- Email: [email protected]
Analysis and reduction of thermal dose errors in MRgFUS treatment
Physica Medica 30 (2014) 111e116
Contents lists available at SciVerse ScienceDirect
Physica Medica journal homepage: http://www.physicamedica.com
Original paper
Analysis and reduction of thermal dose errors in MRgFUS treatment Fabio Zucconi a, *, Paola E. Colombo a, Stefano Pasetto a, Alessandro Lascialfari b, Cristiana Ticca c, Alberto Torresin a a
Department of Medical Physics, Ospedale Niguarda Ca’ Granda, Piazza Ospedale Maggiore 3, 20162 Milano, Italy Department of Physics, Università degli studi di Milano, via Celoria 16, 20133 Milano, Italy c Department of Radiology, Ospedale Niguarda Ca’ Granda, Piazza Ospedale Maggiore 3, 20162 Milano, Italy b
a r t i c l e i n f o
a b s t r a c t
Article history: Received 22 April 2012 Received in revised form 27 February 2013 Accepted 25 April 2013 Available online 5 June 2013
ExAblate 2000 MRgFUS system (InSightec) installed in Ospedale Maggiore Niguarda Ca’ Granda (Milano, Italy) is currently used to treat uterine fibroids. Through the magnetic resonance thermometry (PRF method), it is possible to monitor the temperature in the target in real-time and compute the treated region calculating the thermal dose. The purpose of this work is to investigate the errors in the temperature measurements and their effect on thermal dose. A low pass filtering of temperature maps is proposed to reduce the errors and therefore to improve the reliability of the treated regions calculated. The PRF method was studied through a calibration experiment on ex vivo pig muscle. The outcome resulted to be a very good linearity (p value 0.03) between phase and temperature in the range of interest, and an a value of 0.0109 0.0002 ppm/ C. Temperature statistical uncertainty was evaluated by analyzing the temperature readout variability in specific gel provided by InSightec for daily quality assurance control. It resulted to be 1.89 0.32 C. A Monte Carlo simulation of the MRI temperature measurement and thermal dose calculations in our specific conditions of geometry and statistical uncertainty revealed that a low-pass filtering process on each temperature map can strongly reduce systematic errors in thermal dose evaluations (1.11 overestimation factor instead of 2.62 without filter); consequently the systematic errors on the size of the predicted ablated area are reduced as well. Ó 2013 Associazione Italiana di Fisica Medica. Published by Elsevier Ltd. All rights reserved.
Keywords: HIFU Thermal dose MRgFUS
Introduction MRgFUS (Magnetic Resonance guided Focused Ultrasound Surgery) is a technique that combines the power of a focused ultrasound beam to locally heat biological tissue over a necrotic level and an MRI performed at the same time. Although not necessarily required, MR imaging equipment provides several advantages, such as the guidance of the ultrasound beam in order to bypass critical anatomical structures and makes it possible to evaluate the necrotic volume after the treatment (contrast-enhanced imaging). However, the main potential of a MRgFUS system is the real-time monitoring of temperature. The quantification of heating generated by the ultrasound wave and the determination of its exact location allows the control of clinical effects, even before exceeding the necrotic threshold. After early applications of MRgFUS performed on breast tumors [1,2], the technique was improved by the introduction of phased
* Corresponding author. Tel.: þ39 3388360321. E-mail address: [email protected] (F. Zucconi).
array transducers [3]. MRgFUS was then applied to the treatment of uterine leiomyoma (fibroids), which is currently the major application for MRgFUS treatments (Insightec MRgFUS system ExAblate 2000 received FDA approval for this application) [4e7]. Other applications of MRgFUS, especially in the treatment of liver, prostate and cervical carcinoma [8e16], brain lesions [17e20], bone metastasis [21,22] and the most recent which is the targeted drug delivery methodology [23e25] are under investigation and seem very promising. It has been widely demonstrated that different methods can successfully be used to measure temperature using an MRI [26e 28], but the most relevant for MRgFUS treatments is the Proton Resonance Frequency (PRF) method, based on a dependence of the Larmor frequency on temperature. Several calibrations in vivo of the PRF method in different tissues have been performed [29] resulting in a good linearity between the Larmor frequency and the temperature in the range of interest for ultrasound applications, with a proportionality (Table 1) close to that of pure water found by Hindman [30]. Peters et al. [31] investigated the relationship in ex vivo tissues and reported a substantial independence from the tissue type.
