Cavitation monitoring, treatment strategy, and acoustic simulations of focused ultrasound blood-brain barrier disruption in patients with glioblastoma

Abstract

Purpose:

We report our experience disrupting the blood-brain barrier (BBB) to improve drug delivery in glioblastoma patients receiving temozolomide chemotherapy. The goals of this retrospective analysis were to compare MRI-based measures of BBB disruption and vascular damage to the exposure levels, acoustic emissions data, and acoustic simulations. We also simulated the cavitation detectors.

Methods:

Monthly BBB disruption (BBBD) was performed using a 220 kHz hemispherical phased array focused ultrasound system (Exablate Neuro, InSightec) and Definity microbubbles (Lantheus) over 38 sessions in nine patients. Exposure levels were actively controlled via the cavitation dose obtained by monitoring subharmonic acoustic emissions. The acoustic field and sensitivity profile of the cavitation detection system were simulated. Exposure levels and cavitation metrics were compared to the level of BBBD evident in contrast-enhanced MRI and to hypointense regions in T2*-weighted MRI.

Results:

Our treatment strategy evolved from using a relatively high cavitation dose goal to a lower goal and longer sonication duration and ultimately resulted in BBBD across the treatment volume with minimal petechiae. Subsonication-level feedback control of the exposure using acoustic emissions also improved consistency. Simulations of the acoustic field suggest that reflections and standing waves appear when the focus is placed near the skull, but their effects can be mitigated with aberration correction. Simulating the cavitation detectors suggest variations in the sensitivity profile across the treatment volume and between patients. A correlation was observed with the cavitation dose, BBBD and petechial hemorrhage in 8/9 patients, but substantial variability was evident. Analysis of the cavitation spectra found that most bursts did not contain wideband emissions, a signature of inertial cavitation, but biggest contribution to the cavitation dose – the metric used to control the procedure – came from bursts with wideband emissions.

Conclusion:

Using a low subharmonic cavitation dose with a longer duration resulted in BBBD with minimal petechiae. The correlation between cavitation dose and outcomes demonstrates the benefits of feedback control based on acoustic emissions, although more work is needed to reduce variability. Acoustic simulations could improve focusing near the skull and inform our analysis of acoustic emissions. Monitoring additional frequency bands and improving the sensitivity of the cavitation detection could provide signatures of microbubble activity associated with BBB disruption that were undetected here and could improve our ability to achieve BBB disruption without vascular damage.

Graphical Abstract

A hemispherical ultrasound phased array was used to disrupt the BBB in glioblastoma patients. Left: targeted and disrupted volumes. Right: treatment plan and post-ultrasound contrast-enhanced MRI.

Introduction

The blood-brain barrier (BBB) limits the delivery of most drugs into the central nervous system. Focused ultrasound (FUS), when combined with microbubble ultrasound contrast agents, can temporarily disrupt the BBB [1] and enable targeted delivery of drugs that normally are excluded from the brain. Numerous preclinical studies have shown the procedure can be administered repeatedly with little or no permanent effects to the brain [2–4] and have shown that it can achieve drug concentrations sufficient to improve outcomes in animal models of brain tumors and other central nervous system disorders [5–9]. Several clinical studies of this approach have shown safety and feasibility for different applications [10–21], but clinical efficacy has yet to be proven.

One promising application of FUS-induced BBB disruption (FUS-BBBD) is for glioma, which infiltrates extensively into surrounding brain tissue and has only shown a marginal therapeutic response with a few drugs. Enabling or enhancing drug delivery to significant volumes surrounding the surgical resection cavity where the BBB is intact could improve the therapeutic effect of drugs that are potentially active against glioma but are limited by poor delivery due to the BBB.

FUS systems to flexibly target clinically relevant large volumes face several technical challenges, particularly when the ultrasound exposures (sonications) are performed through the intact skull. One challenge is how to ensure a safe and effective exposure level. If the exposure level is too high, small blood vessels can be damaged, resulting in petechial hemorrhage that are evident in T2*-weighted MRI. The threshold for this vascular damage is not much higher than for BBBD [2]. The attenuation of the ultrasound field from the human skull varies significantly between patients and depends strongly on the location of the focus within the cranium and the angles between the face of the transducer and the bone [22]. Monitoring and feedback control are thus essential to ensure that the correct exposure level is achieved without causing vascular damage or other adverse effects.

One way to monitor the exposure level is to capture the acoustic emissions from microbubbles during the sonications. When microbubbles interact with an ultrasound field (cavitation), they respond nonlinearly and emit sound at distinct frequencies different than the driving frequency [23]. At low intensities, harmonics at integer multiples of the driving frequency increase when microbubbles are introduced and can be accompanied by a small subharmonic at one-half the driving frequency. As the intensity increases, a large increase in subharmonic emission occurs, and then at slightly higher intensities, the bubbles rapidly and violently collapse (inertial cavitation) and emit wideband signals. Wideband emissions have been associated with vascular damage in preclinical studies [24, 25]. The acoustic emissions at different frequency bands are signatures for different levels of microbubble activity and have been associated with the onset and magnitude of FUS-BBBD and vascular damage [24]. Such cavitation data has been used in different ways to control the exposure level [25–30] and have been utilized in clinical studies [31, 32]. These preclinical findings have implications for BBBD in patients but have not yet been systematically investigated in clinical treatments

Achieving a safe and effective exposure level might also be aided by using acoustic simulations of individual patients to estimate the transmission and focusing through the skull [33]. Such simulations could be used to estimate the approximate exposure level to use and to predict the size of the focal region of BBBD. They may also reveal whether reflections or standing waves are expected when the focus is near the skull. Simulating the sensitivity distribution of the cavitation detectors may also be useful in interpreting the acoustic emissions.

