Improved fat suppression using multipeak reconstruction for IDEAL chemical shift fat-water separation: Application with fast spin echo imaging

Authors


Abstract

Purpose

To evaluate and quantify improvements in the quality of fat suppression for fast spin-echo imaging of the knee using multipeak fat spectral modeling and IDEAL fat-water separation.

Materials and Methods

T1-weighted and T2-weighted fast spin-echo sequences with IDEAL fat-water separation and two frequency-selective fat-saturation methods (fat-selective saturation and fat-selective partial inversion) were performed on 10 knees of five asymptomatic volunteers. The IDEAL images were reconstructed using a conventional single-peak method and precalibrated and self-calibrated multipeak methods that more accurately model the NMR spectrum of fat. The signal-to-noise ratio (SNR) was measured in various tissues for all sequences. Student t-tests were used to compare SNR values.

Results

Precalibrated and self-calibrated multipeak IDEAL had significantly greater suppression of signal (P < 0.05) within subcutaneous fat and bone marrow than fat-selective saturation, fat-selective partial inversion, and single-peak IDEAL for both T1-weighted and T2-weighted fast spin-echo sequences. For T1-weighted fast spin-echo sequences, the improvement in the suppression of signal within subcutaneous fat and bone marrow for multipeak IDEAL ranged between 65% when compared to fat-selective partial inversion to 86% when compared to fat-selectivesaturation. For T2-weighted fast spin-echo sequences, the improvement for multipeak IDEAL ranged between 21% when compared to fat-selective partial inversion to 81% when compared to fat-selective saturation.

Conclusion

Multipeak IDEAL fat-water separation provides improved fat suppression for T1-weighted and T2-weighted fast spin-echo imaging of the knee when compared to single-peak IDEAL and two widely used frequency-selected fat-saturation methods. J. Magn. Reson. Imaging 2009;29:436–442. © 2009 Wiley-Liss, Inc.

FAT SUPPRESSION is commonly used to improve tissue contrast in musculoskeletal magnetic resonance imaging (MRI). Fat appears bright on many MR pulse sequences, which can obscure underlying pathology such as edema, inflammation, and neoplasm. For this reason, fat suppression is commonly combined with T2-weighted and intermediate weighted fast spin-echo sequences to increase fluid sensitivity and with T1-weighted fast spin-echo sequences to improve visualization of gadolinium contrast uptake in structures such as bone marrow and subcutaneous tissue. Fat suppression is also useful for cartilage imaging to avoid chemical shift artifact and to optimize the overall dynamic contrast range of the image.

The most commonly used methods of fat suppression in musculoskeletal MR imaging are short tau inversion recovery (STIR) and frequency selective fat-saturation. STIR relies on differences in the T1 relaxation time between fat and water protons. Based on this difference, an inversion time can be chosen to suppress fat signal at precisely the time the inverted longitudinal magnetization of the methylene protons of fat crosses the null point after the application of a nonselective 180° inversion pulse (1). STIR provides uniform fat suppression independent of magnetic field inhomogeneity. However, the technique can only be used with intermediate-weighted or T2-weighted imaging because of the risk of suppressing signal from protons within gadolinium contrast, subacute hemorrhage, and proteinaceous fluid, which have similar T1 relaxation times as fat. In addition, the overall signal-to-noise (SNR) efficiency of STIR imaging is restricted by its reliance on inversion pulses and relatively long inversion times. STIR also causes partial saturation of desired water signal, further reducing SNR performance (2–4).

