AAPM Task Group 333 Report: Quality assurance for MRI‐guided focused ultrasound body systems

Abstract

A quality assurance (QA) procedure for magnetic resonance‐guided focused ultrasound (MRgFUS) body systems was developed and tested at five institutions. The QA procedure is based on a custom‐designed phantom with tissue‐mimicking acoustic, MR, and thermal properties specific for MRgFUS. The data acquisition includes assessments of maximum temperature rise, targeting error, signal‐to‐noise ratio (SNR), and three‐dimensional thermal ablation spot size, which are to be tracked over longitudinal repetitions of the QA procedure. The procedure is appropriate for intermediate‐level QA, for example, every 20 patients or 6 months, and can be completed in approximately two hours. A subset of the full procedure is recommended for system verification either daily or before each patient (in addition to the vendor‐recommended procedure). The phantom and QA procedure have been tested on a variety of custom and commercially available MRgFUS body systems, including systems with 1.5 and 3.0 T static magnetic fields. Normal variability in measurements based on data from five functional MRgFUS body systems over periods of at least 6 months is reported and establishes quantitative criteria for passing the QA test. This QA procedure may be used to increase confidence in the safety and performance of MRgFUS treatments.

Abbreviations

CDRH

Center for Devices and Radiological Health

FDA

Food and Drug Administration

FUS

focused ultrasound

HIFU

high intensity focused ultrasound

MRgFUS

magnetic resonance‐guided focused ultrasound

MRI

magnetic resonance imaging

MRTI

magnetic resonance temperature imaging

PRF

proton resonance frequency

QA

quality assurance

RF

radiofrequency

ROI

region of interest

SNR

signal to noise ratio

T1w/T2w

T1‐weighted/T2‐weighted MRI images

TMMs

tissue mimicking materials

US

ultrasound

1. INTRODUCTION

Magnetic‐resonance‐guided focused ultrasound (MRgFUS) systems use MR thermometry ¹ , ² , ³ , ⁴ , ⁵ , ⁶ , ⁷ , ⁸ , ⁹ , ¹⁰ to guide and monitor high intensity focused ultrasound (HIFU) for thermal ablation of tissues. ¹¹ , ¹² There are many examples of MRgFUS systems approved or cleared by the U.S. Food and Drug Administration (FDA) or European Medicines Agency (EMA) for treating a variety of clinical targets. Still more are currently under development. Current clinical applications that have either gained approval/clearance or are being evaluated include uterine fibroids, ¹³ , ¹⁴ soft tissue tumors (e.g., desmoid tumors ¹⁵ , ¹⁶ ), bone, ¹⁷ prostate, ² , ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²² , ²³ brain, ²⁴ , ²⁵ , ²⁶ muscle, ²⁷ , ²⁸ breast, ²⁹ , ³⁰ , ³¹ , ³² , ³³ , ³⁴ liver, ²⁹ , ³⁵ , ³⁶ , ³⁷ , ³⁸ , ³⁹ and kidney. ³⁸ , ⁴⁰

AAPM Task Group 241 recently published a medical physicist's guide to MRgFUS body systems, which provides a comprehensive tutorial of mechanisms of treatment in addition to practical considerations for operation and maintenance. ⁴¹ The term “body system” is used to refer to systems other than transcranial MRgFUS systems, which differ considerably from body systems. Transcranial MRgFUS systems for ablation typically use much lower frequencies, different transducer geometry (e.g., hemispherical), and must cope with significant attenuation and aberration due to propagation of ultrasound through the skull. Consequently, QA procedures for the two types of systems may differ. The AAPM Task Group 241 report includes recommendations for the frequency of various QA tests to be performed by the medical physicist. The purpose of the present AAPM Task Group 333 report is to provide (1) additional specificity for how to conduct some elements of daily and intermediate QA tasks for MRgFUS body systems recommended in the AAPM Task Group 241 report, (2) a specific QA procedure to be performed at intermediate periodicity (e.g., every 20 patients or 6 months), (3) multi‐institution testing for the QA procedure, and (4) assessment of normal variability in measurements in normally functioning MRgFUS body systems to establish quantitative pass/fail criteria. Given the variety in MRgFUS body system configurations due to the variety of anatomical applications and receiver coil designs, performance metrics might vary from system to system. However, potential problems in an MRgFUS body system might be detected by comparing periodic test results to baseline test results acquired from the same system.

The charge of AAPM Task Group 333 was as follows: (1) Design and develop a QA phantom independent of MRgFUS system manufacturers; (2) develop a platform‐independent QA procedure for MRgFUS thermal ablation and possibly hyperthermia; (3) test the QA phantom and procedure at several sites using major commercially available and custom MRgFUS body systems; (4) analyze the results for interlaboratory comparison; (5) prepare a comprehensive report to include phantom design, QA procedure, results analysis, and recommendations for best practices.

The procedure is primarily intended for intermediate QA, for example, every 20 patients or 6 months, but a subset of the procedure is appropriate for daily or pretreatment QA. For wide applicability, the QA procedure should be (1) relevant to both 1.5 and 3 T systems, (2) possible to complete within a few hours, (3) useful for quick identification of problems in MRgFUS body system performance, and (4) tested at multiple institutions on multiple platforms for intervals spanning at least 6 months (corresponding to the maximum time between consecutive intermediate QA tests recommended by the AAPM Task Group 241 report)

Table 1 lists the institutions participating in the AAPM Task Group 333 study and the general properties of their MRgFUS systems.

TABLE 1.

MRgFUS body systems used in the interlaboratory comparison.

		Ultrasound transducer
Institution	MRgFUS system	Frequency (MHz)	Radius of curvature (cm)	Aperture (cm)	MR system field strength
University of Utah	Image Guided Therapy MRgFUS System	1.0	13	15.4	3 T (SMS a )
Fox Chase Cancer Center (FCCC)	Insightec ExAblate 2000	1.0	16	12	1.5 T (GEHC b )
Focused Ultrasound Foundation/University of Virginia (FUSF)	Insightec Exablate 2000	1.0	16	12	3 T (GEHC b )
University of California at San Francisco (UCSF)	Insightec ExAblate 2100	1.0	16	12	3 T (GEHC b )
University Medical Center Utrecht (UMCU)	Profound Sonalleve	1.2	14	13	1.5 T (Philips c )

^a SMS: Siemens Medical Solutions. Erlangen, Germany.

^b GEHC: GE Healthcare, Waukesha, WI, USA.

^c Philips, Amsterdam, Netherlands.

2. INTERLABORATORY COMPARISON STUDY PHANTOM

2.1. Phantom structure and target physical properties

The development of QA program, especially for an emerging application such as MRgFUS, faces a “chicken and egg” dilemma: A rigorous QA program should rely on a QA procedure and a QA phantom, but both components require the other for validation. Therefore, it was necessary for TG 333 to direct the design, manufacture, and characterization of a new vendor‐agnostic QA phantom tailored for the new vendor‐agnostic QA procedure. TG 333 members worked with an experienced scientist (TG 333 consultant member with declared COI) at Sun Nuclear (Norfolk, VA), a medical imaging phantom manufacturer that provided time and access to the required specialized equipment. The consultant had a well‐established record of service to the medical physics community, including active roles with AIUM, IEC, RSNA/QIBA, in addition to AAPM committees and task groups. These TG activities were conducted transparently and with full disclosure. The IP generated through phantom design and development is open to the public and included in this TG report, which lists all design parameters (i.e., “recipe”) of the phantom. Further, in the United States, through several mergers, there is now only one company producing US phantoms of this type.

