T2-weighted spine imaging with a fast three-point dixon technique: Comparison with chemical shift selective fat suppression


  • Presented in part at the 10th Annual Meeting of ISMRM, Honolulu, 2002.



To develop a phased-array coil-compatible, fast three-point Dixon (TPD) technique, and compare its performance in T2-weighted spine imaging with that of the standard chemical shift selective (CHESS) fat suppression technique.

Materials and Methods

We acquired T2-weighted spine images of 27 patients using essentially identical scanning parameters with the fast TPD technique and standard fast spin echo (FSE) with CHESS fat suppression. A phased-array coil-compatible image reconstruction algorithm was developed to generate separate water and fat images from the data acquired with the fast TPD technique. Three neuroradiologists independently scored the images from the two different techniques for uniformity of fat suppression and lesion conspicuity using a four-point system (1 = poor, 2 = fair, 3 = good, 4 = best).


The reviewers' mean scores were 3.2 and 2.1 for the uniformity of fat suppression, and 3.0 and 2.0 for the lesion conspicuity for the fast TPD and the CHESS fat suppression techniques, respectively. The fast TPD technique was statistically superior to the CHESS technique at P < 0.0005.


The fast TPD technique provides superior fat suppression and lesion conspicuity, and potentially can be used as an alternative to T2-weighted imaging of the spine. J. Magn. Reson. Imaging 2004;20:1025–1029. © 2004 Wiley-Liss, Inc.

FAST SPIN-ECHO (FSE, or rapid acquisition with relaxation enhancement (RARE)) (1) T2-weighted magnetic resonance imaging (MRI) with chemical shift selective (CHESS) (2, 3) fat suppression is currently an essential component of routine MRI of the spine. Despite the widespread use and success of this technique, however, the quality of its fat suppression is sometimes clinically unsatisfactory; it may even occasionally deteriorate to the extent that repeat scanning or scanning without fat suppression is required. Magnetic field (B0) and radiofrequency (RF) field inhomogeneities are among the known causes of this problem, and they may be both scanner- and patient-dependent. For imaging of the spine, the problem is often exacerbated because of a patient's anatomic geometry and the use of a phased-array coil, which is necessary for sufficient signal-to-noise ratio (SNR) and spatial coverage.

The three-point Dixon (TPD) technique (4–9), in which the chemical shift difference between water and fat is encoded into three images with different echo shifts, can provide separate water- and fat-only images, and therefore may serve as an alternative to the CHESS technique for fat suppression. In the TPD technique, field inhomogeneity manifests as image-phase offsets and can be removed by a postprocessing phase correction. As a result, high-quality water- and fat-only images can be generated even in the presence of substantial B0 and RF field inhomogeneities. However, two prominent drawbacks to the TPD technique have hindered its widespread use. The first is the long scanning time required by the multipoint data acquisition. Contributing to this problem—and less well-known—is the additional increase in scanning time or reduction in slice coverage that results from the echo shift required by the Dixon acquisition. The second drawback is that the TPD technique requires a computationally extensive and sometimes error-prone phase correction, which can be quite prohibitive in the case of multicoil acquisitions. Consequently, few of the works published so far using the Dixon technique have been carried out with phased-array coil acquisitions.

To address these drawbacks, we first sought to develop a more efficient TPD technique (herein called the fast TPD technique) that would be suitable for T2-weighted imaging with fat suppression. For data acquisition, we used an FSE-based Dixon technique that achieves echo shift without increased interecho spacing or loss of slice coverage (10). We designed a phased-array coil-compatible image reconstruction algorithm, with special attention paid to processing reliability and speed, especially for multicoil acquisitions. Our second purpose was to compare the performance of the technique with that of the standard CHESS fat suppression technique in spine imaging, using essentially identical scanning parameters. Three neuroradiologists independently scored the fat suppression and lesion conspicuity achieved by the two techniques to determine whether the fast TPD technique can potentially be used as an alternative to T2-weighted imaging of the spine.


