|Home | About | Journals | Submit | Contact Us | Français|
This work reports preliminary results from the first human cardiac imaging at 7 T. Images were acquired using an 8-channel transmission line (TEM) array together with local B1 shimming. The TEM array consisted of anterior and posterior plates closely positioned to the subjects' thorax. The currents in the independent elements of these arrays were phased to promote constructive interference of the complex, short wavelength RF field over the entire heart. Anatomic and functional images were acquired within a single breath hold to reduce respiratory motion artifacts while a vector cardiogram (VCG) was employed to mitigate cardiac motion artifacts and gating. SAR exposure was modeled, monitored and was limited to FDA guidelines for the human torso in subject studies. Preliminary results including short-axis and four-chamber VCG-retrogated FLASH cines, as well as, short-axis TSE images demonstrate the feasibility of safe and accurate human cardiac imaging at 7 T.
Cardiovascular magnetic resonance (CMR) imaging has rapidly evolved as an important tool in the arsenal of techniques employed for the management of cardiovascular disease (1–4). This is largely based on developments in real time imaging (5,6), imaging of myocardial perfusion at rest and under pharmacological stress (7,8) and assessment of regional wall motion (9). However, in other areas critical to diagnosis of cardiac disease, such as coronary artery angiography, MR methods (10–12) remain suboptimal despite continuous improvements over the years, due primarily to low signal-to-noise ratios (SNR) (12). Similarly, while animal studies have indicated the great potential of 31P spectroscopy to investigate in vivo energetics in the myocardium (13,14), studies in the human heart are again limited due to low SNR (15, 16).
Since SNR is approximately proportional to field strength, increasing magnetic fields can alleviate SNR limitations. Results from 3 T have shown some promise when compared to 1.5 T (17,18) due to increased SNR and contrast to noise ratio (CNR). These studies, however, have also highlighted the challenges of working at elevated magnetic fields, such as decreased T2* times, increased B0 inhomogeneities due to magnetic susceptibility effects, non-uniform B1 distribution, increased RF deposition, and the degradation of the electrocardiogram (ECG) as a result of increased magneto-hydrodynamic (MHD) effects. Particularly, the non-uniformities in B1 become a real challenge with increasing field strength beyond 3 Tesla, as the radio frequency (RF) wavelength becomes smaller than the size of the imaged object. Potential solutions to this problem, however, have been recently demonstrated for applications in the head (19–21) and even more recently for prostate imaging at 7 T (22). Consequently, while it was virtually unthinkable to pursue imaging in the human chest at 7 tesla in the past, these new developments suggest that the problem may be tractable. Accordingly, the objective of this investigation was to use and build on some of these developments in order to investigate the feasibility of safe and successful cardiac imaging at 7 tesla.
An eight-channel TEM transceiver was used together with B1 shimming localized in the heart (23). SAR levels in the torso were predicted from models and average SAR was monitored during studies to assure subject safety. VCG-retrogated FLASH cine protocols and VCG-gated turbo-spin echo (TSE) sequences were used to image the heart along both the short and four-chambered axes. A more detailed account of the materials and methods follows.
An eight-channel, flexible, transceiver array (Fig. 1a) was built according to stripline transmission line (TEM) principles (21,24,25). This eight-channel array consisted of a pair of four-element arrays, one located anterior and the other posterior to the chest (Fig. 1b). Four coil elements were attached in a parallel configuration to a flexible polytetrafluoroethylene (PTFE) plate measuring 22.7 cm × 35.6 cm × 0.3 cm for both the anterior and posterior plates. The individual coil elements were 15.3 cm long with a 1.27-cm wide inner conductor and a 5.0-cm wide outer conductor, separated by a 1.9-cm thick bar of PTFE, a dielectric with a low loss tangent and a permittivity of 2.08. A 5.0-cm air gap separated the outer conductor of each coil element.
