. (53) and Matoba et al. (54), although in slightly different patient populations. Conversely, in a consecutive series of 15 patients with proven malignant vertebral lesions Castillo et al. (55) found that these lesions appeared either hypointense (n = 5), hyperintense (n = 3), or isointense (n = 2) and concluded that SSFP DWI does not offer any advantage over conventional MRI. In a recent review article, Baur et al. (56) pointed out that the patients enrolled in the Castillo study are not the primary target group for DWI in spine (ie, patients with sclerotic metastases and previously treated metastases). Furthermore, they suggest the following inclusion criteria: 1) unknown reason for the vertebral collapse, 2) lack of sclerotic metastases, and 3) no prior therapy.


Unfortunately, all the aforementioned studies have been performed without absolute quantification of the diffusion coefficient. Hence, the diagnosis was based solely on DWI examinations where the diffusion information was obscured by concurring effects on the MR contrast, such as T2-shine-through or T2*-effects. As mentioned earlier, SSFP diffusion sequences lack the ability to absolutely quantify diffusion. This is mainly because the final echo formation of SSFP comprises different echo components and the magnitude of these different contributions strongly depends on the underlying relaxation and sequence parameters (eg, TR, TE, and flip angle).


More recently, reports from (semi)-quantitative diffusion measurements have been reported: Spuentrup et al. (51) analyzed 35 lesions (18 acute osteoporotic or traumatic fractures and 17 untreated malignant infiltrations) and found significant differences between malignant lesions with and without fractures in DWI scans that were normalized to unweighted images. Moreover, significant relative signal loss was also found for osteoporotic and traumatic fractures, whereas only small changes were found in metastatic lesions. Quantitative measurements by Zhou et al. (52) have demonstrated that acute benign vertebral fractures (n = 12) have higher diffusivity than malignant lesions (n = 15). However, they also showed that both groups overlap considerably. DWI of the benign fractures demonstrated hypointense (n = 1), isointense (n = 5), and hyperintense lesions (n = 6). The isotropic diffusion coefficient of the group ranged from 240 to 350 × 10-6 mm2/s and was slightly lower than values of normal vertebral bodies (270-330 × 10-6mm2/s) in the same subjects. On the other hand, DWI of the metastatic vertebral lesions demonstrated hypointense (n = 4), isointense (n = 1), hyperintense (n = 9), and heterogeneous (n = 1) lesions. The isotropic diffusion coefficient of this group ranged from 130 to 200 × 10-6 mm2/s and was therefore considerably lower than values of normal vertebral bodies in the same subjects (270-330 × 10-6 mm2/s). Zhou et al. (52) concluded from this that DWI intensity values alone were highly unspecific but that measurement of the diffusion coefficient improved diagnostic specificity. Recently, Herneth et al. (57) also quantified ADC values in 22 patients with vertebral compression fractures and showed that vertebral metastases had ADC values that were significantly lower than normal vertebrae (690 ± 240 vs. 1660 ± 380 × 10-6 mm2/s). Also, pathologic compression fractures had significantly lower ADC than benign compression fractures (710 ± 270 vs. 1610 ± 370 × 10-6 mm2/s). The absolute ADC numbers, however, differed significantly from the results observed by Zhou et al. (52). One possible explanation for this effect is that Herneth et al. (57) used fat suppression for their IEPI acquisition whereas the other group did not.


Quantitative measurements in vertebral bodies of normal volunteers have revealed a volume fraction of fat ranging from 23.9% to 54.2% with an increasing fractional size for older subjects (58). When lipids are suppressed, for typical echo times in DWI, the MR signal from the remaining water can be extremely weak. Consequently, the low SNR can limit the precision of diffusion measurements and the noise characteristics can bias measurement accuracy (ie, non-zero mean noise). Even more important is the fact that water- and lipid-bound protons have significantly different diffusion coefficients. Thus, in a recent study using LSDI, we have seen a clear bi-exponential diffusion decay if no fat suppression is used (Figs. 7, 8) (39). The mean values for the slow and fast mean diffusivity found in vertebral bodies, which most likely represent the ADCs of lipids and water, were 65 ± 21 × 10-6 mm2/s and 1551 ± 297 × 10-6 mm2/s. Here, the slowly diffusing fraction was 89.9 ± 1.97%. Whether a shift between compartments (ie, the replacement of lipid tissue by free water) also adds to diagnostic specificity has not yet been addressed. Irrespective of that, the significantly different diffusion coefficients of lipids and free water make it immediately evident that without fat suppression the quantified ADC is strongly dependent upon the selected b-value if a mono-exponential diffusion model is erroneously chosen. Also, in a preliminary patient study, we were very satisfied with the excellent image quality and robustness of the LSDI approach (Figs. 8, 9).

