Ultrashort TE Spectroscopic Imaging (UTESI): Application to the Imaging of Short T2 Relaxation Tissues in the Musculoskeletal System

JOURNAL OF MAGNETIC RESONANCE IMAGING 29:412 421 (2009) Original Research Ultrashort TE Spectroscopic Imaging (UTESI): Application to the Imaging of Short T2 Relaxation Tissues in the Musculoskeletal System Jiang Du, PhD, 1 * Atsushi M. Takahashi, PhD, 2 and Christine B. Chung, MD 1 Purpose: To investigate ultrashort TE spectroscopic imaging (UTESI) of short T2 tissues in the musculoskeletal (MSK) system. Materials and Methods: Ultrashort TE pulse sequence is able to detect rapidly decaying signals from tissues with a short T2 relaxation time. Here a time efficient spectroscopic imaging technique based on a multiecho interleaved variable TE UTE acquisition is proposed for high-resolution spectroscopic imaging of the short T2 tissues in the MSK system. The projections were interleaved into multiple groups with the data for each group being collected with progressively increasing TEs. The small number of projections in each group sparsely but uniformly sampled k-space. Spectroscopic images were generated through Fourier transformation of the time domain images at variable TEs. T2* was quantified through exponential fitting of the time domain images or line shape fitting of the magnitude spectrum. The feasibility of this technique was demonstrated in volunteer and cadaveric specimen studies on a clinical 3T scanner. Results: UTESI was applied to six cadaveric specimens and four human volunteers. High spatial resolution and contrast images were generated for the deep radial and calcified layers of articular cartilage, menisci, ligaments, tendons, and entheses, respectively. Line shape fitting of the UTESI magnitude spectroscopic images show a short T2* of 1.34 0.56 msec, 4.19 0.68 msec, 3.26 0.34 msec, 1.96 0.47 msec, and 4.21 0.38 msec, respectively. Conclusion: UTESI is a time-efficient method to image and characterize the short T2 tissues in the MSK system with high spatial resolution and high contrast. Key Words: ultrashort TE; projection reconstruction; spectroscopic imaging; short T2 species; deep layers of cartilage; menisci; ligaments; tendons; entheses. J. Magn. Reson. Imaging 2009;29:412 421. 2009 Wiley-Liss, Inc. 1 Department of Radiology, University of California, San Diego, California. 2 Global Applied Science Laboratory, GE Healthcare, Menlo Park, California. Contract grant sponsor: GE Healthcare. *Address reprint requests to: J.D., University of California, San Diego, Department of Radiology, 407 Dickinson St., San Diego, CA 92103-8756. E-mail: jiangdu@ucsd.edu Received November 14, 2007; Accepted May 9, 2008. DOI 10.1002/jmri.21465 Published online in Wiley InterScience (www.interscience.wiley.com). THE HUMAN MUSCULOSKELETAL (MSK) system contains a variety of tissues with relatively long T2 relaxation components that can be visualized with conventional magnetic resonance imaging (MRI) techniques, as well as many tissues with short T2 components such as the deep radial and calcified layers of articular cartilage, menisci, ligaments, tendons, and entheses that cannot be directly visualized (1 5). These short T2 tissues are crucial in the MSK system. For example, the calcified cartilage is intimately associated with superficial cartilage and subchondral bone and serves as transition tissue between the compliant unmineralized superficial cartilage and the stiffer bone. It helps to prevent large stress concentrations at the interface of these two biomechanically diverse tissues (6,7). Recent studies suggest that changes in the calcified layer could compromise the more superficial portion and cause it to degenerate (6 12). However, the deep radial and calcified layers of cartilage have been virtually unexplored using MRI due to their short T2s, and the inability of conventional clinical pulse sequences to acquire data in this range (5). Knee menisci consist of concentrically and radially arranged collagen fibers that play an important role in absorbing impact load. Clinical sequences can detect little or no signal from normal meniscus. The interpretation of increased signal within the meniscus can be clinically challenging and may be a sign of a tear (1,4). Ligaments are short bands of touch fibrous connective tissue composed mainly of long, stringy collagen fibers, connecting the osseous structures of a joint to afford stability. Tendons are tough bands of fibrous connective tissue attaching muscles to bone, and normally appear as low signal with clinical pulse sequences (1 3). Enthesis are mineralized collagen fibers representing the point at which a tendon inserts into bone. It is highly desirable to qualitatively and quantitatively evaluate these short T2 tissues under a clinical MR system. Chemical shift imaging (CSI) or spectroscopic imaging (SI) is a technique that combines acquisition of spectral and spatial information in a single scan (13,14). In conventional CSI the spatial information is obtained through phase encoding, which is followed by free induction decay (FID) sampling to generate spectroscopic information. This spatial encoding scheme is time-consuming and makes it difficult to generate high- 2009 Wiley-Liss, Inc. 412

Ultrashort TE Spectroscopic Imaging 413 resolution images in vivo. Various fast spectroscopic imaging techniques have been proposed, including spectroscopic FLASH (SFLASH) imaging, spectroscopic GRASE, fast gradient echo CSI, echo planar SI, multiple spin echo SI, spectroscopic RARE, and multiecho outand-in spiral sampling (15 19). These techniques provide higher signal-to-noise ratio (SNR) or spatial resolution for long T2 species with shorter scan times. However, spatial resolution is still limited, and the techniques are not suitable for short T2 tissues. It is highly desirable to generate spectroscopic images of the short T2 tissues with high spatial resolution, high spectral resolution, and broad spectral bandwidth coverage. Projection reconstruction is more efficient in achieving high spatial resolution per unit time than Cartesian imaging (21). Spectral resolution can be increased by acquiring more images with longer TE delay. Another option to increase spectral resolution is to acquire multiecho images combined with variable TE delay, as is used in spiral spectroscopic imaging (20). The total scan time can be reduced through angular undersampling without spatial resolution degradation, which has been extensively