Frequency-selective fat-saturation takes advantage of the difference in resonance frequency between fat and water, which is ≈−420 Hz at 3.0T. A 90° saturation pulse tuned to the main resonance frequency of fat is applied that flips the bulk magnetic vector from fat into the transverse plane. Spoiler gradients are then used to destroy the phase coherence of the transverse magnetization (5). Unlike STIR, frequency selective fat-saturation can be combined with most MR pulse sequences and only modifies the signal characteristics of fat. However, the imaging technique is highly sensitive to magnetic field inhomogeneity. Furthermore, the technique fails to suppress all signal arising from fat protons. This occurs because fat consists of multiple chemically distinct moieties that give rise to at least six resonance peaks. Several of these peaks, most notably those arising from olefinic protons and CH2 protons of the glycerol backbone, have resonance frequencies very close to that of water and will not be suppressed by frequency-selective fat-saturation pulses (6). More recent developments in frequency-selective fat-saturation use a partial inversion pulse of ≈110° that partially inverts the main resonance peak of fat (7). The partially inverted magnetization of the main fat peak cancels the magnetization of noninverted fat peaks at different resonance frequencies through a partial volume effect. Despite some improvement in fat suppression, this technique remains highly sensitivity to magnetic field inhomogeneity, which can cause areas of failed fat suppression or even water suppression (4, 8).

Water excitation is an alternative method of fat suppression that has been applied to musculoskeletal MRI (9). This technique uses a series of correctly phased radiofrequency pulses to selectively excite water protons (10). Water excitation is more SNR efficient but equally sensitive to magnetic field inhomogeneity when compared to frequency-selective fat-saturation (9, 11, 12). Furthermore, the technique will excite all protons near the water resonance including any resonance peaks of fat such as those due to olefinic protons and CH2 protons of the glycerol backbone (6). For this reason, water excitation will lead to incomplete fat suppression similar to that which occurs with frequency-selective fat-saturation.

Due to limitations of currently available fat-suppression techniques, there has been a recent resurgence in chemical shift-based fat-water separation methods. These methods are based on the original work by Dixon (13) in 1984 using in-phase and out-of-phase images to separate fat and water signal. The Dixon method was later modified by Glover (14) into a three-point technique designed to eliminate the effects of magnetic field inhomogeneity and then subsequently applied to fast spin-echo imaging by Hardy et al (15). Other versions of chemical shift-based water-fat separation techniques have also been described (16, 17). These methods are relatively insensitive to magnetic field inhomogeneity and provide robust and uniform fat suppression even in the complex magnetic environments commonly encountered during musculoskeletal MR imaging.

Iterative Decomposition of water and fat with Echo Asymmetry and Least squares estimation (IDEAL) is a newly developed chemical shift-based fat-water separation method that can be combined with a variety of MR pulse sequence (18–20). IDEAL uses an iterative approach to estimate the magnetic field map and remove its effects from the fat-water decomposition. After phase shifts caused by magnetic field inhomogeneity are removed through demodulation of the complex images acquired at three different echo times, a simple least squares estimation is used to determine the amount of fat and water signal within each pixel (18, 20–23). IDEAL allows use of arbitrary echo times so that echo spacing can be optimized in order to achieve maximum SNR performance (24).

While a few chemical shift-based fat-water separation techniques have considered the spectral complexity of fat (25, 26), most currently used methods including IDEAL have used a simple reconstruction model in which fat and water have a single resonance peak. An important advantage of IDEAL is that the signal model can be modified to include multiple fat peaks, so long as the resonance frequencies and relative amplitudes of these peaks are known a priori (27). The purpose of this study was to evaluate and quantify improvements in the quality of fat suppression for fast spin-echo imaging of the knee using multipeak fat spectral modeling and IDEAL fat-water separation.

MATERIALS AND METHODS

The study was approved by our Institutional Review Board and was performed in compliance with HIPAA regulations and with a waiver of informed consent.

Ten knees of five asymptomatic volunteers (three males and two female; age range between 22 and 32 years with an average age of 26 years) were imaged on the same 3.0T MR scanner (HDx, GE Healthcare, Waukesha, WI) using a single-channel quadrature transmit/receive extremity coil. All MR examinations consisted of the following pulse sequences performed in the sagittal plane: 1) T1-weighted fast spin-echo using a frequency selective 90° saturation pulse (fat-selective saturation); 2) T1-weighted fast spin-echo using a frequency selective 110° partial inversion pulse (fat-selective partial inversion); 3) T1-weighted fast spin-echo with IDEAL fat-water separation; 4) T2-weighted fast spin-echo using a frequency selective 90° saturation pulse (fat-selective saturation); 5) T2-weighted fast spin-echo echo using a frequency selective 110° partial inversion pulse (fat-selective partial inversion); and 6) T2-weighted fast spin-echo with IDEAL fat-water separation. The imaging parameters of all sequences are summarized in Table 1.