Involvement of a medical imaging phantom manufacturer was expected to increase the likelihood that the phantom could eventually be commercialized and therefore made available to any medical physicists seeking to implement the AAPM TG 333 QA procedure. By getting a commercial phantom manufacturer involved early in the process, the phantom could be designed with appropriate consideration of cost and practicality of mass production in addition to faithfully mimicking relevant physical properties of soft tissues. Although instructions and recipes for making tissue‐mimicking materials (TMMs) for MRgFUS are available in the public domain, ⁴² , ⁴³ , ⁴⁴ many medical physicists and investigators would prefer commercially available phantoms that closely mimic relevant tissue properties, offer good long‐term stability, and are fabricated under rigorous quality control methods. In addition, many medical physicists would prefer to avoid handling toxic chemicals associated with some TMM recipes. The major vendors of MRgFUS body systems, Profound and Insightec, each have system‐specific test phantoms with proprietary properties available for simple QA.

Figure 1 shows a phantom used in the QA procedure. The phantom geometry was based on a previous phantom model. ⁴² The shape (cylindrical), inner diameter (10 cm), and height (15 cm) were chosen to be compatible with most MRgFUS systems designed for body applications (rather than, for example, transcranial applications) while also limiting volume and corresponding cost. The membrane was 0.003′′ thick Saranex, which has excellent vapor barrier properties. Although nylon meshes are commonly used for diagnostic ultrasound phantoms to help withstand pressure from an ultrasound transducer, they were not required here because the FUS transducers under test were not pressed against the phantom surface (but instead separated by a water path).

FIGURE 1.

An example of the cylindrical phantom used for the MRgFUS QA procedure.

A homogeneous phantom design was chosen to minimize phantom cost and QA procedure complexity. While an inhomogeneous phantom design could be more anatomically realistic and simulate effects such as refraction and aberration more closely, the structure would be application specific. The homogeneous phantom design simplified phantom alignment during MRgFUS measurements, expediting the QA procedure. The homogeneous phantom design satisfied the main purpose of the QA procedure, which was to provide a fast and highly repeatable methodology to easily detect deviations from prior system performance that might indicate malfunction.

Further, channels to accommodate temperature probes (e.g., fiber optic probes) were not included. The task group determined that MR temperature mapping would be adequate to assess MRgFUS system performance on a routine basis without the complexity of independent temperature measurements. Moreover, experimental demonstration of MR thermometry accuracy is more straightforward in a nonlocal (rather than focused) heating experiment. Validation of MR thermometry accuracy in an environment with steep temperature gradients in the TMM (e.g., due to focused ultrasound) would require extremely precise knowledge of probe tip location. Adding a port for a fiber optic probe could introduce a channel for air entry and could be a source of phantom degradation. MR temperature mapping can be checked against a fiber optic temperature probe in a fluid or a phantom. ⁴⁵ , ⁴⁶ , ⁴⁷ If a custom MRTI protocol is implemented, comparison with the vendor recommended protocol or a previously calibrated protocol might serve as a useful check.

Similarly, the phantom was designed without fluid‐filled tubes to simulate perfusion. Perfusion is very application‐specific and patient‐specific, making a general clinically relevant model difficult to implement. Finally, to minimize phantom cost and complexity, the TMM was designed to mimic acoustic, MR, and thermophysical properties central to MRgFUS QA (Table 2) but not designed to mimic other ultrasound properties such as ultrasound backscatter coefficient, ⁴⁸ elastic modulus, ⁴⁹ or shear wave velocity. ⁵⁰

TABLE 2.

Target and measured phantom specifications.

Category	Property	Target range	Measured values
Acoustic	Attenuation coefficient at 1 MHz (dB/cm)	0.4–0.6 51 , 52	0.44 ± 0.1
	Frequency exponent of attenuation coefficient	1.0–1.3 44 , 51 , 52	1.2 ± 0.2
	Sound speed (m/s)	1520–1560 44 , 51 , 52	1542 ± 8
MR	T1 @ 1.5 T (ms)	575–1450 53 , 54	959 ± 12 (@1.4T)
	T2 @ 1.5 T (ms)	40–90 53 , 54	76 ± 4 (@1.4T)
Thermal	Specific heat capacity (J/kg/K)	3420–3800 42 , 43 , 44 , 52	3418 ± 5
	Conductivity (W/mK)	0.49–0.59 44 , 52 , 55	0.598 ± 0.002
Other	Density (kg/m3)	1000–1100 42 , 51	1040 ± 20
	Inner diameter (cm)	10
	Height (cm)	15
	Mass (kg)	1.18
	Melting point (°C)	≥85
	Shelf life (months)	≥12

Note: Measured values are expressed as mean ± standard deviation.

Table 2 shows the target and measured ranges of the acoustic, MR, thermal, and other properties of phantom TMMs. Through iterative design and testing, phantoms were manufactured by Sun Nuclear (Norfolk, VA) according to the following recipe: water (85.7%), polyacrylamide (9.7%), 1‐micron alumina polishing powder (3%), cellulose (0.9%), Tween80 (0.7%), methyl 4‐hydroxybenzoate (0.1%). Other TMM recipes for MRgFUS have been proposed and may be suitable for use in QA phantoms for the procedure specified here, assuming that the TMM properties meet the specifications listed in Table 2. ⁴² , ⁴³ , ⁴⁴ The combination of water base and polymer matrix material has been demonstrated to be stable over time (5–10 years, if stored properly and sealed properly as discussed above to inhibit desiccation) and for temperatures not exceeding 100°C. Phantoms were stored at room temperature in airtight plastic bags or containers to prevent desiccation. They were not refrigerated, as refrigerators can have a dehydrating effect.

Phantoms were weighed prior to each QA procedure to test for weight loss due to desiccation. The manufacturer suggested that in the event of desiccation, a wet sponge could be inserted in the airtight bag for storage to prevent further desiccation and perhaps enable some rehydration. Based on the manufacturer's experience, a 1% change in weight would typically cause less than 1% changes in physical properties. However, the bigger concern was that water loss could potentially cause the gel medium to detach from the container wall or cause slight retraction (curvature) of the membrane due to volume changes, potentially having a detrimental effect on ultrasound coupling and measurements. MR and/or ultrasound imaging may be used to confirm (1) phantom integrity, (2) good contact between the membrane and the TMM, and (3) absence of air bubbles.

Phantoms should be robust enough to handle occasional accidental overheating and still retain original properties. It is important that phantoms approximately mimic tissue attenuation and absorption so the FUS‐induced temperature distributions are similar.

For the interlaboratory comparison described herein, a separate phantom was constructed for each laboratory. The standard deviations of measurements of acoustic properties (performed on four samples using four laboratory setups) described in Section 2.2 give an indication of consistency of the formulations.

2.2. Acoustic characterization

Acoustic properties were measured on small cylindrical samples (5.1 cm outer diameter × 4.5 cm thick) of the TMM in water tanks. Measurements were performed on four different samples using four laboratory setups, with two methods (time delay spectrometry and conventional through transmission) at the FDA, with time delay spectrometry at Acertara Laboratories, and with conventional through transmission at University of Utah, as described previously. ⁴² , ⁵⁶ The measured speed of sound was 1542 ± 8 m/s (mean ± standard error). The power law fit to measured attenuation coefficient as a function of frequency (f) was (0.44 ± 0.1) X f ^1.2 ± 0.2 dB/cm. The variability in these measurements was likely due to a combination of variability in measurement methodology, intersample variability, and drift of sample acoustic properties over time. Even with all these sources of variability, the TMM was believed to simulate the acoustic properties of human tissue consistently enough with target values in Table 2 for the purpose of MRgFUS QA measurements.

