Multi-resolution MRI temperature monitoring in a reduced field-of-view

Abstract

Purpose

To develop a new MRI technique for fast temperature monitoring with extended volume coverage.

Materials and Methods

MURPS (Multiple Resolutions along Phase-Encode and Slice-Select Directions) was implemented in both a 2D SPGR sequence and in a multi-shot EPI sequence. Both modified sequences were used to acquire image data from three slices with variable phase-encode resolution and slice thickness. In the SPGR sequence, a 2DRF pulse was also implemented to enable imaging within a reduced FOV and this was used to monitor (at 1.5T) the temperature changes in a live rabbit and in gel phantoms heated by focused ultrasound. A modified EPI sequence was tested during heating of a phantom undergoing motion.

Results

The in-vivo experiments demonstrated that temperature changes in unexpected locations away from the focal plane, such as near bone structures, could be detected due to the extra volume coverage afforded by the MURPS method. Temperature changes in a moving phantom were resolved using the MURPS EPI sequence with an acquisition rate of three slices every 300 msec.

Conclusion

The MURPS method enables temperature monitoring over multiple slices without loss of temporal resolution compared with single-slice imaging and, if combined with multi-shot EPI, enables volume temperature monitoring in moving organs.

INTRODUCTION

MRI allows temperature mapping of thermal ablation by Focused ultrasound (FUS) (1, 2) and can provide guidance to the tumor in all three directions at the time of treatment. Temperature changes can be mapped using the temperature dependence of several MRI parameters, such as the water proton resonant frequency shift (3–8), the diffusion coefficient (D) of water (9–11), the longitudinal relaxation time T1 (12–15), or transverse relaxation time T2 (16). However, the water proton resonant frequency (PRF) shift is the most frequently used among these parameters (17). In general, changes in resonant frequency, which are directly proportional to the temperature change, are calculated from phase maps obtained during the heating time period using a gradient echo sequence (5, 8).

Relatively rapid MRI is required to track the temperature changes caused by FUS heating of the target volume during the thermal ablation. Generally, during FUS heating, the exposures (or ‘sonications’) are applied for from 1 to 60 seconds, producing a relatively small heated zone at a focal point. By steering the focal point to different locations, a volume can be ablated. During each sonication, low-level heating is produced outside of the focal zone in the near- and far-field of the ultrasound beam. This heating, especially in the near field, limits the speed at which one can target neighboring locations without overheating tissue in the beam path. In order to reduce the overall treatment time, which is a current limitation for FUS ablation, one needs to monitor the heating in the near field. This need, as well as a desire to monitor other critical structures outside of the focal zone, creates a strong demand for increased slice coverage.

Generally in MRI, one faces a tradeoff between spatial coverage, temporal resolution, and signal-to-noise ratio (SNR) and a typical tradeoff for MR temperature mapping is to limit the acquisition to a single slice. Several fast imaging techniques, including echo planar (18), spiral (19), PRESTO (20), parallel imaging (21), and reduced Field-of-View (rFOV) (22), have been applied to improve temporal resolution. These fast methods could, however, also be used to increase the slice coverage for temperature mapping during FUS ablation, instead of simply increasing temporal resolution.

Rather than relying solely on fast imaging techniques, one may also be able to take advantage of the spatial heating pattern produced during FUS ablation to further optimize the image acquisition. First, we note that the heated area generally occupies only a portion of the in-plane field-of-view (FOV), thus, it may be possible to reduce the FOV and thereby the number of phase encodes (as long as aliasing can be avoided). Second, the low-level heating that is produced in the near-field of the FUS beam is expected to be relatively broad and spatially diffuse, so we may be able to reduce the spatial resolution in those areas with little loss of information. This reduction in spatial resolution in the near-field might have an additional benefit of increasing the temperature sensitivity in these regions due to the increased SNR. As one will need to track the additive effect of small temperature changes produced during many sonications, temperature accuracy in these regions is critical.