investigated in contrast-enhanced time-resolved MR angiography (22 26). The combination of these techniques may provide a novel approach for high-resolution spectroscopy imaging of short T2 tissues. In this study we devised a new approach termed ultrashort TE spectroscopic imaging (UTESI) for tissues with short T2s. We show that the combination of highly undersampled interleaved projection reconstruction with a multiecho UTE acquisition at progressively increasing TEs is able to provide high spatial resolution spectroscopic imaging of short T2 tissues in the MSK system, including the deep radial and calcified layers of cartilage, menisci, ligaments, tendons, and entheses. T2* was quantified through exponential signal decay fitting of the multiecho images, or line shape fitting of the magnitude UTESI images. This UTESI technique was validated through cadaveric specimen and in vivo healthy volunteer studies. MATERIALS AND METHODS We previously demonstrated that spectroscopic imaging of cortical bone can be realized through UTE acquisition at variable TE delays (27), where a single slice was imaged with a single half pulse and a single echo, or free induction decay (FID). In this study we extended the original UTESI to multislice with multiecho acquisition and double half pulse excitation. Figure 1 shows the revised UTESI pulse sequence, which is based on a two-dimensional (2D) UTE pulse sequence and employs a double half radiofrequency (RF) pulse for signal excitation followed by radial ramp sampling (28 32). Radial ramp sampling was started immediately after the end of the half RF pulse. Up to four echoes with an echo spacing ( TE) of 4 6 msec were collected after each halfpulse excitation. These were delayed progressively with a delay time ( t) of 120 300 s. For each radial k-space line, two acquisitions with reversed slice selection gradient polarity were sampled and summed to form a selective 2D excitation (33). Figure 1. The UTE spectroscopic imaging pulse sequence employs double half pulses (separated by separation time T) for slice selective excitation followed by radial ramp sampling at variable TEs, along with a minimal achievable TE of 8 s. Single free induction decay (FID) or multiple gradient echoes were sampled with variable TE delays to generate spectroscopic information. The UTE sequence is sensitive to eddy currents that may cause out-of-slice signal excitation (34,35). Here a double half pulse rather than a conventional single half pulse was employed to improve slice profile of the long T2 components (32), which experience both half pulses resulting in a conventional full pulse excitation. The short T2 signal mainly comes from the second half pulse, since the excitation from the first half pulse is attenuated by the separation time (T sep ) between the two half pulses. It still requires the summation of two acquisitions with reversed slice selection gradients to form a complete slice excitation for short T2 tissues. The double half pulse technique was first implemented by Josan et al (32) to improve half pulse slice profile for more accurate T2* measurement and temperature mapping. In Josan et al s initial approach, gradient trapezoids with areas of 1, 2, 1 were used. RF was played during the first up-ramp and flattop and during the last flattop and down-ramp with the RF waveforms reversed in time. In our double half pulse design, the same half pulse RF waveform was repeated twice with the bipolar trapezoidal pair of slice selection gradient reversed for the second half pulse. In this way the eddy currents for the first and second half pulses will be similar but opposite, resulting in reduced eddy currents and improved slice profile. The radial projections can be undersampled to reduce the total scan time without spatial resolution degradation (23 27). In the UTESI sequence the radial half projections were highly undersampled and interleaved for each TE. Figure 2 shows the acquisition scheme. The whole set of projections were interleaved into multiple groups with each group at a progressively increasing TE delay which was a multiple of t. For multiecho acquisition, all the echoes were shifted simultaneously by t while keeping the echo spacing constant at TE. The small number of projections in each group sparsely but uniformly sampled the k-space. Since the undersampling streaks are in alignment with the half projections, the oscillation pattern of the streaks can be controlled by adjusting the interleaving scheme. For

414 Du et al. Figure 2. Interleaved variable TE UTE acquisition scheme divided the whole sets of half projections into multiple groups with each group of half projections sparsely but uniformly covering k-space with TE successively delayed by t, while echo spacing TE was kept constant. Here TE ij refers to the echo time for interleave i at echo j. simplicity, let us consider nine interleaved groups with each group having 45 projections sparsely but uniformly covering k-space. The nine interleaved groups of half projections can be sampled in the following way: 1, 4, 7, 2, 5, 8, 3, 6, and 9. In this way the high frequency projection data from neighboring interleaved groups uniformly covers the periphery of k-space, and can be shared to reduce streak artifact (24 27). Meanwhile, the undersampling streaks oscillate periodically every three groups, simulating signals with high temporal frequency. Spectroscopic images were generated through Fourier transformation of the time domain images, which will shift all the streak artifacts to high temporal frequencies, leaving streak artifact-free images around the water peak (31). Institutional Review Board permission was obtained for this study. UTE spectroscopic imaging was performed on six cadaveric specimens and four asymptomatic volunteers. The fresh human knees and ankles were harvested from nonembalmed cadavers. The specimens were immediately deep-frozen at 40 C (Forma Bio-Freezer; Forma Scientific, Marietta, OH). The specimens were then allowed to thaw for 36 hours at room temperature prior to imaging. A 3-inch coil was used for cadaveric specimens. A quadrature knee coil was used for signal reception in volunteer studies. Typical acquisition parameters include: field of view (FOV) of 14 16 cm for volunteers and 10 cm for cadaveric samples, TRs of 60 200 msec, an initial TE of 8 s, and a TE delay of 120 300 s thereafter, one to four echoes with an echo spacing of 4 6 msec, flip angle of 40 60, bandwidth of 62.5 khz, readout of 