2.3. MR characterization

T1 and T2 relaxation times were measured in a Bruker (Berlin, Germany) Minispec mq60 benchtop time‐domain NMR spectrometer equipped with a 1.41‐T permanent magnet and a 32.00‐mm‐diameter probe. TMM samples were poured to a height of about 1 cm into a glass test tube (inner diameter 4 mm) designed for use with the spectrometer. All samples were temperature‐controlled to 28.0 ± 0.1°C. The Minispec software was used for data acquisition. The values obtained for the phantoms were T1 = 959 ± 12 ms and T2 = 76 ± 4 ms. T1 and T2 values vary slightly with field strength. For example, measurements in vivo from human kidney, liver, pancreas, and spleen suggest that at 3.0T, T1 values are 23 ± 11% higher, and T2 values are 15 ± 10% lower (means ± standard deviations) compared with measurements at 1.5 T. ⁵³

2.4. Thermal characterization

Thermal properties were measured using a Decagon (Pullman, WA) KD‐2 thermal property analyzer as described previously. ⁴² , ⁵⁷ Specific heat capacity was 3418 ± 5 J/kg/K (mean ± standard deviation). Thermal conductivity was 0.598 ± 0.002 W/m/K. These values are consistent with the target values shown in Table 2.

3. INTERLABORATORY COMPARISON STUDY DATA ACQUISITION

3.1. Acquisition parameters

The data acquisition and analysis involved assessments of maximum temperature rise (ΔT _max), targeting error, signal‐to‐noise ratio (SNR), and three‐dimensional thermal ablation spot size, which were tracked over longitudinal repetitions of the QA procedure. These measurements required establishment of some preliminary experimental considerations, as will be discussed in this section, regarding alignment of ultrasound transducer to the phantom, acoustic coupling to the phantom, electrical power applied to the ultrasound transducer, amplitude and duration of heating, MR thermometry voxel size, and tests for targeting accuracy. Table 3 shows the data acquisition parameter settings.

TABLE 3.

Experimental configuration and input parameters.

Input parameter	Value
Output electrical power (W)	Adjusted at baseline (initial) QA procedure to achieve ΔT max = 22°C ± 2°C rise in 20 s at geometric focus (same electrical power level should be used for all subsequent longitudinal repetitions of QA procedures)
Acoustic coupling	Degassed, deionized water
Heating pulse duration (s)	20
Number of pulses for each measurement (for statistics)	3
Cooling time duration (min)	3–5 (may be significantly reduced if phantom is translated transversely by at least 2 cm between sonications) a
How far into phantom to apply heating (cm)	As close to 6 cm as system geometry allows
Sequence of points to measure (relative to ultrasound propagation direction)	Geometric focus 5 mm left of geometric focus 5 mm right of geometric focus 5 mm in front of geometric focus 5 mm above geometric focus 5 mm below geometric focus 5 mm behind geometric focus b
Phased array steering test angles (°)	0°, 5°
Temperature mapping plane orientation	Orthogonal to ultrasound beam axis
Maximum slice thickness (mm)	3 (ideal) or 5 (acceptable)
Maximum voxel dimensions in temperature mapping plane (mm)	1 × 1 (ideal) or 1 × 2 (acceptable)
Temporal resolution (s)	≤4.5

^a Cooling times and phantom transverse translations may vary for systems with substantially different transducer frequencies, apertures, and focal distances as well as heating amplitudes and durations compared with the systems and QA procedure described here. See Section 3.1.5.

^b “In front of” and “behind” the geometric focus are along the axis of ultrasound propagation closer to and further from the transducer, respectively. The remaining directions (left, right, above, below) may be defined according to the patient's reference frame.

The QA procedure recommended in this AAPM Task Group 333 report is intended to be applicable to a wide variety of body applications, including breast, liver, kidney, bone, and so forth. This range of applications is associated with a diversity of application‐specific characteristics, including transducer dimensions, transducer frequency, data acquisition parameters, and data analysis parameters. The QA procedure is intended to assess performance of the total system: focused ultrasound source in combination with MR thermometry. While absolute levels of the QA measurements are expected to be system‐dependent, they will provide useful insight into system performance. Moreover, tracking their longitudinal consistency over extended periods of time (e.g., months to a year or two) could help in the detection of system problems when they occur.

3.1.1. Phantom placement

The focus of the ultrasound transducer was placed 6 cm from the phantom surface (or as close to 6 cm as the system geometry would allow). When possible, this focal point was also set at the geometric focus of the transducer array. Other QA procedures have used 5 cm, ⁵⁸ 3 or 5 cm, ⁵⁹ 3–4 cm, ²⁸ or 3 cm. ⁴²

3.1.2. Acoustic coupling

Acoustic coupling of the phantom was achieved using water. Degassed water is typically used in MRgFUS systems for both cooling and acoustic coupling. ⁴¹ Coupling pads have been used previously ⁵⁹ for HIFU QA, but coupling pads can be expensive, can dry out, and introduce additional variability in measurements. While the goal of achieving proper acoustic coupling was important for all systems, the methods to achieve this coupling was system‐dependent. For example, the University of Utah system (Figure 2) had a custom jig that suspended the phantom over the transducer in a degassed water bath. A clear acoustic path was ensured by cleaning the bottom of the phantom with alcohol and ensuring no bubbles in the acoustic path via MR imaging. See also Figures 3, 4, 5.

FIGURE 2.

University of Utah setup.

FIGURE 3.

UCSF and FUSF setup. Phantom placed in receptacle with water coupled with gel on top of the FUS transducer assembly and treatment membrane within MRI bed (left). MRI image (right).

FIGURE 4.

FCCC setup. Phantom placed in receptacle on top of HIFU transducer assembly within MRI bed (left). MRI image (right).

FIGURE 5.

UMCU setup. The phantom was placed in an adapter ring fitting into the standard Sonalleve phantom holder (left). MRI survey of the setup showing the HIFU transducer below.

3.1.3. Heating amplitude and duration

For simplicity, single point sonications were implemented for the QA procedure rather than complex trajectories in which heating would be generated by multiple sonications that move through space in rapid sequence. ⁶⁰ The heating pulse duration was chosen to be 20 s, which is the same as in two previous QA procedures ⁴² , ⁵⁹ and consistent with another (15–25 s). ⁵⁸ The output electrical power was established during the baseline QA procedure experiment by each participating laboratory to achieve maximum temperature rise ΔT _max = 22°C ± 2°C rise in approximately 20 s at the geometric focus for their system. Temperature rises above body temperature of this magnitude are common for MRgFUS because they are associated with rapid cell death. ⁶¹ , ⁶² , ⁶³ The output electrical power setting was recorded and used throughout all subsequent QA procedures at a given laboratory. Other procedures have used preset power levels of 32–120 W, ⁵⁸ 20–100 W, ⁵⁹ and 6.6–20.7 W. ⁴² MR thermometry precision (variability of temperature measurements in response to repeated trials of the same true temperature rise) in the QA procedure was required to be ≤1°C. This should be achievable for many systems provided that the pulse sequence is configured appropriately. An additional QA check could entail longitudinal assessment of consistency of measured calibration curves of ΔT _max versus electrical power over an extended range of applied powers. This could help elucidate amplifier problems if and when they occur.

At UCSF, the Insightec ExAblate was controlled using the research GUI and computer control interface, along with in‐house magnetic resonance temperature imaging (MRTI) monitoring software. At FCCC, the MRgFUS system, consisting of an Insightec ExAblate 2000 and a GE 1.5T MR scanner, was controlled using the bone treatment software, which is routinely used for treatment of bone metastases.

3.1.4. MR thermometry voxel size

To estimate targeting errors and therapeutic spot size accurately, the MR thermometry voxel size should be as small as practically achievable. The therapeutic spot size was characterized by the full‐width half‐maximum (FWHM) in the axial dimension (i.e., ultrasound propagation direction) and the transverse dimension (i.e., perpendicular to the ultrasound propagation direction).

