|Home | About | Journals | Submit | Contact Us | Français|
T1 and T2 were measured for white matter (WM) and gray matter (GM) in the human cervical spinal cord at 3T. T1 values were calculated using an inversion-recovery (IR) and B1-corrected double flip angle gradient echo (GRE) and show significant differences (p = 0.002) between WM (IR = 876 ± 27 ms, GRE = 838 ± 54 ms) and GM (IR = 973 ± 33 ms, GRE = 994 ± 54 ms). IR showed significant difference between lateral and dorsal column WM (863 ± 23 ms and 899 ± 18 ms, respectively, p = 0.01) but GRE did not (p = 0.40). There was no significant difference (p = 0.31) in T2 between WM (73 ± 6 ms) and GM (76 ± 3 ms) or between lateral and dorsal columns (lateral: 73 ± 6 ms, dorsal: 72 ± 7 ms, p = 0.59). WM relaxation times were similar to brain structures with very dense fiber packing (e.g., corpus callosum), while GM values resembled deep GM in brain. Optimized sequence parameters for maximal contrast between WM and GM, and between WM and cerebrospinal fluid (CSF) were derived. Since the spinal cord has rostral-caudal symmetry, we expect these findings to be applicable to the whole cord.
Image contrast in conventional MRI relies on the distinct relaxation behavior of water spins residing in different tissue environments. Quantitative determination of the relaxation time constants is important for the derivation of experimental parameters that optimize image contrast. Furthermore, understanding the nature of relaxivity in different tissues facilitates the development of new imaging methods. As the use of higher field whole body MRI systems (i.e., >1.5T) is becoming more widespread, it should be recognized that tissue relaxation rates are field-dependent and that experimental parameters must be re-optimized to take full advantage of the benefits of higher field strength. In vivo human tissue relaxation parameters have recently been measured in the brain (1,2) and in blood (3) at 3T, but to our knowledge no studies of the human spinal cord have been reported at any clinical field strength. The small size and mobile nature of the spinal cord hamper quantitative measurements, and it has been necessary to assume that spinal cord white matter (WM) and gray matter (GM) relaxation rates will mimic those in the brain, in spite of histological indications that spinal cord tissues differ from brain tissue (4). The same difficulties that have deterred measurement of relaxation behavior have also slowed the development of spinal cord imaging in general (5,6). Recently, a number of techniques for high-resolution imaging have been applied to the spinal cord (7–9), yielding important clinical information about several pathologies (10,11), most notably multiple sclerosis (MS). Further development (and thereby, widespread adoption) of these methodologies may be facilitated by quantitative measures of the relaxation times, as they allow optimization of imaging parameters, potentially yielding improvements in sensitivity and contrast.
Several other approaches require knowledge of water relaxation times. For example, in the field of in vivo spectroscopy, it is necessary to quantify metabolite concentrations from signal intensities that are functions of concentration, relaxation rates and experimental parameters. Since the intensity of the unsuppressed water signal is often used as an internal standard (12), accurate quantification of the relaxation parameters of water is critical. In the case of magnetization transfer imaging, knowledge of the relaxation times is important for the quantification of magnetization transfer effects which can provide parameters that reflect macromolecular interactions with the water signal (e.g., bound pool fraction, exchange rate) (13,14), as well as the optimization of imaging sequences at higher field strength (15)
In this work, T1 and T2 of both GM and WM are reported in normal human spinal cord at the cervical level for dorsal column WM, bilateral lateral column WM and dorso-lateral horn GM at the level of C3. T1 values were measured by two methods for comparison. B1 field mapping of the transmit inhomogeneity was also examined as it may affect measurements made in the spinal cord. Finally, simulations are performed to demonstrate optimal imaging parameters for conventional spinal cord imaging in the clinic.
A total of six healthy volunteers (two male, four female; mean age = 30.8 ± 6.4 years) were enrolled in the study, which was approved by the local institutional review board. Signed, informed consent was obtained prior to examination. All scans were performed on a Philips Intera 3T (Philips Medical Systems, Best, The Netherlands) MRI system with body coil excitation and a 16-channel neurovascular coil (eight head elements, two bilateral c-spine elements, six anterior-posterior lower c-spine/upper thoracic elements) for reception. All scans were centered at the middle aspect of the C3 vertebral body.
