|Home | About | Journals | Submit | Contact Us | Français|
To demonstrate water and fat suppressed proton projection MRI (WASPI) in a clinical scanner to visualize the solid bone matrix in animal and human subjects.
Pig bone specimens and polymer pellets were used to optimize the WASPI method in terms of soft tissue suppression, image resolution, signal to noise ratio (SNR), and scan time on a 3T MRI scanner. The ankles of healthy 2–3 month old live Yorkshire pigs were scanned with the optimized method. The method was also applied to the wrists of six healthy adult human volunteers to demonstrate the feasibility of the WASPI method in human subjects. A transmit/receive coil built with proton-free materials was utilized to produce a strong B1 field. A fast transmit/receive switch was developed to reduce the long receiver dead time that would otherwise obscure the signals.
Clear 3D WASPI images of pig ankles and human wrists, showing only the solid bone matrix and other tissues with high solid content (e.g., tendons), with a spatial resolution of 1.6 mm in all three dimensions were obtained in as briefly as 18 min.
WASPI of the solid matrix of bone in humans and animals in vivo is feasible.
Bone tissues are comprised of bone substance (mineralized matrix); bone cells; marrow; blood vessels; and other soft tissues. Bone substance is a two-phase composite material consisting of organic matrix (mostly collagen I) and mineral (apatite). The brittle calcium phosphate (Ca-P) crystals of apatite convert the pliable organic matrix into a relatively hard, rigid material (1). Osteoporosis and osteopenia are conditions of low bone density, disorders affecting 44 million Americans (2). In practice, they are diagnosed by measuring bone mineral density (BMD) with x-ray based bone densitometry methods such as Dual Energy X-ray Absorptiometry (DXA) and Quantitative or Peripheral X-ray Computed Tomography (QCT or pQCT) (3–6). However, in additional to osteoporosis and osteopenia, other metabolic bone diseases, such as osteomalacia, can result in low BMD. While osteoporotic and osteopenic BMD is lower than in healthy bone, the bone mineralization is normal (7). On the other hand, both BMD and the degree of bone mineralization are lower in osteomalacic bone (8). The critical factor in differentiating osteomalacia from osteoporosis and osteopenia is knowledge of the degree of mineralization (9), which requires knowledge of the matrix content. Today, this information can only be obtained by transiliac bone biopsy after tetracycline labeling, and in vitro study by Fourier Transform Infrared Imaging (FTIRI) (10) and other analytic methods on bone specimens; there is no clinical method to image bone matrix density noninvasively and quantitatively.
Patients who have diminished calcium phosphate mineral content in bone due to osteomalacia may remain misdiagnosed with osteoporosis for many years since x-ray based densitometry, which measures BMD, but not bone matrix density or the degree of bone mineralization, cannot differentiate between these two disease states. Furthermore, due to the various causes of osteomalacia (11–14), blood and urine measurements (e.g., serum vitamin D measurement) frequently fail to distinguish between the two diseases in their early stages (15–17). There is an urgent need to develop new noninvasive methods to obtain information on the degree of bone mineralization.
Bone mineral and organic matrix have generally been regarded as beyond the purview of MRI due to the difficulty of eliciting and acquiring magnetic resonance signals from nuclei (31P in mineral and 1H in matrix) with exceedingly short spin–spin relaxation times (T2s). Imaging bone using conventional (liquid state) high resolution MRI is based on the fact that the interstitial spaces of trabecular bone and the intramedullary cavities of non-trabecular bone are filled with soft tissue (water, fat and marrow). Conventional MRI thus produces a positive image of the soft tissues between trabeculae, and a negative image of the mineralized bone matrix. Such images measure the volume of bone substance, but provide no information on its composition. The images provide a means to study trabecular bone micro-architecture and porosity. However, because composition information is not obtained, neither the bone mineral density, nor the matrix density, can be determined (18–22).