1120-1797/$ e see front matter Ó 2013 Associazione Italiana di Fisica Medica. Published by Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.ejmp.2013.04.003
112
F. Zucconi et al. / Physica Medica 30 (2014) 111e116
Table 1 a Values obtained during PRF thermometry calibrations [4]. Tissue
Temperature range ( C)
a (ppm/ C)
Rabbit muscle Rabbit muscle Rabbit muscle Pig muscle Rabbit muscle
0e20 0e50 0e40 22e40 5e35
0.007 0.00876 0.00909 0.009 0.0104
0.001 0.00069 0.00068 0.001 0.0002
length of the hydrogen bonds between water molecules [30,36], resulting in a change of the chemical shift of the MRI signal. This model can explain the linear behavior observed experimentally between the water proton chemical shift and temperature [37]. The PRF method can be applied in MRI by measuring the phase accumulated by magnetization during the echo time of a gradient echo sequence and comparing it to the phase accumulated in a situation in which temperature is known a-priori [38].
D4 ¼ a$gB0 TE$DT Thermal heating measured while delivering US sonication allows the evaluation of the thermal dose (TD) introduced by Sapareto [32]. This empirical parameter is an indicator of the effects of biological tissue heating. Thermal dose is expressed as equivalent minutes (EM) at 43 C; for a given time-dependent temperature profile, this parameter quantifies how many minutes are necessary at a constant temperature of 43 C to reach the same biological effect. When analyzing hyperthermia treatment data, Sapareto found that a thermal dose threshold of 240 EM is required to obtain a total coagulative necrosis in biological tissue. In recent years, when comparing MRI thermal dose maps of MRgFUS treatments to post-treatment contrast-enhanced MRI images, several studies suggest that a lower thermal dose threshold can be assumed to be a threshold for total necrosis [33,34]. Evaluation of the thermal dose threshold to predict necrosis can be highly dependent on the method used to calculate the thermal dose, since it is well known [34,35] that the statistical uncertainty of temperature measurements generates a systematic error in thermal dose values, because of the strong non-linearity in the relation between temperature and thermal dose. This error will be called “thermal dose bias”. In this work, a quantitatively evaluation of the thermal dose bias was performed in simulated sonications of a uterine fibroid treatment. A method to reduce the bias is proposed and tested in the same simulated environment. Materials and methods MRgFUS system The system used in this work is ExAblate 2000, developed by Insightec (Israel). This system consists of a circular (12 cm in diameter) phased array transducer with 208 active independent piezoelectric elements capable of generating an ultrasound beam with a frequency ranging from 0.95 to 1.35 MHz (in the uterine fibroid mode). The transducer of ExAblate 2000 is placed inside a modified MRI table and the entire system is fully integrated into an MRI scanner (SIGNA HD EXCITE 1.5 T, General Electric Medical Systems). During a uterine fibroid treatment the patient is positioned prone inside the MRI scanner with the lesion facing the transducer. The space between the patient’s skin and the transducer is filled with water and a specific gel pad in order to assure a good acoustic coupling and minimize the loss in energy. Each US sonication has a duration of between 20 s and 30 s and causes a cylindrical-shaped thermal lesion in the focal point having a typical length between 2 and 4 cm, approximately 0.5 cm in diameter and a depth of 12e14 cm (the focal length of the beam can be electronically set from 6 to 20 cm).