Longitudinal relaxation time constants were measured via inversion-recovery (IR) as well as double flip angle gradient echo (GRE) methods. The standard method for quantifying T1 in the brain is by using an IR sequence with multiple inversion times (TIs) and a similar methodology was used in the spinal cord. A single 5-mm slice centered at C3 was acquired at 14 TIs using an IR preparation and multishot EPI (EPI factor = 5) readout (TR/TE = 4000 ms/12 ms). Data were acquired at 14 TIs (100, 200, 300, 400, 500, 600, 700, 800, 900, 1000, 1500, 2000, 2500, and 3000 ms). The nominal resolution was 1 mm × 1 mm (FOV = 212 mm) and reconstructed to 0.5 mm × 0.5 mm. Other parameters were: sensitivity encoding (SENSE) factor = 2, second order shimming, and a total scan time per TI = 1.5 min.
All IR based images were coregistered with a three degrees-of-freedom model using FLIRT (FMRIB's Linear Image Registration Tool, Oxford, UK). The longitudinal relaxation time, T1, was calculated from a monoexponential fit to the signal recovery curve for each voxel using a least-squares calculation routine supplied in Matlab (The Mathworks, Natick, MA, USA).
In the spinal cord, due to its mobility, diminutive size of its structures and the requirement of high-resolution, multiple TI acquisitions can be long and prohibitive for routine multislice imaging. Therefore, we also employed a second method for characterizing the longitudinal relaxation time using a double flip angle approach (16), in which two T1-weighted volumes were obtained with excitation flip angles of 15° and 60°. Each T1-weighted volume was acquired using a three-dimensional spoiled GRE (TR/TE = 100 ms/10 ms), consisting of 10 axial, 4-mm slices centered at C3. The nominal in-plane resolution was 1.0 mm × 1.0 mm (FOV = 212 mm × 212 mm) reconstructed to 0.5 mm × 0.5 mm. Other parameters were as follows: EPI factor = 5, SENSE factor = 2.0, 1 k-space average, second order shimming, and total scan time = 28 s per volume. The second T1-weighted volume (α2 = 60°) was coregistered to the first volume (α1 = 15°) using a six degree-of-freedom, rigid-body transformation algorithm supplied by automated image registration (AIR) (17).
At higher field, a double (or multiple) flip angle approach is hindered by transmit field inhomogeneity (B1) such that the prescribed flip angles are not accurately performed in the tissue of interest, and leading to spurious calculated values of T1. Furthermore, in the spinal cord, transmit variability could be large due to the proximity large bones and flowing cerebrospinal fluid (CSF). Thus, in order to use a double flip angle method for accurate T1 quantification, it is necessary to correct the signal intensities using a B1 mapping experiment. While several methods exist (18–20), we utilized the actual flip angle imaging (AFI) technique (20). Briefly, a three-dimensional (3D) GRE sequence is obtained with a double excitation, each of which is followed by its own GRE readout. Given a prescribed flip angle, αnom, the B1 correction factor (Bcorr) is then calculated as the ratio between the realized (i.e., calculated) and prescribed flip angles. If B1 is known, then the corrected T1 can be calculated from a double flip angle experiment.
Two volumes (identical to the T1-weighted GRE acquisitions) were acquired, one immediately after each of the two excitation flip angles of 60°). The time between the first and second excitation (TR1) was 30 ms and the time between the second excitation and repeat of the first excitation (TR2) was 100 ms. No EPI or SENSE was used, and TE = 2.0 ms. Second order shimming was performed to match the B1 map to the same conditions as the double flip angle approach. The total scan time to acquire data necessary for the actual flip angle map was 1.5 min. The actual (i.e., realized) flip angle was calculated as:
where S1 and S2 are the signal intensities obtained after TR1 and TR2, respectively. Then Bcorr = /αnom. The corrected T1 was then calculated voxel by voxel according to:
where S(α1) and S(α2) are the signal intensities at the first and second flip angle (where α1 < α2), respectively (20).