Our goal in this work has been to develop a method that allows one to monitor temperature changes in the focal plane slice along with the additional slices covering the surrounding regions in the same time needed to monitor the original single slice. In this research, a new hybrid technique is proposed that combines a previously introduced method referred to as Multiple Resolution along Phase-encode and Slice-select-dimensions (MURPS) (23) with rFOV (24–25) using 2D spatially selective RF excitations (2DRF) (26–27).

In the MURPS method presented here, the voxel size of the focal plane slice is maintained while the slice thickness and the in-plane voxel size of the other two slices are increased. This is a direct trade-off of the spatial resolution in some parts of the volume for increased temporal resolution. The original MURPS method required some sacrifice in temporal resolution compared to imaging in a single slice. In the new MURPS method presented here, a 2DRF pulse is incorporated into the imaging sequence to enable reduced FOV imaging without aliasing. Thus, with the reduction in scan time due to the reduced FOV, the new MURPS method allows us to achieve the goal of covering three slices in the time normally taken to image one single slice.

The MURPS method is not generally sequence dependent; therefore, it can be implemented using most imaging sequences, including the fast approaches mentioned above. In this work, MURPS was implemented and tested using both a standard 2D spoiled gradient echo (SPGR) sequence as well as with a multi-shot echo-planar imaging (EPI) sequence.

MATERIALS AND METHODS

Implementation of MURPS in a SPGR sequence

The MURPS method was implemented with the ability to acquire three slices of variable thickness and variable phase-encode (P.E.) resolution as shown schematically at the top of Fig. 1. Typically, the central slice (labeled as A1 in Fig. 1), which has the highest spatial resolution, will be placed to cover the focal zone. The outer slices (labeled as B1 and B2 in Fig. 1) should be placed to cover regions away from the focus where the heating could result in undesirable and potentially dangerous effects. Note that Fig. 1 shows the three slices as being parallel and evenly spaced. It is possible on many modern MRI scanners, however, to prescribe these slices at arbitrary positions and with variable orientations.

Figure 1.

Top: Idealized profiles of the three MURPS slices A1, B1 and B2. Each block schematically represents the relative voxel size along the phase-encode direction. In this case, slice thickness is increased for the outer slices by 1.5 times and resolution in the phase-encoding (P.E.) direction is reduced by a factor of 2. Bottom: Schematic of the 2DRF pulse showing gradients in the slice-select (G_SS) and phase-encode directions (G_PE). The 3 ms pulse consists of 5 Gaussian-shaped sub-pulses of 0.6 ms duration. Weights of each sub-pulse are computed from a cosine-windowed SINC function. The effective time-bandwidth product of the pulse in the phase-encode direction is 4.

In our first implementation, the MURPS acquisition scheme was obtained by modifying a standard SPGR imaging sequence. The modification to the sequence consisted of adding: (1) the capability to change the slice selective gradient strength for each individual slice to give variable slice thickness; (2) the capability to appropriately change the excitation frequency when modifying the slice select gradient in order to maintain the excitation at the desired slice location; (3) including the ability to acquire variable number of phase encodes depending on the slice number; (4) the replacement of the standard slice selective pulse with a 2DRF pulse.

For the first pulse sequence modification, we note that slice thickness is controlled according to the following equation (28):

$$\mathrm{\Delta }z=\frac{2\pi \mathrm{\Delta }f}{\gamma G_z},$$ [1]

where Δf is the RF pulse bandwidth and γ is the gyromagnetic ratio. Any increase in the slice thickness Δz, therefore requires reducing the amplitude of the slice select gradient G_z if Δf is fixed. Thus, in our implementation, the pulse sequence was modified so that G_z was reduced for slices B1 and B2 to give thicker slices than for slice A1.