Each element was individually tuned to 297 MHz (7 T), and matched to a 50-Ω, coaxial signal line. Capacitive decoupling facilitated greater than 18 dB isolation between elements. All bench measurements required for coil tuning, matching, isolation and other circuit characterizations were performed using a calibrated Hewlett-Packard (Palo Alto, CA) HP 4396A network analyzer together with an HP 85046A “S” parameter test set.
Imaging experiments were performed on a 7 T (ωo = 297.14 MHz) magnet (Magnex Scientific, UK) interfaced to a Siemens console. Siemens Avanto whole body gradients were set to a slew rate of 170 T m−1 s−1 and a maximum amplitude of 40 m mT−1. The output of an 8 kW RF power amplifier (Communications Power Corp., Hauppauge, NY) was divided into eight channels with equal magnitude and phase through an 8-way RF splitter (Werlatone, Brewster, NY) to feed the eight element coils. Eight high-power phase shifters (Advanced Technical Materials, Inc., Patchogue, NY) and incremental cable lengths were used to adjust the transmit phase of each transmit coil as required to accomplish B1 shimming for optimal image homogeneity over the heart. Power divider specifications included −0.4 dB insertion loss and 1° phase resolution, while the phase shifters measured −0.5 dB insertion loss over a phase range of 120° at 297 MHz.
For a given location in space, the maximum efficiency of a transmit array will occur when there is complete phase coherence between all channel’s B1+ vectors. In this case, the magnitude of sum of the complex B1+ vectors, which we refer to as the "available B1+", equals the sum of their magnitudes. Opposite to this case, when B1+ vectors are not in phase coherence, the effective B1+, which corresponds to the magnitude of sum of complex B1+ vectors, will only amount to a fraction of the available B1+ due to complex destructive interferences (20).
To optimize the RF transmit efficiency over the heart, the transmit B1 field (B1+) components of the independent coil elements were adjusted, or “shimmed”, to affect an approximate phase coherence or “constructive interference” of the short RF wavelengths (12 cm) over the region of interest (19,20,22,26). In order to perform such B1 optimization, calibration scans were first obtained to determine relative transmit B1 maps for each transmit coil, as described in (20). Based on those relative B1 maps, we then perform a modified version of a local B1 phase shimming technique which has been demonstrated in the head and in the prostate at ultra-high fields (19,20,22,26). A brief description of those methods follows.
Calibration scans obtained to map the relative excitation profiles consist in a series of low resolution, low flip angle gradient echo images. Each image is collected when transmitting through only one array element at a time, while all other transmit elements are terminated on 50 Ω. Relative transmit B1+ phases are then calculated based on those images. It is also possible with the same image series, and without knowing the absolute magnitude of B1+, to estimate for each pixel which fraction of the available B1+ is obtained for a given set of transmit coil phases (20,22). This approximation relies on two conditions: a low excitation flip angle in order to have an approximately linear relationship between B1+ and the actual flip angle (with sin(θ) ~= θ for θ<10°), and minimal T1 weighting in the resulting images.
From the gradient echo image series, a Sum of Magnitude (SOM) image can be written as:
where p is the complex valued signal in a pixel, K is the index of transmit channels (K=1,2,…,M) and J is the index of receiver channels (J=1,2,…,N). For any given set of phases, βK, the result that would occur when pulsing all coils simultaneously can be simulated numerically by applying those phases to the complex images and then computing a Magnitude of Sum (MOS) image:
The ratio of MOS to SOM for a given βK predict for each pixel the fraction of available B1+ actually obtained when experimentally transmitting with all channels simultaneously. For a small sized target, it is possible to calculate a set of phases βK yielding transmit phase coherence through all channels within the target, resulting in very high local transmit efficiency. Successful local B1 shimming based on this approach has been demonstrated in the prostate (22) at ultra high field.