GraphicFigure 7

GraphicFigure 8

GraphicFigure 9

Besides attempts to increase diagnostic specificity, a recent study has proposed that DWI can also improve the monitoring of therapy for metastatic vertebral spread. In general, both radiation therapy and chemotherapy are used to destroy neoplastic cells or at least stop their growth to provide relief from pain and to preserve bone stability (59). Just recently, vertebroplasty treatment, using substances such as polymethylmethacrylate (PMMA) (Fig. 10), has become increasingly popular. The mechanism responsible for the palliative effect of vertebroplasty is unknown, but it may partly be the result of neural damage that is caused by heat released during polymerization of the bone cement. Of course, in areas treated with PMMA, MRI measurements are no longer meaningful as the water is replaced by bone cement. With other treatments, however, a change in ADC of infiltrated areas may reflect therapeutic effects that are otherwise difficult to establish in an objective manner (59). In this context, Byun et al. (59) demonstrated that DWI shows decreasing signal intensity of metastatic disease of the vertebral bone marrow (23 of 24 patients) with successful therapy. Unfortunately, these results were compared against only one patient with treatment failure. Thus, DWI as an adjunct in monitoring treatment response appears potentially promising, but studies of larger patient groups and more quantitative results are certainly still necessary for more solid conclusions.

Figure 10



Back to Top




As in the brain, DWI of the spinal cord could contribute both to the diagnostic process, such as for delineating acute spinal cord ischemia, and to the pathophysiologic understanding of many disorders. Damage to the spinal cord may be caused by a wide range of pathologies and can result in profound functional disability. Characterization of the structural integrity of the spinal cord can be assessed using diffusion tensor imaging methods. Development and application of this technique may improve our understanding of the nature and evolution of structural damage in various types of spinal cord disease. Possible developments include improved detection of ischemic lesions, clarification of the relationship between clinical disability and structural damage to the cord (including degenerative disorders), and monitoring of anti-inflammatory or neuroprotective therapies.


Quantitative diffusion measurements in healthy volunteers confirmed the assumption that diffusion coefficients in the spinal cord are comparable to those of the brain and demonstrate diffusion anisotropy (33,60–63). The authors found in their own studies that the ratio between diffusion along the fibers and across them differed by at least a factor of two (Table 1). Similar findings were observed by other researchers. Using diffusion anisotropy, it is possible to visualize different fiber tracts, such as the corticospinal tract or transverse bundles in the pontine region, very well (Fig. 1).


The small size of the spinal cord and the adjacent CSF space, however, makes it sometimes difficult to quantify diffusion and to distinguish between gray and white matter. Partial volume effects have to be considered and can affect measurements, while the rapid transition from CSF space to the spinal cord can cause ringing artifacts which, if uncorrected, can falsify DWI further (62). Anisotropic diffusion is characterized most accurately by DTI. In a recent study using navigated diffusion-weighted IEPI, Riess et al. (62) were the first to demonstrate DTI in the human spinal cord. However, due to the aforementioned problems, spinal DTI appears much more difficult than conventional DWI. Hence, healthy skepticism is appropriate in assessing new results from DTI of the spine.


Back to Top


Spinal Cord Infarction


Although the exact pathomechanism is still not fully clear, DWI has demonstrated high sensitivity to early ischemic changes in acute cerebral stroke (3,4). Within a few minutes after the onset of stroke, a reduction of the ADC can be observed. Therefore, DWI might be equally well suited to detect spinal cord infarction. With conventional MR sequences, it may take days to observe intramedullary signal changes following spinal cord ischemia. Even then it is often hard to discriminate such changes from those caused by other etiologies such as myelitis. Such information, however, would appear mandatory to probe more aggressive therapies such as rtPA treatment. Thus far, the authors (61) and three other reports (64–66) have shown findings of altered diffusion in spinal cord infarction (Figs. 10, 11). Using diffusion-weighted EPI, Gass et al. (64) reported significantly reduced ADC values 30 hours after onset of symptoms (490 × 10-6 mm2/s), whereas at day 11 the ADC was found to be higher than normal (950 × 10-6 mm2/s). Similar results were observed by Stepper and Loevblad (65), although their absolute ADC measurements (acute lesion ADC = 900 × 10-6 mm2/s, adjacent healthy cord = 1200 × 10-6 mm2/s, and lesion at follow-up after 5 days = 1000 × 10-6 mm2/s) were higher than those of Gass et al. (64) and comparable to values for ischemic brain parenchyma (3). All of the studies that have shown altered diffusion in spinal cord infarction have investigated no more than two patients. This reflects existing difficulties in applying DWI to the spinal cord in the routine clinical setting. Therefore, conclusions about the exact sensitivity and specificity of DWI for spinal cord infarction can not yet been drawn.

