Over the past two decades, MR imaging has evolved into the preferred imaging modality for evaluating pathology of the vertebrae and soft tissue structures within and adjacent to the spine (1). The normal spine contains red marrow rich in adipocytes, and as the patient ages, fat content increases progressively (2-4). Meanwhile, abundant adipose tissues exist in perispinal structures (2, 5). Fat in the marrow and perispinal structures appears bright on fast spin-echo (FSE) T1-weighted (T1W) and T2-weighted (T2W) images. Furthermore, a variety of pathologies usually appear in a high signal intensity similar to fat signal on T2W images (1, 2, 6-8). Therefore, the normal fat may obscure underlying pathologies that can involve the spine, including metastasis, primary neoplasm, direct extension of adjacent neoplasm, diffuse or focal degeneration, or inflammatory conditions, and so on. For these reasons, robust and uniform fat suppression, especially on T2W images, is extremely important for radiologists to achieve accurate diagnoses MR imaging of the spine.

Many techniques exist for obtaining fat-suppressed images. FSE imaging with frequency-selective fat saturation pulses is routinely used in spine MR imaging. However, achieving uniform and reliable fat suppression in the spine with this method is difficult due to main magnetic field (B₀) and radiofrequency field (B₁) inhomogeneities. This is particularly true in some situations such as off-isocenter imaging, large field of view (FOV) imaging, air-tissue interfaces and the presence of metallic hardware (9). Short-inversion-time inversion-recovery (STIR) imaging provides uniform fat suppression; however, STIR is sensitive to B₁ inhomogeneities, requires additional time for inversion pulses and suffers from a low signal-to-noise ratio (SNR) (10). Spectral spatial pulses can be used with FSE for fat suppression. Although this method is insensitive to B₁ inhomogeneities, it is relatively sensitive to B₀ inhomogeneities (11).

The original two-point Dixon method was first described by Dixon (12) in 1984, and was used for fat and water separation by utilizing differences in chemical shifts, but was sensitive to field inhomogeneities. In order to demodulate field inhomogeneity effects from the acquired source images, a three-point method with symmetrically acquired echoes was developed to measure the inhomogeneities directly, which renders this method insensitive to both B₀ and B₁ inhomogeneities (13-15). It was recently demonstrated, however, that decomposition of water from fat with symmetrically acquired echoes could not be achieved when the proportions of water and fat within a voxel are approximately equal (16).

Iterative decomposition of water and fat with echo asymmetry and least-squares estimation (IDEAL) is a cutting edge modified three-point method choosing asymmetric echo shifts to create relative water-fat phase shifts of -π/6, π/2, 7π/6 (9, 17). This approach has been demonstrated to provide uniform and reliable fat suppression and has been successfully combined with FSE (9), gradient-recalled acquisition in steady-state (GRASS) imaging (18), spoiled gradient-recalled echo (SPGR) (19), and balanced steady-state free precession (SSFP) techniques (20) to create high-quality fat-suppressed MR images.

Iterative decomposition of water and fat with echo asymmetry and least-squares estimation (IDEAL) imaging has been applied with robust water-fat separation throughout the body, including the head and neck (9, 21), breast (22), heart (22, 23), abdomen (9, 22), pelvis (9), and extremities (9, 18, 20, 24). However, to our knowledge, only a few studies have focused the application of IDEAL in spine MR imaging (6, 9, 25), and no previous MR imaging study of the spine with T2-weighted IDEAL fast recovery fast spin-echo (FRFSE) at 3.0 Tesla (T) has been reported.

The aim of this work was to evaluate T2W IDEAL FRFSE imaging of the spine at 3.0T under different limited situations. In thirty-five patients, we compared T2W IDEAL FRFSE with fat-saturated T2W FRFSE imaging. Quantitative measurements of SNR efficiency and qualitative scoring of diagnostic image quality and fat suppression were performed.