While the aforementioned technique works well for shimming over the entire prostate, it does not work well for shimming over the entire heart; this is mainly due to the size of the heart and its asymmetric placement with respect to the array. These two conditions result in large variations of transmit B1 phase and amplitude for each coil within the heart. Ideally, B1 shimming calculations over such large ROI's should be based on complex B1+ profile (magnitude and phase) of each transmitting coil. However, obtaining magnitude B1+ map for each of the eight transmit channels would take an extensive time during a volunteer session. Therefore, a modified version of the local B1 phase shimming algorithm was implemented, which is referred to as “local phase, and magnitude-approximation B1+ shimming". Instead of utilizing actual magnitude B1+ profiles, this approach utilizes the magnitude of the images series obtained with one coil transmitting at a time, merged with the calculated relative transmit B1+ phases (we refer to those hybrid images as "approximation B1+ maps"). The modified algorithm consists in a non linear minimization problem which incorporates each pixel in the ROI. The minimization criteria is:
where F(βK) is the magnitude of the sum of the approximate B1+ maps with, for a given βK:
with being the approximation B1+ map for each transmit coil K. The minimization critera given in Eq. 3 was determined empirically and favors increased transmit efficiency over homogeneity. A global optimization, which avoids local minima, was computed using a Multilevel Coordinate Search (27) to solve for the phases of the individual transmit elements.
To determine the improvement in transmit efficiency, due to the B1+ shimming, the predicted relative transmit efficiency (RTE) and estimated B1+ nonuniformity were calculated from combinations of the approximated B1+ maps with different phase sets (22).
Remcom’s Finite Difference Time Domain software (XFDTD, Remcom, Inc., State College, PA) was used for numerical modeling of SAR before and after B1 shimming was performed.
The complex electric field maps (E) were also solved for each transmit channel. The resulting complex E field components (Ex, Ey, and Ez) were determined by summing the E-field maps after applying the corresponding RF phases for each phase set (equal phase or local B1+ phase shim). SAR was calculated in accordance to the standard formula,
where ρ is the proton density and σ is tissue conductivity. The ratio of the average B1+ magnitude of the equal phase condition to the local B1+ phase shim was calculated. The square of this ratio was used to rescale the SAR maps (22).
Volunteers were recruited and imaged under a protocol approved by the University of Minnesota’s Institutional Review Board. For all imaging experiments, subjects were supine with the two halves of the transceiver coil placed anterior and posterior to the chest wall. For additional comfort, the volunteers were insulated from the array by a 1.3 cm thick, closed-cell foam pad. All images were acquired during a breath hold to reduce respiratory motion artifacts while the Siemens’ VCG was used to gate image acquisition.
Low-resolution, multi-slice, gradient echo scout images were first acquired with the following parameters: TR/TE =7.1 ms/2.05 ms; matrix = 192 × 192; FOV = 40.0 cm × 40.0 cm, slice thickness = 0.5 cm. To map the relative excitation phase and approximate magnitude in the body generated by each coil element, a series of eight cardiac gated, low resolution, low flip angle gradient echo images were acquired. The image parameters were: TR/TE = 11.54 ms/4.1 ms; Matrix=192 × 122; FOV=40.0 cm × 40.0 cm; slice thickness=0.5 cm. The short repetition time was used in order to acquire all the calibration data in a single breath hold. While the amount of T1 weighting is spatially varying across the heart, it is expected to be minimal in these data due to the extremely low power used and the distance the heart is from the coil elements when used individually for RF transmission. Each image was collected by transmitting with only one array element at a time, while the other array elements were properly terminated on the RF power amplifier input port of each channels’ transmit/receive switch. All eight transceive coils were used simultaneously during reception, and the corresponding signals were sampled on eight receiver channels of the Siemens console. Following localized B1+ shimming based on the previous data, transverse, short axis and four-chamber views were acquired with VCG-retrogated FLASH cines: TR/TE = 6.1 ms / 3.06 ms; matrix = 192 × 132; FOV = 34.0 cm × 23.0 cm; slice thickness = 0.5 cm. Sixty successive phases of the cardiac cycle were interpolated from data acquired over the course of 12 heartbeats.