Figure 11



Back to Top




While DWI of the human spine in vivo is still in its infancy, DWI of animal spinal cords has already been performed for a longer period and has focused on spinal cord injury. This is because spinal cord trauma is the most frequent cause of acute paraplegia or tetraplegia. In the United States, approximately 20,000 patients per year will suffer from paralyzing spinal cord injury (67). Traumatic injury may result in cellular swelling and degeneration as well as in the disruption of myelin membranes or even more severe damage. To this end, the initial cord injury and axonal transection only in part cause the functional deficits that ultimately occur. Increased functional loss is related to “secondary injury,” an immune response, which can persist for several days and results in increased lesional size, swelling, and ultimately the additional degeneration of axonal fiber tracts (68).


The exact stage of traumatic injury is often difficult to characterize by conventional MRI as it is the ultimate goal of testing for the functional integrity of the axons within the white matter tracts of the spinal cord. Similarly, conventional MRI cannot detect possible therapeutic responses to neuroprotective drugs. Only histologic examinations can precisely reveal the severity of tissue damage or recovery. In this context, Wallerian degeneration above and below the site of injury is known to be indicative of axonal loss. However, Wallerian degeneration as seen by conventional MRI occurs only with advanced progression of tissue damage and, furthermore, it is not differentiable from edema; diffusion-related parameters obtained from DWI might be better able to define the type and extent of spinal cord injury than conventional MRI, as different pathophysiologies may affect diffusion properties differently.


It is postulated that spinal injury causes intracellular swelling and increased permeability as the myelin sheaths degenerate. In experimental animal models of spinal cord injury, the efficacy of DWI has been studied and a decrease of longitudinal ADC and an increase of transverse ADC was observed (69). In so-called weight drop experiments in rodents, it has been shown that diffusion anisotropy is a very sensitive marker for damage to the spinal cord. Moreover, reduced anisotropy loss was observed if neuroprotective therapy was applied. Similarly, another experimental study reported that the diffusion anisotropy in the spinal cord was higher if neuroprotective drugs had been administered than without (70). This outcome was explained by the possible enhancement of myelin survival and the diminished occurrence of cysts, which have increased isotropic diffusivity.


A spinal trauma can be complicated further if syringomyelia develops. This is a cystic cavitation within the center of the spinal cord (Fig. 11), and it has been shown in animal models that changes can be seen on ADC maps already after 1 week, while conventional MRI was first positive 4 weeks after the injury (71).


Recently, in a case of acute spinal cord injury with type II odontoid fracture, it has also been shown that DWI is sensitive to traumatic lesions in vivo (72). Within this lesion, the isotropic diffusion-coefficient dropped to 660 × 10-6 mm2/s, while the normal brainstem showed an isotropic diffusion coefficient of 890 × 10-6mm2/s.


Overall, the animal research clearly demonstrates that DWI is more sensitive for the evaluation of spinal cord injury and the outcome after neuprotection and regeneration. Nevertheless, successful application to humans is still lacking. The potential drawbacks for DWI in the presence of hemorrhagic components of traumatic tissue damage have been discussed earlier.


Back to Top


Multiple Sclerosis


Because of the dense arrangement of long fiber tracts to and from the extremities within the spinal cord, focal lesions to this part of the central nervous system tend to have great clinical impact (73). Likewise, remote fiber changes, resulting from multifocal supratentorial diseases or from motor neuron disorders, concentrate in the spinal cord. This has led to speculations that disability, such as from multiple sclerosis (MS), may correlate best with spinal cord abnormalities (74). The author’s have found that MS lesions usually present increased rates of diffusivity, which was in agreement with findings by Clark et al. (75). They found that the isotropic diffusion coefficient was significantly higher than in a group of healthy controls (1180 ± 120 × 10-6 mm2/s [n = 4] vs. 910 ± 50 × 10-6 mm2/s [n = 3]), whereas differences in diffusion anisotropy, although present, did not reach statistical significance. However, some phenotypes of MS, eg, primary progressive MS, can develop profound clinical disability with very little lesion load and, therefore, speculations have been raised that MS is accompanied by a diffuse disease component. In this context, DWI and especially DTI have been shown to be sensitive to diffuse changes in normal-appearing white matter of the brain (6). Although the assumptions of a diffuse disease component to MS are supported by reports of diffuse signal changes of the spinal cord on conventional MRI in secondary progressive MS (74), there exist no comprehensive DWI studies in this setting thus far.