Beyond exposure control, additional technical challenges for transcranial FUS-BBBD arise from the need to steer the focus to target hundreds or thousands of individual sonication targets to deliver drugs across clinically relevant volumes and to correct for aberrations/defocusing caused by the skull. These challenges can be overcome using a phased array transducer that can steer the focus to different targets instantaneously.

Multiple clinical trials with such a phased array system are ongoing, and there is little hard data to guide those efforts. In this paper we describe our experience it over 38 sessions of FUS-BBBD in nine patients with newly diagnosed glioblastoma receiving temozolomide chemotherapy. This retrospective study had several goals. First, we describe how our treatment strategy evolved based on clinical measurements and yielded uniform BBBD with minimal MRI-evident petechiae. Second, we examined how well the acoustic exposure levels and cavitation data used for real time control of the procedure predicted the extent and magnitude of BBB disruption and microvasculature damage. Finally, we performed numerical simulations of the treatments to estimate the acoustic exposure levels at the focus, the effects of reflections and standing waves, and explore how the skull may have impacted the cavitation monitoring. Overall, our findings suggest that improvements in treatment planning and cavitation monitoring could lead to safer and more effective BBBD. Methods

Patients

FUS-BBBD was performed in ten patients with glioblastoma as part of a multi-center clinical trial [36]. IRB approval was obtained for the clinical trial. The treatments in our first patient used different sonication planning, treatment software, and microbubble administration rates and were not included here. All patients received the standard of care treatment for newly diagnosed glioblastoma, which included maximal safe surgical resection, radiotherapy with concurrent low dose temozolomide, and monthly adjuvant temozolomide chemotherapy [37]. The treatments aimed to enhance temozolomide delivery to regions surrounding the resection cavity during maintenance chemotherapy, which began approximately 4 weeks after the completion of radiation. During this period, patients received daily chemotherapy for 5 days every four weeks. FUS-BBBD was performed immediately before the administration of TMZ on day one, two, or three of each monthly temozolomide cycle for up to six treatments. Initial feasibility and technical results of the multicenter trial have been published elsewhere [38]; clinical outcomes will be reported in a future paper.

Device

The treatments used the Exablate Neuro low-frequency system (InSightec), which was integrated into the detachable table of a 3T MRI (GE750W, GE Healthcare). This system uses a 220 kHz 1024-element hemispherical phased array transducer (1018 active elements) to electronically steer the focal region and to correct for aberrations caused by the skull. The transducer is attached to a manually operated positioning system that allows for translation in three dimensions and rotation about the left-right axis of the patient.

The FUS system adds delays/phase shifts to different elements to cause their fields to coincide at different locations away from the geometric focus. This electronic beam steering can be performed instantaneously and allowed for sequential targeting of many “subsonication” targets during each sonication. The system also adds phase shifts to different elements to correct for differences in sound speed between the bone and soft tissue to correct for skull-induced aberrations. These phase shifts were estimated using a ray tracing acoustic model that was based on CT scans and empirically derived relationships between the skull’s density and acoustic properties [34, 35]. The system uses the shear sound speed in this model when the external angle between the skull and element exceeded 30°.

In the first 17 treatments, the patient’s head was shaved; in the remaining treatments hair up to 10 cm was allowed. In most treatments, the patient was placed in a stereotactic frame that was then fixed to the MRI table. In the final four treatments, the stereotactic frame was replaced by a bite bar immobilization system. A flexible membrane was attached to the patient’s head and the open face of the transducer; the space between the transducer face and the scalp was filled with degassed water. The location of the transducer within the MRI reference frame was confirmed using MRI tracking coils. The system also obtained images before each sonication to confirm that the head had not moved. The manual positioner was used to place the geometric focus of the transducer within the targeted volume; the phased array provided additional electronic steering up to ±25 mm in each direction. If needed, the transducer was moved to a second location to treat areas outside of this range.

Imaging

Before treatment, we obtained an MRI study using a 24-channel head coil and the 3T scanner where the FUS system was installed. This study was obtained 4–18 days (median: 5) before the FUS-BBBD session. This exam included T2-weighted FLAIR, T1-weighted Propeller Fast Spin Echo (FSE), and SWAN T2*-weighted sequences. Imaging parameters are listed in Supplemental Table 1. The T1-weighted FSE sequence was obtained before and after administration of gadobutrol contrast agent (Gadavist, Bayer) at a dose of 0.1 ml/kg. The median time between contrast administration and the final T1-weighed image acquisition was 20.4 min [quartiles: 12.1–23.9 min]. Two of the pre-treatment exams for the second patient were acquired using a Siemens 3T scanner due to scanner availability.

We also obtained a CT scan of the head before the treatment. The images were reconstructed using a bone kernel for use during treatment planning. Before sonication we segmented the edges of the craniotomy, areas with metallic hardware implanted in the skull, air sinuses, and calcifications in the brain on the CT scan using the FUS system software. Transducer elements in direct line with these structures were disabled, so the number of active elements of the transducer varied between treatments.