Pulse Sequence

The Dixon technique (4) is used primarily for spectroscopic imaging of an object with two or more distinct chemical species (such as water and fat), and is compatible with many different pulse sequences. In the original FSE implementation (11), the echo shift required by the Dixon technique is achieved by increasing the interecho spacing, which lowers scanning efficiency and may introduce additional image blurring due to increased signal modulation during the prolonged echo train. In this work, we used a modified FSE implementation (10) to acquire T2-weighted images and reduce the scanning time. The modifications were made so that echo shifting was induced by “sandwiching” in time the readout gradients with a pair of small gradients of equal area and opposite polarity. The area of the small gradients is determined by the desired phase shift between water and fat. Since these gradients can usually be played out during the dead time on the readout axis, neither an increase in echo spacing nor a loss of slice coverage is incurred.

Image Reconstruction

When a phased-array coil is used, the two-dimensional Fourier transform of the raw data in a multipoint Dixon acquisition from a single-component coil can be expressed as (7):

equation image(1)

where n and k are the indices for the Dixon acquisition and the coil number, respectively; W and F represent the relative content of water and fat at a given spatial location or image pixel; C(k) is the magnitude of the coil sensitivity; ωo is the known chemical shift difference between water and fat (≈3.5 ppm); ω is the off-resonant frequency due to the magnetic field B0 inhomogeneity; ϕmath image is the frequency-independent phase offset, including the contribution of the complex coil sensitivity; and τ is the user-selectable time shift in the data acquisition window for the Dixon acquisition.

The ϕmath image term in Eq. [1] can easily be determined from the images when τ = 0 (corresponding to the first Dixon acquisition). The major challenge in processing Dixon data lies in determining ωτ, especially for regions with a low SNR. A straightforward way of processing the multicoil data is to treat the images from each coil separately and then combine the resultant water and fat images (12). However, such an approach would invariably lead to a total processing time that is proportional to the number of coils. More importantly, the robustness of the processing may be compromised because the component images often contain different regions of very low SNR.

To remedy these problems, we performed phase correction by first summing the component images from all individual coils after ϕmath image was removed:

equation image(2)

In Eq. [2] it is possible to factor out C(k) because W, F, and ωτ do not vary for different coils. Since Eq. [2] has the same functional form as Eq. [1], we can use the same algorithm that is appropriate for processing data from a single coil to determine ωτ in Eq. [2]. Once oτ is determined, we can then apply it to signals from individual coils to obtain corresponding separate water and fat images. In this way, the phase correction processing time becomes independent of the number of coils used. Further, processing becomes more reliable because the combined images as represented in Eq. [2] contain in general few regions of low SNR. The algorithm for phase correction we used for this work is based on a region-growing process without direct phase unwrapping, as detailed in Refs. 9 and10.

Patients and Data Collection

A total of 27 patients (mean age = 54 years, range = 20–80 years, standard deviation = 14) were enrolled in the study, which was approved by our institutional review committee. All of the subjects provided written informed consent. For each patient, both fast TPD images and conventional FSE images with CHESS fat suppression were acquired. The scanning parameters used for both techniques were kept identical except as noted below, and varied only slightly from patient to patient to accommodate the differences in spatial coverage. The typical protocol was as follows: TR = 3100 msec, TE = 90 msec, echo train length = 16, receiver bandwidth = ±16 kHz, field of view (FOV) = 34 cm, matrix size = 384 × 224, slice thickness = 4.0 mm, and slice gap = 1.0 mm. The number of signal averages (NEX) typically used for the conventional FSE sequence with CHESS fat suppression was six to ensure sufficient SNR. For a fair comparison, the same equivalent NEX was used for the TPD technique. Using this protocol, 10 slices were covered with the CHESS technique, with a TR of 3100 msec and within 4:27 minutes. Using the same protocol, the same 10 slices were covered with a TR of 2925 msec, and with the fast TPD technique they were covered within 4:11 minutes. The small increase in TR and scanning time for the CHESS technique resulted from the extra sequence time needed for the CHESS pulses. In all cases, data were collected with a 1.5-Tesla scanner (GE Medical Systems, Waukesha, WI) and a six-element CTL phased-array coil (USA Instruments, Aurora, OH).