For the second pulse sequence modification, when changing G_z, we know that (28):

$$Z_{\mathit{slice}\mathit{position}}=\frac{2\pi f_{\mathit{offset}}}{\gamma G_z}.$$ [2]

Therefore, to guarantee the slice excitation remains at the same location as initially prescribed, Z_{slice position}, the frequency offset f_offset for each slice must be recalculated based on the new gradient amplitude G_z.

A third modification to the sequence was made to the phase encoding looping structure to enable us to acquire a reduced number of phase encodes for slice B1 and B2. As a result, the spatial resolution in the phase encode direction for these slices could be set separately from slice A1. Table 1 shows the scheme used whereby the phase-encode spatial resolution in slices B1 and B2 is one half of the spatial resolution for slice A1. Note that, in the modified excitation scheme, A1 is excited on every second TR and slices B1 and B2 on every fourth (where TR is defined as the shot-to-shot time). Thus the T1-weighting is different for slice A1 and the B slices. This will affect the relative SNR of the different slices, however, because we are using the PRF method for temperature mapping, it will not affect temperature measurements.

Table 1.

The ordering of slice excitations for the MURPS acquisition scheme. Each column in the table represents a different shot or TR period. The same excitation scheme is used for both the SPGR and multi-shot EPI sequences. In the original sequences, excitations for the three slices are interleaved (and the effective TR for each slice is 3*TR). In the MURPS excitation Slice A1 is excited every second shot (effective TR = 2*TR) whereas Slices B1 and B2 are excited every fourth shot (effective TR = 4*TR). Thus there is a different T1-weighting for Slice A1 compared to the B slices. In the EPI acquisition, Slice A1 was acquired in four shots while slices B1 and B2 were acquired with only two shots. For both the SPGR and the multi-shot EPI sequences, slice A1 will have twice the spatial resolution of slices B1 and B2.

Original slice index	A1	B1	B2	A1	B1	B2	A1	B1	B2	A1	B1	B2
Original shot index	1	1	1	2	2	2	3	3	3	4	4	4
Modified slice index	A1	B1	A1	B2	A1	B1	A1	B2	-	-	-	-
Modified shot index	1	1	2	1	3	2	4	2	-	-	-	-

Lastly, the pulse sequence was also modified to implement a 2DRF excitation pulse with echo planar trajectory to enable control over both slice thickness and FOV reduction in the phase encode direction. The 2DRF pulse consists of multiple Gaussian-shaped RF sub-pulses of short duration accompanying gradients blips for the gradient in the phase encode direction. The slice and phase gradients determine the echo-planar k-space trajectory and, along with the RF pulses, produce the spatial frequency weighting. A schematic of the 2DRF pulse with gradient and RF waveforms is shown at the bottom of Fig. 1. The pulse consists of 5 sub-pulses each of duration 0.6 ms (for a total pulse duration of 3 ms). The amplitude weights of the train of Gaussian sub-pulses describe a function whose Fourier transform defines the shape of the section profile in the phase-encoding direction. These weights were determined from a cosine-windowed SINC function. Based on these weights, the effective time-bandwidth product for the pulse in the phase-encoding direction was 4. The time-bandwidth product of the Gaussian sub-pulses was 2.8.

Using the 2DRF pulse, we only excite the in-plane region where we are interested to monitor temperature changes. In our particular implementation, we reduced the FOV in the phase encode direction by a factor of two without aliasing. Thus, the number of phase-encode steps required is reduced by one-half.

Implementation of MURPS in an EPI sequence

To achieve high temporal resolution suitable for temperature monitoring in moving organs, the acquisition scheme of a multiple-shot EPI sequence was also modified. The MURPS method was implemented in the EPI sequence to acquire three interleaved slices with multiple numbers of shots for the three slices. The focal plane slice A1 was encoded with four shots while slices B1 and B2 were encoded with only two shots (resulting in lower spatial resolution in those slices compared to slice A1). The interleaved acquisition scheme for the three MURPS slices is the same as for the SPGR sequence and is summarized in Table 1. The reduced FOV method with 2DRF was not implemented in the EPI approach, although this is the focus of current work.