On the treatment day, we obtained T2-weighted images with the patient in the FUS device using either the body coil (29/38 treatments) or a disposable two-channel coil developed by the FUS system manufacturer. Treatment volumes were defined on these images; T2-FLAIR images obtained before treatment were used as a reference for this planning. After FUS-BBBD, the patient was allowed to get up from the treatment table, the FUS device was removed, and the same MRI exam obtained for pre-treatment baseline was repeated using the 24-channel head coil.

Treatments

The focal spot was electronically steered during each sonication to up to 32 subsonication targets. The treatments consisted of 20–726 subsonication targets and volumes 1.9–32 cm³ (Figure 1). The sonications were each arranged so that all subsonications were on one plane of the planning T2-weighted FSE images; they were spaced 2.5 mm apart within each plane and were placed on planes 4 mm apart. As described below, we often repeated each sonication more than once. We typically shifted the subsonication targets within the T2 FSE imaging plane by 1–2 mm between the repetitions. MRI showing the treatment volumes for the nine patients are shown in Figure 1. The volumes targeted in the fifth patient were comparatively small because the treatments were aborted after a few sonications due to the patient becoming nauseous after administration of microbubbles.

Figure 1:

T2-weighted FLAIR MRI obtained after the last FUS-BBBD session. The nine patients received up to six monthly sessions; contours indicate the targeted volumes for different treatment sessions. Targets were selected in hyperintense regions on FLAIR sequences. In some cases, we were unable to target the entire hyperintense region due to time constraints (such as the anterior regions shown above in patient 4) or when areas were out of the treatment range of the device (such as the posterior region shown above in patient 8). Plots show the treatment volume, number of subsonication targets, and the median number of active transducer elements for each treatment.

The sonications were combined with intravenous infusions of Definity microbubbles (Lantheus). Microbubbles were diluted in a 250 ml bag of saline and infused using a continuous drip at a rate of approximately 0.24 μl Definity per kg-min; the infusion rate was monitored using a drip counter (DripAssist, Shift Labs). We waited at least two minutes after the initiation of the infusion before sonication. Each subsonication consisted of 5 ms bursts applied at 1 Hz.

Cavitation monitoring and feedback control

The Exablate Neuro used four receivers to monitor the cavitation activity during the sonications. The detectors were sensitive around 110 kHz, the subharmonic of the 220 kHz FUS transducer. The system acquired the time signals, filtered and amplified them, and calculated their spectra using Fast Fourier Transform. It combined the signals using a weighted average over different spectral bands and the four detectors to produce a cavitation score. This weighting took into account differences in skull attenuation for each element. Exact details of the spectral bands, weighting factors, and calibrations of the detectors were proprietary; cavitation measurements are thus reported in this paper as arbitrary units (AU). In addition to evaluating the cavitation levels provided by the system, we performed our own analysis of the cavitation signals using spectra saved by the device. We identified whether wideband emissions and/or a discrete subharmonic peak were evident during each burst.

The cavitation dose, defined as the sum of the cavitation scores, was used to dynamically control the power level. The user selected a cavitation dose goal, sonication duration, gain factor, and maximum acoustic power for each sonication. The system modified the power level so that the sum of the cavitation scores would reach the dose goal by the end of the sonication. The gain factor controlled how quickly the system could increase the power. Initially the controller used the average cavitation scores of all the subsonications to control the power. Later, the system controlled the power for each subsonication target individually. Before the system provided subsonication-level control, we would often observe that one or a few subsonication targets had a comparatively high cavitation dose with the others receiving little or none. In this situation, we would stop the sonication, remove the “hot” subsonication targets and repeat the sonication. Thus, the sonication duration varied for the different subsonications.

Simulations

Numerical modelling of the acoustic field was performed using k-Wave, an open-source MATLAB toolbox [39]. The simulations used the O2 High Performance Compute Cluster, supported by the Research Computing Group at Harvard Medical School. We simulated the acoustic field separately for the 1018 active transducer elements of the phased array transducer (Figure 2), as described previously [40]. The elements were modeled as square 1×1 cm flat pistons centered on the locations provided by the manufacturer. The dimensions of the individual elemental simulations had a grid size of 76 × 76 × 201 with a spacing of 0.974 mm (one-seventh of a wavelength), which resulted in a simulation space of 74.0 × 74.0 × 195.8 mm per element. We used a perfectly matched layer with an attenuation of 2 Np per grid point and a size of 10 grid points.

Figure 2.

Element-wise simulations. Coronal T2 FLAIR image displayed on a diagram of the hemispherical transducer. The simulated acoustic field for one transducer element is superimposed. The insets show the combined field in axial and coronal views. They were created by interpolating the individual simulations into the ROI shown in the white box, adding phases to the individual simulations for beam steering and aberration correction, and summing them. The skull is indicated in green; the field of the individual element is displayed on a log scale.

Simulations were performed for each transducer location used in the treatments. Supplemental Table 2 lists the acoustic parameters used in the simulations for the skull, brain, and water. The acoustic density of the skull was estimated from the CT scan. We used a linear relationship between Hounsfield units and density, and −1000 and 57 Hounsfield units for air and soft tissue, respectively. Skull acoustic sound speed and attenuation were estimated from the density using relationships found previously [40]. We did not include shear mode propagation.