New methods for visualizing bone mineral and organic matrix have been developed in the past decade. The ultra short TE (UTE) pulse sequence was developed to image short T2 components in lung tissue by Bergin, Pauly and Macovski (23) and later to examine collagen in tendons and knee menisci by Gold et al (24). Bydder and his colleagues applied this method to proton imaging of collagen and bound water in bone and 31P imaging of bone mineral (25–27). The UTE sequence generally employs a radial acquisition. To shorten the echo time, slice selection is accomplished by combining two consecutive acquisitions: the first RF excitation is performed using a self-refocusing half pulse under the slice selection gradient, and the data acquisition starts at the beginning of the ramp-up of the readout gradient (which becomes the center of k space due to the self-refocusing pulse) and extends radially in k space; a second rf half pulse excitation is then performed with a reversed slice selection gradient. The first TE is limited only by how quickly the MR scanner’s RF electronics can switch from transmission to reception. Typical second and subsequent TEs are longer, usually on the order of milliseconds. The RF excitation pulse is on the order of 100μs either when using the scanner body coil for proton imaging or using a surface transmit/receive coil for 31P imaging. Wehrli and colleagues combined fat and water suppression pulses with a UTE sequence to image short T2 water in tendons, ligaments and cortical bone. They also developed a hybrid 3D imaging method with radial in-plane encoding (x, y) and phase encoding in the z direction to reduce the minimally achievable scan time relative to true three dimensional radial imaging. To lessen the TE penalty due to the time required for z phase encoding, they utilized a variable-TE approach in which the minimum TE (equal to the receiver dead time, 90 μs) was set to the kz = 0 line and the maximum TE (230 μs) for the highest spatial frequency kz. The excitation pulse was on the order of 100 μs (28). Wehrli’s group also applied solid-state 3D radial imaging to measure phosphorus content in normal and osteomalacic rabbit bone specimens (29).
The relatively long excitation pulses (which may fail to excite the full spectral bandwidth of solid substances) and receiver dead times (which may obscure rapidly dephasing signals from solid substances) can prevent UTE pulse sequences from detecting solid bone signals with high SNR and quantitative accuracy. Garwood and co-workers developed a method termed “SWIFT” to overcome these limitations by combining short and intense excitation RF pulses with limited average RF power. The excitation pulse is divided into N sub-pulses with amplitude and frequency modulations equivalent to an adiabatic sweep through resonance, approximating an adiabatic sweep performed with continuous wave NMR. The work was carried out on a scanner with a research console with solid state MR capabilities, and therefore exhibited a short receiver dead time, which is required for this method. Adding water and fat suppression pulses before each RF excitation to create pure images of the solid substances may be challenging with the SWIFT technique (30, 31).
Another approach, Solid State MRI (SMRI), has been developed to detect MR signals from all but the most molecularly immobile constituents (32, 33). SMRI is a “zero TE” method, as neither spin nor gradient echoes are employed. The difference between this and other MRI methods of imaging solid bone components is the high intensity and short duration of the RF pulse (10 μs for an excitation angle of 15°), which excites the full solid signal from bone substance, and the very short time delay (also about 10 μs) between the end of the RF pulse and the beginning of data acquisition, which ensures the detection of the very short T2 solid signal. Another key component is a method to recover the data points lost in the receiver dead time (34), which reduces image artifacts and improves quantitative accuracy of the measured bone density. Additionally, water and fat suppressed proton projection imaging (WASPI), a method that combines SMRI with suppression of fluid signals, has been developed to obtain pure images of the solid bone matrix signal so that the matrix density may be readily measured (35). WASPI has been demonstrated to image bone matrix and quantitatively measure bone matrix density in bone specimens (36, 37).
In this study the feasibility of WASPI applied to live animal and human subjects with clinical scanners was explored.
All animal and human images were obtained in full compliance with federal regulations and the guidelines of the respective clinical research committees of the relevant institutions.
Whole pig legs, disarticulated from the knee, were obtained from 3 months old Yorkshire pigs (Cummings School of Veterinary Medicine, Tufts University, North Grafton, MA, USA) following euthanasia of the animal. The legs were wrapped with aluminum foil and frozen until being thawed immediately prior to MRI scanning for the optimization of the WASPI protocol.
A 2 cm long section was cut from the midshaft cortex of the tibia of a pig leg for the study of the proton MR properties (T1 and spectral linewidth) of the solid bone matrix. The specimen was kept frozen until MR measurement.
Three healthy 2–3 months old Yorkshire pigs, weighing 30–40 lbs, were the subjects for the in vivo study. During MRI scanning, the pigs were anesthetized and laid on either their right or left side on the scanner bed with their right or left ankle inside the coil.
Six healthy human volunteers (4 males and 2 females), ages 18–57 years, were the subjects for the in vivo human study. During MRI scanning, the subject laid in the scanner with either the right or left wrist positioned horizontally inside a cylindrically shaped MRI transmit/receive coil.