Table 1. Imaging Parameter for T1-Weighted Fast Spin-Echo and T2-Weighted Fast Spin-Echo Sequences With Fat-Selective Saturation, Fat-Selective Partial Inversion, and IDEAL Fat-Water Separation
Imaging parameterSequence
T1-weighted fast spin-echoT2-weighted fast spin-echo
Fat-selective saturationFat-selective partial inversionIDEALFat-selective saturationFat-selective partial inversionIDEAL
Repetition time850 ms850 ms850 ms5000 ms5000 ms5000 ms
Effective echo time19.3 ms19.3 ms12.7 ms81.0 ms81.0 ms85.0 ms
Field of view14.0 cm14.0 cm14.0 cm14.0 cm14.0 cm14.0 cm
Matrix384 × 224384 × 224384 × 224384 × 224384 × 224384 × 224
Slice thickness3.0 mm3.0 mm3.0 mm3.0 mm3.0 mm3.0 mm
Bandwidth31.25 kHz31.25 kHz31.25 kHz41.67 kHz41.67 kHz41.67 kHz
Excitations331331
Echo train length222161616
Scan time4:48 min4:48 min4:48 min3:35 min3:35 min3:35 min

During reconstruction of the IDEAL T1-weighted and T2-weighted fast spin-echo images, fat-water separation was performed using a conventional single-peak method, a precalibrated multipeak method, and a self-calibrated multipeak method. All fat-water separation methods were performed using the same IDEAL source data. The single-peak IDEAL signal model assumes that fat consists of a single resonance peak separated from the water peak by −420 Hz at 3.0T. The precalibrated and self-calibrated multipeak IDEAL signal models take into account the multiple resonance peaks of fat. The signal model of multipeak IDEAL requires accurate knowledge of the frequencies and relative amplitudes of the multiple peaks in the fat spectrum. The frequencies of the fat peaks can be determined a priori using MR spectroscopy and are considered to be relatively constant for fat-containing tissues of various origins. In precalibrated multipeak IDEAL, the relative amplitudes of the fat peaks are measured using a multiecho scan by fitting the signal to the six most prominent spectral peaks of fat. This spectrum precalibration needs only to be performed once and can be used for fat-water separation on all additional IDEAL acquisitions. In self-calibrated IDEAL, the relative amplitudes of three of the main resonance peaks of fat are measured directly from the source data during each IDEAL acquisition (27).

All MR examinations were transferred to a computer workstation (Advantage Windows, v. 4.2, GE Healthcare) for analysis. For each MR examination the SNR of subcutaneous fat, bone marrow, cartilage, fluid, and muscle were calculated for all sequences. The region of interest (ROI) used for each signal and noise measurement was placed at identical locations on subcutaneous fat, bone marrow, cartilage, fluid, and muscle on the MR images. Each ROI contained ≈50 pixels for signal measurements of subcutaneous fat, bone marrow, and cartilage and ≈20 pixels for signal measurements for fluid. Each ROI contained ≈50 pixels for noise measurements. The standard deviation of the background ROI was used as an estimate of image noise. SNR values were calculated using the following equation:

equation image

The factor 0.655 in Eq. [1] accounts for the fact that the SNR and CNR measurements were obtained using magnitude images (28).

Student t-tests were used to compare SNR measurements for all sequences. Differences in SNR measurements were considered to be statistically significant if the P-value was less than 0.05.

To demonstrate potential clinical advantages of multipeak IDEAL, sagittal T2-weighted fast spin-echo sequences with fat-selective saturation and IDEAL fat-water separation were performed on the knees of 10 patients (seven males and three females; age range between 19 and 35 years with an average age of 22 years) with acute posttraumatic contusions. The sequences were performed on the same 3.0T MR scanner (HDx, GE Healthcare) using an eight-channel phased array transmit/receive extremity coil and the imaging parameters summarized in Table 1. IDEAL fat-water separation was performed using the conventional single-peak method and the self-calibrated multipeak method as previously described.

