Combined ultrasound and MR imaging to guide focused ultrasound therapies in the brain

A clinical MRgFUS system (ExAblate 4000 low frequency, InSightec), which was developed for high-intensity sonications for transcranial thermal ablation in the brain (Martin et al 2009), was modified to provide low-power sonications for BBB disruption (Arvanitis et al 2012, McDannold et al 2012). This system has a phased array transducer with 1024 elements arranged in a 30 cm diameter hemisphere and operates at a central frequency of 220 kHz. It was driven in burst mode via a gating signal provided by an arbitrary waveform generator (model 396, Fluke). It was integrated with a clinical 3T MRI unit (GE Healthcare). For this work, the hemisphere transducer was oriented 90° from its normal clinical use so that it could be simply filled with degassed water like a bowl. Imaging was performed using a 15 cm surface coil with the animal placed supine on the MRI scanner table with its head partially submerged in water (figure 1). The animal's head was tilted back so that the focal plane of the MRgFUS device, which in this configuration was parallel to the water surface, corresponded to an approximately axial plane in the brain. FUS beam aberration correction (Clement and Hynynen 2002), which is needed to compensate for beam distortions in a human skull, was not necessary with the thinner primate skull (McDannold et al 2012). Prior characterization of the ultrasound field in water and after transmission through an ex vivo monkey skull found that an acoustic power level of 1 W with this device produces an estimated peak negative pressure amplitude in the brain of approximately 223 kPa (Arvanitis et al 2012).

Zoom In Zoom Out Reset image size

The ultrasound imaging probe (L382, Acuson) used for passive cavitation imaging was a 128-element (82 mm) linear array with a 3.21 MHz central frequency and a bandwidth of approximately 75%. It was incorporated into the therapeutic MRgFUS phased array with an acoustic mirror (a 6 mm thick brass plate) placed at 45° with respect to the face of the imaging array. Brass was chosen to minimize image artifacts at the MR images. This reflector enabled the collection of axial ultrasound images of the monkey brain (figure 1). The imaging array and reflector were mounted so that the ultrasound imaging plane was at a depth within a few millimeters of the geometric focus of the MRgFUS device. This plane and the animal position were also selected to be at a depth where the cingulate cortex (see below) was included in ultrasound imaging. The focal point of the MRgFUS device was steered electronically to different targets in the ultrasound imaging plane using its phased array.

The ultrasound imaging engine (Verasonics) was triggered along with the MRgFUS device using an external trigger from an arbitrary waveform generator (Fluke) and was programmed in Matlab (Mathworks) to operate in passive mode. The ultrasonic RF data from 64 elements of the imaging array were recorded synchronously at a sampling rate of 12.84 MHz. Only half of the elements (every other element) were used, since the imaging engine could only synchronously read data from 64 channels. Sonications were delivered as a series of 10 ms bursts (see below); the first 180 µs of ultrasonic RF data were recorded for each burst. Forty datasets were recorded for each 50 s sonication. Power spectra were displayed in real-time on the computer that controlled the system during sonication; cavitation maps were generated offline on the same computer. The ultrasonic RF data were filtered with a 300 kHz high-pass Butterworth filter in software in order to remove the signal from the fundamental frequency of the MRgFUS transducer (220 kHz).

The acoustic emissions were also recorded with two MRI-compatible piezoelectric transducers using a separate acoustic emissions monitoring system. The ability of this system, which utilized narrow-band receivers with maximal sensitivity at 620 ± 10 kHz, was demonstrated previously in experiments on nonhuman primates to control the exposure level during BBB disruption (Arvanitis et al 2012). We used this system, which is capable of transcranially detecting broadband and harmonic emissions, to confirm that the ultrasound imaging probe was recording the same spectral content.

