PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
 
Magn Reson Med. Author manuscript; available in PMC 2011 November 9.
Published in final edited form as:
PMCID: PMC3212435
NIHMSID: NIHMS313492

Real Time MR Thermometry for Monitoring HIFU Ablations of the Liver

Abstract

A high resolution and high speed pulse sequence is presented for monitoring high intensity focused ultrasound (HIFU) ablations in the liver in the presence of motion. The sequence utilizes polynomial-order phase saturation bands to perform outer volume suppression, followed by spatial-spectral excitation and three readout segmented EPI interleaves. Images are processed with referenceless thermometry to create temperature rise images every frame. The sequence and reconstruction were implemented in RTHawk and used to image stationary and moving sonications in a polyacrylamide gel phantom (62.4 acoustic W, 50 sec, 550 kHz). Temperature rise images were compared between moving and stationary experiments. Heating spots and corresponding temperature rise plots matched very well. The stationary sonication had a temperature standard deviation of 0.15°C, compared to values of 0.28°C and 0.43°C measured for two manually-moved sonications at different velocities. Moving the phantom (while not heating) with respect to the transducer did not cause false temperature rises, despite susceptibility changes. The system was tested on non-heated livers of 5 normal volunteers. The mean temperature rise was −0.05°C with a standard deviation of 1.48°C. This standard deviation is acceptable for monitoring HIFU ablations, suggesting real time imaging of moving HIFU sonications can be clinically possible.

Keywords: real time MR-thermometry, MRg-FUS, HIFU, referenceless thermometry

Introduction

High intensity focused ultrasound (HIFU) is a promising therapy tool capable of noninvasively depositing high amounts of energy deep inside tissue while minimally depositing energy elsewhere along its path. The amount of energy deposition at the focus of the transducer can be many times greater than the energy at the surface of the individual transducer array elements [1]. MRI has been valuable in guiding and monitoring these treatments, particularly for treatments of the prostate [2] and uterine fibroids [3], utilizing the temperature dependent proton resonant frequency (PRF) shift that occurs in water molecules. As these regions of treatment remain relatively stationary, PRF thermometry via baseline subtraction is simple and effective.

Unlike in the pelvic region, respiratory motion complicates MR-guided focused ultra-sound (MRgFUS) of the upper abdomen. HIFU sonications typically take many seconds to a minute to create enough heat inside tissue to cause coagulation necrosis. During this time, respiratory motion can cause the liver to move 10-26 millimeters in the cranio-caudal direction during quiet inspiration, and even more during deep inspiration [4]. This free breathing, coupled with the rapid heating time, has implications for temperature monitoring. First, temperature monitoring can no longer use the straightforward baseline subtraction technique. Second, imaging must be rapid, not only to prevent blurring due to motion, but to provide feedback and control for slewing the ultrasound beam to follow the desired sonication region.

Various approaches have been implemented to monitor abdominal ablations in previous work. General anesthesia combined with breath holds [5] or single lung ventilation [6] makes conventional temperature MRI techniques applicable, but not all patients may qualify for breath holds or general anesthesia, or both. If no breath holds is a criterion, another strategy is a multi-baseline approach [7,8]. In this approach, background phase information prior to heating is collected at various stages of the respiratory cycle so that baseline data exists for all positions of the organ; the baseline subtraction is done by first matching the image acquired during heating with the corresponding baseline data acquired prior to heating. In this way, the appropriate baseline subtraction is performed, mitigating motion-induced mis-registration. This approach requires additional setup time and computation, and the scan plane is constrained to the plane in which the baseline data is collected; it does not allow changes on the fly during sonication. A real time monitoring method that does not require the acquisition of multiple baseline data would be a simpler and more flexible approach.

Real time MR thermometry requires several components: baseline-free temperature reconstructions, a rapid pulse sequence, and real time control of imaging parameters. With respect to baseline-free temperature reconstructions, our group [9] and others [10] have developed referenceless algorithms to calculate temperature changes from a single image. Real time control can be implemented with RTHawk (HeartVista, Inc., Los Altos, CA), a software environment capable of controlling slice orientation, slice parameters, and reconstructions without needing to stop a scan or sonication [11].