After the completion of the individual simulations for each treatment, we interpolated the resulting complex pressure field from each element’s individual reference frame to a 76 × 76 × 71 element volume in the xyz frame of the hemispherical array. For a given sonication, the simulations of 1018 elements were loaded simultaneously into memory, and the magnitude and phase corrections supplied by the manufacturer were applied to the corresponding elements along with phase values needed to electronically steer the focus to each subsonication location. We then preformed a complex sum over the elemental simulations to estimate the pressure field for that location. We also calculated the pressure fields using “ideal” phase correction. For this calculation, we simply subtracted off the phase interpolated to the subsonication target location. Finally, we repeated the simulations without the skull. Those simulations were used to calculate the phase needed to steer the focus and to compare the size and shape of the focus to transcranial simulations.

We obtained the time history of the acoustic power for each subsonication from the treatment log files. We used the power levels to compile the maximum pressure amplitude delivered during all the overlapping sonications in each treatment. To account for different sonication durations, we calculated the accumulated exposure at each location in J/cm² by summing the simulated intensity of each burst multiplied by the burst time of 5 ms. Finally, the maps of peak pressure amplitude and accumulated exposure were interpolated to the same reference frame as the MRI.

We also simulated the cavitation detectors’ sensitivity profile for each treatment to evaluate the effect of the skull on the cavitation monitoring. Since the transmitted field of a transducer is proportional to its sensitivity profile, we simulated the transmitted field from the detectors. The detectors were modeled as 1 cm square element transducers operating at a single frequency of 110 kHz at the locations provided by the manufacturer. Simulations covered a 17.0 × 17.0 × 27.5 cm volume. Other parameters were the same as those used with 220 kHz transducer elements. To evaluate the relative impact of the skull on the cavitation measurements, we normalized the sensitivity profile to those obtained in additional simulations performed without the skull present.

Data Analysis

Registration between the reference frames of the pre-treatment CT and MRI and the treatment MRI and transducer coordinate system were performed by the FUS system software and obtained from the treatment logs. Registration between other imaging sessions was performed using the open-source Advanced Normalization Tools (ANTS) software package. All other data analysis was performed using MATLAB (MathWorks). We interpolated the imaging for all sessions to the reference frame of the T2 FLAIR sequence used to plan the first treatment. A 4 × 4 × 10 mm ellipsoid region of interest was centered on each subsonication target and angulated based on the relative position of the transducer with respect to the imaging. These dimensions were chosen to be similar to the simulated 50% isopressure contours of the focal region. The combined volume of these ellipsoids was used to define the treatment volume. We manually segmented the resection cavity and ventricles adjacent to the target to exclude them from the analysis, as they are avascular.

We measured the signal enhancement resulting from extravasation of MRI contrast across the BBB with the T1-weighted FSE sequence. To correct for signal changes unrelated to contrast extravasation that was evident after image registration, we assumed the signal enhancement in a region surrounding the treatment volume was zero. In each image plane, we dilated the treatment volume by 25 voxels; regions 5–25 voxels from the edge of the treatment volume were used to define this background region. Signal changes evident in the background region were fit to a smooth surface [41], extrapolated into the targeted area, and subtracted off. Areas in the background with enhancement greater than 3%, such as blood vessels and enhancing tissue at the margin of the resection cavity, along with nonvascular tissue such as the resection cavity and ventricles were excluded from the background. Note that low-level enhancement unrelated to FUS-BBBD arising from the presence of contrast agent in the microvasculature was subtracted off. To validate this method, we mirrored the treatment volumes to the contralateral hemisphere and confirmed that the net contrast enhancement was zero. We also measured signal changes in the SWAN sequence between images obtained the week prior and immediately after FUS-BBBD. Before image subtraction, the images were normalized using the N4 normalization [42].

We calculated the percent of each ellipsoidal ROI that had BBB disruption and/or petechiae. For BBB disruption, we calculated the percentage of each elliptical region of interest that had signal changes greater than one standard deviation above the enhancement in the contralateral hemisphere. For petechiae, we manually segmented the hypointense areas in the SWAN imaging. To look for long-lasting changes, we segmented hypointense regions evident between SWAN images obtained after treatment and those obtained approximately three weeks later.

Results

Evolving treatment strategy

Initial treatments used a cavitation dose goal suggested by the manufacturer to control the acoustic power level during each sonication (Figure 3A). To reduce the amount of petechiae that was evident as hypointense regions in SWAN MRI, we gradually reduced the goal in each treatment while increasing the sonication duration starting in patient 4. We ultimately settled on a strategy of reducing the cavitation goal by a factor of approximately five, using the maximum sonication duration of the system of 3 min, and repeating each sonication twice (Figure 3B). This revised strategy, along with subsonication-level control (shaded regions in Figure 3), yielded substantially less variability in the cavitation measurements (Figure 3D-F). By the last treatment, we achieved BBB disruption over a similar fraction of the targeted volume as we did during the initial treatments (Figure 3G), but with substantially fewer petechiae evident in SWAN MRI (Figure 3H-I).

Figure 3:

Evolving treatment strategy maximized BBBD while minimizing petechiae. (A-C) Cavitation dose goal, sonication duration, and acoustic power used per sonication. Data are presented in the order in which the different patients were treated. The shaded region indicates treatments where subsonication-level control was enabled. (D-E) Total cavitation dose and maximum cavitation score measured during each treatment. (F) Ratio of the total dose delivered to the dose goal per sonication. (G) Percentage of the targeted volumes where BBBD was observed in contrast-enhanced T1-weighted imaging. (H-I) Percentage of the targeted volume where hypointense regions were observed in SWAN imaging immediately after FUS and 3–4 weeks later. Reducing the goal and extending the sonication duration resulted in similar total cavitation doses delivered to the treated region BBBD as our early treatments while reducing the probability for wideband emissions and petechiae evident in SWAN. Subsonication-level control also improved consistency. Data shown in (A-F) are median values of all subsonication targets ± 25–75% quantiles.