Phantom experiments

A 2D spoiled gradient-echo (SPGR) sequence was modified for both MURPS and 2DRF excitation as described in the previous section. Three coronal slices of varying thickness and in-plane resolution were acquired in sequential mode using the modified SPGR sequence on a 1.5T MRI system (Signa, General Electric Healthcare, Milwaukee, WI) system to monitor heating by FUS using the proton resonance shift method. In the MURPS experiments, dynamic series of images were acquired at the rate of three multi-resolution slices every 3.5 seconds (20 time points in total).

The experimental setup is shown in Fig. 2. Slices B1 and B2 were spaced 5 mm and 16 mm away from the focal plane slice A1 as shown in the figure. The gel phantom (1), the tank containing degassed water (2) and the curved spherical transducer (frequency: 1.5 MHz, diameter/radius of curvature: 10/10 cm) (3) are also shown. For comparison, separate heating experiments were conducted and temperature imaging was done using an SPGR sequence with: (1) a full FOV, (2) with reduced FOV only and (3) with MURPS in a reduced FOV. Imaging parameters for the full FOV acquisition were TR/TE = 55/22 ms, flip angle = 30°, bandwidth = ±3.57 kHz, matrix size of 256×64, FOV = 20 cm. For the reduced FOV experiments without MURPS, the matrix size was 256×32 and the FOV in the phase encode direction was 10 cm. Table 2 compares imaging parameters for the different experiments with and without MURPS for each of the three slices. The voxel size of the focal slice (A1) remained unchanged in all three sets of experiments while the voxel size of the outer MURPS slices (B1 and B2) was increased from 7.3 mm³ to 29.3 mm³. Phase maps from heating experiments were obtained by taking the phase of each complex image set after it was divided by the complex baseline image obtained before heating began. The phase maps of the SPGR sequence were unwrapped using a Matlab function (Mathworks, Natick MA) and were converted into thermal maps using the proportionality relation between the temperature change and the proton resonance frequency shift (PRF) of approximately −0.01 ppm/°C (29).

Figure 2.

Axial view of gel phantom and setup used for a FUS experiment. A1, B1, and B2 are three unevenly spaced slices with variable thickness and location that were acquired during the FUS. Arrows point out (1) the gel phantom, (2) the degassed water tank, and (3) the curved spherical transducer.

Table 2.

Imaging parameters for three gel experiments: (1) full FOV, (2) 0.5 rFOV and (3) 0.5 rFOV with MURPS

Imaging Parameters	Slice A1				Slices B1 and B2
	Thickness (mm)	Phase Encodes	Scan Time (s)	Voxel size (mm3)	Thickness (mm)	Phase Encodes	Scan Time (s)	Voxel size (mm3)
Full FOV	3	64	3.52	7.3	3	64	3.52	7.3
rFOV	3	32	1.76	7.3	3	32	1.76	7.3
MURPS/rFOV	3	32	1.76	7.3	4.5	16	0.88	29.3

In other phantom experiments, the modified EPI sequence was used to monitor FUS heating of a moving gel phantom in a GE 1.5T Signa system. Three slices were prescribed in the coronal plane, slices B1 and B2 were spaced 10 mm on opposite sides of the focal plane slice A1. The imaging parameters used were flip angle = 30°, bandwidth = 62 kHz, matrix size of 256 × 128, FOV = 20 cm and TR/TE 107/12.8 ms. A dynamic series of images was acquired at the rate of three multi-resolution slices every 300 ms for over 300 acquisitions. The heat sonication time was 33 seconds with 25 W electrical power. The motion of the gel phantom was created using a manufacturer-provided utility function that rocks the MRI table during scanning, for our experiments, a distance of 5 mm along the bore with a velocity of 13 mm/s.