FLASH cines were acquired for three different shim configurations: 1) all coil elements in both halves of the array set to equal transmit phase (equal phase), 2) elements in the anterior half set in 180° phase opposition with the posterior half (opposite phase) and 3) coil element phases adjusted to optimize the local B1+ in the heart based on experimentally measured B1+ phases for each array element for each individual subject (B1 shimmed). These cines were used to evaluate the effectiveness of shimming the B1 field over a local ROI around the heart.
Additionally, TSE images were acquired along the short axis (TR/TE = 909.7 ms/45.0 ms; matrix = 192 × 156; FOV = 34.0 cm × 23.0 cm; slice thickness = 0.5 cm, ETL = 9; NEX = 2; iPAT = 2 with 26 reference lines). Refocusing was achieved with a nominal 180° pulse and Siemen’s standard weak fat saturation preparation pulse was employed. Normalization filters were applied for image intensity correction.
It should be mentioned that standard pre-scan calibration routines supplied by manufacturers for clinical use are not accurate for ultra-high field cardiac imaging experiments. For example, the standard power calibration strategy which integrates signal over an entire transverse slice at iso-center works well for clinical field strengths but is not appropriate for 7 T imaging of the body because of the increasingly inhomogeneous B1+ (28). Therefore, in this study, the transmit power was incrementally increased to find power levels that maximized contrast and signal homogeneity.
SNR was calculated for three different shim configurations listed above: equal phases, opposite phases, and B1 shimmed. Signal levels were computed from single time-frame images, during systole, from the short axis FLASH cines by drawing an oblique line plot across the entire chest. Noise levels were computed by determining the standard deviation of the signal intensity in identical images acquired without applying transmit RF power.
All volunteers were given exit interviews regarding their experience during the 7 T cardiac exam; three of the volunteers commented on “sleepiness”, two reported “metallic tastes” and one note each was recorded for feeling “warm” and “cold.”
Experimentally, it was found that reversing the RL and LL VCG leads had the greatest impact on increasing the R wave with respect to the MHD effect in the scanner; this modification to the standard VCG placement provided the most reliable method for accurate cardiac gating and VCG triggering. The VCG traces from outside and within the magnet are shown in Figure 2 with the RL and LL leads reversed. While MHD distortion of the VCG signal from within the magnet is apparent; the accentuated T waves do not preclude accurate and effective cardiac monitoring or gating as indicated by the red arrows positioned above the appropriate QRS complex.
Experimentally attained MOS/SOM maps are shown in Figure 3 for the same transmit power with equal phases across all transmit channels (Fig 3a) and after subject specific B1+ shimming (Fig 3c). The maximal attainable B1+ in a given voxel is defined as the B1+ that would be achieved if in that voxel the contribution of each coil array element would be in-phase and thus add constructively. In Figure 3a, approximately only 20% of the B1+ field in the heart adds constructively, as opposed to approximately 66% for the B1+ shimmed case in 3c. Figures 3b and 3d show the gradient echo image before and after shimming, respectively. For the example shown in Figure 3, the optimal phases obtained for this subject for the different elements of the array coil were determined to be 146, 38, 0, 61 degrees for coils 1 through four, respectively, and 75, 346, 210, 208 degrees for coils 5 through 8 respectively.