This paper focuses on the development of a fast pulse sequence for MR thermometry and its implementation into the real time system. The key elements of the high-speed pulse sequence are high time bandwidth polynomial-phase spatial saturation bands to limit the field of view, a spectral spatial excitation pulse to excite only water in the slice, and a fly-back readout segmented EPI pulse sequence to read multiple k-space lines at once. Two EPI sequences, conventional flyback and flyback readout-segmented, are considered and discussed. Integrating this sequence with a referenceless thermometry algorithm and RTHawk, the method was tested during HIFU sonications in a moving phantom and compared to stationary sonications. In vivo experiments were also performed in healthy volunteers in order to evaluate the precision of temperature measurements in the liver with no heat applied.

Methods

Real time temperature monitoring of focused ultrasound ablations during free breathing has the following requirements: high spatial and temporal resolution, selective water excitation, and a long TE (compared to normal real time sequences). The pulse sequence we designed to meet these criteria consists of three key elements: outer volume suppression, spatial-spectral excitation, and an interleaved flyback echo planar (EPI) readout. These elements are described in more detail, as is the real time reconstruction and display system.

Excitation

Outer volume suppression was performed with cosine modulated Shinnar-Le Roux designed polynomial phase saturation pulses [12]. Two saturation bands of 15 cm restricted the image FOV to 8 cm in the phase encode direction. Polynomial phase saturation pulses allow for sharper saturation boundaries while staying within peak B1 constraints (20 μT). However, saturation pulses are susceptible to B1 inhomogeneity and T1 relaxation. Thus, to compensate for these effects, multiple saturation pulses were utilized with gradient spoiling in between, as suggested by Wilm [13]. The spatial-spectral (SPSP) excitation was a 1-2-1 binomial subpulse RF pulse with a flyback design. Its specifications and application to imaging at 3T are discussed by Santos et al [14].

Readout

The readout sequence was subject to two constraints in addition to speed: there must be sufficient time for phase to evolve, and off-resonance effects should be minimized. For EPI sequences, off-resonance effects can be mitigated through the use of multiple interleaves. Two methods of echo planar imaging were investigated for real time thermometry: an interleaved flyback EPI (IFB-EPI) and an interleaved flyback readout segmented EPI (RS-EPI) [15,16]. In the RS-EPI approach, the interleaves, or blinds, are shortened by breaking up each readout line, while in IFB-EPI a complete phase encode line is collected, but subsequent phase encode lines are skipped for the next interleaves. Thus, for an RS-EPI readout with N interleaves, there are no breaks in the phase encode direction and (N-1) breaks in the frequency encode direction. For an N interleaf IFB-EPI readout, there are continuous breaks along the phase encode direction, but no breaks in the frequency encode direction.

Both readout trajectories were simulated with MATLAB (The Mathworks, Natick, MA), to investigate the effect of a moving impulse. The impulse was modeled as a 2D delta function moving with a constant speed in one direction, passing through the origin halfway through the simulated readout time. During quiet inspiration, the liver moves 10-26 millimeters, and assuming an average of 12 breaths per minute, the liver would on average move between 4-10.4 mm/sec. Thus, constant speeds of 4 mm/sec and 10.4 mm/sec were simulated for both trajectories, moving in both the frequency and phase encode direction. The simulation was subject to the following sequence parameters for IFB-EPI: 3 interleaves, TR = 97 ms, 42.25 ms of which was used for reading out the image, resolution = 1.4×1.4 mm2, and a grid size of 57×57. For RS-EPI, the TR was lengthened to 117ms, with a readout time of 64.25 ms. For each k-space pixel, the time at which that pixel was collected was determined based on the particular readout trajectory, TR, and portion of TR devoted to readout. This time was then used to calculate the displacement of the impulse and the resultant phase shift for the pixel. The inverse Fourier transform was then applied to observe the blurring and artifact effects.

Additionally, both sequences were tested in a phantom on a GE Signa Excite 3T scanner (GE Healthcare, Waukeshaw, WI). The sequence parameters matched the simulation and for the IFB-EPI were: TE = 16.1 ms, TR = 97 ms, 8×20 mm2 FOV,1.4×1.4 mm2 resolution, 3 interleaves, offset by 14 phase encodes, 3.43 fps. For the RS-EPI sequence: TE = 15.9 ms, TR = 117 ms, 8×20 mm2 FOV,1.4×1.4 mm2 resolution, 3 interleaves, offset by 17 phase encodes, 2.84 fps. The scanner's rocker program was used, moving the phantom back and forth 20 mm at 13 mm/sec. Image quality during motion and rest were compared between the two sequences.