Zero‐padded interpolation can be used during image reconstruction to create finer spacing in the reconstructed voxel grid in MR thermometry images such that one voxel in the finer grid more accurately aligns with the center of the therapeutic spot. This results in more accurate and precise measurement of the spatial distribution of temperature rise in the therapeutic spot

The minimum FWHM of the therapeutic spot is related to the FWHM of the acoustic beam (i.e., in the limit of no thermal diffusion). The transverse FWHM of the acoustic beam under linear propagation conditions through tissue‐like‐attenuating media for a focused circular transducer is 1.41λF/#, ⁶⁴ , ⁶⁵ where λ is the wavelength and F/# is the f‐number (ratio of transducer focal length to diameter). For moderately nonlinear propagation, this formula, which is based on linear diffraction theory, ⁶⁴ has been shown to agree well with measurements for HIFU transducers in water (1.05 and 3.3 MHz) with F/# as low as 1.0 ⁶⁶ even though the conditions for the underlying linear diffraction theory are not always strictly met (see tab. 1 in Wear ⁶⁶ ). For the systems considered in the QA procedure interlaboratory comparison (see Table 1), the theoretical minimum acoustic beam transverse FWHM values were 1.82 mm (Utah), 1.95 mm (Utrecht), and 2.89 mm (FCCC, FUSF, and UCSF). In practice, thermal spot FWHM values will be greater because of thermal diffusion and acoustic center frequency downshift due to propagation through attenuating media. Based on these considerations, the maximum MR thermometry voxel dimensions in the transverse dimensions were chosen to be 1 mm × 1 mm ideally, with 1 mm × 2 mm acceptable if 1 mm × 1 mm is not achievable. (For circularly symmetric beams, like the ones used in this study, the orientation of the voxel in the transverse plane is not critical.) For measurements at higher acoustic frequencies or lower F/#s than those shown in Table 1 or exposure times shorter than 20 s, finer sampling might be necessary. To minimize spatial averaging along the ultrasound propagation direction, maximum MR thermometry slice thickness was chosen to be 5 mm.

3.1.5. Tests for targeting accuracy

Measurements were targeted to seven locations (using combinations of steering and focusing) throughout the ablation volume: the geometric focal point and six other locations ± 5 mm in three orthogonal directions. At a depth of 6 cm, 5 mm corresponded to steering angles up to ±5° (5 mm ≈ 60 mm × tan 5°) in the transverse plane. While steering angles greater than 5° are often utilized in clinical treatments, we believed that 5° adequately captured typical repeatability of system steering performance. A 5° phased array steering angle was thought to be sufficient to see power drop off due to directivity or crosstalk. Due to geometric constraints of the UCSF system, the focus was at a depth of 112 mm (instead of 60 mm), resulting in angles of ±2.5° for offsets of ±5 mm.

Three sonications were performed at each of the seven target locations to obtain means and standard deviations of ΔT _max measurements. This resulted in 3 × 7 = 21 measurements, which was consistent with the goal that data acquisition should require no more than 2 h. Other system characterization procedures have used 6–10, ⁵⁸ at least 3, ⁵⁹ 24, ²⁸ or 1–9 ⁴² measurements. Three measurements were chosen for the present QA procedure to achieve a reasonable balance between practicality of implementing the QA procedure and accuracy of estimates of means and standard deviations.

Repeated measurements could be performed after waiting for a “cooling time” for the phantom temperature to return to ambient level. Previous studies have used intersonication cooling times of 5 ⁵⁸ and 10 min. ⁴² Our measurements indicated that 3 min would be adequate for our systems (see Section 4.1). The intersonication cooling time can be significantly reduced if the phantom is translated transversely by several transverse focal spot sizes (e.g., 2 cm for our systems; see Section 4.4) between sonications.

Table 4 shows the MR thermometry parameters.

TABLE 4.

MR thermometry parameters.

Parameter	Utah	UCSF	FCCC	FUSF	Utrecht
MRI scanner	Siemens 3T	GEHC 3T	GEHC 1.5T	GEHC 3T	Philips 1.5T
Sequence	3D GE/EPI	FSPGR	FSPGR	FSPGR	2D GE/EPI
TR (ms)	28	26	25	26	39
TE (ms)	12	12.9	12.3	12.8	19
Slice thickness (mm)	4	3	5	5	5
Coronal plane spatial resolution (mm) a	2.0 × 2.0	1.09 × 2.18	1.09 × 2.1	1.09 × 1.09	1.19 × 1.26
FOV (mm)	384 × 216	280 × 280	280 × 280	280 × 280	300 × 300
Matrix (frequency × phase)	192 × 108	256 × 128	256 × 128	256 × 128	252 × 238
Flip angle (°)	12	30	30	30	16
Temporal resolution (s)	4.47	3.3	3	3.7	3.5
Bandwidth (Hz/pixel)	Readout: 745 Phase: 106	44.4	44.4	44.4	583
Echo train length	7	1	1	1	11
Frequency direction	SI	SI	SI	SI	SI
Number of excitations	1	1	1	1	1
MRI coil(s) used	Single 16 cm diameter loop around phantom in geometric focus plane.	OEM body coil	Insightec anterior posterior (A/P) pelvic coil	OEM body RF coil	5‐channel Sonalleve A/P array

Note: In some cases, matrix was zero filled to 512 in one or more dimensions.

Abbreviations: 2D, two‐dimensional; 3D, three‐dimensional; FOV, field of view; FSPGR: fast spoiled gradient echo; GE/EPI, gradient echo with echo‐planar imaging readout; SGE, spoiled gradient echo; TE, echo time; TR, repetition time.

^a Coronal plane spatial resolution row provides acquired, not interpolated, values. Some sites used interpolation to decrease pixel spacing. For example, Utah interpolated to achieve 0.5 mm × 0.5 mm pixel spacing while UCSF interpolated to achieve 1.09 mm × 1.09 mm pixel spacing.

Figures 2, 3, 4, 5 show experimental setups.

3.2. Measurements

Table 5 shows the measurements performed during the QA procedure.

TABLE 5.

Output measurements.

Measurement	Method	Recommended measurement precision
Maximum temperature rise ΔT max (°C) at geometric focus and at six points targeted 5 mm away from geometric focus	Highest temperature rise (ΔT) pixel, average of ROI surrounding the highest temperature rise (ΔT) pixel, or functional fit to local spatial distribution of temperature rise (e.g., Gaussian)	±1°C
Temperature versus time profile at geometric focus	Record of ΔT(t) at location of ΔT max throughout 20 s of heating time followed by 60 s of cooling time.	±1°C at each time point
Targeting error in three dimensions (mm, mm, mm) and magnitude of targeting error (mm) at geometric focus and at six points targeted 5 mm away from geometric focus	Distance from center of prescribed spot to ΔT max position (digital calipers on MR) Measure Δx, Δy, Δz, and $\sqrt{{(\mathrm{\Delta }x)}^2+{(\mathrm{\Delta }y)}^2+{(\mathrm{\Delta }z)}^2}$. Zero‐padded interpolation may be used to improve localization accuracy (see Section 3.1.4).	±1 mm
SNR	μ/σ in unheated region. σ is measured by taking a sequence of images (measuring the variation over time not over space).	–
Three‐dimensional sonication spot size at geometric focus	Focal spot full width half maximum (FWHM) in axial and transverse dimensions	±1 mm
Phantom weight	Phantom must be weighed prior to each QA test to check for desiccation.	±1 g
Room temperature	Room temperature must be measured prior to each QA test	±1°C

3.2.1. Maximum temperature rise (ΔT _max)

Maximum temperature rise ΔT _max was measured from the highest temperature rise voxel of the image (Utah, FCCC, FUSF, UMCU), or alternatively from an average ΔT over a 3 × 3 pixel ROI centered around the pixel with the maximum ΔT (UCSF). Averaging over multiple voxels might produce more stable results when images are noisy but can underestimate ΔT _max. ⁵⁹ The optimal ROI size depends on application‐specific considerations including beam dimensions, thermometry grid dimensions, and system noise. Fitting a function (e.g., Gaussian) to spatially dependent ΔT measurements is another alternative. Whichever approach was adopted initially by each laboratory, the same approach was used for each repetition of the QA procedure at each site to ensure consistency. Maximum temperature rise ΔT _max was measured at 7 points: at the transducer geometric focus (determined by the curvature of the transducer surface) and at six points with the acoustic beam targeted 5 mm away from geometric focus.