(1)
where D4 is the difference in the accumulated phase in the voxels between the image acquired at an unknown temperature and the baseline image at a known temperature, DT is the difference in temperature in the sample between the two acquisitions, g is the gyromagnetic ratio of the proton, B0 is the static magnetic field of the MRI scanner and TE is the echo time of the gradient echo sequence. Parameter a is the proportionality constant obtained through a calibration process. To experimentally find the value of a in Equation (1), a pig dorsal muscle sample (nearly a cube of 8 cm per side) was used. This sample was heated in a water bath and then placed inside the MRI scanner during the cooling process. Inside the scanner, the temperature of the sample was monitored by two fiber optic temperature sensors (OTG-A fibers connected to TempSens signal conditioner, developed by Opsens Inc.) with an accuracy of 0.8 C and a resolution of 0.1 C. During the cooling down of the sample 9 scans were performed with a Fast Spoiled Gradient Echo sequence (FSGR) (Echo Time: 12.3 ms, Repetition Time: 25.1 ms, Flip Angle: 30 , Bandwidth: 44 kHz, matrix size: 256 128, FOV: 28 cm 28 cm). For each scan, real and imaginary components of the signal were used to build the phase image. An 8 8 pixels ROI (Fig. 1) was chosen in a position between the temperature sensor terminations. Fatty regions, where the PRF method might face issues, were carefully avoided by visual analysis. The mean and standard deviation of phase values in the ROI were considered and plotted in a phase vs. temperature graph to fit the data and obtain the calibration parameter. Thermal dose evaluation During a uterine fibroid treatment several sonications are performed with different beam orientations and transducer positions to cover the entire target volume, the aim is to create a coagulative
PRF thermometry The MRI thermometry method used in this work is the Proton Resonance Frequency shift method, called PRF method. This method is based on the dependence of the water proton chemical shift on temperature; the physical reason of this dependence is related to modifications, driven by temperature, occurring in the average
Figure 1. MRI scan of the muscle sample. Calibration 8 8 pixels ROI is highlighted.
F. Zucconi et al. / Physica Medica 30 (2014) 111e116
113
necrosis region (as large as possible) inside the fibroid. In the months following the treatment this region will gradually be absorbed and the fibroid volume will decrease, thus reducing the patient’s symptoms. During each sonication, 9 to 12 thermal MRI acquisitions are performed in order to build a temperature vs. time profile of the tissue exposed to ultrasound using the same FSGR protocol as in the calibration experiment. MRI scans are usually sagittally oriented and the position of the plane scanned is in the middle of the focal spot of the ultrasound beam. On the basis of these data thermal dose (TD) is calculated e pixel by pixel e from the Sapareto’s original definition:
Z TD ¼
R43TðtÞ dt
(2)
where R is 0.5 for temperature values above 43 C or 0.25 for temperatures below 43 C, t is the time measured in minutes, T(t) is the temperature at time t. Thermal dose TD is expressed in EM (43 C equivalent minutes). A linear interpolation of the temperature versus time profile has been made for the calculation. Equation (2) is used to transform the temperature maps of a sonication, pixel by pixel, into a single thermal dose map. If x is a generic value of thermal dose, in a thermal dose map we defined a quantity called sonicated spot (x) as the region overcoming the threshold x of thermal dose (generally 240 EM is assumed as the threshold to reach coagulative necrosis). Evaluation of temperature statistical uncertainty Because of the exponential relation between temperature and thermal dose, a statistical uncertainty in MRI temperature measurement will cause the presence of the thermal dose bias [34,35]. A homogeneous gel was scanned and temperature in sagittal orientation was used to measure the temperature statistical uncertainty of the system. The gel was provided by Insightec for Daily Quality Assurance (DQA) and mimics biological tissue in the acoustic impedance and a value (equation (1)). The statistical uncertainty of temperature measure requires to consider a ROI in each temperature map, far enough from the heated region so that a homogeneous temperature can be assumed, and