A solid polyethylene cylinder (Laminated Plastics, Billerica, MA, USA), diameter 90 mm and length 183 mm, was used as a test phantom. Three polymer pellets, 25 mm diameter and 12 mm thick, made of 20% poly(ethylene oxide)/80% poly(methyl methacrylate) (PEO/PMMA) blends, were used as test phantoms. The polymer densities of the pellets were 1.093 and 0.750 and 0.509 g cm−3 respectively. The polymer blend was designed to have MR properties similar to those of bone matrix (36).
The SMRI method utilizes a projection MRI pulse sequence consisting of a single, fixed-amplitude magnetic field gradient pulse, during which a brief, hard RF pulse is issued after a suitable delay following the start of the gradient pulse to allow eddy currents to decay. The free induction decay (FID) is acquired after the RF pulse and a very short receiver dead time, while the gradient is held constant. This sequence is repeated for a set of gradient directions distributed in a uniform pattern about the unit sphere (34). SMRI data was acquired under the following protocol unless otherwise specified. The regular FID projection data were acquired under fixed 19.57 mT/m gradient magnitudes in 8148 directions at a sampling rate of 5 μs per complex point. The short hard pulse used to excite the signal was 10 μs in duration (15°), the receiver dead time was 10 μs, the repetition time TR was 65 msec and the FIDs averaged over 2 acquisitions. A second set of data was acquired with the same MRI parameters except that the number of projections was 20, and the gradient magnitude was 10 mT/m with a sampling rate of 10 μs per complex point. This second set of data was acquired to recover the data points lost in receiver dead time (two points very close to the k-space origin) and was integrated into the regular data set for image reconstruction (34). The measurement time, including both sets of data acquisition, was approximately 18 min. One hundred and twenty eight complex points of the FID were used in the reconstruction, effectively creating a 120 mm field of view (FOV) in a 1283 cubic lattice.
WASPI data was acquired with pairs of π/2 pulses of 2–3 ms in duration, with frequencies set at the water and fat chemical shifts to suppress water and fat signals respectively, before the SMRI part of the pulse sequence. The separation of these pulses was carefully arranged to reduce the possibility of water or fat echo formation.
SMRI and WASPI imaging resolution is limited by the projection pixel size (Δxp, the resolution limit due to the reconstruction) and intrinsic pixel size (Δxi, the resolution limit due to the spectral properties of the resonance). For efficient imaging, these quantities should be comparable in magnitude. The Δxp is determined by the FOV and the number of independent pixels NIP, which in turn is determined by the total number of projections (P) (34):
If P = 8148, NIP = 51;Δxp is ~ 2.4 mm for FOV = 120 mm, and ~1.6 mm for FOV = 80 mm. If P = 5216, NIP = 40; Δxp is 3 mm for FOV = 120 mm, and 2 mm for FOV = 80 mm.
To produce the desired FOV with a given number of data points per scan (Nps), a minimum size of the reconstruction matrix (MMAT) is required (34):
where Ws is the spectral width, G the projection gradient strength, and γ the magnetogyric ratio. In this study, we used MMAT = 128 and Nps = 128, respectively.
The Δxi is determined by the resonance line width (WL) and the gradient strength:
For WL = 2 kHz and G = 20 mT/m or 30 mT/m, Δxi is ~2.3 mm or ~1.5 mm, respectively.
Water and fat suppressed proton spectroscopy was utilized to measure the proton resonance line widths with acquisition parameters similar to WASPI without applying the projection gradients. T2* was derived from this measurement as:
This experiment was carried out only on pig bone specimens and the polyethylene phantom, due to concern over possible contamination of the poorly-suppressed soft tissue signal at the edge of the FOV of in vivo human imaging.
Progressive saturation WASPI measurements were carried out to measure the T1 of the solid bone matrix with TR varying from 0.05 sec to 0.8 sec. Since T2 is much shorter than the repetition time TR and its weighting effect on the signal can be neglected, the data was fitted to the Ernst angle formula (38). This measurement was carried out only on pig bone specimens because of the long scan times.