Transverse relaxation times were measured using a 16-echo spin-echo sequence (TE = 10–160 ms). To account for imperfections in the 180° refocusing pulse, off-resonance effects, and to eliminate possible stimulated echo contributions, only the eight even echoes were used for exponential curve-fitting (20–160 ms) (21). A single 5-mm slice centered at C3 was acquired, with nominal in-plane resolution of 1.0 mm × 1.0 mm (FOV = 212 mm × 212 mm). One k-space average was acquired with SENSE reduction factor = 2 and the data were reconstructed to a resolution of 0.5 mm × 0.5 mm.
To eliminate CSF in-flow effects associated with cardiac pulsation in the longer echo T2 acquisitions, the sequence was triggered using a peripheral pulse-oximeter placed on the right index finger of each volunteer, at a time point 10 ms into systole. The minimum TR was set to 2.5 s, corresponding to three cycles of a 70-bpm heartbeat, and each scan was started at the next trigger point thereafter. As this TR is much shorter than 5*T1, the signal intensities in the individual scans may vary slightly depending on the heartbeat. However, the decay rate in each scan will be approximately the same and thus reflect an accurate assessment of the measured T2. The total scan time for each data set ranged from 4.5 to 5 min depending on the actual pulse rate of the volunteer. The transverse relaxation time constant T2 was extracted from a monoexponential fit to the signal decay curve for each voxel using a least-squares calculation routine supplied in Matlab (The Mathworks, Natick, MA, USA).
For each of the volunteers, four regions of interest (ROI) were analyzed at the level of C3 (Fig. 1a and b): left and right lateral column WM, dorsal column GM, and dorso-lateral horn GM. For IR-based T1 calculations, ROIs were placed on the TI = 100 ms (Fig. 2a) acquisition as the contrast was sufficient for structure determination. Each ROI was then propagated to the absolute T1 maps. For the B1-corrected double flip angle calculation of T1, ROIs were placed manually on the T1-weighted (α = 15°) image and propagated to the absolute T1 maps. Due to local susceptibility artifacts and mismatch between spin echo (SE) and GRE acquisitions, ROIs for T2 calculation were manually placed on the first echo (in a manner similar to Fig. 1b) as the contrast was sufficient to visualize each column and the GM horns. The mean and SD of each T1 acquisition and T2 were calculated for each ROI and over all volunteers. To test the fitting quality of this voxel-by-voxel approach in both T1 IR and T2 calculations, a recovery and decay curve was reconstructed from the average fitted parameters within each ROI and compared with a recovery and decay curve representing the raw data averaged over the ROI in a single volunteer. To compare relaxation times between lateral and dorsal column WM and between WM and GM over all volunteers, a Wilcoxon rank sum (Mann-Whitney) test was performed. This comparison technique was also employed to test between WM and GM T1 values obtained from IR- and B1-corrected GRE acquisitions. This final comparison tests the hypothesis that a B1-corrected double flip angle GRE acquisition yields results indistinguishable from a multiple TI IR experiment in the spinal cord.
Figure 2a shows representative T1-weighted/spin density images as a function of TI (from upper left to bottom right) acquired using the IR approach. There is appreciable contrast at short TI (largely proton density–weighted [PDw]), and that contrast inverts as the TI increases (becoming more T1-weighted). Figure 2b shows the absolute T1 map resulting from a voxel-by-voxel fit to the signal recovery as a function of TI. There is good discrimination between GM and WM in the resulting T1 map, which is highlighted in Table 1. Using the IR method, significant differences were observed between the measured T1 values in WM and GM (p = 0.002), and in the lateral and dorsal columns (p = 0.01).
Figure 2c–e shows representative signal recovery curves (mean ± SD) for three ROIs: GM (Fig. 2c); lateral column WM (Fig. 2d); and dorsal column WM (Fig. 2e). Overlaid in red on the signal intensities is the fitted monoexponential model. Visually there is good agreement between model and data, and chi-squared goodness-of-fit analysis reveals an excellent agreement (p = 0.99); a high p value (p > 0.05) would indicate reason to accept the null hypothesis that the observed values equal the fitted values.