The images of the EPI sequence were corrected using the PLACE method combined with UNFOLD (30) to remove Nyquist ghosting artifacts. In the PLACE (31) method, the polarity of the sampling trajectory is alternated from one frame to the next and k-space lines in each frame are separated into negative and positive lines to form two set of images. The two image sets are processed using an UNFOLD (32) filter and combined to produce the final images. The phase maps were unwrapped and corrected for motion using 3D quality guided path algorithms in which a quality map for all pixels is computed (33) and used as a guidance path for the flood-fill unwrapping algorithm (33–34).

In-vivo experiment

An animal experiment was conducted according to a protocol approved by our institutional animal committee. A male New Zealand white rabbit (weight: 4 kg) was anesthetized using intramuscular injections of a mixture of 12 mg of sodium xylazine (Xyla-ject; Phoenix Pharmaceuticals, St Joseph, MO, USA) and 48 mg of ketamine hydrochloride (Abbott Laboratories, North Chicago, IL, USA) given per kg of body weight per hour. The fur on the skin was removed using clippers and depilatory cream. A transducer of 10 cm radius of curvature and 1.5 MHz resonant frequency was used to sonicate the thigh of the rabbit for 69 seconds at 50 watts sonication power. These parameters were selected to create a temperature rise of approximately 55°C. The same SPGR pulse sequence that was used in the phantom experiments was used in the in-vivo imaging for 20 time points. The imaging time for each volume of three axial slices was 7.7 sec, for a total imaging time of 2.34 min. The imaging parameters used were TR/TE = 60/22.8 ms, flip angle =30°, bandwidth = ±3.57 kHz, matrix size of 256×128, FOV = 20 cm.

The imaging slices were oriented along the direction of the ultrasound beam. Slice A1 was positioned to include the focal point of the FUS beam. Slices B1 and B2 were placed 11.2 and 8.7 mm from slice A1. As in the phantom experiments, the voxel size of A1 was the same in the three experiments at 3.7 mm³ while the voxel size of slices B1 and B2 in the MURPS experiments was 14.6 mm³.

SNR verification experiments

The signal to noise ratio (SNR) varies between slices according to slice thickness and number of phase encodes. The theoretical SNR value in any slice is proportional to:

$$\mathit{SNR}\propto \left(\frac{{\mathit{FOV}}_y}{\sqrt{P_y}}\right)·S,$$ [3]

where S is the slice thickness, P_y and FOV_y are the number of phase encodes and the FOV in the y-direction. In all experiments, both the number of frequency encodes and the FOV in the x-direction were fixed. To compare the SNR in the three slices in the MURPS acquisition to that of the standard SPGR sequence, we calculated their ratio as follows:

$${\left(\frac{{\mathit{SNR}}_{\text{Standard}}}{{\mathit{SNR}}_{\text{MURPS}}}\right)}_{\mathit{focal}\mathit{plane}}={\left(\frac{{\mathit{FOV}}_y}{\sqrt{P_y}}\right)}_{\text{Standard}}/{\left(\frac{{\mathit{FOV}}_y}{\sqrt{P_y}}\right)}_{\text{MURPS}}\times S_{\text{Standard}}/S_{\text{MURPS}}.$$ [4]

Note that the above equation ignores the different T1-weighting in the different slices. In the standard SPGR sequence, the single slice is excited on every shot (TR). In the MURPS acquisition, slice A1 is excited every second TR and slices B1 and B2 are excited every fourth TR. Thus, there will be a different T1-weighting that will have an effect on the SNR in each of the slices, which will be more significant at short TR’s. The actual temperature measurements should, however, not be affected because the PRF method is based on the phase of the images.

A phantom was used in the SNR verification experiments. In the first set of these experiments, the SNR of a single slice imaged by the standard sequence (SPGR) was compared to the SNR of slice A1 imaged using the MURPS sequence. Both slices were positioned at the focal plane. Slice thicknesses for slices in the experiments were the same. The FOV was reduced by factor of two in slice A1 which reduced the number of phase encodes (P_y) by a factor of two.