MR experiments were carried out with a Siemens 3.0 Tesla Trio clinical scanner. The maximum available gradient strength for this study was 30 mT/m. The proton resonance line width (full width at half height) of solid bone matrix was measured by water and fat suppressed proton spectroscopy of a pig bone specimen to be 1.5–2.0 kHz at 3T (Fig. 1a), while that of polyethylene at 3T was measured to be 2.0–2.4 kHz (Fig. 1b), corresponding to T2* = 159 – 212 μs for bone matrix and T2* = 133 – 159 μs for polyethylene, respectively. Like many solid polymers, polyethylene has amorphous and crystalline regions distributed in microscopic domains throughout the material; these domains have somewhat different T2s due to differences in molecular order and motion (39). The dual width character of the lineshape is clearly seen in the figure, further confirming our ability to correctly capture solid signals. The excitation pulse was 10 μs by a local 91 mm inside diameter transmit/receive coil, corresponding to a 10° flip angle. To test if the clinical scanner’s body coil can excite the broad solid signal, SMRI images of a solid cylindrical polyethylene phantom (diameter 90 mm, length 183 mm) were acquired and received with the body coil. The excitation pulse length was 30 μs (11°) and acquisition spectral width was 200 kHz (dwell time 5 μs). Other acquisition parameters were: FOV 400 mm, projections 998, TR 100 msec, signal averages 4, total scanning time 14 min. The images (Fig. 1c and d) showed that the 30 μs pulse did excite part of the MR signal from the solid polyethylene, but not the whole phantom. The SNR was low. When a shorter excitation pulse, such as 10 μs (~ 3°), was used, the SNR was lower and the image of polyethylene cylinder could not be recognized.
A quadrature birdcage proton transmit/receive coil, constructed on an acrylic tube (diameter 91 mm; length: 110 mm) (Fig. 2b) was built to generate strong B1 fields and increase the filling factor for imaging pig ankles and human wrists. Hard pulses as short as 10 μs corresponding to a 15° flip angle can be generated by this coil. SMRI images (FOV 300 mm) of a 1 cm diameter tube of water inside the coil showed that the acrylic tube of the coil and the dielectric and jackets of coax cables were visible in the images (Fig. 2a). Although this is a very positive indication that the B1 field was strong enough to excite very short T2 FIDs, these signals represent interferences that must be eliminated. A new coil was therefore constructed using a Teflon tube and Teflon insulated coax cables, eliminating the interfering signals (Fig. 3a). WASPI images of a solid uniform polyethylene cylinder were acquired with this new coil and shown in Figs. 3b and c. The FOV was set to 120 mm, larger than the diameter of the coil. As can be seen, the coil tube and cables are no longer visible. The extent of the region of strong B1 field is about 75 mm in the z direction and 90 mm in diameter. The images of the polyethylene cylinder serve to visualize the B1 field distribution, which exhibits a variation of about 5% within the desired cylindrical volume of 40 mm along the coil axis, and 70 mm diameter. (The images in Figs. 3b and c were acquired with the crossed diode T/R switch described below.)
Following the end of an RF pulse, the coil rings down with stored RF energy, and the scanner receiver is briefly overloaded. There is also a finite time required for the T/R switch to change from transmit mode to receive mode. During these events, the MR signal is obscured with large, wildly varying transient signals, which if included in the raw data would severely distort the image. A minimum time delay is therefore imposed by the scanner software between the end of RF pulse and signal acquisition, permitting full recovery of the receiver. This recovery delay serves to prevent overload signals from influencing the image data and also to protect the receiver hardware from the RF pulse energy. Because solid state MRI methods that acquire FIDs require rapid receiver recovery so that the earliest time points of the FID can be measured with high fidelity, it is critical to reduce the receiver recovery to as short a time as possible. This requires modification of the transmit/receive (T/R) switch to permit faster switching, possibly lowering of the coil Q (to reduce the ring down time), and reduction of the software-imposed minimum recovery delay so that signals may be sampled sooner following the RF pulse than normally permitted.
The default setting for the recovery delay for this scanner is 20μs. However, it was found that although the scanner receiver itself was able to recover in a shorter time when the T/R coil was used, the switching time of a conventional actively driven PIN diode T/R switch was far longer than 20 μs, which obscured too much of the FID and distorted the image. An alternative T/R circuit using passively switched standard 1N914B diodes was found to switch extremely quickly and not extend the receiver recovery time. This circuit utilizes crossed (back-to-back paired) diodes between the transmitter output and the RF coil, followed by a quarter wave cable connecting to crossed diodes to ground at the preamplifier input. This circuit has been in standard use in spectrometers for fluid and solid state NMR spectroscopy for decades. Standard diodes switch between on and off states in nanoseconds. PIN diodes, while having far superior on/off ratios and higher power handling capacity, have inherently slow on/off switching so that the RF is not rectified; they require DC drive signals that may themselves exhibit slow transitions.