512 (actual sampling points 278), 3 8 slices, slice thickness of 2 3 mm, 1980 2025 projections interleaved into 45 72 groups. The total scan time was 8 13 minutes. The oblique sagittal plane was used to evaluate the calcified layer of cartilage in the femorotibial joint. The oblique coronal plane was used to image the lateral collateral ligament. The oblique sagittal and axial planes were used to interrogate the Achilles tendon and enthesis. MR images were reviewed by a subspecialized musculoskeletal radiologist to identify region of interest (ROI) placement in normal appearing calcified layer of the femorotibial joint, the ligament, the meniscus, the Achilles tendon, and enthesis. Normal calcified layer cartilage was identified by its location just superficial to subchondral bone, its thin linear morphology and bright signal intensity. Normal-appearing meniscus was identified by its triangular morphology in the sagittal plane and the absence of any regions of superimposed linear signal that extended to an articular surface. Normal-appearing ligament, Achilles tendon, and enthesis were defined by its uniform signal and linear morphology. Raw data were transferred to a Linux computer for offline image reconstruction after the UTE spectroscopic imaging data acquisition. The half projection data at each TE was regridded onto a 512 512 matrix followed by 2D fast Fourier transformation. Then the complex images at multiple TEs (45 144) were zeropadded to 512, yielding a spectroscopic imaging series with a matrix size of 512 512 512 after Fourier transformation in the time domain. Voxel or ROI-based spectra were generated by simply plotting the signal intensity across the 512 spectroscopic images. T2* was derived through exponential fitting of the UTE images at progressively increasing TEs, or line shape fitting of the magnitude spectra using the following equation (27): S r,f s 0 r [1] 2 4 f f T 2 * 0 2 1 where f 0 is the peak resonance frequency, r is the location in image space, s 0 r is the effective observable MR signal, and s r,f is the MR signal intensity at location r and resonance frequency f. To evaluate the accuracy of the T2* estimation using time domain exponential fitting and spectral domain line shape fitting, UTE images at progressively increasing TEs were acquired with full sampling for each TE. In total, 811 projections were acquired with a readout matrix of 512 and TEs ranging from 8 s to 10 msec. Exponential fitting was performed to calculate T2* values, which were used as a reference standard in evaluating the accuracy of UTESI T2* quantification. Descriptive statistical analysis was performed to determine average T2* values in all cadaveric specimens and asymptomatic volunteers, interspecimen variation, and variation between the UTESI and standard UTE T2* quantification using constant TR and varying TEs. RESULTS Figure 3 shows UTE spectroscopic images of the femorotibial articular cartilage and meniscus from a healthy volunteer with a quadrature knee coil for signal reception. The imaging FOV of 16 cm, readout of 512, and 3 mm slice thickness resulted in an acquired voxel size of 0.3 0.3 3.0 mm 3, providing excellent depiction of the knee structure such as the superficial layers of cartilage, deep radial, and calcified layers of cartilage and meniscus. The deep layers of cartilage appear bright over a broad range of spectrum, consistent with their short T2 relaxation time. Fat signal is shifted to

Ultrashort TE Spectroscopic Imaging 415 Figure 3. Selected UTE spectroscopic images of the human knee from a healthy volunteer show clear definition of the deep layers of cartilage (thin arrows), superficial cartilage, meniscus (thick arrows), as well as excellent fat water separation. 420 Hz at 3T, suggesting that UTESI provides accurate fat water separation. The undersampling streak artifact was shifted to high spectral frequencies, leaving streak artifact-free images near the water resonance frequencies. Figure 4 shows the water peak images from two of eight slices. Both the articular cartilage and meniscus are depicted with high spatial resolution, high SNR, and excellent fat suppression without any streak artifact. However, the deep layers of cartilage are not well depicted due to the poor contrast over the superficial layers of cartilage. UTESI was also performed with fat signal suppression using a long duration Gaussian pulse focused on fat resonance frequency. The image dynamic range was increased, providing high contrast for the cartilage and meniscus in the time domain image series, as shown in Fig. 5. The undersampling streak artifact is significantly reduced due to the view sharing reconstruction strategy (31). However, the contrast between the superficial layers and deep layers of cartilage is quite limited. Fourier transformation of these time-domain image series provides excellent spectroscopic images shown in Fig. 6. The deep layers of cartilage (thin arrow) and Figure 5. Selected time-domain UTESI images of the knee with fat suppression provide excellent depiction of the articular cartilage and meniscus (thick arrow). There is little contrast between the deep layers of cartilage (short thin arrow) and superficial layers of cartilage (long thin arrow). Streak artifact was suppressed by sharing high frequency projection data from neighboring interleaved groups during the image reconstruction. meniscus (thick arrow) appear bright over a broad range of spectrum, suggesting its short T2 relaxation time. Maximal contrast was achieved for the deep layers of cartilage at around 200 Hz away from the water Figure 4. Selected water peak images from two of eight slices show excellent depiction of the articular cartilage (thin arrows) and meniscus (thick arrows) with a high spatial resolution of 0.3 0.3 3mm 3 and high contrast with excellent fat signal suppression. But there is little contrast between the deep layers and superficial layers of cartilage. Figure 6. Selected spectral domain UTESI images of the knee show excellent definition of the deep layers of cartilage (short thin arrow), superficial layers of cartilage (long thin arrow), and menisci (thick arrow). The deep layers of cartilage have much shorter T2 than the superficial layers of cartilage, creating increased contrast from 0 Hz up to 200 Hz. The streak artifact was shifted to high spectral frequencies due to the interleaved acquisition scheme (the first and last images are rescaled in order to show the streaks better).