The second set of experiments compared the SNR of a single slice imaged by the standard SPGR sequence to the SNR of the MURPS B1 slice as follows: The number of phase encodes in the MURPS slice was reduced by a factor of four (a factor of two due to FOV reduction and a factor of two due to resolution reduction). Slice thickness was set at 6 mm for both slices B1 and the standard sequence slice. The position of the B1 slice was located at the focal plane of heating for comparison with the result obtained using the SPGR sequence. In a third set of experiments, the SNR in slice B2 was compared with the SNR for the result with the standard SPGR sequence using a similar setup as for the experiment with slice B1. The SNR in the temperature maps in phantom experiments was measured by taking the mean value in an ROI placed at the location of heating and dividing this by the standard deviation from another ROI selected outside the phantom in the background noise region.

RESULTS

Phantom results

The temperature changes in a selected ROI plotted for the three MURPS slices acquired from gel phantom using modified SPGR sequence is shown in Fig. 3a. Note that, as expected, the measured temperature was highest in the focal plane slice nearest the transducer and lowest in the slice farthest from the focal plane. Figure 3b shows temperature maps for the three MURPS slices of the gel phantom at a single time point. There is an evident SNR difference in the temperature maps of the three slices (i.e. lower SNR in slice A1 as predicted). It can also be noted, as expected, that the heat distribution in the outer slices is more spatially diffused.

Figure 3.

Results obtained from gel experiments depicted in Fig. 2 using the modified SPGR sequence. (a) Temperature curves for the gel experiments. As expected, the temperature rise was highest in the focal plane slice and lowest in slice B2, which is the farthest slice from the location of the focal heating spot. (b) Comparison of the three MURPS temperature maps acquired from the gel phantom after heating for 18 seconds. As expected, temperature change is lower and more spatially diffuse in slice B2.

The temperature change with time of the three MURPS slices that were acquired with multi-shot EPI after motion and Nyquist ghosting correction are shown in Fig. 4. As expected, the temperature rise varies based on the location of each slice. The fluctuations in the temperature curves are due to the much wider bandwidth used in the EPI acquisition and the fast imaging used in monitoring the temperature, which introduces more noise. As predicted, however, the fluctuations due to noise are lower in slices B1 and B2 compared to slice A1 because of the lower spatial resolution in these slices.

Figure 4.

Temperature curves of the three MURPS slices acquired with a modified multi-shot EPI sequence during heating in a gel phantom. In the fast multi-shot EPI-based MURPS sequence, as expected, a wide bandwidth leads to greater noise in the temperature measurements.

In-vivo result

Figure 5a shows an axial MRI image of the rabbit’s thigh from the in-vivo experiment. Temperature maps computed from the phase difference maps at one time point are overlaid on the three MURPS slices. The temperature map for slice A1 in Fig. 5a shows intensive heating around the focal spot of the sonication as expected. It is interesting to examine the temperature map for slice B1, which was placed 11.2 mm from the focal plane. The temperature is distributed over larger region and away from the focus spot and there is a noticeable heating around the bone structure and at the boundary between the skin and the water bath. Also shown is a temperature map for slice B2, which is at distance of 8 mm from the focal plane. As seen in slice B1, there is also significant heating in slice B2 away from the focus of treatment, especially around the bone. Figure 5b shows the temperature change plotted versus the imaging time in an ROI for each MURPS slice. For comparison, the temperature change in slice A1 is plotted for the standard SPGR temperature-mapping sequence (TMAP) based on data that was acquired in separate heating experiments. The temperature change in slice A1 with MURPS reached a value of more than 60°C compared to a maximum temperature change of 55°C in the TMAP experiment. The temperature curves are qualitatively similar; however, there is a systematic difference that could be due to the fact that we were sonicating the same location twice. Once the tissue is thermally coagulated, its US absorption coefficient increases and larger temperature rise will be achieved if the same location is sonicated with the same parameters. The difference could also be due to an imperfect repeatability in the conditions of the two in-vivo experiments. The temperature changes within an ROI in the focal area of slices B1 and B2 are also shown in Fig. 5b. The temperature change in slice B1 shows heating above baseline (presumably due to a slight heating effect of the sonication) while in slice B2 the temperature change curve was little above the noise level in the focal area ROI.