The predicted RTE (22) is the ratio of the MOS after B1 shimming (B1+ shim phase) to the MOS prior to shimming (equal phase) averaged over the heart; it measures the increase in transmit B1 efficiency due to reduced destructive interferences resulting from B1+ shimming. The average predicted RTE over 6 subjects was 2.62 ± 0.5 meaning there was, on average, 162% more transmit B1+ available averaged over the heart after B1+ shimming for a given input power. It was shown previously that the predicted RTE derived from the B1+ shimming data is a good estimate of the measured RTE based on actual B1+ maps measured in vivo (22). Similarly, an estimate of B1+ nonuniformity can be determined by the standard deviation over the mean of the MOS over the ROI used for optimization in the heart. The B1+ nonuniformity dropped from 42% to 20% after B1+ shimming. The theoretical and experimental data show that improvements in B1+ efficiency, B1+ homogeneity and signal intensity can be gained by “B1 shimming” over the cardiac ROI marked by the red oval in (Fig. 3a).
The approximation B1+ map is clearly inherently biased by spin density, receive B1 profiles, spin saturation and excursions from the low tip-angle approximation. Despite these limitations our experimental results show that sheer improvements in transmit efficiency, and thus in SNR and image quality, can be obtained based on this approach. Possible reasons for this apparent robustness against multiple sources of bias may include the fact both transmit and receive magnitude B1 profiles feature somewhat similar spatial decaying gradients as a function of the distance to the RF coils. Naturally, this does not alleviate the differences existing between receive and transmit B1 profiles at very high field, and this latter peculiarity may reflect that the limits of the B1+ approximation are most noticeable near the chest wall where the B1+ and B1− are strongest and significantly differ from each other. Recently Cunningham et al (29) proposed a rapid B1+ mapping technique; it is expected that a B1+ map will not be bound to the same limitations as the approximate B1 map and will allow for better B1 shimming in the heart.
The B1+ shimming procedure, including cable and phase shifter manipulation, took approximately 15–20 minutes per slice. Due to the time requirement of this procedure, only a single slice could be shimmed during a study. A central axial slice was found to provide the greatest global B1+ shim when compared to shimming on either short axis or four-chambered views. New improvements in transmitter phase and gain control will allow for slice-by-slice shimming. It is expected that slice-by-slice shimming will allow for greater B1 efficiency, homogeneity and signal intensity as well as reductions in SAR, when compared to single slice shimming.
Modeled SAR distributions are shown in Figure 4. To compare the pre- and post-shim SAR distributions, the input voltages have been adjusted so that the average flip angle over the FOV within the heart is constant between Fig 4a and 4b. For a desired B1 transmit field over the heart, Figures 4 shows that SAR is significantly lower in the subject after B1 shimming, as compared to the equally-phased array. The ability to reduce local SAR with phase-only based B1+ shimming does not extend to techniques attempting to achieve a homogeneous B1+ where optimization of each channel’s transmit amplitude is required. In such cases, large increases in local SAR can result (30). Regardless of the B1+ shimming method employed, the SAR will be proportional to the square of the flip angle with a given set of transmit phases and amplitudes.
Figure 5 shows heart images resulting from different phasing of the coil elements along with associated SNR profiles. Figure 5a shows a short-axis slice in heart imaged when all eight elements equally phased (equal phase); figure 5b is the same slice, however, the anterior half of the array is set in 180° phase opposition with the posterior half (opposite phase); finally figure 5c is when the array elements are phased for maximum B1+ efficiency through the heart (B1+ shimmed). Figure 5d shows the SNR line plot for all three cases (fig 5a–c). The SNR of the blood in the right ventricle in the B1+ shimmed case is approximately three times greater than either the equal phased or opposite phased cases; the SNR in the septum of the B1+ shimmed case has improved by a factor of 2.4 over the equal phased or opposite phased cases.
More importantly, the contrast between the blood and the myocardium is tremendously improved with B1+ shimming for a given transmit power. While the SNR is sufficiently high in all three cases, only the B1+ shimmed case clearly delineates the structure of the heart. Clinically, ejection fraction, myocardial mass quantification, and valvular function depend on this difference in signal intensity between blood and the myocardium; therefore, of the three images in figure 5, figure 5c has the most clinical relevance. It may be possible to achieve the same contrast in the opposite phase case, but at the cost of significantly increasing the transmit power. Therefore, for cardiac imaging, accurate phasing of the transmit channels increases the efficiency of the B1+ leading to both increases in SNR and contrast between the structures within the heart.