416 Du et al. Figure 7. Line shape fitting of the magnitude UTE spectra of the deep layers of cartilage (a) and meniscus (b) show a short T2* of 1.39 0.25 msec and 4.96 0.35 msec, respectively. Exponential decay fitting of the multiple echo images of the deep layers of cartilage (c) and meniscus (d) shows similar values of 1.24 0.19 msec and 4.64 0.23 msec, respectively. peak resonance frequency, where the superficial layers of cartilage appear dark due to its long T2 relaxation time and narrow spectrum. The spectroscopic images near the peak resonance frequency did not show any streak artifact that was shifted to high spectral frequencies due to the interleaved acquisition scheme. Figure 7 shows typical UTE spectra from a small ROI (3 pixels) drawn in the deep layers of cartilage and a large ROI (100 pixels) in meniscus, respectively. Line shape fitting of the magnitude UTE spectrum of the deep layers of cartilage shows a short T2* of 1.39 0.25 msec, which was comparable with the value of 1.24 0.19 msec derived from exponential signal decay fitting in the time domain. There is a significant signal fluctuation in the signal decay curve mainly due to the residual undersampling streak artifact. Line shape fitting of the magnitude UTE spectrum of the meniscus shows a short T2* of 4.96 0.35 msec. Exponential signal decay fitting shows a similar T2* value of 4.64 0.23 msec. Figure 8 shows UTESI images of a meniscus sample in the time domain (upper row) and spectral domain (lower row). The imaging FOV of 10 cm, readout of 512, and 2 mm slice thickness resulted in a high spatial resolution of 0.2 0.2 2.0 mm 3 (acquired voxel size), providing excellent depiction of the meniscus structure. Figure 8. Selected UTE images of a meniscus in vitro in the time domain (upper row) and spectral domain (lower row) show excellent depiction of the meniscus structure with a high spatial resolution of 0.2 0.2 2.0 mm 3 (acquired), a high spectral resolution (29 Hz) and broad spectral bandwidth coverage of 5 khz. Figure 9 shows the corresponding exponential signal decay curve and spectra from a small ROI. There are some Gibbs ring artifacts at high spectral frequencies that can be suppressed by increasing spectral resolution or applying lowpass filtering (20). The latter approach is typically used in conventional spectroscopy. However, this filtering also leads to line broadening, increasing errors in T2 quantification through line shape fitting, and thus not used in our analysis. Excellent line shape fitting was achieved for this UTE spectrum, providing a T2* of 3.69 0.16 msec, which is slightly larger than the value of 3.58 0.13 msec derived through exponential signal decay fitting. Figure 10 shows axial UTESI images of the Achilles tendon in a cadaveric ankle specimen with a 3-inch coil for signal reception. The imaging FOV of 10 cm, readout of 512, and 2 mm slice thickness resulted in a high spatial resolution of 0.2 0.2 2.0 mm 3 voxel size, providing excellent depiction of the tendon structure, including the fascicular pattern. Fat signal was shifted 456 Hz away from the water peak, providing robust fat suppression in the tendon peak images. Figure 11 shows the sagittal UTESI images of the same cadaveric sample. Again, the tendon structure was demonstrated with high spatial resolution, high SNR without any streak artifact, and fat signal contamination. Figure 12 shows T2* evaluation using three approaches. The first approach is based on single component exponential signal decay fitting of fully sampled fat saturated UTE images at a series of TEs ranging from 0.1 10 msec (Fig. 12a), and shows a short T2* of 1.56 0.06 msec. The second approach is based on two components (fat and tendon) exponential signal decay fitting of the UTESI images in time-domain, and shows a comparable short T2* of 1.59 0.15 msec. The third approach employs line shape fitting of the magnitude UTE spectrum from a small ROI drawn in tendon, and shows a short T2* of 1.66 0.24 msec.