Successful separation of water and fat was achieved in all studies for T2W IDEAL FRFSE imaging. The IDEAL water-only images showed robust and uniform fat suppression and high SNR-efficiency performance (Fig. 1A). Separate IDEAL fat-only images and recombined IDEAL in-phase images had corrected chemical shift artifacts (Fig. 1B-D).

Signal-to-noise ratio-efficiency of the vertebral body was significantly different between IDEAL (2.83 ± 0.28) and fat-saturated FRSE (1.82 ± 0.28), and SNR-efficiency of spinal cord was also significantly higher for IDEAL FRFSE (8.69 ± 0.63) compared with fat-saturated FRFSE imaging (5.99 ± 0.35) (p < 0.05).

Overall image quality score for T2WI IDEAL FRFSE (2.9) was higher (p < 0.01) than that for fat-saturated T2W FRFSE (2.1) imaging. Fat suppression was rated as fair to good on fat-saturated T2W FRFSE images (1.8), but fat-water separation was excellent (2.9) on the T2W IDEAL FRFSE images. The difference between fat suppression and fat-water separation using IDEAL was statistically significant (p < 0.01).

Fat-water separation with IDEAL FRFSE imaging performed particularly well in the presence of metal (Fig. 2) and some challenge regions (Fig. 3), showing fewer artifacts than fat-saturated FRFSE imaging. In some cases, areas of poor fat suppression on fat-saturated FRFSE images were seen mimicking traumatic bone marrow edema (Fig. 4) or metastasis (Fig. 5), but superior fat-water separation in the corresponding water-only images showed that these were artifacts caused by poor fat saturation.

It is well known that T2W sequences with fat suppression are very important for radiologists to discriminate a variety of pathologies in the spine. Unfortunately, we occasionally have situations in which, despite careful shimming, fat signal intensity is inadequately suppressed using conventional methods. Murakami et al. (25) have demonstrated that IDEAL can provide more uniform fat suppression and spinal canal projection with fewer artifacts from magnetic inhomogeneity compared with fat-suppressed FSE using chemical shift selective saturation (CHESS). In our work, T2W IDEAL FRFSE imaging reliably achieved excellent fat suppression in areas susceptible to consistent failure of fat suppression on the fat-saturated T2W FRFSE images-near not only the presence of metallic hardware, but also the air-tissue interfaces, off-isocenter and top margin of FOV. Meanwhile, IDEAL images showed superior image quality and higher SNR-efficiency performance compared with fat-saturated T2W FRFSE images.

In addition to water-only images, IDEAL provides multiple contrasts including fat-only and recombined fat/water in-phase and out-of-phase images from one acquisition. Although our study did not focus on the recombined images, the recombined in-phase IDEAL FRFSE images with chemical shift correction are of high quality and provide anatomic reference similar to conventional T2W images, which may permit elimination of conventional T2W image from the MRI protocols. Thus, a single acquisition with IDEAL imaging has the potential to simplify spine MRI protocols by replacing separate acquisitions that use fat-saturation and chemical shift techniques. Furthermore, because all data emanate from a single acquisition, the resulting diverse image sets are perfectly coregistered. IDEAL imaging possibly reduces "on-the-table" time for patients and improves the cost-effectiveness of MRI. Further study is required to determine whether removing the T2 or T1 weighted images from a spine protocol can result in the loss of important diagnostic information or not. Another benefit of IDEAL imaging is that in-phase and out-of-phase technique may be useful for differentiating benign from neoplastic processes within bone marrow by detecting small amounts of microscopic fat within benign lesions (26). Advantages of the IDEAL in-phase and out-of-phase images need to be fully explored in the future study.