Examples of BBBD and petechiae from the two treatment strategies are shown in Figure 4. Using either a high cavitation dose goal, or a lower one with repeated sonication resulted in BBBD that covered most of the targeted volume, while the lower dose approach greatly reduced the amount of petechiae. Most petechiae were reduced in severity or no longer present after one month (Figure 5), while severe petechiae were evident throughout the subsequent sessions. In two patients, the vessel damage seen in SWAN was sufficient for blood to accumulate in surrounding sulci. One example is the patient shown in Figure 5; the blood in the sulci is indicated by the arrow. This finding was asymptomatic in both patients and cleared before the next treatment.

Figure 4:

Contrast-enhanced T1-weighted FSE (T1C) and SWAN MRI showing BBBD and petechiae, respectively, in our first and last patient. Using a lower cavitation dose goal and repeating the sonication twice as we did in our last patient achieved BBBD across the targeted volume with fewer petechiae after the procedure. Contours indicate the treated volumes. The insets show areas with BBBD (yellow), hypointense regions in SWAN MRI immediately after FUS (red), and hypointense regions that were still evident a month later (cyan)

Figure 5:

SWAN MRI acquired before and after six sessions of FUS-BBBD in the fourth patient. Hypointense spots were evident after each session, presumably resulting from petechiae. A month later, the spots either disappeared or their severity was largely reduced. Bleeding into a sulcus was evident immediately after session 2 (arrow) and was resolved in images obtained the next month; it was asymptomatic. Images shown are minimum-intensity projections over five 2 mm slices; bar: 1 cm.

Results were improved with the addition of subsonication-level control. In the early treatments that used sonication-level control, we often observed individual subsonication targets with substantially higher cavitation scores (Figure 6A). In such cases, we would halt the sonication, remove the high cavitation subsonications, and then repeat the sonication with the remaining subsonication targets. Despite these efforts, the resulting BBBD was often nonuniform. With subsonication-level control (Figure 6B), the individual control led to a uniform cavitation dose and BBBD. In the example shown in Figure 6B, we also had extensive petechiae evident in SWAN MRI. In subsequent treatments, we further reduced the cavitation dose goal per sonication and repeated the sonications twice. Spectra and cavitation measurements for individual subsonication targets in Figure 6 are shown in Supplemental Figure 1.

Figure 6:

Feedback control for two consecutive treatments in the fourth patient. (A) In the first treatments, the FUS device controlled the acoustic power level using the average cavitation score for all the subsonications targeted in each sonication. It modified the power to achieve a user-specified cavitation dose by the end of the sonication. We often observed some subsonications with high cavitation scores. With an aim of minimizing petechiae and producing more homogeneous BBBD, we halted the sonication, removed the high cavitation targets, and sonicated again. Despite this effort, non-uniform BBBD sometimes occurred, such as in this example. (B) With subsonication-level control, the FUS device controlled the power level individually for each subsonication target, resulting in a uniform cavitation dose and BBBD. In this example, we also observed significant petechiae in SWAN MRI. Using a lower cavitation goal in later patients and repeating each sonication twice and resulted in BBBD with less severe petechiae.

Acoustic simulations

We simulated the acoustic field for each treatment. An example of a simulated pressure field for one subsonication is shown in Figure 7. When the subsonication target was not close to the skull, the size and shape of the focus in simulations performed with and without the skull and with and without phase correction were similar (Figure 7C). Reflections and standing waves were evident when the focus was located close to the skull laterally (Figure 8A-C), superiorly (Figure 8D-F), or near the skull base (Figure 8G-I). These effects appeared to be mitigated when the simulations used ideal focusing instead of the phase corrections used during the treatment, although reflections were still evident.

Figure 7.

Acoustic simulation of one subsonication in patient 7. (A) FLAIR MRI in three orientations showing the targeted volume, the position of the subsonication (dotted lines) and geometric focus (crosses), and the region covered by the simulation (white box). (B) Simulated pressure field shown in three orientations using the phase aberration corrections employed during the treatment. The skull and 50% isopressure contours are indicated. Note the reflections from the skull base in the xz and yz views. (C) 50% isopressure contours for acoustic simulations with phase aberration corrections used during the treatment, ideal phase corrections, and no phase corrections. Also shown are isopressure contours obtained in simulations that did not include the skull with the 694 active elements used during the treatment and all 1018 elements active. (D) Map of the transducer with the active elements indicated in red.

Figure 8:

Simulations of subsonication targets close to the skull laterally (A-C), near the skull base (D-F) and at a superficial target (G-I). Effects of standing waves and reflections are evident in simulations using the aberration corrections used during the treatments and were mitigated using ideal focusing. Dotted lines indicate the location of subsonication targets; the geometric focus of the transducer is noted in each image with a cross. Contours indicate regions within 50% of the peak pressure amplitude. Bar: 1 cm

The median estimated pressure amplitude and accumulated acoustic exposure for each treatment are shown in Figure 9A-B for simulations using the phase aberration corrections used during the treatments and in Figure 9E-F for simulations performed with ideal phase corrections. Estimated peak pressure values varied between patients by more than a factor of three and by an order of magnitude for the total acoustic exposure in J/cm². The distance between the location of the peak pressure amplitude in the simulation and the subsonication targets is shown in Figure 9C for simulations that used the treatment phase corrections. This mismatch was apparently due to not correcting the effects of reflections; the median target mismatch with ideal focusing was less than 1 mm in all patients (Figure 9G). Since the FUS system used shear mode to calculate the phase correction in many elements (see below) and the simulations did not include shear mode, the mismatch in Figure 9C was likely not present in the treatments. We also calculated the volume of the simulated 50% isopressure contours for each sonication. Figure 9D and Figure 9H plot the median volumes of 50% isopressure contours relative to those simulated without the skull for the different treatments. This relative volume was greatest in the second patient.