Ultrashort TE Spectroscopic Imaging 417 Figure 9. T2* estimation of the meniscus sample using exponential signal decay fitting of the UTE images at variable TEs and line shape fitting of the magnitude spectrum show comparable results of 3.58 0.13 msec and 3.69 0.16 msec, respectively. Figure 13 shows sagittal UTESI imaging of the enthesis in a cadaveric ankle specimen with a 3-inch coil for signal reception. The imaging FOV of 10 cm, readout of 512, and 2 mm slice thickness resulted in a high spatial resolution of 0.2 0.2 2.0 mm 3 voxel size, providing excellent depiction of the enthesis structure. All the streak artifacts were shifted to high temporal frequencies, leaving high contrast enthesis images near the water peak. Figure 14 shows the magnitude UTESI spectra from a small ROI drawn in the enthesis. Line shape fitting shows that enthesis has a short T2* of 4.63 0.26 msec. Figure 15 shows coronal UTESI imaging of the lateral collateral ligament in the knee joint from a 30-year-old healthy male volunteer with a 3-inch coil for signal reception. The imaging FOV of 14 cm, readout of 512, and 2 mm slice thickness resulted in a high spatial resolution of 0.27 0.27 2.0 mm 3 voxel size, providing excellent depiction of the ligament structure (short arrows). Meanwhile, the deep layers of articular cartilage are also well depicted (long arrows). Line shape fitting of the UTESI spectra from a small ROI drawn in the ligament shows that lateral collateral ligament has a short T2* of 3.44 0.18 msec. Table 1 summarizes the mean and variation in T2* measurements for the deep layers of articular cartilage, menisci, ligaments, tendons, and entheses from four healthy volunteers and six cadaveric specimens. All these tissues have a short T2* of about 1 4 msec, providing little or low signal with conventional MR pulse sequences. DISCUSSION It has been demonstrated that the UTESI technique is capable of producing high-resolution images of species with short T2 relaxation time. Examples shown here include the deep radial and calcified layers of articular cartilage, which has never been imaged before with conventional pulse sequences. Also included are the tendons and menisci from both healthy volunteers and cadaveric samples, as well as ligaments, tendons, and entheses. The high-quality images of these short T2 components were generated mainly due to the use of a minimal TE of 8 s achieved through the combination of half pulse excitation, VERSE, radial ramp sampling, and fast T/R switching (29 31). This TE is significantly shorter than all previously reported sequences implemented on clinical MR systems, leading to significantly reduced signal decay and increased SNR in both timedomain images and spectroscopic images. This single digit TE also minimizes the undesirable baseline distortion which is a significant challenge in conventional CSI based on FID acquisition method (13 20). UTESI provides a spatial resolution on the order of 0.08 0.27 mm 3, which is much higher than that of the conventional spectroscopic imaging techniques (typically tens Figure 10. Selected UTESI images of a cadaveric ankle specimen in the axial plane show excellent depiction of the Achilles tendon with a high spatial resolution of 0.2 0.2 2.0 mm 3 (acquired). Fat signals peaked at 456 Hz, leaving excellent image contrast for the Achilles tendon around the water peak. Figure 11. UTESI imaging of a cadaveric ankle specimen in the sagittal plane shows excellent depiction of the tendon with accurate fat water separation. The fibrous tissues in tendon are also well depicted (arrow).

418 Du et al. Figure 12. T2* estimation of Achilles tendon using single component exponential signal decay fitting of fully sampled UTE images with fat suppression at progressively increasing TEs (a), two components exponential signal decay fitting of the UTESI images without fat suppression in the time domain (b) and line shape fitting of the magnitude spectrum (c). Comparable T2* values were derived at 1.56 0.06 msec, 1.59 0.15 msec, and 1.66 0.24 msec, respectively. of mm 3 ). This small voxel size may lead to a significant increase in field homogeneity and T2* over the lowresolution acquisitions and lead to a less than linear loss in SNR with increasing resolution. Furthermore, the interleaved variable TE UTE acquisition strategy significantly improved the spectral resolution and spectral bandwidth by acquiring a relatively large number of images at variable TEs within clinically acceptable scan times. The highly undersampled acquisition produces strong streak artifact at each TE. However, these streaks oscillate at variable TEs, simulating a high temporal frequency signal and are thus shifted to high spectral frequencies. As shown in Figs. 3 12, all the UTE spectroscopic images show minimal or no streaks near the water resonance frequencies. UTESI not only provides high-quality images in both the time domain and spectral domain, it also provides fast estimation of T2* relaxation time. T2* values can be derived through exponential fitting of the UTE images at multiple echo times, as shown in Figs. 7, 9, and 12. The other approach is line shape fitting of the UTE spectroscopic images. We did not employ the conventional Lorentzian line shape fitting of the real spectra, which is susceptible to line broadening and distortion due to field inhomogeneity, susceptibility, eddy currents, and gradient anisotropy. A variety of algorithms have been proposed to correct these phase errors and line shape distortion, which would be quite time-consuming since UTESI contains a large number of spectra (here we have 512 512 262,144 spectra) (37,38). Further, it would be challenging to correct all the phase errors and line broadening effects since UTE imaging is especially sensitive to these errors due to its half pulse excitation and radial ramp sampling scheme (34,35). Instead, a modified Lorentzian line shape fitting shown in Eq. [1] was applied to the magnitude spectra, which Figure 13. UTE spectroscopic imaging of a cadaveric ankle specimen in the sagittal plane shows excellent depiction of the entheses (arrows). Figure 14. Line shape fitting of the magnitude spectrum shows a relatively short T2* of 4.63 0.26 msec for the entheses.