3.2.2. Targeting error

Targeting error could arise due to factors such as imperfect positioning of transducer, refraction, and/or changes to the transducer or RF amplifier subsystems. Digital calipers were used to measure the distances in three dimensions (Δx, Δy, Δz) from the center of the prescribed target to the position of measured ΔT _max. ²⁸ , ⁵⁸ The targeting error was $\sqrt{{(\mathrm{\Delta }x)}^2+{(\mathrm{\Delta }y)}^2+{(\mathrm{\Delta }z)}^2}$.

3.2.3. Signal‐to‐noise ratio (SNR)

SNR was computed as the ratio of μ/σ, where μ was the mean signal and σ was the standard deviation of the noise. Both μ and σ were measured from a sequence of at least 10 MRTI images that were acquired over a 2‐min period (without application of focused ultrasound). If image noise is spatially uniform, then SNR could alternatively be measured from the difference of two images using the NEMA subtraction method. ⁶⁷ The NEMA method has its origin in determining the SNR of diagnostic MRI images, such as a T1‐weighted MRI of the brain. Historically, these had imaging times of a few minutes, and performing duplicate scans would then give a feasible and valuable SNR assessment. It requires averaging over an ROI to determine SNR in some region of the image. For many advanced MRI methods, however, the NEMA method is inadequate. Examples include fMRI, MR‐Thermometry, and MR ASL‐perfusion imaging. These methods all acquire images repeatedly in rapid succession, and allow for spatial heterogeneity of SNR, by analyzing SNR based on the signal variation in the temporal dimension. Hence the term tSNR is frequently used in literature. tSNR maps give direct insight into the variation for SNR across the object, and an ROI analysis can be used to obtain a stable figure of merit for the data.

If measured properly, SNR is inversely related to the phase noise and can be used to estimate temperature uncertainty. Although field drift may occur during a 2‐min MRTI imaging sequence, the effect on SNR measurement will be minimal. However, field drift could be assessed by analyzing the sequence of phase images. Phase noise, whether measured directly or indirectly from the SNR, is directly related to noise in the temperature measurements. It is essential to use the same image filters, normalization, coil intensity corrections, and so forth, in all subsequent QA tests. In the recommended QA procedure, the mean μ over time for each voxel was computed within the ROI (including at least 10 × 10 voxels) as shown in Figure 6. The standard deviation over time of every voxel within the ROI was computed. The SNR was computed from the ratio of average mean to average standard deviation. The effects of spatially varying MRI receiving coil array response can be minimized by using multiple ROIs.

FIGURE 6.

Signal‐to‐noise ratio (SNR) measurement using two ROIs. (Left) Mean MR signal intensity for each voxel. The spatial variation in MR signal intensity is typical of the spatial variation in sensitivity of an MRI receiving coil array. This can be normal or an indication of an issue with the MRI scanner receiver coil array. (Middle) The standard deviation of MR signal intensity for each voxel. (Right) The SNR for each voxel. These MR images were acquired without heating. The pixel values in the MR image are related to temperature but not equal to temperature.

3.2.4. Focal spot full‐width‐half maximum (FWHM)

Spatial resolution was characterized by FWHM in the axial dimension (ultrasound propagation direction) and the transverse dimension (perpendicular to ultrasound propagation direction). The FWHM was calculated by mean measured distances between locations of 50% ΔT _max and ΔT _max (multiplied by 2). Alternative methods for characterizing sonication spot size include fitting a two‐dimensional Gaussian function to spot, ⁵⁹ and counting the number of voxels surpassing target ΔT for more than 20% of the time. ²⁸ For a Gaussian function centered at (x0, y0),

$$G\left(x,y\right)=exp\left[\frac{-{\left(x-x0\right)}^2}{2{\sigma }_x^2}\right]exp\left[\frac{-{\left(y-y0\right)}^2}{2{\sigma }_y^2}\right]$$

FWHM _x ≈ 2.35σ _x and FWHM _y  = 2.35σ _y (see Section II.C in Wear ⁶⁸ ). In some cases, thermometry images might have to be exported from the MRI system to an offline platform for custom processing. The critical point here, as with all metrics in this QA procedure, is to use the same method for every longitudinal assessment for a given MRgFUS system. The ability to measure FWHM requires sufficient spatial samples per spot dimension and may not be practical in some cases.

3.2.5. Phantom and room temperature check

Before the QA procedure, phantoms were left out at the testing room temperature for at least a few hours to equilibrate to room temperature. Phantom interfaces were visually inspected for irregularities that could interfere with ultrasound transmissions. Phantoms were weighed prior to each QA procedure to test for desiccation, and room temperature was measured.

4. INTERLABORATORY COMPARISON STUDY RESULTS

4.1. Maximum temperature rise (ΔT _max)

Figure 7 shows temperature versus time curves acquired in the phantom using an Insightec ExAblate 2000. The cooling portions (i.e., for times after about 18 s in Figure 7) were used to determine the minimum time lag between repeated measurements to ensure that each measurement begins at a temperature reasonably close to baseline. Cooling portions for nine trials for the ExAblate 2000 were fit to decaying exponentials. This yielded an estimated time constant of 28 ± 6 s (mean ± standard deviation). This implied that a cooling time of 4.6 time constants or about 2 min would be adequate to allow the temperature rise ΔT(t) to decay down to about 1% of its maximum value before a repeated measurement should be performed at the same location, consistent with a 3–5 min cooling times recommended in Table 3 and reported elsewhere. ⁵⁸

FIGURE 7.

Real‐time temperature measurement with MR thermometry during sonication (red curve: maximum temperature measured at the focal spot in a single voxel of 0.86 × 1.72 × 3.00 mm; green curve: average temperature measured over 3 voxels). The curves show temperature rise ΔT(t) above an assumed value for initial temperature at time = 0 set in system software: 37°C, which corresponds to typical human body temperature. Actual temperatures in phantoms would initially be at room temperature and therefore would be approximately 17°C lower. The curves show maximum temperature rises ΔT _max of 29°C and 21°C.

Figure 8 shows ΔT _max at geometric focus versus the number of months from baseline for the five participating institutions. FUSF1 and FUSF2 refer to two different copies of the phantom interrogated with the same system and QA procedure. When using the initial constant transmit power for all measurements over the test period (at least 6 months), ΔT _max values were typically within 5°C of the first set of measurements.

FIGURE 8.

Maximum temperature rise ΔT _max after approximately 20 s at geometric focus. Month 0 corresponds to baseline measurements for each participating institution.

Figure 8 suggests the possibility of a decreasing trend of ΔT _max over time, potentially due to temporal drift in MRgFUS system or phantom properties. However, when linear regressions were applied to (1) temperature rise versus time data and (2) change in temperature rise relative to baseline versus time data, 95% confidence intervals for correlation coefficients and slopes were found to include zero. Although the trends did not achieve statistical significance in this limited dataset, ultrasound phantoms are known to exhibit some drift in acoustic properties over time. The possibility of phantom drift may be countered by (1) replacing phantoms periodically (e.g., annually) and/or (2) emphasizing changes in QA metrics compared to measurements from the immediately previous QA session rather than earlier QA session(s).