Figure 9:

Summary findings for the acoustic simulations per treatment. (A, E) Acoustic pressure amplitude. (B, F) Total exposure delivered to each target. (C, G) Distance between the peak pressure amplitude and the target. Note the simulations did not include shear propagation, while the FUS device used the shear sound speed when the incidence angle was greater than 30°; the targeting mismatch with the treatment phase corrections was thus likely less than the simulations suggest. (D, G) Volume of 50% isopressure contours shown relative to the volume of the same isopressure contours simulated without the skull. Simulations using ideal phase corrections (E-H) had higher pressures, more accurate targeting, and tighter focusing than those for the corrections used during treatment (A-D) Data shown are median values of all subsonication targets in each treatment ± 25–75% quantiles.

In addition to the sonications, we also simulated the sensitivity profiles of the cavitation detectors. These simulations suggest that the sensitivity distribution at 110 kHz varied spatially within the targeted volume and between patients (Figure 10). The estimated insertion loss (the ratio of acoustic pressure in transcranial sonications to that in simulations without the skull) of the ultrasound transmission and cavitation detectors per subsonication are shown in Figure 11A. The median estimated insertion loss at 110 kHz on the cavitation detectors ranged from 60–75%.

Figure 10:

Simulated sensitivities of the cavitation detectors at a single frequency of 110 kHz. (A) Simulated sensitivity pattern for one cavitation detector from a treatment in the sixth patient. (B) Combined sensitivity patterns of the four detectors for treatments in the nine patients. The targeted regions are indicated. The crosses indicate the locations of the geometric focus of the FUS transducer for each treatment; circles indicate the allowed electronic steering ranges. Acoustic simulations are shown on a log scale. Note that shear propagation was not included in the simulations, and only one frequency was simulated.

Figure 11:

(A) Ratio of simulated pressure amplitude during transcranial sonication to that obtained in simulations without the skull for the cavitation detectors at 110 kHz and the 220 kHz FUS transducer. For the FUS transducer, data are shown using the phase corrections used during the treatment, ideal focusing, and no correction. The skull density ratio (SDR) for each patient is noted [50]. Data shown are median values of all subsonications for each treatment ± 25–75% quantiles. (B) Plot of the maximum acoustic pressure amplitude estimated in simulations with ideal phase correction vs. those with the corrections used during the treatments for individual subsonications. The results in the second patient did not follow the same trend as the other patients.

The estimated insertion loss for the 220 kHz FUS transducer was around 90%. The high insertion loss at 220 kHz was likely due to the higher frequency and oblique incidence angles between the transducer elements and the skull. The number of transducer elements that had incidence angles at the skull surface greater than 30° ranged from 15–66% for the nine patients (median: 33%). Insertion loss measured in elemental simulations are shown as a function of external incidence angle in Supplemental Figure 2. Above this threshold the FUS system used the shear wave sound speed to calculate the phase corrections, but we did not include shear mode conversion in our simulations. Not including shear mode might explain why the simulations predicted little or no improvement in focusing when the treatment phase corrections were used. The pressure amplitude was 1.2–1.9 times higher in simulations using ideal focusing; this ratio was greatest in the second patient (Figure 11B).

Subsonication-level analysis

We examined the relationship between BBBD, petechiae, acoustic exposure levels, and cavitation metrics on a subsonication level. We analyzed the MRI and maps of the maximum and total exposure from the simulations in elliptical regions of interest centered on each subsonication target. While there was substantial variation, trends were evident. Figure 12A-D shows the median percentage of each subsonication target where BBBD and petechiae were evident in contrast-enhanced T1-weighted and SWAN MRI, respectively, as a function of the applied and the simulated acoustic levels. While a correlation was observed on average for both, qualitatively the simulated exposure levels appeared to do a better job in predicting these outcomes.

Figure 12:

Subsonication-level analysis of BBBD and petechiae vs. exposure conditions (A-D) and cavitation metrics (E-L). (A-H) Percent of each subsonication that had BBBD and hypointense regions in SWAN MRI. (I-L) Signal intensity changes in T1-weighted imaging after Gadavist administration and between SWAN MRI obtained immediately after sonication and a week earlier. Circles indicate results from individual subsonications. Squares represent medians of the data in 15 equally numbered bins (indicated by horizontal lines). Vertical bars indicate 25–75% quartiles in each bin. Patient 2 did not follow the same relationships and was not included; a version of this figure with the second patient included is shown in Supplemental Figure 2.