Figure 5.

Results from the in-vivo experiment heating experiment. (a) Three axial MURPS slices of a rabbit’s thigh with overlaid color images showing the heat distribution due to FUS sonication at 50 watts power for about 69 seconds. Left: slice B1 was positioned 11.2 mm away from the focal plane slice and was acquired with reduced spatial resolution in the phase encode direction and double the slice thickness. The arrows indicate locations of undesired heating near bones and the skin surface, showing the need to monitor a larger volume than just at the focus of the heat deposition. Middle: slice A1 was located at the focal point of the FUS beam to monitor the location of maximum heat deposition. The highest spatial resolution was used for slice A1. Right: slice B2 was located 8.7 mm from the focal plane slice. As with slice B1, there is heating around the bone and at the skin surface. The legend gives the temperature scale for the maps in the three slices where light yellow represents the highest temperature reached and dark red is the lowest temperature. (b) Plots of temperature changes with time for the in vivo rabbit experiment, measured by both MURPS and the standard gradient echo sequence (TMAP). Temperature rise in the focal plane A1 by both sequences gives a qualitatively similar temperature rise. However, there is a systematic difference probably due to some irreversible change in the tissue between repetitions of the heating at the same point.

SNR verification results

The results of SNR from the phantom experiment are given in Table 3. The measured SNR values of SNR_standard/SNR_A1 = 75.78/53.30 = 1.42 matched the expected value of the SNR given by Eq. 4 which predicts that SNR_standard/SNR_A1 = 1.41. Temperature changes in a region-of-interest obtained by both sequences are compared to each other and the results are shown in Fig. 6a for both slices.

Table 3.

SNR of MURPS slices (A1,B1,B2) compared to the SNR of original SPGR sequence

	Experiment 1			Experiment 2			Experiment 3
Slice label	SPGR	A1	Ratio SPGR/A1	SPGR	B1	Ratio SPGR/B1	SPGR	B2	Ratio SPGR/B2
SNR	75.8	53.3	1.42	146	148	0.99	146	141	1
Predicted SNR	-	-	1.41	-	-	1.0	-	-	1.0

Figure 6.

Results of temperature measurement experiments in gel phantoms for validation of MURPS as compared with standard SPGR sequence. (a) The temperature rise in MURPS slice A1 is compared to the temperature rise in the original SPGR focal slice where thickness of both slices was 3 mm. Very similar temperature values were measured by the two sequences. (b) The temperature rise in MURPS slice B1 is compared to temperature of the same slice imaged by the standard SPGR sequence. (c) The temperature rise in MURPS slice B2 is compared to the temperature of the same slice imaged by the standard SPGR sequence.

The predicted SNR in the temperature maps obtained with the standard SPGR sequence compared to the MURPS modification is given according to Eq. 4 such that, SNR_standard/SNR_{B1, B2} = 1. The measured SNR_standard/SNR_B1 was 146.35/147.90 = 0.99, which is very close to the predicted value. Figure 6b shows the temperature change with time for this experiment. Figure 6c shows a comparison of the temperature curves for slice B2 and the temperature mapping result using the reference sequence. The figure demonstrates that the temperature curves for slice B2 and the slice imaged using the reference SPGR sequence are in close agreement. The measured SNR_standard/SNR_B2 was 146.36/140.60 = 1.04, which is very close to the predicted value SNR_Standard/SNR_B2 = 1.