The sagittal T2-weighted fast spin-echo sequences with fat-selective saturation, single-peak IDEAL, and self-calibrated multipeak IDEAL were independently reviewed by two fellowship-trained musculoskeletal radiologists who had 6 years and 10 years of clinical experience. For each MR examination, the radiologists ranked the images from the various acquisitions and reconstruction methods, with one being the best and three being the worst, according to the following subjective criteria of image quality: 1) quality of fat suppression, 2) overall tissue contrast, and 3) conspicuity of bone marrow edema. Exact binomial tests were used to compare the ranks given to the sequences. Differences between sequences were considered to be statistically significant if the P-value was less than 0.05.

RESULTS

Figures 1 and 2 compare T1-weighted and T2-weighted fast spin-echo images with fat-selective saturation, fat-selective partial inversion, conventional single-peak IDEAL, precalibrated multipeak IDEAL, and self-calibrated multipeak IDEAL. The T1-weighted and T2-weighted fast spin-echo images with precalibrated and self-calibrated multipeak IDEAL had noticeably less signal within subcutaneous fat and bone marrow indicating greater fat suppression.

Figure 1.

Sagittal T1-weighted fast spin-echo images of the knee in an asymptomatic 25-year-old male performed with (a) fat-selective saturation, (b) fat-selective partial inversion, (c) single-peak IDEAL, (d) precalibrated multipeak IDEAL, and (e) self-calibrated multipeak IDEAL. The precalibrated and self-calibrated multipeak IDEAL images have noticeably darker signal within Hoffa's fat pad (arrows) and bone marrow (arrowheads).

Figure 2.

Sagittal T2-weighted fast spin-echo images of the knee in an asymptomatic 25-year-old male performed with (a) fat-selective saturation, (b) fat-selective partial inversion, (c) single-peak IDEAL, (d) precalibrated multipeak IDEAL, and (e) self-calibrated multipeak IDEAL. The precalibrated and self-calibrated multipeak IDEAL images have noticeably darker signal within Hoffa's fat pad (arrows) and bone marrow (arrowheads).

Figure 3a compares SNR values for the T1-weighted fast spin-echo sequences. Precalibrated and self-calibrated multipeak IDEAL had significantly greater suppression of signal (P < 0.05) within subcutaneous fat and bone marrow than fat-selective saturation, fat-selective partial inversion, and single-peak IDEAL. For subcutaneous fat, self-calibrated multipeak IDEAL had 84%, 65%, and 79% greater suppression of signal when compared to fat-selective saturation, fat-selective partial inversion, and single-peak IDEAL, respectively. For bone marrow, self-calibrated multipeak IDEAL had 86%, 66%, and 82% greater suppression of signal when compared to fat-selective saturation, fat-selective partial inversion, and single-peak IDEAL, respectively. Precalibrated and self-calibrated multipeak IDEAL also had significantly higher (P < 0.01) SNR of cartilage, fluid, and muscle than fat-selective saturation and fat-selective partial inversion.

Figure 3.

Mean SNR with standard deviations of subcutaneous fat, bone marrow, cartilage, fluid, and muscle for the (a) T1-weighted fast spin echo sequences and (b) T2-weighted fast spin-echo sequences performed with fat-selective (FS) saturation, fat-selective (FS) partial inversion, single-peak (SP) IDEAL, precalibrated multipeak (MP) IDEAL, and self-calibrated multipeak (MP) IDEAL.

Figure 3b compares mean SNR values for the T2-weighted fast spin-echo sequences. Precalibrated and self-calibrated multipeak IDEAL had significantly greater suppression of signal (P < 0.05) of subcutaneous fat and bone marrow than fat-selective saturation, fat-selective partial inversion, and single-peak IDEAL. For subcutaneous fat, self-calibrated multipeak IDEAL had 81%, 54%, and 66% greater suppression of signal when compared to fat-selective saturation, fat-selective partial inversion, and single-peak IDEAL, respectively. For bone marrow, self-calibrated multipeak IDEAL had 75%, 21%, and 66% greater suppression of signal when compared to fat-selective saturation, fat-selective partial inversion, and single-peak IDEAL, respectively. However, precalibrated and self-calibrated multipeak IDEAL also had significantly lower (P < 0.05) SNR of cartilage, fluid, and muscle than fat-selective saturation and fat-selective partial inversion.