For the rest of the study, we used the RS-EPI readout. As seen in the phantom readout experiment, given the desired resolution the magnet's slew rate and gradient amplitude limitations, the readout time for a flyback RS-EPI sequence could be long. One remedy was an increase in the number of interleaves. This was less desirable since the sequence efficiency (time collecting data compared to time scanning) would be less, given the extra time needed to run saturation and excitation pulses. Therefore, we shifted the readout trajectory by 20 lines in k-space (59% coverage) to reduce the TE to 16.2 ms. Optimum phase SNR occurs around TE = T2* [17,18], which at 3T for normal liver tissue has been found to be 14.5 ± 6.7 ms [19]. Similarly, off-center EPI has been shown to be an effective and accurate method for acquiring temperature data [20], as well as other phase data [21]. No partial-Fourier corrections were applied, and the phase rolls resulting from the off-center trajectory were corrected with zero padding before the gridding reconstruction.

The pulse sequence, shown in Figure 1, was designed with the following specific pulse sequence parameters: TE = 16.2 ms, TR = 122 ms, 8×8 cm2 FOV, 1.4×1.4×4.7mm3 voxel size, 3 off-center flyback RS-EPI interleaves, and a frame rate of 2.73 fps. Even though it was designed for 8 cm, the readout FOV size was variable, and rectangular FOVs of up to 8 × 20 cm2 were used for reconstruction to allow for more anatomical information in the frequency encode direction. The sequence was flow compensated along kx and at ky = 0. Since the scanner was capable of a maximum gradient strength of 4 G/cm and maximum slew rate of 150 mT/m/ms, the pulse sequence was designed for maximum gradient strengths of 2.83 G/cm and 106 mT/m/ms to allow for real time free slice rotation (no oblique slice will cause the gradient strength and slew rate on a single physical gradient to be greater than the maxima).

Figure 1
Real time pulse sequence, consisting of spatial saturation, spatial spectral excitation, and a flyback RS-EPI readout trajectory. The inset shows the typical RS-EPI trajectory k-space plot.

Real Time Imaging

The pulse sequence was programmed using RTHawk [11], a real time environment for our scanner. RTHawk allows for continuous imaging and reconstruction, while also allowing instant changes in pulse sequence parameters such as FOV, slice thickness, TR, and flip angle. Additionally, slices can be selected on the fly in any orientation, ideal for picking an optimal orientation to monitor sonications in motion and the reason for designing the pulse with smaller and slower gradients.

The reconstruction pipeline in RTHawk is outlined in Figure 2. K-space data were gridded and Fourier transformed into complex images. For data from multiple coils, the images from each coil were added together using a weighted mean approach [22], multiplying each by a phase offset specific for the imaged area of interest. This offset is calculated from the average unwrapped phase in a region of interest at or near the hot spot location. The phase of the combined image was unwrapped, and referenceless thermometry images were created from user-drawn ROIs, as described by Rieke [9]. Two ROIs were defined: a region encompassing the area where the sonication and heating would occur, and a surrounding region where heating would not occur. For our experiments, the surrounding, nonheated band began at about 2 cm from the hot spot with a width of about 3 mm, although this can be arbitrarily modified by the user. A 5th order polynomial fit was determined from the surrounding, non heated frame and was then subtracted from these ROIs to determine temperature. The polynomial fit was determined for each new image from the region and subtracted from the entire ROI. This temperature information was overlain onto a magnitude image and displayed to the user.

Figure 2
Temperature rise images are created in RTHawk via the above reconstruction pipeline. For the phantom sonication experiments, when the an automatic tracking mode could be activated, and a new ROI is estimated from the data and the temperature calculations ...

Additionally, the user could turn on a tracking mode in RTHawk. In this case, the amount of motion between frames was assumed to be small in relation to the drawn ROIs (despite the motion, the maximum peak was still in the region deemed to contain the sonication). An image would be acquired and the temperature information was processed for the current ROI, which while not centered on the sonication now still contained it. The location of the maximum temperature rise was determined, and both ROIs were recentered at this position. The polynomial fitting and temperature calculations were then repeated. Maximum temperature rises as well as pixel locations for these temperatures were exported for analysis. This tracking mode was not designed for in vivo experiments, but merely served as a rapid way of shifting the referenceless ROIs around during the phantom motion.