Ultrashort TE Spectroscopic Imaging 419 Figure 15. UTE spectroscopic imaging of the knee of a 30-year-old male volunteer in the oblique coronal plane shows excellent depiction of the lateral collateral ligament (short arrow) and deep layers of articular cartilage (long arrow). eliminates all the phase errors. T2* values generated through line shape fitting of the magnitude spectrum is almost completely unaffected by the undersampling streak artifact. Its accuracy in T2* estimation has been demonstrated through comparison with fully sampled UTE acquisition at progressively increasing TEs, as shown in Fig. 12. Therefore, UTESI is a comprehensive technique that provides a wealth of information including time domain images with high spatial resolution and SNR, spectroscopic domain images with moderate spectral resolution, and accurate fat/water separation, as well as fast estimation of T2* relaxation times. Gold et al (1) proposed a projection reconstruction acquisition combined with half RF pulse excitation and variable TE delays to generate imaging and spectroscopic information of the short T2 tissues in the knee. A minimum TE of 200 s was used with the capacity to detect signals from tendon and menisci. Four to eight spectral interleaves were generated with a total scan time of 8 minutes and a relatively high spatial resolution of 1 7.8 mm 3, a moderate spectral resolution of 61 120 Hz, and limited bandwidth coverage of 1330 Hz. Compared with the original spectroscopic imaging technique proposed by Gold et al, the UTESI technique provides a significant higher spatial resolution, spectral resolution, and spectral bandwidth coverage (1). Voxel size was reduced from 0.5 0.5 3.5 mm 3 in the old approach down to 0.3 0.3 3.0 to 0.2 0.2 2.0 mm 3 in the new approach. The number of acquired multiple echo images increased from 4 to 8 in the old approach, to 45 to 144 in the new approach, providing more detailed spectroscopic images. The improved image quality in UTESI is partly due to the higher field strength (3.0T over 1.5T), and stronger gradient system (4.0 G/cm vs. 2.2 G/cm), as well as the increased chemical shift difference between water and fat (440 Hz at 3.0T vs. 220 Hz at 1.5T) (39). Normal collagen-containing structures such as the deep layers of cartilage, meniscus, ligament, tendon, and enthesis appear as signal voids on conventional MR pulse sequences, and exhibit signal only with increased T2 relaxation times due to injury of degenerative changes or magic angle effects (1 4). UTESI allows direct imaging and quantification of these short T2 tissues, thus providing a novel approach for detection of earlier or more subtle changes. UTESI has great potential for evaluating the short T2 tissues in the MSK system. It provides excellent fat water separation, producing water only images with high spatial resolution, as shown in Figs. 3 6. Conventional fat suppression techniques based on a 90 saturation pulse centered on fat resonance frequency typically provide imperfect suppression of fat signals due to B 1 inhomogeneity, as well as off-resonance due to B 0 field inhomogeneity, shimming, and susceptibility. UTESI generates images at a series of resonance frequencies, therefore providing robust fat water separation. UTESI allows short T2 imaging without long-t2 suppression pulses that may partly saturate short T2 signals. High contrast can be generated for short T2 tissues, such as the deep layers of cartilage, at certain ranges of off-resonance frequencies where water and fat have very low signal because of their long T2 and narrow spectrum, while the short T2 tissues still have relatively high signal because of their short T2 and broad spectrum. However, efficient suppression of long T2 tissues is still helpful in increasing the dynamic range since many short T2 tissues, especially cortical bone, have a proton density much lower than that of the surrounding long T2 tissues, such as muscle and fat (27). The anti-aliasing property of radial sampling allows small FOV imaging without aliasing artifact, which is another advantage of UTESI over the conventional Cartesian imaging. Furthermore, UTESI has potential for fat quantification since both fat and water spectra are generated for each voxel. The integration of the fat peak and water gives an estimation of fat content as well as fat distribution. Table 1 T2* Relaxation Times for the Deep Radial and Calcified Layers of Articular Cartilage, Menisci, Ligaments, Tendons, and Entheses Measured Through Line Shape Fitting of the Magnitude UTESI Spectra Tissues Deep Radial and Calcified Cartilage Menisci Ligaments Tendons Entheses T2* (ms) 1.34 0.56 4.19 0.68 3.26 0.34 1.96 0.47 4.21 0.38

420 Du et al. Table 1 shows quite a big variation in T2* quantification of the short T2 tissues in the MSK system, especially for the deep layers of articular cartilage, which are very thin ( 100 m) and subject to significant partial volume effect. Magic angle effect further complicates the T2* measurement. Collagen fibers in menisci, ligaments, tendons, and entheses are highly ordered. The protons within the bound water are subject to dipolar interactions whose strength depends on the orientation of the fibers to the static magnetic field B 0. These dipolar interactions cause a rapid dephasing of the MR signal, and are dependent on 3cos 2-1, where is the angle of the fibers relative to B 0. When 3cos 2-1 0, or 55, 125, etc, dipolar interactions are minimized, resulting in an increase of T2 as well as MR signal. Fullerton et al (40) reported an increase in T2 of Achilles tendon from 0.6 at 0 to B 0 to 22 msec at 55. Henkelman et al (41) described an increase from 7 to 23 msec when the orientation of the tendon to B 0 was changed from 0 to 55. All these studies were performed on a small