For both T1-weighted and T2-weighted fast spin-echo sequences, self-calibrated multipeak IDEAL had greater suppression of SNR of subcutaneous fat and bone marrow and higher SNR of cartilage, fluid, and muscle than precalibrated multipeak IDEAL. However, differences between precalibrated and self-calibrated multipeak IDEAL were significantly different (P < 0.05) only for SNR of subcutaneous fat and muscle for the T2-weighted fast spin-echo sequences.

On subjective analysis of T2-weighted fast spin-echo sequences performed on 10 patients with acute posttraumatic contusions of the knee, multipeak IDEAL provided significantly better (P < 0.05) fat suppression, overall tissue contrast, and conspicuity of bone marrow edema than fat-selective saturation and single-peak IDEAL (Figs. 4, 5). In one patient with prior anterior cruciate ligament reconstruction surgery, multipeak IDEAL also showed more robust and uniform fat suppression than fat-selective saturation in areas of magnetic susceptibility created by metallic orthopedic hardware (Fig. 6).

Figure 4.

Sagittal T2-weighted fast spin-echo images of the knee in a 21-year-old male with acute posttraumatic contusion performed with (a) fat-selective saturation, (b) single-peak IDEAL, and (c) self-calibrated multipeak IDEAL. Note that the bone marrow edema within the lateral femoral condyle (arrows) and lateral tibial plateau (arrowheads) are more conspicuous on the multipeak IDEAL image than on the fat-selective saturation and single-peak IDEAL images.

Figure 5.

Sagittal T2-weighted fast spin-echo images of the knee in a 26-year-old male with acute posttraumatic contusion performed with (a) fat-selective saturation, (b) single-peak IDEAL, and (c) self-calibrated multipeak IDEAL. Note that the bone marrow edema within the medial femoral condyle (arrows) is more conspicuous on the multipeak IDEAL image than on the fat-selective saturation and single-peak IDEAL images.

Figure 6.

Sagittal T2-weighted fast spin-echo images of the knee in a 27-year-old female with a history of prior anterior cruciate ligament reconstruction surgery performed with (a) fat-selective saturation, (b) single-peak IDEAL, and (c) self-calibrated multipeak IDEAL. Note that the fat suppression in areas of metallic orthopedic hardware (arrows) is more robust and uniform on the single-peak IDEAL and multipeak IDEAL images than on the fat-selective saturation images.

DISCUSSION

Our study has shown that multipeak IDEAL provides superior fat suppression for fast spin-echo imaging of the knee when compared to currently used techniques. In our study, precalibrated and self-calibrated multipeak IDEAL had significantly greater suppression of signal of subcutaneous fat and bone marrow than fat-selective saturation, fat-selective partial inversion, and single-peak IDEAL for both T1-weighted and T2-weighted fast spin-echo sequences. The improved suppression of fat signal by multipeak IDEAL is due to its ability to more accurately model the NMR spectrum of fat. Most currently used fat-suppression techniques consider fat signal as arising from a single species of protons from the methylene group of fatty acids with contribution from smaller species generally neglected. However, Brix et al (6) have shown that fat-suppression techniques which account for only a single fat resonance peak suppress only 85% of signal arising from fat protons. The residual signal from the neglected fat peaks leads to incomplete fat suppression and causes fat to appear “gray” instead of “black” on MR images. Multipeak IDEAL suppresses signal from multiple fat resonance peaks which results in greater fat suppression in tissues such as subcutaneous fat and bone marrow.