Focused Ultrasound Experiments in a Phantom

HIFU sonications were performed on a polyacrylamide gel phantom using an InSightec Ex-Ablate 2000 (InSightec Ltd, Haifa, Israel) HIFU system in our scanner. The transducer was a 1000 element extracorporeal planar transducer with center frequency of 550 kHz, and 62.4 W acoustic power was delivered for 50 seconds during each sonication. First, sonications were performed at one location without motion. Following this, the experiment was repeated at the same location, moving manually both the transducer and phantom along the z-axis as shown in Figure 3, keeping the sonication within the field of view. The experiment was then repeated with the scanner's rocker program, moving at a minimum speed of 13 mm/sec. After six seconds of initial sonication, the rocker was programmed to move 2.5 cm back and forth in the scanner at 13 mm/s, pausing for 6 seconds at each terminus. The motion was along the phase encode direction of the image to minimize errors from ringing artifacts from interleaving the RS-EPI. Temperature rise images and maximum temperature rise plots were created using RTHawk for all of the phantom results.

Figure 3
Diagram of transducer and phantom setup (left). The transducer and phantom move in unison into and out of the magnet bore while imaging the slice of the sonication. For the manual sonications, a wooden rake extended into the bore to push the setup in ...

Although the polynomial fit and baseline calculation and subtraction are performed for each time point, the effects of susceptibility changes was investigated by placing the transducer underneath the rails that guide the motion for the manual motion experiment. A coronal slice about 12 cm from the transducer was acquired through the phantom. Without heating, the phantom was moved in and out of the bore, although at a slower rate than the prior experiment. The transducer was stationary during this motion. ROIs similar to those used in the heating experiment were placed in the phantom and moved to follow the phantom's change in location. Temperature rise maps were created in these ROIs to assess temperature as a function of position.

Real Time Tests In Vivo

Five normal volunteers were scanned using an 8-channel cardiac array coil centered on the liver. The volunteers placed their arms behind their heads to limit the need for an additional 10 cm of outer volume saturation per arm. Images of the liver were acquired with only the real time sequence, and the temperature was calculated from ROIs similar to the heating experiments, even though no heat was being applied. The regions were manually moved to follow a section of the liver as the volunteer breathed. As no out of plane motion correction was done, these measurements were only performed on sagittal and coronal images. The average temperature rise, standard deviation, and any contributions from overall noise and phase from blood vessels were analyzed.

Results

Pulse Sequence Simulation Experiments

The effects of the discontinuities between interleaves are shown in Figure 4. The primary difference between the two readout trajectories are the ghosts apparent in the IFB-EPI approach. These ghosts are of the same order of magnitude as the blurred impulse, and they are evident both in the low speed and higher speed simulations. For the RS-EPI trajectories the artifact produced by the discontinuities is ringing that decreases with distance from the original impulse location. At the image full range the ringing artifacts are barely evident; they are at least an order of magnitude less than the center. For the experimental motion results, regardless of motion direction in the IFB-EPI sequence ghosts are present along the phase encode direction. For the RS-EPI sequence there is no ghosting for motion in either direction, but motion causes blurring when it is along the frequency-encode direction. In light of the fact that ghosts could complicate the referenceless thermometry polynomial fit algorithm, RS-EPI was chosen for the readout trajectory of the pulse sequence. Furthermore, for optimal performance during motion, a RS-EPI sequence should have the motion aligned with its phase encode direction.

Figure 4
Response of motion simulations for two speeds for both IFB-EPI and RS-EPI readout trajectories are shown in the first four columns. In the right two columns are phantom images acquired with the two readouts. The direction of motion is indicated on the ...

Phantom Experiments

Plots from the susceptibility experiment are shown in Figure 5. Over a range of 10 cm of motion, the apparent temperature, with a mean rise of -0.11°C, appeared to have no correlation to location.

Figure 5
Results of susceptibility test. The top plot shows the position of the phantom (and ROI) over time while the transducer remains stationary at approximately 11 cm, and the bottom plot shows the mean temperature rise in the ROI as a function of location. ...

The pixel with the greatest temperature rise was plotted over time for each of the phantom experiments, as well as the motion velocity, which was determined automatically from the pixel shift of the maximum temperature rise. Plots from moving the phantom with the scanner rocker and manually moving are shown in Figure 6. For velocity measurements, the increase in noise after the ultrasound was turned o was attributed to the fact that during heating, a single pixel was the hottest area. During cooling the hottest area was a broader region as the heat dissipated, hence the maximum pixel location would fluctuate more.