To compare the performance of the conventional actively driven PIN diode and passively activated T/R switches the scanner recovery delay was set to 10 μs, such that signal digitization would occur beginning 10 μs following the RF pulse. The RF coil was disconnected, and a continuous wave artificial signal from an RF synthesizer was applied to the receiver while the scanner was in operation; the transmit pulse amplitude was set to zero to protect the synthesizer. Fig. 4a compares the recovery of the digitized artificial signal of the PIN and crossed diode switches. The sampling interval is 5 μs per complex point. With the PIN diode switch the signal is completely suppressed for at least 10 μs and is not fully recovered until 30 μs into the sampling, a total of 40 μs following the RF pulse. Except for a small amplitude transient, the crossed diode switch permits full recovery at the first sampled data point (10 μs at most following the RF pulse). To lose 40 μs of an FID with a T2 on the order of 100 – 200μs results in intolerable distortion and signal loss.
Using the conventional PIN diode switch, a WASPI image of a pig leg specimen with polymer pellet attached was acquired and reconstructed based on the assumption that the receiver dead time was 10 μs and the number of data points (at a 5 μs dwell time) lost in the recovery time was two. The images contained severe artifacts: a smear of background signal spread across the FOV, including the marrow space (which should be dark in WASPI images), and overwhelming the polymer pellet signal, which was not visible (Fig. 4b). In vivo WASPI images on an ankle of a live pig acquired with the crossed-diode switch showed that the background artifact was suppressed and the solid bone matrix and the polymer pellet were visualized, while the signal of the soft tissue was suppressed (Fig. 4c). A conventional spin echo image of the same ankle was acquired with a protocol of FOV 120 mm, slice thickness 3 mm, TR 30 ms, TE 4 ms and 1 average, for comparison (Fig. 4d). The correlation between the dark regions in the spin echo image and the bright regions in the WASPI images justifies the conclusion that these regions represent solid bone matrix.
SMRI (non-suppressed projection images) and WASPI images of a bottle of water and a bottle of corn oil were acquired to evaluate the degree of suppression in the WASPI method (Figs. 5a–d). The SMRI and WASPI images in the figures were windowed identically so that their brightness may be directly compared. In the water image, the water frequency was set at zero offset; in the corn oil image, the fat frequency was set at zero offset and the water frequency at 3.5 ppm upfield. In both cases, the water and fat signals were suppressed to less than 5% of full intensity over the entire volume of uniform B1 (nearly the entire FOV). At the edge of the WASPI images, the suppression is compromised due to the closeness of the bottles to the coil where the B1 field is not uniform. Because of the high Larmor frequency (123 MHz), dielectric resonance effects likely contribute to the B1 distortion. The fluid-air interface issue found in in vitro WASPI study on bone specimens (34), which causes failure of marrow signal suppression due to the magnetic susceptibility discontinuity at the fluid-air surface, is greatly alleviated in vivo where the bone is surrounded by soft tissue, and the artifact did not appear in the in vivo pig ankle study (Fig. 4C) and in the in vivo human study.
WASPI images of three polymer pellets separated by 3 mm thick Teflon disks were acquired with FOV of 120 mm (projection gradient strength of 19.57 mT/m) and projection number 8148 (Fig. 6a). The images of the polymer pellets were clearly separated, indicating that the resolution was better than 3 mm. When the FOV was set to 100 mm, (gradient strength 23.48 mT/m) with no change in the projection number, the image resolution improved, indicated by the widening of the apparent separation (Fig. 6b). These observations validate the expressions for spatial resolution (Eqs. 1 and 2).
The T1 of solid bone matrix was found to be 1.8 s in pig bone specimens. Based on the relationship between excitation angle and TR (38), a TR of 65 ms and an excitation angle of 15° was chosen for in vivo human studies. Water and fat suppression was accomplished with 2 ms 90° pulses using the settings found by the scanner. WASPI images were acquired with the subjects lying on their side. The left or right wrist was placed in the coil and stabilized by an air-filled cushion made of thin Teflon sheets. Fig. 7 shows WASPI images (Figs. 7a, b, c) and SMRI (non-suppressed projection images, Figs. 7d, e, f) of the left wrist of a 57-year-old male volunteer, acquired with the protocol described in the Materials and Methods section (FOV 120 mm; number of projections 8148). The cortical bone was not visible (dark) in the non-suppressed images because of the much brighter fluid signals, but clearly visible in the WASPI images with signal to noise ratio (SNR) ~ 45. Trabecular bone was visible in the distal radius epiphysis with SNR ~ 10. The spatial resolution Δxp is ~2.4 mm. The brachioradialis and carpi ulnaris tendons also appear because of the high content of highly organized (motionally restricted) collagen. The marrow and muscle signals were suppressed to less than 5% of the non-suppressed signal. Without air filled cushions to stabilize the wrist inside the coil, motion artifacts appeared in the images.