Table 6 shows ΔT _{max, 5 mm} − ΔT _{max, at focus} where (1) ΔT _{max, 5 mm} = ΔT _max measured at a target point 5 mm away from the geometric focus (left, right, in front, above, below, or behind) when beam was steered and/or focused to the target point and (2) ΔT _{maxat focus} = ΔT _max measured at the geometric focus when beam was focused to the geometric focus. Temperature consistency was within 3°C for all positions, except at UCSF where limitations on off‐axis focusing with the sectored‐concentric ring array generated grating lobes and lower temperatures for the shallow off‐axis points.

TABLE 6.

ΔT _{max, 5 mm} − ΔT _{max, at focus} (°C).

	UCSF	FCCC	Utah	FUSF1	FUSF2	Utrecht
5 mm left	−12.4 ± 0.2	0.2 ± 0.4	−1.4 ± 1.4	−0.1 ± 0.2	−0.2 ± 0.1	−1.2 ± 1.0
5 mm right	−12.2 ± 0.4	−0.4 ± 1.3	−1.6 ± 0.4	−0.3 ± 0.3	0.2 ± 0.7	−0.6 ± 0.4
5 mm front	−12.2 ± 0.3	0.0 ± 0.5	−2.4 ± 1.9	0.4 ± 0.7	0.8 ± 0.6	−1.3 ± 0.9
5 mm above	−0.9 ± 0.6	−0.9 ± 0.7	1.5 ± 0.7	−0.4 ± 0.3	−0.4 ± 0.6	0.5 ± 0.7
5 mm below	−0.1 ± 0.6	1.8 ± 0.4	−1.6 ± 2.1	0.2 ± 0.6	−0.2 ± 0.4	−0.9 ± 0.9
5 mm behind	−12.3 ± 0.3	0.2 ± 1.3	−2.9 ± 0.4	−0.2 ± 1.1	−1.0 ± 0.2	1.1 ± 1.8

(1) ΔT _{max, 5 mm} = ΔT _max measured at a target point 5 mm away from the geometric focus (left, right, in front, above, below, or behind) when beam was steered and/or focused to the target point and (2) ΔT _{max, at focus} = ΔT _max measured at the geometric focus when beam was focused to the geometric focus. Values expressed as mean ± standard deviation.

The outlier results from UCSF deserve some discussion. In setting up the experiments at UCSF, the phantom with receptacle was placed directly on the treatment platform surface (Figure 3). The UCSF Insightec ExAblate 2100 body transducer was a spherically curved 208 element sectored concentric ring phased array with a natural geometric focus at 16 cm (12 cm diameter, f‐number = 1.3), with inherent limitations to off‐axis beam steering. The experiments were performed at 6 cm depth within the phantom as specified, but with the transducer in its home depth, the focal distance was 11 cm from the transducer surface. The array generated grating lobes and reduction of peak intensity, and hence lower temperature compared to on‐axis focusing (Table 6).

It is difficult to achieve a “one size fits all” QA procedure, considering the variety in designs and anatomical applications of body MRgFUS systems. Medical physicists might have to adapt the MRgFUS QA procedure presented here to manage unusual aspects of their own systems.

4.2. Targeting error

Figure 9 shows targeting error when targeting to the geometric focus. Absolute levels of measurements were system‐dependent, but for each system, measurements remained consistent to within 0.25 mm over time. Table 7 shows off‐axis targeting error for the six remaining positions investigated for those systems that achieved steering electronically with a phased array. (Note FCCC achieved steering by mechanical positioning of the transducer).

FIGURE 9.

Targeting error when targeting the geometric or primary focus point. FCCC data are not included on this plot as the method of target spot manipulation was mechanical rather than electronic.

TABLE 7.

Targeting error when targeting away from geometric focus (mm).

	UCSF	Utah	FUSF1	FUSF2	Utrecht
5 mm left	0.6 ± 0.1	0.5 ± 0.1	1.5 ± 1.1	1.5 ± 0.6	1.2 ± 0.2
5 mm right	0.5 ± 0.1	0.5 ± 0.2	1.5 ± 0.9	0.8 ± 0.3	1.4 ± 0.4
5 mm front	0.5 ± 0.1	0.3 ± 0.1	0.7 ± 0.2	1.0 ± 0.2	0.8 ± 0.1
5 mm above	0.2 ± 0.2	0.6 ± 0.2	1.0 ± 0.6	1.0 ± 0.5	1.8 ± 0.2
5 mm below	0.8 ± 0.3	1.4 ± 0.6	0.7 ± 0.4	1.3 ± 0.2	1.7 ± 0.3
5 mm behind	0.5 ± 0.1	0.4 ± 0.1	1.0 ± 0.4	0.8 ± 0.2	1.5 ± 0.6

Note: Values expressed as mean ± standard deviation.

4.3. Signal‐to‐noise ratio (SNR)

Figure 10 shows SNR measurements at the geometric focus versus the number of months from baseline. The range of SNR values was rather broad, indicating diversity in designs of MRgFUS systems. Absolute levels of measurements were system‐dependent, but for each system, measurements remained consistent to within approximately 24% (root mean squared difference from baseline) over time. As in Figure 8, Figure 10 shows an unusually large change between measurements at 0 and 8 months for the FUSF1 system. Such an inconsistency might signal a system problem. In cases like this, it may be useful to examine the signal and noise data separately to test for potential interference from the FUS equipment or determine if receiver coil failure has occurred.

FIGURE 10.

Signal‐to‐noise ratio (SNR) at geometric focus.

4.4. Focal spot full‐width‐half maximum (FWHM)

Figure 11 shows FWHM of sonication spot size in the transverse dimension (perpendicular to the ultrasound propagation direction) and the axial dimension (ultrasound propagation direction). As expected, the FWHM in the transverse dimensions (x and y) were much smaller than the FWHM in the axial dimension (z). Absolute levels of measurements were system‐dependent, but for each system, measurements remained consistent to within ±15% over time. The FWHM were considerably greater than the MR thermometry voxel dimensions (approximately 1 mm × 1–2 mm in transverse dimensions and 5 mm in axial dimension; see Table 3), which justified our requirements for MR thermometry sampling. However, for measurements at higher acoustic frequencies or lower F/#s than those shown in Table 1 or exposure times shorter than 20 s, finer sampling might be necessary because of considerations discussed in Section 3.1. Transverse FWHM were somewhat larger than the theoretical (in the absence of thermal diffusion) minimum values calculated in Section 3.1, indicating the effects of thermal diffusion and frequency downshift due to linear ultrasound attenuation. Due to MRTI limitations, FWHM measurements at UCSF were only obtained in the transverse plane. FWHM measurements were not performed on data from the Focused Ultrasound Foundation due to inadvertent deletion of datafiles from the system.

FIGURE 11.

FWHM of sonication spot in transverse dimension (perpendicular to ultrasound propagation) (left) and axial dimension (parallel to ultrasound propagation) (right).

To check for potential redundancy of QA procedure measurements, correlation coefficients between changes in measurements of ΔT _max, targeting accuracy, SNR, transverse FWHM, and axial FWHM were computed. The 95% confidence intervals for estimates of correlation coefficients for temporal changes of all parameter pairs included zero.

4.5. Phantom and room temperature check

The weight change of the phantoms averaged over all the laboratories between first and last QA procedure tests (time spans ranging from 6 to 30 months) was −0.85% ± 0.39% (mean ± standard deviation), suggesting negligible desiccation throughout the study. No irregularities that could interfere with ultrasound transmission were found on phantom interfaces. The room temperature averaged over all laboratories and QA procedure tests was 19.4°C ± 2.0°C.