Figure 6
Maximum temperature rise and calculated velocity plots for both the rocking motion experiments with a constant speed of 13 mm/s (left) and manual motion experiments with variable speeds (right). Both experiments were repeated. All trials compared favorably ...

Even with calculated velocities of 13 mm/s and greater, all of the moving sonications corresponded well with the stationary sonications. The stationary plot was smoothed and subtracted from the stationary and two manual sonication plots, from which temperature standard deviations were calculated. Temperature standard deviation for the stationary sonication was calculated from all the time points, and for the moving sonications only during the period of motion (approximately from 20 to 100 seconds). The standard deviation was 0.15°C for the stationary sonication, 0.28°C for the “Manual 1” sonication, and 0.43°C for the “Manual 2” sonication. All of these were deemed acceptable accuracies.

The top row of Figure 7 shows representative images from the phantom experiments. The left image was taken with the phantom at rest. The middle image was acquired while the phantom was moving at 13 mm/s with the table rocker control, while the right column depicts that of a sonication in the phantom during manual motion. The images from the rocker experiments were often disturbed by artifacts from electronic noise during motion, usually in the form of high frequency sinusoid patterns superimposed on the image. In spite of this, with the images overlaying temperature rises between 5-20°C, the three sonications look nearly indistinguishable, despite the in-plane motion.

Figure 7
Representative images from ultrasound sonication while the setup is stationary (a), rocking automatically via the scanner (b), and moving manually (c) are shown along the top row. Axial (d), coronal (e), and sagittal (f) images of the abdomen are shown ...

In Vivo Experiments

The bottom row of Figure 7 shows representative images of the liver, with a color overlay for a rectangular regions of interest. Even though the temperature measurements were not performed on the axial slices, an image is provided for comparison. The saturation bands have made these images near-alias free.

Example plots of the mean, standard deviation, and percentage of pixels over 5 degrees temperature rise in the region of interest are shown in Figure 8. The top plot shows the motion of the ROIs as it tracked the particular location in the liver. Jumps in the X location signified a user initiated slice location change and the ROI was moved to a different section of the liver for tracking. Average statistics from this study are presented in Table 1. These are combined from the data from both coronal and sagittal image sets per person. The mean is around 0°C with a standard deviation in the range of 1 to 2°C. The percentage of pixels over 5 degrees (and 10 degrees) rise appeared to be a function of vasculature in the region of interest. To illustrate this and to test the magnitude of the artifact, two regions in one data set from volunteer #5 were analyzed: one area with very few vessels and another with large ones. Images showing the ROI and sample temperature rise overlays are shown in Figure 9, as well as a time plot of the percentage of pixels greater than 5 and 10 degrees temperature rise. The vessel region has more pixels with a “higher” temperature rise, but that percentage is still low: 3.95% under 5°C rise and 0.37% under 10°C rise.

Figure 8
Example time course plots for a volunteer (#4) for the sagittal (left) and coronal (right) plane. The top graphs show the relative positions of the ROIs drawn inside the image. Sharp changes in the X location indicate a change in slice by the user's modification ...
Figure 9
The plots (right) show the percentage of pixels in the ROI that register above 5°C or 10°C. This was done for an ROI without vessels (top) and an ROI with vessels (bottom). The image (left) of an oblique slices through the liver shows ...
Table 1
Statistics from in vivo studies.

Discussion

The results of this study demonstrate one approach to real time MR thermometry of the liver. With this pulse sequence, we are now capable of producing high-resolution temperature images at a high frame rate. The regions of elevated temperature moved with the phantom while maintaining their shapes and matching the temperatures of stationary spots without the need to collect baseline images for all possible phases of the motion cycle. Additionally, the in vivo results had relatively low noise (<2°C) compared to the desired sonication temperature rises, which are on the order of 15-30 degrees above baseline.

The results of the study show that the sequence is capable of providing alias free images of the abdomen for thermometry. Outer volume suppression worked well, even in the superior/inferior direction, as did the selective water excitation, most evident by the suppression of fat under the skin. Further improvement of water excitation can be performed by dynamic shimming of regions of interest to ensure the frequency is correct for this region [14]. The RS-EPI sequence showed no ghosting artifacts from the motion, as expected from the simulation and readout experiments, and the ringing seen in simulations did not appear to affect the temperature measurements either. Furthermore, the referenceless reconstruction was fast enough for the sequence to continuously acquire data and display it.