Figure 12E-H shows the percentage with BBBD and petechiae as a function of the acoustic emissions metrics. Again, while there was substantial variation, a clear trend was observed between these outcomes and both the maximum cavitation scores and the total cavitation dose. A similar result was evident in analysis of the MRI signal changes due to contrast agent extravasation in T1W FSE and petechiae in SWAN MRI (Figure 12I-L). A window was also evident where it was possible to achieve BBBD without petechiae, albeit at a low level in a relatively small portion of the subsonication targets. The relationships shown in Figure 12 did not hold for the treatments of the second patient, which required higher exposure levels than the other patients (Figure 3C, Figure 9A-B) and had substantially lower BBBD despite having a relatively high cavitation dose. A version of Figure 12that includes the second patient is shown in Supplemental Figure 2.

Comparing the percent of the subsonication volumes with BBBD and SWAN and the signal intensity changes in contrast enhanced T1W and SWAN MRI (Supplemental Figure 3) showed that petechiae evident in SWAN MRI was typically accompanied by BBBD. We did not observe many targets where petechiae were evident without BBBD, suggesting that vessel damage produced by the sonications did not halt local blood flow.

Burst level analysis

Analysis of the individual cavitation spectra revealed that while most bursts did not have wideband emissions, most of the cavitation dose came from bursts with wideband emissions (Figure 13). While only 21% of the bursts delivered over all the treatments had wideband emissions, 84% of the total cavitation dose was contributed by those bursts. The revised treatment strategy of using low-dose, long-duration sonications resulted in a substantial reduction in the percentage of bursts with wideband emissions. In treatments where we used a cavitation dose goal above 0.5, 31% of the bursts had wideband emissions; in treatments that used a goal less than 0.5, this was reduced to 15%. Only a small portion of the bursts had an identifiable peak at the subharmonic. However, examination of the average spectra in bursts that did not contain wideband or subharmonic emissions in analyses of individual spectra revealed a small peak at the subharmonic (Supplemental Figure 4). The presence of a subharmonic without wideband emissions is a signature for stable cavitation, so these findings suggest that stable cavitation may have been present in bursts that did not contain wideband emissions but below the noise floor of the cavitation monitoring system.

Figure 13:

Analysis of acoustic emissions. For each burst, we categorized the spectra as having wideband (WB) and/or subharmonic (SUBH) emissions. (A) Most bursts did not contain wideband emissions, and later treatments that used a lower exposure level and a longer overall sonication duration had a lower percentage of bursts with wideband emissions. (B) Most of the cavitation dose was contributed by bursts with wideband emissions.

Discussion

The treatment strategy evolved over the course of these treatments, and our results suggest that reducing the overall exposure level through reducing the cavitation goal while increasing the duration results in BBBD with significantly reduced petechiae. These findings are in line with a recent publication by Huang et al., who also settled on a cavitation dose goal of approximately 0.2 in FUS-BBBD in patients with Parkinson’s Disease and found only minor petechiae [32]. While the clinical significance of the petechiae is unknown, reports in animals suggest that the small number of extravasated red blood cells that can be detected in T2*-weighted imaging are not associated with significant tissue damage or neuronal loss [43]. Nevertheless, it would be desirable to avoid petechiae if possible, especially in patients receiving multiple treatments.

Experiments in animals have shown that FUS-BBBD does not require wideband emissions, and that when such emissions are detected, vascular damage is typically observed histologically [24, 25]. The treatments in the current study all had bursts with wideband emissions and varying levels of apparent vascular damage evident in SWAN MRI. We have observed consistent BBBD with no evident petechiae in both small animals and nonhuman primates with the device used here [2, 44]. In those studies, we reduced the power if either wideband or subharmonic emissions were detected and used the second and third harmonic to determine the correct power level to use. While the higher harmonics typically have a stronger signal than the subharmonic, they will be attenuated significantly by the human skull, and the utility of this approach in clinical treatments needs to be confirmed. Another approach increases the exposure level and uses the onset of subharmonic or ultraharmonic (ultrasound frequency times 1.5, 2.5, etc.) emissions as a signature of a specific pressure threshold and reduces the power by a fixed ratio [26]. That approach is promising as it uses a binary signature instead of a signal strength that could be confounded by skull attenuation and other factors. However, the subharmonic and ultraharmonic emissions signals may be small, and our results suggest that there was often a subharmonic peak below the noise floor of the cavitation detectors. The prior preclinical work, along with our finding that most of the cavitation dose was produced by bursts with wideband emissions, suggest that cavitation monitoring can be improved by including metrics of stable cavitation and reducing the exposure level when wideband emissions are detected.

Whatever cavitation metric is used to control the procedure, the effects of the skull on the emissions should be considered. Here, simulations predicted variability in detector sensitivity at 110 kHz in the targeted regions and between patients, at least for discrete frequencies. This variability will impact the apparent threshold for different acoustic signatures and affect the apparent strength of the cavitation signals. Increasing the sensitivity and/or number of detectors and estimating the impact of the skull on the signals might improve reliability and produce more consistent results. Understanding the sensitivity profile of the detectors will be more important if higher frequency bands are used for feedback control.

The attenuating effects of the skull and other factors made cavitation detection in this study more challenging than preclinical studies of FUS-BBBD. The microbubble dose was less than has been typically used in animal models. Most preclinical studies have used a bolus injection before each sonication at the dose used for ultrasound imaging (10 μl/kg for Definity) or higher. Here, we administered the microbubbles as a continuous infusion at a rate of approximately 0.24 μl/kg/min, meaning a dose of 0.72 μl/kg was delivered during a 180s sonication. While the steady-state concentration will be higher, given the rapid clearance of microbubble agents, both the peak and steady-state level of microbubbles was likely lower in the patients than in most preclinical studies. Further, most preclinical studies have not been in white matter structures, which have a low vascular density. Glioma tends to infiltrate along white matter structures [45], and consequently most targets here were in white matter.