DISCUSSION

Ideally, in MR temperature mapping for monitoring of thermal therapies, one would like to have high spatial and high temporal resolution, excellent temperature sensitivity and extensive volume coverage. Unfortunately, due primarily to imaging time constraints, it is necessary to make tradeoffs. A typical tradeoff, for example, is to reduce slice coverage to a single slice centered at the focus of heating where temperature changes are expected to be the greatest. The problem with this is that other regions throughout the volume, where there may be a potential for damage due to heating, tend to go unmonitored [35,36]. In this work, we have chosen to implement an approach that aims to increase slice coverage without adversely affecting temporal resolution. The approach can be combined with fast imaging methods, for example, with EPI, which was also implemented and was tested with heating in a moving phantom.

In the MURPS implementation reported here, we extended coverage beyond a single slice but, in order to avoid loss in temporal resolution, we reduced the spatial resolution in the outer slices where little or no temperature change was expected. We also reduced the in-plane FOV to shape slices for more optimal coverage of the volume affected by heating, further reducing scan time. This was possible because the new MURPS imaging scheme we implemented allows us to vary, for each slice, both the slice thickness and the resolution in the phase-encoding direction as well as to reduce the in-plane FOV without aliasing. In the first phase of this work, a SPGR pulse sequence was modified for MURPS acquisition and 2DRF excitation so that we were able to image two additional slices away from the focus of heating with no increase in imaging time over single-slice imaging. The predicted SNR based on imaging parameters matched the measured SNR. Temperatures measured by the standard SPGR sequence and the MURPS sequence were also in good agreement in the phantom experiments.

It should be noted that by using a 2DRF pulse some tradeoffs are necessary in terms of the selectivity in the slice direction if one wants to keep the total pulse duration reasonably short. The sub-pulses we implemented were only 0.6 ms and the overall pulse length only 3 ms. This was achieved by using Gaussian-shaped pulses, which give a less-sharp slice profile. Having such short sub-pulses also limits the minimum achievable slice thickness. Another tradeoff is that options for fat suppression are more limited if a 2D spatially selective pulse is being used. Potential options for combining fat suppression with 2D spatially selective RF pulses include using approaches that exploit the natural chemical-shift sensitivity of 2DRF pulses for reduced FOV imaging [37–39] or a temporal filtering approach such as that proposed in [40].

An inherent assumption in adopting the MURPS approach to increase slice coverage is that lower spatial resolution is acceptable for the outer slices. This would seem to be justified based on the expectation that temperature profiles away from the focus are likely to be quite smooth spatially. It is possible that at boundaries, such as between bone and soft tissue, there could be relatively sharp transitions in the spatial profile of the temperature change and, therefore, higher spatial resolution would be desirable in such cases. However, given the constraint of imaging time along with the benefit of higher SNR (and thus better temperature sensitivity) at lower spatial resolution, we would argue that the MURPS variable resolution approach for outer slices is still a very reasonable compromise.

As shown in the in-vivo rabbit experiment, where there was unpredictable heating around bone and at the tissue boundary, the addition of two more slices can be important to monitor for undesirable heating effects. Of course, one should be careful about drawing conclusions based on a single case. At the same time, it should also be acknowledged that, even thought the method is applied in just this single in-vivo case, an example of its utility has been revealed.

In the in-vivo experiment we applied heat to the thigh of an anesthetized animal, thus, motion was not an issue. For cases where motion is a problem such as in the liver or kidney, the multi-shot EPI-based MURPS method that we developed could be used. The temporal resolution increased in the modified EPI sequence by a factor of one third when using the MURPS method, but it could have been increased further if reduced FOV imaging was included. Future implementations of the EPI approach will include 2DRF excitation to enable reduced FOV imaging so that, as with the SPGR sequence, three slice images can be acquired in the time normally taken to acquire a single slice. A modification to reduce imaging time even further or, alternatively, to enable increasing coverage beyond three slices, would be to include a parallel imaging approach.