The pulse sequence is relatively simple and requires little preprocessing to start calculating temperature rises in the images. An alternative to outer volume suppression would be to utilize parallel imaging, which can increase frame rates for full FOV imaging. As we are currently only interested in a small region of the abdomen, the region primarily around the sonication, we chose our reduced FOV method to benefit from the reduced reconstruction time of using a non-parallel imaging algorithm.

The simulations and readout phantom experiment highlighted the important benefit of readout segmented over traditional interleaved EPI approaches: the absence of ghosts in images. Ghosts could affect the referenceless algorithm's ability to determine the background phase and corrupt the temperature data itself. RS-EPI was also selected over other considered trajectories, including flyback segmented (non-interleaved) EPI, spiral sequences, and non-flyback EPI sequences. These others were not selected due to their off-resonance effects. These sequences do not have the ability to reduce off-resonance effects to pixel shifts like both IFB and RS-EPI. Thus, even when there is no motion these sequences would suffer from artifacts. Additionally, these sequences (except flyback segmented EPI) require more processing in the real time environment compared to the other flyback approaches to correct for gradient and timing errors. RS-EPI does exhibit a couple drawbacks: ringing artifacts and T2* effects from a long readout trajectory. Ringing artifacts could potentially be reduced with a slight overlap at the boundaries between shots. The overlap would allow for a smoother transition between data from one shot and the subsequent one. With respect to T2*, the readout time is long (>60ms), and with T2* in the liver potentially under 20ms, the actual resolution can potentially be much worse than the theoretical. The resolution could thus be optimized with high order shimming or calculating T2* and prescribing the readout sequence that will best minimize its effect.

With our sequence, the images from the phantom experiments showed a good depiction of the circularly growing hot spots, even in the presence of motion that can potentially distort the shape. The tracking algorithm kept the ROI centered on the sonication during motion, and the referenceless algorithm was able to subtract the background phase as it varied with the objects changing position inside the magnet.

For the rocker experiments in the top plot of Figure 6, there was a slight dip in the temperature rise during part of the motion. The effect was more pronounced in the second experiment than the first. This is probably due to a combination of effects. In the rocking experiments, the cradle mover introduced electronic noise that corrupted the images while moving. This could cause an under- or overestimation of temperature change depending on the artifact's effect on the image. Additionally, partial volume effects and blurring could lead to an underestimation. However, a velocity of 13 mm/sec is on the high end of normal respiratory motion. Again, assuming average motion between 4-10.4 mm/sec, motion-related errors would be 30-80% less than in the rocker case.

The first manual motion experiment was generally at a slower speed than the rocker, as evidenced by the lower left plot of Figure 6, and since the setup was moved manually there was no electronic noise in the phantom images. In this case, the rate of motion was less than 10 mm/sec during most of the sonications, and the moving temperature plot appeared to match the stationary temperature plot almost identically. The low temperature standard deviation confirms this. The second manual motion experiment (bottom right) was moved at higher speeds, and more variability can be seen in the temperature plot comparison and by the higher temperature standard deviation. However, it is still acceptable and promising for imaging HIFU ablations in vivo.

Figure 5 showed that the referenceless reconstruction was relatively immune to susceptibility changes at a distance likely for treatment. This was over a range of 10 cm, much larger than typical respiration induced motion. Furthermore, in addition to having no dependence on position, the susceptibility changes did not cause false temperature rises: the maximum apparent temperature rise during this motion was 0.62°C.

The in vivo temperature data had lower SNR than the phantom experiment, as expected given the coil geometries and MR parameters of liver compared to the gel phantom. However, the mean temperature rise in vivo remained close to 0 degrees, with a standard deviation around 1.5 degrees, shown in Table 1. This standard deviation is sufficient for monitoring temperature, and overall these results are very promising for monitoring an in vivo HIFU ablation. Besides mean and standard deviation, maximum temperature rise is also a concern for measuring thermal dose. False temperature spikes did occur, especially in and around larger vessels, as shown by Figure 9. This is expected given the fact that flowing vessels accumulate phase during magnetic field gradients. The thermal dose equation for determining coagulative necrosis is given by:

equation M1
(1)

where k = 2 for temperatures greater than 43°C and k = 4 for temperatures less than 43°C. A 10 degree temperature rise for a duration of 50 seconds results in an equivalent thermal dose of only 13.3 minutes, compared to greater than 240 minutes that we expect in the ablated region. Thus, these low apparent temperature errors should be insignificant when determining the damage from an actual ablation.