The coherent acoustic emissions recorded by the linear imaging array were summed and back-propagated to form maps (Gyongy and Coussios 2010a, Norton et al 2006) proportional to the strength of the emissions produced by the microbubble oscillations. In particular, the wave emitted from each bubble can be modeled as being emitted by a point source s(r, t) originating from the position r(x, y, z). By making some simple assumptions, we can locate this point source using the signals recorded at the locations r_n of the N transducer elements. Assuming a constant sound speed c between the bubble and the transducer, the location of the point source can be estimated by coherently summing the recorded acoustic emissions u(r_n, t) in (r, t) space, with $t = \frac{{| {r - r_n } |}}{c}$. Then, the reconstruction of the point source $s( r ) = \sum\nolimits_{i = 1}^N {| {r - r_n } | \cdot u( {r_n ,| {r - r_n } |/c} )}$ is performed by back-propagating the recorded wavefront to multiple point source coordinates and summing over the array's elements. The |r − r_n| term describes geometric wavefront loss. For the linear array used in this study, the three-dimensional vectors r are converted to image coordinates x (transverse) and z (axial) $( {| {r - r_n } | = \sqrt {( {x - x_n } )^2 + z^2 } } )$ (y = 0). The relative point source intensity I(r) = |s(r)|², after subtracting incoherent background noise (i.e. the dc component) (Norton et al 2006) is given by

$$\begin{equation*} I( r ) = \left| {\sum\limits_{i = 1}^N {| {r - r_n } | \cdot \overline u \left( {r_n ,\frac{{| {r - r_n } |}}{c}} \right)} } \right|^2 - \sum\limits_{i = 1}^N {\left| {| {r - r_n } | \cdot \overline u \left( {r_n ,\frac{{| {r - r_n } |}}{c}} \right)} \right|^2 } . \end{equation*}$$

This back-propagation algorithm requires knowledge of the speed of sound of the media between each point in the image and the transducer (Gyongy and Coussios 2010a). For points in the brain, this was estimated by the linear combination of the thicknesses and sound speeds in the water (95 mm, 1480 m s^–1 at 19 °C), brain tissue (31.5 mm, 1541 m s^–1) and bone (3.5 mm, 3080 m s^–1), which resulted in an average speed of sound of 1537 m s^–1 (Del Grosso and Mader 1972, Fry and Barger 1978). The brain and water path thicknesses were measured from the MR images, whereas the bone thickness was measured from head scans obtained with a portable CT scanner (Ceretom). It is important to note that due the small fraction of the bone with respect to beam path, relatively small variations in the speed of sound of the skull are not expected to substantially affect the apparent location of the cavitation activity. Maps were normalized relative to data obtained few minutes earlier without the microbubble agent to remove harmonics or other nonlinear terms present during sonication without the ultrasound contrast agent. The resulting images were expressed in dB.

The axial×transverse field of view was set to 100×80 mm², which included the entire monkey head in the image. With the 8.2 cm aperture of the ultrasound imaging probe, the 13 cm distance from the focal targets in the brain (see below) and a 3.21 MHz central frequency, the theoretical axial and transverse resolution of the cavitation maps is expected to be 0.6 and 7.7 mm, respectively (Gyongy and Coussios 2010b). As the device was also expected to record higher frequency content (and for visualization purposes), we reconstructed the images at a higher resolution (axial×transverse pixel dimensions: 0.5×0.5 mm²). Therefore, due to the geometry used, the cavitation maps will appear elongated along the axial direction of the array, which is perpendicular to the FUS beam axial direction.

The high-pass filtered RF data that were recorded from the array were used to construct the cavitation maps. In this paper, stable cavitation map refers to the absence of broadband emissions from the RF data recorded by the array. The absence of broadband emissions was also confirmed by the single-element passive cavitation detection system that was operated in parallel to the passive mapping. This system was able to capture the entire 10 ms RF waveform and determine the spectral signature of the emissions from the entire sonication, as described previously (Arvanitis et al 2012). Thus, the reports below of inertial cavitation in the passive maps refer to cases where obvious broadband emissions were evident in the spectra obtained by either system. No comb filtering or specific spectral band has been used to form the cavitation maps in the data presented here. Therefore, in the presence of broadband emissions, all possible types of oscillations, including inertial, might be included.