WASPI images with stronger projection gradients were acquired. The available maximum gradient strength on this scanner is 30 mT/m. If the spectral bandwidth and dwell time are fixed, the stronger the projection gradient, the smaller the FOV that can be achieved. With a smaller FOV fewer projections and therefore less time are needed to achieve comparable spatial resolution. Fig. 8 shows transverse WASPI views of the left wrist of a 24-year-old male volunteer. In Fig. 8a the FOV is 100 mm (projection gradient strength 23.48 mT/m), and the number of projection is 5216. According to Eqs. 1 and 2, Δxp is 2.5 mm in this case, comparable to the 2.4 mm obtained with 8148 projections for FOV 120 mm. However, the total acquisition time with 5216 projections was 12 min, shorter than the 18 min with 8148 projections. On the other hand, higher resolution can be achieved with a smaller FOV by keeping the number of projections (hence the total acquisition time) the same. In Fig. 8b the FOV is 80 mm (projection gradient 29.57 mT/m), while the number of projections is 8148, resulting in an Δxp of ~1.6 mm for FOV 80 mm, much better than an Δxp of 2.4 mm for FOV 120 mm.
Unlike the images in Fig. 7, the WASPI images in Fig. 8 were acquired with water and fat suppression pulses 3 ms in duration, with the pulse amplitude manually adjusted to achieve the best suppression, reducing the soft tissue intensity to 2–3% of the non-suppressed intensity. However, the superior soft tissue suppression resulted in a somewhat lower SNR of the solid signals. The corresponding slice of a conventional spin-echo image matching the 80 mm FOV of Fig. 8b is shown in Fig. 8c.
The challenges of conducting in vivo WASPI are two folds: special requirements for the transmit and receive hardware (a strong B1 field and a very short receiver dead time) which unmodified clinical MR scanners do not provide, and suppression of fluid state signals.
Even with a high powered transmit amplifier providing the order of 35 kW of pulse power, the B1 field of the body RF coil is insufficient to properly excite the wide spectral bandwidth of solid substances with a single hard pulse in the range of 10 – 20 μs. A local transmit coil is therefore essential. The small diameter quadrature birdcage T/R coil utilized in this study is suitable for WASPI of human extremities, but imaging of the central skeleton will require improvements in technology.
Because clinical MR scanners use pulse sequences based almost exclusively on spin or gradient echoes, the long receiver recovery time is never a limitation. However, the recovery time is a critical element when imaging FIDs. The 40 μs of lost data (equivalent to 8 complex data points at a 5 μs dwell time) when using the standard PIN diode T/R switch resulted in unrecoverable distortion, even when supplemented with acquisition of additional projections at low gradient strength to recover the lost data points (34). The improvement achieved by acquiring additional projections is limited when the T2 of the tissue is short and the number of missing data points is considerable (eight points in this case), because the measurement of these many missing central k-space data would occur far in time following the RF pulse where the loss in SNR due to T2 decay is considerable. The crossed diode T/R switch reduced the lost data points to two, resulting in a dramatic improvement in image quality.
Both projection pixel size Δxp and intrinsic pixel size Δxi determine the resolution of WASPI images. In general, with stronger projection gradients, Δxi is smaller (Eq. 3), and the image resolution is improved (Fig. 6b). When the maximum available gradient strength is applied, Δxi reaches its minimum, and any improvement of the image resolution would be limited by Δxp, achieved either with smaller FOV or more projections. There is a trade-off between resolution, SNR, and total acquisition time. In this study, the maximum gradient strength is 30 mT/m, and the corresponding intrinsic Δxi is 1.5 mm for a resonance linewidth of 2 kHz. Δxp would be 1.6 mm if the FOV was 80 mm and the number of independent pixels was 51 (8148 projections). The total acquisition time would be 18 min with TR = 65 msec, and the number of averages = 2. If the number of independent pixels was 40 (5216 projections), Δxp would be 2 mm, and the total imaging time, including the second acquisition for the missing initial FID data points, would be reduced to 12 min. Our study subjects found that it was easier for them to keep their wrists still inside the coil during a 12-min scan than during an 18-min scan, reducing discomfort and motion artifacts. The protocol of 5216 projections and a 12 min scanning time was considered a reasonable choice for human wrist WASPI scanning.