Data Analysis

Three experienced neuroradiologists who were blinded to the imaging techniques and parameters independently reviewed the water-only images from the fast TPD technique and the images from the standard FSE with the CHESS fat suppression technique. Of the 27 patients, 14 had one or multiple lesions. In all, 38 lesions (29 in the vertebral bodies, six in the epidural space, two in the vertebral discs, and one in soft tissue) were found. Using a four-point system (1 = poor, 2 = fair, 3 = good, 4 = best), the three neuroradiologists scored the uniformity of fat suppression for all of the patients, and the conspicuity of the lesion for the patients with lesions. A score of 1 (poor) was assigned when the fat suppression was severely nonuniform over a large region of interest (ROI) or impaired lesion detection (such as by inadvertent water suppression). On the other hand, a score of 4 (best) was assigned when no noticeable fat suppression nonuniformity was present over the entire images. The scores of each observer were then averaged, and differences between scores were evaluated by means of Wilcoxon's signed-rank test.

The same three neuroradiologists were also asked to provide an overall assessment of the images generated by the two different techniques, and to choose among the following for each clinical case: CHESS fat suppression was superior to the fast TPD technique, the fast TPD technique was superior to the CHESS fat suppression technique, or the two techniques performed equally well. The determinations of the reviewers were paired (A vs. B, B vs. C, and A vs. C), and unweighted kappa statistics were calculated.


Processing of the Dixon images from all patients was automated, and none of the processed images had any noticeable misidentification of water and fat, demonstrating the robustness of the algorithm. The processing times for phase correction of the multicoil and single-coil data were essentially identical, which is a substantial improvement in computational efficiency considering that typically three to four component coils are used in spine imaging.

The qualitative improvement in fat suppression uniformity and lesion conspicuity offered by the fast TPD technique was statistically significant at the 0.0005 level (Table 1). Three reviewers were in complete agreement that the fast TPD technique was better than the conventional technique in 16 patients, the conventional technique was better than the fast TPD technique in one patient, and both techniques performed the same in one patient. Two reviewers agreed that the fast TPD was better than the conventional technique in two patients, the conventional technique was better than the fast TPD technique in three patients, and the two techniques performed the same in two patients. For the remaining two patients, the three reviewers disagreed on whether one technique was better than the other, or whether they performed equally well. The kappa coefficients (κ) calculated between the three possible pairs of reviewers were 0.51, 0.70, and 0.58, respectively, indicating moderate to good agreement among the different reviewers. The advantage of the fast TPD technique was most obvious when the CHESS fat suppression became so unreliable that it inadvertently suppressed the water signals (noted to occur in 10 of the 27 patients).

Table 1. Comparison by Three Neuroradiologists of Images Obtained With Fast TPD and Standard FSE with CHESS Fat Suppression Techniques in 27 Patients
 Fast TPDaFSE with CHESSaP
  • a

    Data are the mean scores for the three independent neuroradiologists; the scoring scale used was 1 = poor, 2 = fair, 3 = good, 4 = best.

Fat-suppression uniformity3.2 ± 0.72.1 ± 0.8<.0005
Lesion conspicuity3.0 ± 1.12.0 ± 1.0<.0005

A case for illustration is shown in the image of Fig. 1a, which was obtained by the conventional CHESS technique from a 20-year-old female patient with plasmocytoma. Poor fat suppression and unintentional water suppression is especially notable in the superior and inferior regions of the image. For comparison, the water-only image from the fast TPD technique of the same patient, and using the same parameter, is shown in Fig. 1b. Clearly, much better fat suppression is achieved with the fast TPD technique throughout the entire FOV. The solitary lesion at the location of the T10 vertebra is also better visualized in the fast TPD image than in the image with CHESS fat suppression.