Fiducial markers (MR-Spots, Beekley Medical) visible in both MRI and B-mode ultrasound imaging were used before the animal experiments to register the two imaging modalities. The registration accuracy was limited by the size of fiducial markers (<1.5 mm) (Arvanitis and McDannold 2011). Standard T1-, T2- and T2*-weighted imaging sequences were used to select the brain targets and to evaluate the sonication effects (see table 1 for scan parameters) (Hynynen et al 2001). BBB disruption was assessed in T1-weighted imaging via the detection of signal enhancement after intravenous administration of the MRI contrast agent Gd-DTPA (Magnevist, Berlex) at a dose of 0.1 mmol kg^–1. T2*-weighted MRI was used to detect petechiae that occur when inertial cavitation is produced.

All experiments were performed in accordance with procedures approved by the Harvard University Institutional Animal Care and Use Committee. The animals were anesthetized during all the procedures and were constantly monitored throughout and after recovery. No pain or suffering was evident as a result of the procedures. Monkeys were housed, fed, watered, socially housed and provided with environmental enrichment according to USDA, OLAW and AAALAC regulations. The first monkey tested was an adult female Macaca mulatta (weight: 6 kg); the other two were juvenile male Macaca nemistrina (weight: 4.5–6 kg). Each animal was anesthetized with ketamine (15 mg kg^–1 h^–1 i.m.) and xylazine (0.5 mg kg^–1 h^–1 i.m.), or with ketamine (4 mg kg^–1 h^–1) and dexmeditomidine (0.01–0.02 mg kg^–1 h^–1 i.m.) and intubated. The head was shaved, and a catheter was placed in a leg vein to inject the ultrasound and MRI contrast agents. Heart rate, blood oxygenation levels and rectal temperature were monitored during the experiments. Body temperature was maintained with a heated water blanket.

The sonications consisted of 10 ms bursts applied at a pulse repetition frequency of 1 Hz. Two 50 s sonications were delivered in sequence with a delay between sonications of ∼5 s. These parameters were selected to be similar to a previous work (McDannold et al 2012) with this device. In each animal, four targets were sonicated: the cingulate cortex, two in each hemisphere. This brain structure was selected because it is an anatomically large and homogeneous gray matter target that was aligned with the axial MRI planes and the ultrasound imaging plane. Each sonication was combined with the microbubble ultrasound contrast agent Definity (Lantheus Medical Imaging), which was infused over the entire sonication via an MRI-compatible infusion pump (Spectra Solaris EP, Medrad). The microbubble agent was diluted in 5 ml sterile phosphate-buffered saline and infused at a variable rate (1 ml at 0.1ml s^–1 for 10 s and 2 ml at 0.02 ml s^–1 for 100 s). A dose of 40 µl kg^–1 of Definity was used for each target (four times the recommended clinical dose for ultrasound imaging). This dose was chosen to maximize the strength of the emissions (Arvanitis et al 2012) while remaining close to the clinical dose.

The acoustic power level varied among the different animals and targets. It was set initially to achieve BBB disruption without inertial cavitation using feedback from the acoustic emissions data, as described previously (Arvanitis et al 2012). At the subsequent sonications and targets, we then explored different acoustic power levels, including those slightly above the inertial cavitation threshold (where minor vessel damage is expected). Overall, 20 targets were sonicated over five experiments in the three monkeys (two sessions with the first two monkeys; one with the third). The acoustic power level ranged from 0.5–2.2 W, which yielded an estimated pressure amplitude in the brain of 190–330 kPa (table 2) (Arvanitis et al 2012).

After the experiments, we investigated if the location of the cavitation maps was colocalized with the site of the BBB disruption in MRI, and whether the data were consistent with prior work relating the acoustic emissions' spectral content to BBB disruption and tissue damage (Arvanitis et al 2012). The location of the cavitation activity was determined automatically by finding the pixel with peak image intensity in each cavitation map. The location of the BBB disruption was found manually by selecting the center of the MR enhancing region. Each targeted location was classified as having BBB disruption only or BBB disruption and tissue damage using T2*-weighted imaging, which is hypointense when significant red blood cell extravasation (petechiae) occur (McDannold et al 2012).