Back to Top




In myelitis, acute inflammation affects gray and white matter over one or more adjacent vertebral segments. Often the cause is unknown, but some cases follow nonspecific viral infection or vaccination, suggesting an immunologic cause; others have been associated with Lyme disease, syphilis, TB, or parasitic or fungal agents. As noted earlier, DWI might be potentially helpful to differentiate myelitis from other disorders, especially spinal cord infarction and neoplasms. In two studies of the authors, cases of myelitis that underwent DWI in the acute phase uniformly demonstrated a moderately elevated ADC suggestive of edema in parallel to swelling and signal hyperintensity on T2-weighted scans. At follow-up, a few months later, the swelling and ADC changes were no longer present (Fig. 12).

GraphicFigure 12



Back to Top


Amyotrophic Lateral Sclerosis


Conventional MRI demonstrates only poor specificity for amyotrophic lateral sclerosis (ALS), a chronic degenerative disorder of unknown etiology that affects the first and second motor neurons. This might be improved by DWI because of its unique sensitivity for the dissolution of neuronal tracts. Also, with several new therapeutic agents on the horizon, effective and objective disease markers for diagnosis and surrogate outcome measures in clinical trials would be crucial. Recently, Ellis et al. (76) showed that isotropic diffusion of the internal capsule was significantly increased, whereas fractional anisotropy was significantly decreased in patients with ALS relative to healthy controls. Since ALS affects the whole corticospinal tract, similar changes in diffusion properties might also be found in the spinal cord.


Back to Top


Spondylotic Myelopathy


Degenerative spondylosis may cause significant narrowing of the spinal canal and ultimately can lead to compression of the spinal cord. Compromise of the spinal cord can also result from protruded or herniated discs, especially in the case of a narrow spinal canal. Clinically, such patients usually present with chronic progressive signs of myelopathy. Early diagnosis is of paramount importance to prevent severe disability. The authors found in one of their studies (61) that spondylotic myelopathy can present with reduced ADC values, whereas the surrounding cord demonstrated elevated diffusivity. The former is presumably either due to cord compression or due to vascular compromise while the latter is due to surrounding edema (Fig. 13). Ries et al. (62) recently showed a patient suffering from severe spinal stenosis with cord compression. The spatial extent of diffusion abnormality was much larger than the area seen on conventional MRI. In this case, DWI scans appeared hypointense at the level of the stenosis and quantitative measurements indicated significantly increased ADC values. However, it is not clear if and to what extent susceptibility differences (caused by protruded discs or osteophytes that moved toward the anterior aspect of the cord may) have also led to strong signal dephasing in this particular patient.

Figure 13



Back to Top




Current MR imaging sequences using contrast material are already quite effective for the correct diagnosis of different mass lesions. Because of the altered cellular matrix of neoplastic tissue that is picked up by DWI, this technique may add to the staging of tumors or at least could help to differentiate different types of mass lesions. Promising results have been shown for the brain, where researchers reported DWI’s ability to differentiate between cerebral tumor types (77) and similar results might be anticipated for the spine. In one patient suffering from an astrocytoma in the cervical cord (Fig. 14), for example, we found that the lesion had a significantly elevated ADC. However, in high-grade, heterogeneous tumors, such as glioblastoma multiforme or high-grade astrocytomas, ADC values can vary over a large range (78,79) and a general differentiation based on ADC could be difficult. Another approach is to use the orientational information obtained from DTI to perform fiber tracking (13–15) and, thus, perhaps shed light onto whether a tumor invades or displaces fiber tracts. The latter might have consequences for the planning of surgical intervention. However, to our knowledge, no peer-reviewed study has been published that explores specifically the utility of DTI in spinal tumors, thus far.

GraphicFigure 14



Back to Top




When reviewing the DWI techniques that have been applied to the spine so far, the inherent problems and limitations become readily apparent. However, with constant advances in scanner hardware and pulse-sequence design, DWI of this region of the body has already entered clinical routine, or will at least soon do so. A number of extremely useful applications can be envisioned from existing experimental data and first clinical experience. A greater body of data will be needed, however, to define the true utility of DWI of the spine. Hopefully this review will stimulate efforts in this direction.

This entry was posted in Uncategorized. Bookmark the permalink.



WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / 變更 )

Twitter picture

You are commenting using your Twitter account. Log Out / 變更 )


You are commenting using your Facebook account. Log Out / 變更 )

Google+ photo

You are commenting using your Google+ account. Log Out / 變更 )

連結到 %s