5. RECOMMENDED QA PROCEDURE

As explained further in Section 6, the quantitative criteria proposed in this section correspond to measured values less than two standard deviations above the mean values measured in the interlaboratory study. These are “soft” criteria in the sense that there is not a dramatic difference between a measurement slightly below or slightly above the criterion level. Rather, measurements near or above the criteria should raise concern.

5.1. Maximum temperature rise (ΔT _max)

Objective: To measure maximum temperature rise, ΔT _max (1) at geometric focus with beam focused to the geometric focus and (2) at six other target points 5 mm away from the geometric focus (left, right, in front, above, below, and behind) with beam steered and/or focused to the target point.

Applicability: All MRgFUS systems.

Phantom(s): A phantom with similar characteristics as described in Section 2 or a phantom tailored to the relevant clinical application and characterized according to the methods of Section 2.

Data acquisition: Use a heating pulse duration of 20 s. For initial baseline measurements, adjust output electrical power to achieve a ΔT _max at the focus of approximately 22°C ± 2°C. Use the same output electrical power for all subsequent QA procedures. (An additional QA check could entail longitudinal assessment of consistency of measured calibration curves of ΔT _max vs. electrical power over an extended range of applied powers. This could help elucidate amplifier problems if and when they occur.) Temperature precision for MR thermometry should be ≤1°C. Maximum MR thermometry voxel dimensions should be no greater than 1 mm × 2 mm (ideally 1 mm × 1 mm) in the transverse dimensions. Maximum MR thermometry slice thickness should be 5 mm. Use water for acoustic coupling. The geometric focus should be placed 6 cm from the phantom surface (or as close to 6 cm as the system geometry allows). Perform ΔT _max measurements (1) at geometric focus with beam focused to geometric focus and (2) at six target points 5 mm away from the geometric focus (left, right, in front, above, below, and behind) with beam steered and/or focused to the target point. Perform at least three ΔT _max measurements at each location (for a total of 3 × 7 = 21 ΔT _max measurements) so that means and standard deviations at each location may be computed. Make sure that each treated point begins with temperature close to ambient temperature at time = 0 by repositioning the phantom or by waiting for sufficient cooling time (e.g., 3 min for systems similar to ones investigated in this interlaboratory investigation) between measurements.

Data analysis: Record output electrical power. ΔT _max may be measured from the hottest voxel of the image, from an average temperature over a small ROI (e.g., 3 × 3 pixel) centered around the pixel with the maximum ΔT, or from a Gaussian fit. Zero‐padded interpolation may be used to improve localization accuracy (see Section 3.1.4). Averaging over multiple voxels might produce more stable results when images are noisy but can underestimate ΔT _max. Whichever approach is adopted initially (baseline measurements), the same approach should be used at each subsequent QA procedure to ensure consistency.

Criteria: For a constant output electric power (determined at baseline QA procedure), typical ΔT _max at geometric focus after 20 s is within approximately 4.0°C of baseline measurements. Typical difference between ΔT _max with beam targeted at six points 5 mm away from geometric focus and ΔT _max at geometric focus (ΔT _{max, 5 mm} − ΔT _{max, at focus}) is within approximately 2.2°C.

Frequency: Daily or pretreatment QA: Measurements with beam focused to geometric focus and subsequently at one target point with beam steered and focused 5 mm away from geometric focus in a transverse direction. Intermediate QA (e.g., every 20 patients or 6 months): Measurements with beam targeted to geometric focus and at six target points with beam steered and/or focused 5 mm away from geometric focus.

5.2. Targeting error

Objective: To measure targeting error (1) at geometric focus when beam is focused to the geometric focus and (2) at six target points 5 mm away from the geometric focus (left, right, in front, above, below, and behind) when beam is steered and/or focused to the target point.

Applicability: All MRgFUS systems.

Phantom(s): A phantom with similar characteristics as described in Section 2 or a phantom tailored to the relevant clinical application and characterized according to the method of Section 2.

Data acquisition: Same as Section 5.1.

Data analysis: Targeting error may be characterized by the distance from the center of prescribed spot to the position of the ΔT _max. Using digital calipers is the most straightforward method to measure the distance in three dimensions (Δx, Δy, Δz) from the center of the prescribed target to the position of measured ΔT _max. The overall targeting error would then be $\sqrt{{(\mathrm{\Delta }x)}^2+{(y)}^2+{(\mathrm{\Delta }z)}^2}$.

Criteria: Typical targeting errors are within 1.3 mm at geometric focus and 1.9 mm at points when beam is steered and/or focused 5 mm away from the geometric focus.

Frequency: Daily or pretreatment QA: Measurements with beam focused to geometric focus and subsequently at one point with beam steered and/or focused 5 mm away from geometric focus in a transverse direction. Intermediate QA (e.g., every 20 patients or 6 months): Measurements at geometric focus and at six points steered and/or focused 5 mm away from geometric focus.

5.3. Signal‐to‐noise ratio (SNR)

Objective: To characterize the noise level of the MR thermometry imaging (MRTI).

Applicability: All MRgFUS systems.

Phantom(s): A phantom with similar characteristics as described in Section 2 or a phantom tailored to the relevant clinical application and characterized according to the method of Section 2.

Data acquisition: SNR is measured without application of focused ultrasound. A sequence of at least 10 images (and therefore, 10 noise realizations) is acquired over a 2‐min period. Temperature precision for MR thermometry should be ≤1°C. Maximum MR thermometry voxel dimensions should be 1 mm × 2 mm (or ideally, 1 mm × 1 mm) in the transverse dimensions. Maximum MR thermometry slice thickness should be 5 mm. If the image noise is spatially uniform, then SNR could alternatively be measured from the difference of two images using the NEMA subtraction method. ⁶⁷

Data analysis: The mean μ over time for each voxel is computed within the ROI (including at least 10 × 10 voxels) as shown in Figure 6. The standard deviation σ over time of every voxel within the ROI is computed. SNR is computed from the ratio of average mean to average standard deviation. The effects of spatially varying MRI receiving coil array response may be minimized by using multiple ROIs. Record both signal and noise values to help trace causes of unexpected changes in SNR.

Criteria: Typical variation from baseline is within 48%.

Frequency: Intermediate QA, for example, every 20 patients or 6 months.

5.4. Focal spot full‐width‐half maximum (FWHM)

Objective: To characterize the precision of the treatment spot in axial and transverse dimensions.

Applicability: All MRgFUS systems.

Phantom(s): A phantom with similar characteristics as described in Section 2 or a phantom tailored to the relevant clinical application and characterized according to the method of Section 2.

Data acquisition: Same as Section 5.1 but only applied at the geometric focus.

Data analysis: FWHM may be calculated by mean measured distances between locations of ΔT _max and 50% ΔT _max (and multiplying by 2). A two‐dimensional Gaussian function may be fit to measurements of spatially varying ΔT. Whichever approach is adopted initially (baseline measurements), the same approach should be used at each subsequent QA procedure to ensure consistency. The ability to measure FWHM requires sufficient spatial samples per spot dimension and may not be practical in some cases.

Criteria: Typical variations from baseline for FWHM are within approximately 0.7 mm in the transverse dimension and within approximately 3.4 mm in the axial dimension.

Frequency: Daily or pretreatment QA

5.5. Phantom and room temperature check

Objective: To assess phantom changes between QA measurements across many months.

Applicability: All MRgFUS systems.

Phantom(s): A phantom with similar characteristics as described in Section 2 or a phantom specifically tailored to the relevant clinical application and characterized according to the method of Section 2.