A potential disadvantage of more robust fat suppression is loss of signal from desired tissues which may reduce overall image quality. In our study, multipeak IDEAL provided improved fat suppression for T1-weighted fast spin-echo imaging and also had significantly higher SNR of cartilage, fluid, and muscle when compared to fat-selective saturation and fat-selective partial inversion. However, for T2-weighted fast spin-echo imaging, the improvement in fat suppression of multipeak IDEAL was associated with a small but significant reduction in SNR of cartilage, fluid, and muscle. Differences in SNR values between sequences in our study may be explained in part by differences in their effective echo times. The higher SNR of cartilage, fluid, and muscle for the multipeak IDEAL T1-weighted fast spin-echo sequences when compared to the fat-selective saturation and fat-selective partial inversion sequences may be due to their shorter effective echo time. Likewise, the lower SNR of cartilage, fluid, and muscle for the IDEAL T2-weighted fast spin-echo sequences may be secondary to their longer effective echo time. Although we attempted to achieve the same effective echo times for all T1-weighted and T2-weighted fast spin-echo sequences in our study, this was not possible due to the need to shift readout gradients in IDEAL which resulted in increased spacing between refocusing pulses. In addition, the echo times used for the IDEAL sequences were based on the echo combination known to maximize SNR performance when fat is modeled as a single peak (18). The lack of echo time optimization may also be responsible for the reduction in SNR of cartilage, fluid, and muscle for the multipeak IDEAL T2-weighted fast spin-echo sequences. Future work is needed to determine the best echo combination to maximize SNR performance for IDEAL fast spin-echo imaging when using the multipeak fat spectrum model.

In our study, both precalibrated and self-calibrated multipeak IDEAL provided superior fat suppression for T1-weighted and T2-weighted fast spin-echo imaging when compared to fat-selective saturation, fat-selective partial inversion, and single peak IDEAL. However, the self-calibrated method had significantly better performance than the precalibrated method in some SNR categories. The slightly better performance of self-calibrated multipeak IDEAL may be due to the fact that the method takes into account potential changes in the relative amplitudes of the multiple fat resonance peaks from one acquisition to the next. Since the protons in the multiple fat peaks do not have identical relaxation parameters, their relative amplitudes may change due to sequence dependent differences in T1 and T2-weighting and possibly J-coupling effects (28). In self-calibrated multipeak IDEAL, the relative amplitudes of the main fat peaks are measured directly from the IDEAL source data during each acquisition which requires no increase in acquisition time and only a minimal increase in reconstruction time (27).

Our study has demonstrated potential clinical advantages of multipeak IDEAL fat-water separation in musculoskeletal MR imaging. On subjective analysis of T2-weighted fast spin-echo sequences performed on patients with acute posttraumatic contusions of the knee, multipeak IDEAL provided better fat suppression, overall tissue contrast, and conspicuity of bone marrow edema than fat-selective saturation and single-peak IDEAL. Multipeak IDEAL also provided more robust and uniform fat suppression in areas of magnetic field inhomogeneity when compared to fat-selective saturation. This advantage of multipeak IDEAL was demonstrated in one of our patients with metallic orthopedic hardware from prior anterior cruciate ligament reconstruction surgery. However, similar failure of frequency-selective fat-saturation may occur during other musculoskeletal applications due to the sharp geometric variation of the extremities, off-isocenter imaging, and the use of large field of views.

Our study has several limitations. One limitation was that the T1-weighted and T2-weighted fast spin-echo sequences performed during the study did not have identical effective echo times. In addition, the SNR performance of IDEAL fat-water separation was not optimized for multipeak fat spectrum modeling. Another limitation of our study was that multipeak IDEAL fat-water separation was not compared with other currently used fat suppression techniques such as STIR and water excitation. Furthermore, our study did not address whether the superior fat suppression of multipeak IDEAL allows for better visualization of gadolinium contrast uptake on T1-weighted fast spin-echo images.

In summary, our study has shown that multipeak IDEAL provides superior fat suppression for both T1-weighted and T2-weighted fast spin-echo imaging of the knee when compared to single-peak IDEAL and two widely used frequency-selective fat-saturation techniques. Additional studies are needed to determine whether the superior fat suppression provided by multipeak IDEAL can improve the detection of musculoskeletal pathology in clinical practice.

Ancillary