bore spectrometer using small tendon samples. UTESI sequence is able to depict signal from short T2 tissues, enabling magic angle imaging and quantification of the whole knee of cadaveric sample or healthy volunteers/patients on a clinical system. Our next step will focus on a systematic study of the magic angle effect using UTESI, and measure their baseline T2* values (at 0 C). One of the major technical challenges for UTESI is related to gradient distortion (34 36). UTE sequences employ two half pulse excitations with opposite slice selective gradient polarities (33,42). Distortion in slice selection gradient results in slice profile broadening. The double half pulse scheme employed here improves the slice profile for long T2 tissues, but the short T2 tissues still suffer from slice profile broadening (32). Distortion in readout gradients results in sampling trajectory errors, leading to suboptimal image reconstruction. Errors can be reduced by measuring the slice selection gradient followed by precompensation, and measuring the readout gradient and using it for regridding (34 36). There are some Gibbs ring artifacts in the UTE spectrum, which can be suppressed by applying a temporal filter. MR spectroscopy typically employs a Hanning filtering and zero filling before FFT in the time domain, which is effective in improving the perceived spectral resolution and reducing noise level. However, Hanning filtering also affects the spectral line shape, therefore introducing some errors in T2 estimation based on line shape fitting. We did not therefore apply any temporal filtering. Furthermore, signal from short T2 tissues decays very rapidly. UTESI appears to be less affected by Gibbs ringing than long T2 spectroscopic imaging. Figure 12 confirms that the Gibbs ringing artifacts do not significantly affect the accuracy in T2 quantification. The UTESI technique is also able to estimate T2 of long T2 tissues, such as these from the superficial layers of cartilage. But a longer TE delay or more echoes are desirable for more accurate estimation. A conventional pulse sequence may be more suitable since ultrashort TE based on half pulse excitation is not necessary for long T2 tissues, and thus eliminates the scan time penalty due to two NEX acquisitions associated with half pulse excitation. In conclusion, spectroscopic imaging of the short T2 tissues in the MSK system can be generated using the novel UTESI sequence, which employs a multislice multiecho UTE acquisition combined with an interleaved variable TE acquisition scheme. This technique provides images with high spatial resolution and moderate spectral resolution, together with T2* estimation and robust fat water separation in a single scan. ACKNOWLEDGMENT The authors thank Graeme Bydder for helpful discussion. REFERENCES 1. Gold GE, Pauly JM, Macovski A, Herfkens RJ. MR spectroscopic imaging of collagen: tendons and knee menisci. Magn Reson Med 1995;34:647 654. 2. Robson MD, Benjamin M, Gishen P, Bydder GM. Magnetic resonance imaging of the Achilles tendon using ultrashort TE (UTE) pulse sequences. Clin Radiol 2004;59:727 735. 3. Fullerton GD, Rahal A. Collagen structure: the molecular source of the tendon magic angle effect. J Magn Reson Imaging 2007;25:345 361. 4. Gatehouse PD, He T, Puri BK, Thomas RD, Resnick D, Bydder GM. Contrast-enhanced MRI of the menisci of the knee using ultrashort echo time (UTE) pulse sequences: imaging of the red and white zones. Br J Radiol 2004;77:641 647. 5. Du J, Sinha S, Takahashi AM, et al. Imaging of the deep radial and calcified layers of the cartilage using ultrashort TE (UTE) sequence at 3T. In: Proc 15th Annual Meeting ISMRM, Berlin, 2007; p 3813. 6. Ferguson VL, Bushby AJ, Boyde A. Nanomechanical properties and mineral concentration in articular calcified cartilage and subchondral bone. J Anat 2003;203:191 199. 7. Li BH, Marshall D, Roe M, Aspden RM. The electron miscroscope appearance of the subchondral bone plate in the human femoral head in osteoarthritis and osteoporosis. J Anat 1999;195:101 110. 8. Burr DB. Anatomy and physiology of the mineralized tissues: role in the pathogenesis of osteoarthrosis. Osteoarthritis Cartilage 2004; 12:S20 30. 9. Martel-Pelletier J. Pathophysiology of osteoarthritis. Osteoarthritis Cartilage 2004;12:S31 33. 10. Muir P, McCarthy J, Radtke CL, et al. Role of endochondral ossification of articular cartilage and functional adaptation of the subchondral plate in the development of fatigue microcracking of joints. Bone 2006;38:342 349. 11. Squires GR, Okouneff S, Ionescu M, Poole AR. The pathobiology of focal lesion development in aging human articular cartilage and molecular matrix changes characteristic of osteoarthritis. Arthritis Rheum 2003;48:1261 1270. 12. Donohue JM, Buss D, Oegema TR, Thompson RC. The effects of indirect blunt trauma on adult canine articular cartilage. J Bone Joint Surg Am 1983;65:948 957. 13. Brown TR, Kincaid BM, Ugurbil K. NMR chemical shift imaging in three dimensions. Proc Natl Acad Sci USA1982;79:3523 3526. 14. Maudsley AA, Hilal SK, Perman WH, Simon HE. Spatially resolved high resolution spectroscopy by four-dimensional NMR. J Magn Reson 1983;51:147 152. 15. Haase A, Matthaei D. Spectroscopic FLASH NMR imaging (SPLASH imaging). J Magn Reson 1987;71:550 553. 16. Park HW, Kim YH, Cho ZH. Fast gradient-echo chemical-shift imaging. Magn Reson Med 1988;7:340 345. 17. Guilfoyle DN, Blamire A, Chapman B, Ordidge RJ, Mansfield P. PEEP a rapid chemical shift imaging method. Magn Reson Med 1989;10:282 287. 18. Dreher W, Leibfritz D. A new method for fast proton spectroscopic imaging: spectroscopic GREASE. Magn Reson Med 2000;44:668 672. 19. Duyn JH, Moonen CTW. Fast proton spectroscopic imaging of human brain using multiple spin-echoes. Magn Reson Med 1993;30: 409 414.