Figure 1.

a: T2-weighted FSE image with CHESS fat suppression of a 20-year-old patient with plasmocytoma. A total of 3:25 minutes were used for the image acquisition. b: The water-only image corresponding to the same slice location as in image a using the fast TPD technique. Better fat suppression and visualization of the lesion are achieved in image b because of the ability of the fast TPD technique to compensate for the field inhomogeneity effects.

Figure 2a shows a sagittal T2-weighted FSE image with conventional chemical saturation obtained from a 60-year-old female patient with squamous cell carcinoma of the larynx. Figure 2b shows the fast TPD image of the same patient at the same anatomic location. As with Fig. 1, both images were acquired with essentially the same scan parameters and within the same total time. A comparison of the two images (Fig. 2a and b) again shows noticeable improvement in the uniformity of the fat suppression. The loss of the C7-T1 disc space and the collapse of the adjacent vertebral bodies are depicted better in Fig. 2b than in Fig. 2a.

Figure 2.

a: T2-weighted FSE image with CHESS fat suppression of a 60-year-old patient with squamous cell carcinoma of the larynx. b: The water-only image obtained with the fast TPD technique in the same patient at the same anatomic location. Again, better overall fat suppression and improved visualization of the loss of the C7-T1 disc space and the collapse of the adjacent vertebral bodies were achieved in image b compared to image a.


One approach for reducing the sensitivity of fat suppression to field inhomogeneity is to use a short tau inversion recovery technique (13). However, this technique increases the scan time, and, more importantly, may lower the image SNR and alter the image contrast (14, 15). The major advantage of the Dixon technique over the CHESS technique is that it can make the fat suppression relatively insensitive to both B0 and RF field inhomogeneity (7) while it preserves the desired image contrast. However, previous implementations of the Dixon technique have been limited to spin-echo or gradient-echo sequences and the use of a single receiver coil. Consequently, only T1-weighted imaging of a few body parts has been practical. Because T1-weighted images generally have higher SNRs and fewer artifacts compared to T2-weighted images, the robustness of the phase correction algorithms used has not been evaluated as thoroughly for images with lower SNRs. The FSE implementation of the Dixon technique by Hardy et al (11) demonstrated the feasibility of using the Dixon acquisition for T2-weighted images. However, the scanning efficiency was degraded because of the required increase in echo spacing. As a result, it is not easy to directly compare the performance of the Dixon implementation by Hardy et al (11) with that of existing techniques (such as CHESS), and different scanning parameters must be used to accommodate the needs of clinical scanning (16).

These issues were addressed in this work with a more efficient Dixon data acquisition scheme and a robust phase correction algorithm that handles multicoil data. Our ongoing experience confirmed that the fast TPD technique can be used in most cases to meet the clinical requirements for scanning parameters that are possible with the CHESS technique. This is especially true considering that in comparison with the CHESS technique, the fast TPD technique actually saves a small amount of scanning time by eliminating the CHESS pulses, and does not need any manual prescanning adjustments. The increased robustness and reliability of the fast TPD technique could also lead to better patient throughput than the conventional CHESS method because less failure in fat suppression is expected.

Although we evaluated only spine imaging in this study, the fast TPD technique could certainly be used in other anatomic sites where T2-weighted imaging with reliable fat suppression is needed. Previous studies using the original FSE implementation of the Dixon acquisition with a single coil demonstrated the advantage of the Dixon technique in imaging pediatric musculoskeletal systems and the retrobulbar space (15, 16). With its increased data-collection efficiency and phased-array coil compatibility, our implementation should render this technique more readily applicable to other areas of interest, such as imaging the shoulder, knee, and body.

In conclusion, we have designed and implemented a phased-array coil-compatible fast TPD technique for T2-weighted spine imaging. The technique is both efficient and robust compared to conventional imaging with CHESS fat suppression. Clinical patient imaging and image review by radiologists demonstrated that the technique potentially can serve as an alternative to conventional T2-weighted imaging with CHESS fat suppression.