Along with the highly attenuating skull, the low doses of microbubbles and their low tissue concentrations resulted in weak cavitation emissions. Increasing the microbubble dose will likely increase the cavitation signal strength and allow for improved detection of small signals. We anticipate that increasing the microbubble dose and reducing the exposure level to minimize the number of bursts with wideband emissions could increase the magnitude of the resulting BBBD while avoiding vessel damage and could be a way to achieve similar results to increasing the sonication duration and repeating the sonications, as we did here. Care should be taken, however, to avoid wideband emissions when increasing the microbubble dose, as the severity of the resulting vascular damage would likely increase.

While there was a clear relationship between the cavitation dose and BBBD, there was significant variability. Differences in tissue vascularity could explain this variability. Methodology issues likely also played a role. These issues include registration errors, effects of overlapping sonications and off-target effects from reflections and incomplete aberration corrections, and limitations on the acoustic simulations. These issues could result in subsonications affecting neighboring tissue locations and errors in relating outcomes to the exposure levels and cavitation metrics. Segmentation of hypointense areas in SWAN MRI was also somewhat subjective and was sometimes hampered by persistent effects from previous treatments.

It is unclear why the relationships between BBBD, petechiae, the exposure levels, and cavitation measures in the second patient did not follow the relationships of the others. The exposure level and cavitation dose were the highest, but the resulting BBBD was modest. The target volume in the second patient was more superficial and anterior than the other cases, and the simulations suggest that the focal volume was larger and more sensitive to which phase correction was used. Perhaps the acoustic emissions arose from a larger volume and the peak microbubble activity was lower than in the other patients. Since it was superficial, the pressure amplitude in the skull, subdural space, and water were high. Perhaps emissions arising from these areas unrelated to bubble activity in the brain confounded the measurements. Differences in tissue vascularity or other factors may have played a role as well. Using MRI sequences sensitive to blood volume or flow might be useful in elucidating the impact of variable vascular density and microbubble concentrations.

The acoustic simulations did not include shear mode conversion, while the phase corrections calculated by the FUS software used during treatment used the shear sound speed when the external angle between the transducer element and the skull was greater than 30°. Since many elements exceeded this angle (Supplemental Figure 2), it is not surprising that the device phase corrections did not appear to improve the simulated acoustic fields. Not including transcranial shear mode conversion may have also affected our simulations of the detectors; low sensitivity in some regions may be overstated by the simulations. Our ability to simulate shear transmission through the skull is limited by the resolution of the clinical CT scans. Complex microstructures in the trabecular bone that lead to high levels of scattering and shear absorption [46] are below the resolution of clinical CT scanners. To estimate transmission, we use empirically derived relationships between the skull density and attenuation. These relationships often have a high estimated attenuation in the trabecular bone that presumably includes scattering from and shear absorption in unresolved structures [47]. A recent study that used micro-CT of cadaver skulls in acoustic simulations suggests that shear transmission is likely low in regions with significant trabecular bone [48]. More work is needed to understand the role of shear propagation in FUS-BBBD.

The simulations suggest that reflections and standing waves can have a significant impact on the focal region when the sonication target is close to the bone if the phase is not corrected. These simulations are consistent with a study by Song et al. that mapped the acoustic field of the low frequency Exablate Neuro system within cadaver skulls [49]. Since the treatment phase corrections used the shear sound speed and our simulations did not, we could not determine how well the phase was corrected in the treatments. A ray tracing approach like that used in the FUS system might be improved by including reflections from the inner skull surface.

This work did not study clinical outcomes, only BBBD and vascular damage measured using MRI-based surrogates. While future work is needed to relate these biomarkers to drug delivery and disease progression, we expect that this study will be relevant to other clinical studies using BBBD in brain tumors and other disorders in the central nervous system.

Conclusions

A strategy of extending the sonication duration and aiming for a low cavitation dose that minimized wideband emissions yielded BBBD that covered most of the targeted volume with only minor MRI-evident petechiae. Subsonication-level control of the exposure level also improved homogeneity of the disruption. While most bursts did not have wideband emissions, most of the cavitation dose came from bursts that did, suggesting that including metrics for stable cavitation into the feedback control might provide more information and improved control. These metrics could be monitored using detectors sensitive at higher harmonics or with improved low-frequency detectors that can detect small subharmonic peaks. Acoustic simulations of the treatments suggest that reflections and standing waves were often present when the focus was placed near the skull, which could lead to off-target BBBD or petechiae. Modeling the cavitation detectors suggests that their sensitivity varied across the targeted volumes and between patients and should be accounted for. Comparing MRI-evident BBBD and petechiae with exposure parameters and cavitation measurements on a subsonication level showed correlations but significant variability, which was likely the result of confounding factors such as overlapping targets and differences in local tissue vascularity. Overall, these results point to potential improvements in FUS-BBBD that might be achieved by including acoustic models in treatment planning and improving the sensitivity and bandwidth of the cavitation monitoring. They also show the need for improved understanding of how the cavitation signals relate to drug delivery and risks for vascular damage.

Highlights.

• We used focused ultrasound to disrupt the blood-brain barrier in glioma patients.

• We found parameters that produced effective disruption with minimal side effects.

• We demonstrated the potential utility of acoustic modeling of the treatments.

• We show how acoustic emissions were related to the disruption and vascular damage.