Data acquisition: Weigh phantom with a scientific balance. Check phantom interface for irregularities that could interfere with ultrasound transmission.

Data analysis: Record phantom weight.

Criteria: Phantom weight should change by less than 1.7% from baseline.

Frequency: Intermediate QA, for example, every 20 patients or 6 months.

6. CONCLUSIONS AND RECOMMENDATIONS

6.1. Quantitative criteria for passing QA procedures

Given the diversity in MRgFUS body system configurations due to the range of anatomical applications and receiver coil designs, performance metrics might vary from system to system. However, potential problems in an MRgFUS body system might be detected by comparing periodic test results to baseline test results acquired from the same system. Medical physicists may wish to repeat baseline measurements over multiple days to increase confidence in reference values for their MRgFUS body system.

QA Pass Criteria in the right column of Table 8 correspond to measurements that would be less than two standard deviations above the mean values observed in the interlaboratory comparison. These values were used to generate the criteria listed in Section 5. Assuming normally distributed random variables, there is only about a 2% chance of exceeding these levels unless there is a change in the underlying distributions (which could signal a malfunction). As stated at the beginning of Section 5, these are “soft” criteria in the sense that there is not a dramatic difference between a measurement slightly below or slightly above the criterion level. Rather, measurements near or above the criteria should raise concern.

TABLE 8.

MRgFUS QA—typical measurement variation for normally functioning systems ^a .

Measurement (x)	Magnitude $\mathit{x}$ (μ ± σ)	Magnitude of change relative to baseline $|\mathit{x}-{\mathit{x}}_{\mathit{B}}|$ (μ ± σ)	QA Pass Criteria (≤μ + 2σ except where otherwise noted)
Maximum temperature rise ΔT max at geometric focus after 20 s minus baseline maximum temperature rise ΔT max at the same reference transmit voltage		2.0 ± 1.0°C	≤4.0°C
Maximum temperature rise ΔT max at 6 locations at target points 5 mm away from geometric focus minus maximum temperature rise ΔT max at geometric focus		0.8 ± 0.7°C b	≤2.2°C
Targeting error at geometric focus	0.5 ± 0.4 mm		≤1.3 mm
Targeting error 5 mm away from geometric focus	0.9 ± 0.5 mm		≤1.9 mm
SNR		20 ± 14%	≤48%
Focal spot transverse FWHM		0.3 ± 0.2 mm	≤0.7 mm
Focal spot axial FWHM		1.6 ± 0.9 mm	≤3.4 mm
Phantom weight		−0.9 ± 0.4% c	≥−1.7% (μ – 2σ)
Room temperature	19.4 ± 2.0°C

^a Values are expressed as mean (μ) ± standard deviation (σ). x _B = baseline measurement. Pass/fail criteria may be system‐ and application‐dependent.

^b Values from UCSF system were excluded from this calculation because of anomalous design features. See Section 4.1.

^c Unlike other entries in this column, the value is the change relative to baseline rather than the magnitude of change relative to baseline.

Except for targeting errors, the QA procedure is mostly about consistency of system performance over time rather than absolute values. However, medical physicists can decide, based on their clinical application, if some parameters must meet certain absolute values before system performance is considered acceptable. For example, the proximity of sensitive off‐target tissues (e.g., nerve and bone) with respect to target tissue might impose absolute limitations for FWHM (axial and transverse) to minimize risk of collateral damage.

6.2. The AAPM Task Group 333 procedure in the broader context of complete MRgFUS QA

As recommended in the AAPM Task Group 241 report, QA measurements such as those described in this AAPM Task Group 333 report are part of a larger QA framework that includes three levels: (1) daily or before each patient, (2) every 20 patients or 6 months, and (3) every 100 patients or 1 year. ⁴¹

For daily or pretreatment QA tests, a subset of the AAPM Task Group 333 intermediate periodicity procedure reported here is recommended. This includes measurements of maximum temperature rise, focal spot dimensions (FWHM), targeting error, and steering/focusing accuracy at only one point away from the geometric focus (but not all 6 points listed in Table 3). ⁴¹ The steering test point should be one of the four points in a direction transverse to the beam axis. Since the transverse FWHM is much smaller than the axial FWHM (see Figure 11) targeting errors will be more noticeable in the transverse direction. Additional recommended daily or pretreatment QA steps include (1) visual inspections for fluid leaks, integrity of the oil or water tank, cables, hoses, and connectors, and (2) functionality checks of MR imaging coil, transducer positioning system, operator safety device, and patient emergency safety device (e.g., “stop button” for ultrasound energy delivery). ⁴¹ It should be possible to complete these tasks in 1 h.

As recommended in the AAPM Task Group 241 report, the tests reported here should be complemented with other intermediate QA procedures such as testing the motor system and measuring MR temperature imaging accuracy (e.g., by comparison to an invasive fiber optic probe). Checking MR temperature mapping against a fiber optic temperature probe in a fluid or a phantom ⁴⁵ , ⁴⁶ , ^{47 )} would be especially important if excessive changes in temperature rise are measured in the QA procedure described in this AAPM Task Group 333 report.

Less frequent testing (every 100 patients or 1 year) recommended by AAPM Task Group 241 includes full acoustic output characterization, which requires pressure field mapping using hydrophones. ⁶⁹ , ⁷⁰ , ⁷¹ An International Electrotechnical Commission hydrophone standard ⁷² contains recommendations for hydrophone sensitivity deconvolution methods ⁷³ , ⁷⁴ , ⁷⁵ , ⁷⁶ and hydrophone spatial averaging correction methods ⁶⁶ , ⁷⁷ for correcting hydrophone measurements for artifacts associated with nonlinear waves, such as those produced by MRgFUS body systems.

While there may be some overlap between the AAPM Task Group 333 QA procedure with vendor‐recommended QA procedures, the AAPM Task Group 333 QA procedure benefits from documented performance testing on multiple systems over an extended time. It is plausible that this QA procedure could be adapted to hyperthermia systems and their associated longer sonication times. Validation would require testing using dedicated hyperthermia systems.

The intermediate‐periodicity QA procedure reported here is based on a phantom that has been custom designed for MRgFUS QA by the expert panel assembled for Task Group 333. The data acquisition includes measurements of maximum temperature rise, targeting error, SNR ratio, and three‐dimensional thermal ablation spot size. It has been tested on systems with 1.5 and 3.0 T static magnetic fields. Normal variability in measurements based on data from five normally functioning MRgFUS body systems reported here provides a basis for quantitative criteria for passing a QA test. This 2‐h QA procedure may be used to increase confidence in the safety and effectiveness of MRgFUS treatments.

CONFLICT OF INTEREST STATEMENT

The following members of Task Group‐333: MRI‐Guided Focused Ultrasound Quality Assurance attest that they have no potential conflicts of interest related to the subject matter of materials presented in this document: Chris Diederich, Lili Chen, Xiaoming Chen, Keyvan Farahani, Kisoo Kim, Eugene Ozhinsky, Douglas Christensen, Mathew Eames, Clemens Bos, and Renee Hovenier. The following members of Task Group‐333: MRI‐Guided Focused Ultrasound Quality Assurance disclose the following potential conflict(s) of interest related to subject matter or materials presented in this document: Keith A. Wear and Subha Maruvada are employees of the US Food and Drug Administration. Chrit Moonen has research collaborations with Profound Medical, with Philips Healthcare, and with Celsion Corporation. Nicholas Ellens is an employee of Alpheus Medical, Inc. Keith A. Wear has a research collaboration with Siemens Healthineers. Dennis L. Parker and Allison Payne have research collaborations with Siemens Medical Solutions and Image‐Guided Therapy. Ted Lynch is an employee of Sun Nuclear, Inc.