Figure 3a–d shows representative data from the GRE T1 data set and accompanying B1 acquisition consisting of a small flip-angle T1-weighted/spin density image (α = 15°, Fig. 3a) a large flip-angle T1-weighted image (α = 60°, Fig. 3b), the resulting B1-corrected absolute T1 map (Fig. 3c), and B1 map (Fig. 3d) used for correction. The low flip angle image shows maximum contrast, in agreement with the literature (22), while the larger flip angle shows very little contrast within the cord. Consequently, GM and WM can be discriminated in the resulting T1 maps. A summary of the tissue relaxation parameters is given in Table 1. Using the B1-corrected GRE method, differences between the measured T1 for WM and GM were found to be statistically significant (p = 0.002); however, no such difference was observed between dorsal and lateral column WM (p > 0.05).
Figure 4a shows the T2-weighted images as a function of TE (from upper left to bottom right). There is appreciable contrast between GM and WM at short TE, which is lost at longer TE. Figure 4b shows a representative T2 map resulting from a voxel-by-voxel monoexponential fit to the signal decay curve. Note that compared to the T1 maps, there is little visual contrast between GM and WM in the T2 maps and the difference between GM and WM fails to reach significance. Figure 4c–e shows the signal decay curves (mean ± SD) for three ROIs: GM (Fig. 4c); dorsal column WM (Fig. 4d); and lateral column WM (Fig. 4e). Overlaid on the signal intensities is the resultant monoexponential fit shown as a dashed line. Visually, there is good agreement between fit and data, and chi-squared goodness-of-fit analysis reveals an excellent agreement (p = 0.85–0.91); a high p value (p > 0.05) would indicate reason to accept the null hypothesis that the observed values equal the fitted values.
It has generally been assumed that the tissue relaxation time constants in the spinal cord will mimic those in the brain, or more specifically those regions of the brain that are structurally similar to the spinal cord. The WM of the spinal cord consists of densely packed fiber bundles and, of the tissues in the brain, is structurally most similar to WM found in structures such as the internal capsule and corpus callosum rather than the less dense WM (e.g., frontal WM) (4). The comparison in Table 1 shows that T1 values observed by both methods in the lateral and dorsal column WM at the level of C3 approximates reported values for callosal WM and, to a lesser extent, frontal WM (1,2).
The GM in the spinal cord is thought to be similar to that of deep GM (basal ganglia and brain stem) of the brain. The cytoarchitecture of cortical GM is largely vascular and lamellar and forms later in the development of the central nervous system, whereas the basal ganglia and spinal cord GM are more primitive and less lamellar (4). The data in Table 1 show that spinal cord GM relaxation times are closer to published values for the deep GM rather than cortical structures (e.g., frontal cortex) of the human brain (1,2). However, for the T1 data, there is still a large difference with deep GM, which we attribute to partial volume effects with WM lowering the observed GM relaxation time, an effect mainly visible for T1 because T2 values are quite similar between WM and GM.
While a multiple TI IR sequence is considered the gold standard for mapping absolute T1 values, it would be advantageous if a faster imaging method to estimate T1 in the spinal cord could be used. To test this, a B1 corrected double flip angle approach is presented here. Comparison of T1 values in WM and GM as observed from the two approaches (Table 1) show no significant difference (WM: p = 0.1, GM: p = 0.6) and thus are considered comparable. One reason this is the case is that the B1 field inhomogeneity (i.e., variability) in the spinal cord was observed to be quite small (1–1.08, ~7% intracord maximum variability, Fig. 3d), owing mainly to the fact that the body coil was used for excitation. Consequently, the T1 values calculated from the double flip angle approach can be easily corrected. One surprising result is that the contrast observed in the T1 maps arising from multiple TI IR scan is visibly less than seen in the double flip angle approach. This is most likely due to misregistration errors over multiple volumes, which are kept to a minimum when using only three volumes for calculation of T1, or possibly a signal-to-noise ratio (SNR) difference between acquisitions. The T1 IR experiment presented here takes ~20 min per slice whereas the B1 corrected double flip angle approach takes 2.5 min per 10 slices (15 s/slice). The time benefit of the double flip angle approach can be harnessed to measure a larger extent of the spinal cord in one sitting, or to boost SNR through signal averaging.