Theoretically, the best suppression of water and fat signal would be achieved by two consecutive low power long duration chemically selective 90° pulses each followed by spoiler gradient pulses (35). In practice, it is not easy to set the exact flip angle inside bone tissue, particularly when part of the wrist is very close to the coil. It was found that the scanner-calculated 2 ms 90 pulse was not the most effective suppression pulse. The suppression pulse was empirically optimized by adjusting the pulse amplitude and duration on pig leg specimens. It was found that optimum suppression results in pulses that are larger in amplitude than what the scanner calculates to be a 90° pulse. Figs. 8a and b were acquired with these manually optimized 3 ms pulses. However, these stronger pulses also reduced the solid bone matrix signal. In WASPI experiments on the polyethylene cylinder, it was found that the WASPI signal intensity of the polyethylene acquired with 2 ms scanner calculated 90° suppression pulses was reduced to 75% of the non-suppressed signal level, while the 3 ms optimized pulses (which had higher RF amplitude) reduced the WASPI intensity of the polyethylene to 50% of its non-suppressed signal level. This is another reason that the SNR of Figs. 8a and b was lower than the SNR of Figs. 7a and b. Longer suppression pulses, which are more spectrally selective for fluid state signals, will reduce but not eliminate the loss of solid state MR signals because the solid state resonances are homogeneously broadened and are significantly affected by off resonance radiation. It is easy to suppress narrow water resonances with long suppression pulses, but for fat resonances, which are typically on the order of 400 Hz in width, suppression pulses longer than 3 ms will not suppress the fat signal down to less than 5% of its initial signal intensity. We attempted to narrow the fat resonance line width with an image-based analytic shimming technique (40), but the resulting line widths were no narrower than could be achieved with manual shimming. The trade-off between soft tissue suppression and matrix SNR therefore needs to be considered when optimizing the WASPI sequence, and the trade-off may be different for applications requiring clean visualization versus good quantitative matrix density. In the bone specimen experiments, a tube of bone marrow was used to serve as an indicator of the suppression (34). For in vivo human wrist imaging, the suppression of the water signal of the forearm muscle and subcutaneous fat can serve as intrinsic indicators of the suppression.
Concern over inflow of fresh blood into the FOV compromising the soft tissue suppression proved to be unfounded. Although the blood vessels are very bright in spin-echo images and caused artifacts in the phase encoding direction (Fig. 8c), these vessels did not appear in WASPI images and did not create artifacts. This could be due to a combination of the absence of echoes in the WASPI sequence, the short time between the suppression pulses and the FID acquisition, and manner in which motion artifacts affect radially acquired k-space data.
It is striking how the WASPI images resemble plane film x-ray radiographs, with soft tissue dark and solid bone bright, but of course the sources of WASPI and x-ray signals are quite different. The WASPI signal arises from a portion of collagen protons, tightly bound water, and other motionally restricted molecules. The x-ray signal arises predominantly from the mineral content. Although these constituents have similar spatial distributions in bone, the distributions are not identical, which makes WASPI useful for measuring matrix density, which in combination with mineral density information yields the degree of mineralization. Note that the clear visibility of tendons in WASPI images may prove useful in evaluating these tissues by MRI.
In Conclusion, WASPI imaging of human wrist bone matrix with clinical MR scanners is feasible, provided that the transmit/receive system meets the requirements of strong B1 field and short receiver recovery time. A physical support and restraint system that is invisible in solid state MR images is required to stabilize the wrist during scanning to eliminate motion artifacts. WASPI images with resolution of 2 mm can be acquired in 12 min, showing only solid bone matrix and tendons. If calibration phantoms are imaged with the wrist, bone matrix density can be quantified.
We thank Dr. Andreas Potthast of Siemens Medical Systems, who provided considerable technical assistance in addressing the receiver dead time and other scanner issues.
Grant Sponsors: NIBIB-NIH, NIA-NIH