Figure 6 illustrates four FLASH cine images of the sixty successive phases of the cardiac cycle along the short axis and four-chamber views. These data were collected in approximately 20 seconds with isotropic in-plane resolution equal to 1.8 mm and a slice thickness of 5.5 mm. These cine frames show excellent spatial and temporal resolution throughout the cardiac cycle. Despite the fact that B1 shimming was performed on a single axial slice, Figure 6 demonstrates that single slice B1+ is sufficient to provide a reasonable global B1 shim for the heart. The short axis images accurately show the ventricles, pericardium and pericardial fluid as well as the posterior descending artery, anterior descending artery and great cardiac vein. Strong signal and contrast are seen between the blood and the posterior wall of the heart. In the four-chamber view the atria, ventricles, mitral valve and tricuspid valve are clearly identifiable.
FLASH cines, as in Figure 6, are often used to assess cardiac function. High SNR and myocardial/blood contrast in the short and four-chambered axes would easily allow for calculation of ejection fraction, valvular function, and other clinically relevant measurements.
Figure 7 shows the short axis (a) FLASH cine and (b) TSE from the same subject. The image resolution of the TSE image is 1.8 mm by 1.5 mm by 5.5 mm. Despite the lower signal at the posterior wall of the heart, these images show accurate cardiac morphology.
TSE imaging sequences are used for measuring cardiac morphology. The high myocardial signal and nearly complete suppression of the inflowing blood in the TSE sequence facilitate imaging of anatomic features as demonstrated in the short axis image of Figure 7. TSE images are extremely sensitive to B1+ non-uniformities. Fig.7b demonstrates that, however, even this sequence can be used for cardiac imaging at 7 Tesla with the approach described in this paper. The degraded signal in the posterior wall of the heart in Figure 7b, is likely caused by the extreme sensitivity of the TSE sequence to B1+ inhomogeneities. Better B1 shimming using arrays with larger number of elements and/or better optimization algorithms is expected and this in turn should improve the signal homogeneity and intensity along the posterior wall of the heart.
A comparison with what can be attained at other fields is not provided in this paper. Due to the significant methodological differences it is difficult, to compare these initial cardiac results at 7 T to those acquired at clinical field strengths. Different field strengths require different imaging sequences. At 1.5 T, SSFP sequences are the standard for functional cardiac imaging, these sequences, however, work with only moderate success at 3T and are currently impractical at 7 T. Functional imaging at 7 T is currently limited to FLASH cines. Additionally, coil configurations are also different at the different field strengths. Clinical cardiac studies (both 1.5 T and 3 T) employ a volume transmit coil used in conjunction with local receiver arrays. This combination is known to improve both spatial coverage and SNR when compared to local transceive coils. A surface transceive array was used in this study because volume transmit-only 7 T body coils have not been tested. Finally, B1+ and B0 inhomogeneities are significantly greater at 7 T when compared to clinical field strengths. Despite remarkable advances in B1+ and B0 shimming technique, (at both 3 T and 7 T) further advances are requisite before an accurate comparison can be made. Undoubtedly, 7 T cardiac imaging will improve as hardware, protocols and methods are optimized for higher field strengths.
The feasibility of human cardiac MRI is demonstrated at 7 T. These early results have been achieved by solving or compensating for a number of known challenges for cardiac and ultra-high field whole body imaging related to: 1) increased MHD artifacts in the ECG waveform, 2) nonuniform RF fields due to complex destructive interference patterns, and 3) increased SAR due to increased high frequency energy losses due to the tissue permittivity and conductivity. High-resolution FLASH cines and TSE images have been acquired with an eight-channel local transceiver array phased by a local B1+ phase shimming technique.