Ultrashort TE Spectroscopic Imaging 421 20. Hiba B, Faure B, Lamalle L, Decorps M, Ziegler A. Out-and-In spiral spectroscopic imaging in rat brain at 7 T. Magn Reson Med 2003; 50:1127 1133. 21. Joseph PM, Whitley J. Experimental simulation evaluation of ECGgated heart scans with a small number of views. Med Phys 1983; 10:444 449. 22. Peters DC, Korosec FR, Grist TM, et al. Undersampled projection reconstruction applied to MR angiography. Magn Reson Med 2000; 43:91 101. 23. Du J, Carroll TJ, Wagner HJ, et al. Time-resolved, undersampled projection reconstruction imaging for high resolution CE-MRA of the distal runoff vessels. Magn Reson Med 2002;48:516 522. 24. Du J, Carroll TJ, Brodsky E, et al. Contrast enhanced peripheral magnetic resonance angiography using time-resolved vastly undersampled isotropic projection reconstruction. J Magn Reson Imaging 2004;20:894 900. 25. Song HK, Dougherty L. Dynamic MRI with projection reconstruction and KWIC processing for simultaneous high spatial and temporal resolution. Magn Reson Med 2004;52:815 824. 26. Mistretta CA, Wieben O, Velikina J, et al. Highly constrained backprojection for time-resolved MRI. Magn Reson Med 2006;55:30 40. 27. Du J, Hamilton G, Takahashi AM, Bydder M, Chung CB. Ultrashort TE spectroscopic imaging (UTESI) of cortical bone. Magn Reson Med 2007;58:1001 1009. 28. Robson MD, Gatehouse PD, Bydder M, Bydder GM. Magnetic resonance: an introduction to ultrashort TE (UTE) imaging. J Comput Assist Tomogr 2003;27:825 846. 29. Brittain JH, Shankaranarayanan A, Ramanan V, et al. Ultra-short TE imaging with single-digit (8 microsecond) TE. In: Proc 12th Annual Meeting ISMRM, Kyoto, 2004; p 629. 30. Takahashi AM, Lu A, Brittain JH, et al. Ultrashort TE (UTE) imaging at 8 sec with 3D vastly undersampled isotropic projection reconstruction (VIPR). In: Proc 13th Annual Meeting ISMRM, Miami, 2005; p 2405. 31. Du J, Hamilton G, Takahashi AM, Bydder M, Hinks S, Bydder GM. Ultrashort TE (UTE) spectroscopic imaging of cortical bone using a variable TE acquisition and sliding window reconstruction. In: Proc 15th Annual Meeting ISMRM, Berlin, 2007; p 421. 32. Josan S, Lu A, Pauly J, Daniel B, Butts K. Double half RF pulse for reduced sensitivity to linear eddy currents in ultrashort T2 imaging. In: Proc 14th Annual Meeting ISMRM, Seattle, 2006; p 3004. 33. Conolly S, Nishimura D, Macovski A, Glover G. Variable-rate selective excitation. J Magn Reson 1988;78:440 458. 34. Lu A, Daniel BL, Pauly KB. Improved slice excitation for ultrashort TE imaging with B 0 and linear eddy current correction. In: Proc 14th Annual Meeting ISMRM, Seattle, 2006; p 2381. 35. Wanspaura JP, Daniel BL, Pauly JM, Butts K. Temperature mapping of frozen tissue using eddy current compensated half excitation RF pulses. Magn Reson Med 2001;46:985 992. 36. Du J, Lu A, Block WF, Thornton FJ, Grist TM, Mistretta CA. Timeresolved undersampled projection reconstruction MR imaging of the peripheral vessels using multi-echo acquisition. Magn Reson Med 2005;53:730 734. 37. de Graaf AA, van Dijk JE, Bovee WM. QUALITY: quantification improvement by converting lineshapes to the lorentzian type. Magn Reson Med 1990;13:343 357. 38. Stoyanova R, Kuesel AC, Brown TR. Application of principal-component analysis for NMR spectral quantification. J Magn Reson Series A 1995;115:265 269. 39. Gold GE, Thedens DR, Pauly JM, et al. MR imaging of articular cartilage of the knee: new methods using ultrashort TEs. AJR Am J Roentgenol 1998;170:1223 1226. 40. Fullerton GD, Cameron IL, Ord VA. Orientation of tendons in the magnetic field and its effect on T2 relaxation times. Radiology 1985;155:433 435. 41. Henkelman RM, Stanisz GJ, Kim JK, Bronskill MJ. Anisotropy of NMR properties of tissues. Magn Reson Med 1994;32:592 601. 42. Pauly JM, Conolly SM, Macovski A. Suppression of long T2 components for short T2 imaging. In: Proc 10th Annual Meeting SMRI, New York, 1992; p 330.