In addition, the few pixels that sometimes spike over 10 degrees have an apparent thermal dose footprint much smaller than the actual sonication, and potentially could be filtered out with further signal processing. Additionally, flow compensating the excitation pulse could help mitigate some false temperatures from vessels, at the cost of a small increase in scan time.

The saturation bands appeared to adequately limit the field of view, despite potential B1 variations. A current limitation of the outer volume suppression is that it utilizes cosine modulation to create the distinct saturation bands. Because of this, we cannot apply different flips to each individual band. By making these pulses independent of each other, the flips could be even better tuned to saturate the regions, should the B1 profiles in each region be very different. Regardless, the combination of the saturation bands (15 cm) and the sensitivity of the coil significantly reduced aliasing, even with saturation along the S/I direction.

During the volunteer study, the Specific Absorption Rate (SAR) was monitored and was dependent on the flip angles of the saturation bands. SAR could become a problem if the pulse sequence were to be run continuously without end. In our experience the sequence could be run for several minutes before the saturation band flip angle required paring down. This is not likely to be a limitation, since sonications will require intermittent cool down periods afterwards.

The pulse sequence could be prescribed and run independently of RTHawk, but its real advantages are evident when combined. Besides slice orientation and selection, RTHawk can act as a complete HIFU therapy control system. It has the capability of additional inputs like respiration monitoring. The feedback from the MRI and physiological monitoring can potentially be used to control the ultrasound, particularly for beam steering for motion tracking or treating larger volumes in ways shown by Mougenot [23]. Additionally, RTHawk can instantly switch between different pulse sequences, allowing for an efficient treatment session from initial planning to final verification. This sequence, combined with RTHawk, will play an important role in directing HIFU therapy and is the focus of our future work.

Conclusion

The synergy of our fast pulse sequence, referenceless thermometry, and RTHawk will create the potential for real time therapy sessions. Our sonications experiments have shown that high-resolution images with accurate temperature overlays can be created for motion velocities common during normal respiration. Similarly, the in vivo results demonstrated its suitability for in vivo imaging by calculating a zero mean temperature rise data with minimal variation in non-heated healthy volunteers. The combined experiments suggest that real time thermometry in the presence of motion can be possible and accurate, and is a promising technique for monitoring HIFU ablations.

Acknowledgements

We would like to acknowledge Dr. Will Grissom from Stanford for assisting with some of the data collection. Also, we thank Dr. Sonal Josan from Stanford and Dr. Yoav Medan from InSightec for helpful discussions.

This work was supported by NIH RO1 CA121163 and P41RR09784.