A final point of interest is that the T1 values arising from the IR scans show significant differences between lateral and dorsal column WM. This can be explained in one of two ways. First, it is hypothesized that the measured difference between WM tracts seen when using the IR method arises from CSF flow in the median fissure, which can artificially bloat the observed dorsal column T1 value. This is alleviated in the 3D GRE scans since 3D scans are largely flow insensitive. A second hypothesis is that since the 3D GRE approach is simply a two-point method, the measured values are less precise, and can be justified by the increased SD seen in Table 1. However, even though the dorsal and lateral column structures are slightly different in their composition (4), it is not expected that at this resolution the underlying tissue differences can be appreciated with any degree of confidence using either method.
It should be kept in mind that T1 and T2 values are field dependent. The T1 relaxation time varies with the 1H Larmor frequency, and is ~20% longer at 3T as compared to 1.5T (2). Conversely, T2 values decrease with increasing field, with T2 being ~10% shorter at 3T than at 1.5T.
Since T1 and T2 values are sufficiently different between GM and WM, it is possible to optimize spinal cord imaging to highlight the GM:WM and WM:CSF contrast. Using the steady-state signal equations appropriate for SE, and IR sequences (23), we simulated the effect of TE, TR, and TI on WM:GM contrast and also WM:CSF contrast (assuming CSF T1 = 4000 ms (2), T2 = 2500 ms [estimated]). As much of the contrast in MRI is due to spin-density, we further allowed GM, WM, and CSF to have their own associated spin-densities: 0.7, 0.8, 1.0, respectively (23). Using this formalism, the sequence parameters that give rise to maximal contrast in real-life situations can be appreciated. Figure 5 shows contour plots that demonstrate the interaction between each sequence parameter, where the number over each contour represents the relative signal difference between the simulated signal intensity for WM, GM, and CSF.
The SE shows the greatest GM:WM contrast at short TE and longer TR (Fig. 5a). In fact, the maximum contrast region is seen for shortest TE and TR > 3000 ms. It is interesting to note that at such a long TR, T1-based contrast is minimized and similarly at short TE, T2 contrast is also minimized. Thus, the resulting contrast is driven primarily by the differences in spin density between GM and WM and can be observed in Fig. 4a (top left panel). On the other hand, contrast between WM and CSF, is optimal for long TE (TE > 100 ms) and long TR (Fig. 5b). The inflection point in Fig. 5b should be noted as the point at which there is approximately no contrast between WM and CSF. This can be explained via a T1 effect. The short TR/TE SE generates T1-weighted contrast (signal intensity(CSF) < signal intensity(WM)) while longer TR/TE SE achieves T2-weighted (signal intensity(CSF) > signal intensity(WM)) and thus at a point in-between, signal intensity (CSF) = signal intensity (WM). Optimal sequence parameters are outlined in Table 2.
For the IR sequence, optimal WM:GM contrast is predicted for a TI at which WM is nulled (TI = ln(2)*T1WM ~ 600 ms), where some GM signal is retained and for a TR that is sufficiently long to allow for almost complete T1 recovery for WM (TR > 3000 ms), as shown in Fig. 5c. A similar contour plot is shown for WM:CSF contrast (Fig. 5d). A TI ~ 1800 ms and a long TR accentuates the difference between cord tissue and surrounding CSF. These sequence parameters are similar to those used for fluid-attenuated IR (FLAIR) sequences in routine brain examinations. It should be noted that the TE was assumed to be very short.
It is prudent to mention some of the limitations of the methods used in this study. The T1 calculation resulting from a two-point fit (even with B1 correction), bears a larger SD (~10%) as compared to the IR experiment. Furthermore, the method presented here will fail in regions where B1 variability is extremely large or when using small transmit-receive coils (from which the B1 profile is much less homogeneous than from the body coil). A shortcoming of the IR sequence is ultimately its sensitivity to motion (long TR) and CSF-related pulsation artifacts, which can artificially inflate the observed T1 in small structures. However, the similarity of the measured IR-based T1 values to those previously published for deep gray and dense WM in the brain suggests that it adequately captures the T1 of spinal cord tissue, and a finding of this work is that the B1-corrected double flip-angle experiment is adequate for estimates of spinal cord T1 values.