In recent years, many institutions have increasingly operated 3.0T MR imaging systems. One challenge of 3.0T spine imaging is the increased chemical artifacts due to a doubling of the difference in the resonance frequency between fat and water protons from 1.5T to 3.0T. Imaging at 3.0T often requires increases in bandwidth to avoid the increased chemical shift artifacts, which result in low SNR performance. Correction of chemical shift with IDEAL imaging at 3.0T eliminates the need to increase bandwidth and better preserving the SNR benefit of 3.0T imaging. Another challenge of 3.0T is that susceptibility to magnetic field inhomogeneities increases, which give rise to more artifacts and failure of fat suppression in some situations. IDEAL imaging can overcome the effect of increased susceptibility to magnetic field inhomogeneities and provides robust fat suppression and recombined images. The full benefits of these advantages of IDEAL at 3.0T will require additional evaluation on more subjects in future studies.

In general, SNR and subjective quality comparisons are best done for 2 sequences with nearly the same imaging parameters and the same amount of time. However, IDEAL requires as much acquisition time as do 3 numbers of excitations because it needs 3 echo times, hence taking about triple the time than the fat-saturated sequence for 1 scan under identical imaging parameters. In order to achieve impartial and accurate comparisons, we used a NEX of 4 in a fat-saturated sequence to double the scanning time approximately to the time of IDEAL, which led to an increase of SNR in a fat-saturated sequence. If we changed NEX to 2 and imaging time to 1:06 for the fat-saturated FRFSE sequence, the SNR would be lower, image quality would be worse and SNR-efficiency would not change according to the effect of imaging parameters on SNR. However, in routine practice, we can expect the advantage of fast imaging for the fat-saturated FRFSE sequence, and diagnosis may be easily made on fat-saturated T2 FRFSE images with a NEX of 2.

The main disadvantage of IDEAL is an increase in the scanning time because of the requirement of three acquisitions to separate water from fat. There have been several methods employed to reduce scanning time including parallel imaging, reduced sampling strategies, and the use of half k-space acquisitions and homodyne reconstruction (6, 9, 24). However, the scanning times at 1.5T were still 5-6 minutes in some literatures (6, 24). The FRFSE sequence used in our study is a modified FSE sequence. In FRFSE, transverse magnetization is still present at the end of the echo train, and then a -90° pulse refocuses protons back into the longitudinal axis rather than allowing them to undergo T1W recovery. This reduces the transverse magnetization in the longitudinal axis, resulting in an acceleration of the relaxation of longitudinal magnetization. FRFSE sequences have shorter TR with comparable contrast and a reduced imaging time. This reduction in TR improves SNR efficiency and also offsets the increased minimum scanning time required for the IDEAL technique, and in the meantime, benefited from the comprehensive technical progresses using a GE HDx MR scanner. Moreover, IDEAL combined with FRFSE can be effective in alleviating the long scanning time to approximately 3 minutes, even with 2 NEX and SNR-efficiency improvement. Although the imaging time of the IDEAL sequence is still longer than that of fat-saturated FRFSE imaging due to IDEAL's inherent characteristic, IDEAL imaging has the potential to simplify spine MRI protocols by replacing some acquisitions, and the overall scanning times of spine MRI may be reduced along with excellent fat suppression and high SNR-efficiency. Times are, so to speak, no longer clinically limiting for IDEAL FRFSE imaging.

For the fixed bandwidths received in our study, T2 decay will increase at the later echoes and may worsen blurring for imaging because of the increased time required to acquire a train of echoes. The increased T2 weighting with echoes acquired at the end of the echo train may actually increase the edge enhancement that occurs with T2W FSE imaging. However, different from FSE, FRFSE uses -90° refocusing pulses that elongate echo trains to generate better contrast between long and short T2 species with little blurring and low radiofrequency power. The overall effect of the longer echo train with IDEAL depends on the particular k-space ordering and effective TE chosen. The longer echo train has no appreciable effect on image quality on either T1W or T2W IDEAL-FSE imaging (6), and also on FRFSE imaging.

In conclusion, T2W IDEAL FRFSE imaging is a promising technique for spine MR imaging at 3.0T because it provides consistent, robust and uniform fat suppression compared with conventional fat-saturated FRFSE imaging, while possessing high SNR-efficiency and image quality. Additional studies should be done to assess the performance of IDEAL in discrimination of different pathologies and fat quantification.