References

1. Kennedy JE, TerHaar GR, Cranston D. High intensity focused ultrasound: surgery of the future? Br J Radiol. 2003;76:590–599. [PubMed]
2. Murat FJ, Gelet A. Current status of high-intensity focused ultrasound for prostate cancer: Technology, clinical outcomes, and future. Current Urology Reports. 2008;9:113–121. [PubMed]
3. Chapman A, ter Haar G. Thermal ablation of uterine fibroids using mr-guided focused ultrasound-a truly non-invasive treatment modality. European Radiology. 2007;17:2505–2511. [PubMed]
4. Clifford MA, Banovac F, Levy E, Cleary K. Assessment of hepatic motion secondary to respiration for computer assisted interventions. Comput Aided Surg. 2002;7:291–299. [PubMed]
5. Gedroyc WM. New clinical applications of magnetic resonance-guided focused ultrasound. Top Magn Reson Imaging. 2006;17:189–194. [PubMed]
6. Illing RO, Kennedy JE, Wu F, ter Haar GR, Protheroe AS, Friend PJ, Gleeson FV, Cranston DW, Phillips RR, Middleton MR. The safety and feasibility of extracorporeal high-intensity focused ultrasound (hifu) for the treatment of liver and kidney tumours in a western population. Br J Cancer. 2005;93:890–895. [PMC free article] [PubMed]
7. Vigen KK, Daniel BL, Pauly JM, Butts K. Triggered, navigated, multi-baseline method for proton resonance frequency temperature mapping with respiratory motion. Magn Reson Med. 2003;50:1003–1010. [PubMed]
8. Denis de Senneville B, Quesson B, Desbarats P, Salomir R, Palussiere J, Moonen C. Atlas-based motion correction for on-line MR temperature mapping.. Proceedings of the IEEE International Conference on Image Processing (ICIP 2004); Singapore. 2004.pp. 2571–2574.
9. Rieke V, Vigen KK, Sommer G, Daniel BL, Pauly JM, Butts K. Referenceless prf shift thermometry. Magn Reson Med. 2004;51:1223–1231. [PubMed]
10. Kuroda K, Kokuryo D, Kumamoto E, Suzuki K, Matsuoka Y, Keserci B. Optimization of self-reference thermometry using complex field estimation. Magn Reson Med. 2006;56:835–843. [PubMed]
11. Santos JM, Wright GA, Pauly JM. Flexible real-time magnetic resonance imaging framework. Conf Proc IEEE Eng Med Biol Soc. 2004;2:1048–1051. [PubMed]
12. Schulte RF, Henning A, Tsao J, Boesiger P, Pruessmann KP. Design of broadband rf pulses with polynomial-phase response. J Magn Reson. 2007;186:167–175. [PubMed]
13. Wilm BJ, Svensson J, Henning A, Pruessmann KP, Boesiger P, Kollias SS. Reduced field-of-view mri using outer volume suppression for spinal cord diffusion imaging. Magn Reson Med. 2007;57:625–630. [PubMed]
14. Santos JM, Cunningham CH, Lustig M, Hargreaves BA, Hu BS, Nishimura DG, Pauly JM. Single breath-hold whole-heart mra using variable-density spirals at 3t. Magn Reson Med. 2006 Feb;55:371–379. [PubMed]
15. Robson MD, Anderson AW, Gore JC. Diffusion-weighted multiple shot echo planar imaging of humans without navigation. Magn Reson Med. 1997;38:82–88. [PubMed]
16. Holdsworth SJ, Skare S, Newbould RD, Guzmann R, Blevins NH, Bammer R. Readout-segmented epi for rapid high resolution diffusion imaging at 3 t. Eur J Radiol. 2008;65:36–46. [PMC free article] [PubMed]
17. Cline HE, Hynynen K, Schneider E, Hardy CJ, Maier SE, Watkins RD, Jolesz FA. Simultaneous magnetic resonance phase and magnitude temperature maps in muscle. Magn Reson Med. 1996;35:309–315. [PubMed]
18. Kuroda K, Mulkern R, Oshio K, Panych L, Nakai T, Moriya T, Okuda S, Hynynen K, Jolesz F. Temperature mapping using the water proton chemical shift: self-referenced method with echo-planar spectroscopic imaging. Magn Reson Med. 2000;43:220–225. [PubMed]
19. Storey P, Thompson AA, Carqueville CL, Wood JC, de Freitas RA, Rigsby CK. R2* imaging of transfusional iron burden at 3t and comparison with 1.5t. J Magn Reson Imaging. 2007 Mar;25:540–547. [PMC free article] [PubMed]
20. Stafford RJ, Price RE, Diederich CJ, Kangasniemi M, Olsson LE, Hazle JD. Interleaved echo-planar imaging for fast multiplanar magnetic resonance temperature imaging of ultrasound thermal ablation therapy. J Magn Reson Imaging. 2004 Oct;20:706–714. [PubMed]
21. Szarf G, Dori Y, Rettmann D, Tekes A, Nasir K, Amado L, Foo TKF, Bluemke DA. Zero filled partial fourier phase contrast mr imaging: in vitro and in vivo assessment. J Magn Reson Imaging. 2006 Jan;23:42–49. [PubMed]
22. Thunberg P, Karlsson M, Wigström L. Comparison of different methods for combining phase-contrast images obtained with multiple coils. Magnetic resonance imaging. 2005;23:795–799. [PubMed]
23. Mougenot C, Quesson B, de Senneville B, de Oliveira P, Sprinkhuizen S, Palussiere J, Grenier N, Moonen C. Three-dimensional spatial and temporal temperature control with MR thermometry-guided focused ultrasound (MRgHIFU). Magnetic Resonance in Medicine. 2009:603–614. [PubMed]