The T2 experiment has a few weaknesses: it relies on a monoexponential fit of data without a zero time-point and with a limited maximum TE of ~2.5T2. The former is unavoidable since it is not possible to acquire image data at an echo time of zero. Secondly, B1 inhomogeneities can produce a cumulative effect on each of the 180° pulses used in the experiment which ultimately lead to an underestimate of the T2 values. However, as measured here, the B1 variability is less than 10% in the spinal cord and realistically only 6%. Thus, the effect on each of the 180° pulses is less than 1%. Also, due to time constraints, T2 was measured in a single slice-selective fashion. Thus the slice profile of excitation and refocusing pulses can also have an effect on the signal intensity measured at each echo. However, since the B1 variability is small in the tissue of interest, we expect that this has a negligible effect on the measured T2 values of the spinal cord. Finally, since the T2 imaging experiment was cardiac gated, it is possible that heartbeat variations could lead to a variable TR, causing the magnitude of the initial magnetization to be slightly different for each phase encoding step. However, the decay rate in each scan will be the same and thus the measured T2 is correct. The triggering was needed because T2 experiments necessitate long TEs and the absence of cardiac gating could lead to large pulsatile artifacts that outweigh the effect of small differences in TR.
Spinal cord imaging with currently accessible resolution is sensitive to partial volume effects between neighboring tissues and CSF. We attempted to avoid this in the WM by choosing ROIs in the lateral and dorsal columns that appeared to be clear of GM and CSF. The close agreement between spinal cord and literature values for brain WM relaxation times implies this goal was sufficiently accomplished. However, in the GM, the luxury of discarding border regions was not available and quantification was somewhat compromised. Partial volume contact with WM would tend to lower both T1 and T2. This effect is more pronounced in the T1 maps since the disparity between WM and GM T1 values is larger than between T2 values.
In quantifying T2 for each anatomical region, signal decay curves were fitted voxel-by-voxel, giving rise to a quantitative T2 map. ROIs were then placed in the WM and GM, as seen from the shortest TE image, to give a mean and SD over the selected voxels. In comparison, the alternative approach of averaging the signal over an ROI and fitting a single decay curve per region showed no significant difference. This implies that the SNR was sufficient to analyze the individual voxels.
In this work, only a single 5-mm section of the spinal cord at the level of C3 was chosen for further analysis; however, the results are expected to be representative of a much larger segment of the cord. The spinal cord is an essentially linear organ for which changes in tissue morphology over the cephalo-caudal direction are minimal. At the cervical level, superior to the branching of the brachial plexus, spinal cord WM and GM are known to differ only in shape rather than in composition. However, it is possible that future studies of the spinal cord at different levels (e.g., thoracic, lumbar, conus medularis, cauda equina) might reveal differences in relaxation parameters.
Longitudinal and transverse relaxation time constants were reported in the human spinal cord in vivo. A B1-corrected double flip angle approach was shown to provide T1 values that agree with standard IR techniques. The reported values should be useful in parameter optimization for clinical spine imaging at 3T, and in quantitative methods, such as those to assess metabolite concentrations in MR spectroscopy and quantitative magnetization transfer studies in the cord.
We thank Asif Mahmood and Manus Donahue for their assistance in data analysis and statistical methodology. This publication's contents are solely the responsibility of the authors and do not necessarily represent the official view of NCRR or NIH. Dr. Peter van Zijl is a paid lecturer for Philips Medical Systems. This arrangement has been approved by Johns Hopkins University in accordance with its Conflict of Interest policies.
Grant sponsor: National Institutes of Health (NIH)/National Institute of Biomedical Imaging and Bioengineering (NIBIB); Grant number: EB000991; Grant sponsor: NIH/National Center for Research Resources (NCRR); Grant number: RR015241.