to calculate the standard deviation of the pixels inside (Fig. 2). This value will be considered the measure of temperature uncertainty of our system [39]. Simulation To study the bias in the thermal dose maps caused by the temperature uncertainty, two Monte Carlo programs were developed using MatLab software in order to simulate MRI temperature images and calculate thermal dose maps under ideal conditions known a priori. A Monte Carlo simulation is a statistical nonparametric method useful to determine numerical outcomes related to a complex physical process that can be decomposed in several elementary variables, each with a known probability distribution. It consists in iterative random sampling from the distribution of all the variables to build the distribution of outcomes of the physical process. In this specific case, the aim was to evaluate the difference in thermal dose values caused by the bias (first Monte Carlo simulation) and to measure the discrepancy in the sonicated spot (240 EM) size (second Monte Carlo simulation). The elementary variables were temperature values collected in 256 256 simulated temperature maps and randomly sampled from a gaussian distribution. The mean of this distribution is the real temperature (known a priori as a simulation parameter) while the standard deviation represents the statistical uncertainty
Figure 2. MRI temperature map (sagittal plane) of the gel phantom during a sonication. The 11 11 pixels ROI chosen for the statistical uncertainty test is located at a certain distance from the focus of the ultrasound beam in order to assume that temperature is uniform.
affecting temperature measurement. Both Monte Carlo simulations were performed with six different values of standard deviation: 0.5 C, 1 C, 1.5 C, 2 C, 2.5 C and 3 C. Since the thermal dose bias is expected to depend strongly on the temperature statistical uncertainty, the method proposed to reduce the bias effect is a low pass filter applied to the temperature images, with the aim of reducing the noise in the temperature values. The temperature images produced by the simulations were therefore processed through a uniform 3 3 low pass convolutional filter. The kernel size 3 3 was a priori chosen as a compromise between variance reduction and conservation of spatial resolution; larger kernel was not considered because of the higher loss in spatial resolution. Outcomes of thermal dose bias and spot size discrepancy performed with filtered images were compared to the same analysis with unfiltered images. First Monte Carlo simulation, thermal dose bias In the first simulation the aim was to evaluate the difference in thermal dose values caused by the bias. In the implementation a homogeneous sample was exposed to a uniform and constant temperature of 57 C for a duration of 1 s. In this situation the thermal dose accumulated was known a priori to be 273 EM. At each iteration a temperature map was created and the thermal dose
Figure 3. Phase vs temperature plot for PRF thermometry calibration. aValues results to be 0.0109 0.0002 ppm/ C.
114
F. Zucconi et al. / Physica Medica 30 (2014) 111e116
Table 2 Results of the first Monte Carlo simulation for different standard deviations of the temperature values distribution. The thermal dose (TD) measurement is compared to the actual thermal dose accumulated (273 EM) and the discrepancy factor is calculated. Unfiltered and filtered temperature maps results are compared. Standard errors are reported. Standard deviation simulated ( C)
0.5
Unfiltered maps: TD measured (EM) Unfiltered maps: discrepancy factor (%) Filtered maps: TD measured (EM) Filtered maps: discrepancy factor (%)
289.85 6.17 274.82 0.67
1.0
0.06 0.02 0.06 0.02
347.1 27.16 280.3 2.66
1.5
0.2 0.06 0.1 0.05
map was computed. The mean thermal dose was calculated among the pixels and the result was stored. At the end of the simulation, the average of the stored results was considered, along with its standard error. As a compromise between precision of the simulation and computing time, we performed 1000 iterations, with a relative error lower then 1%. Second Monte Carlo simulation, spot size discrepancy In the second simulation the aim was to measure the discrepancy in the sonicated spot (240 EM) size caused by the thermal dose bias. In the implementation a homogeneous sample was exposed (at each iteration) to a simulated MRgFUS sonication; the model used for the temperature distribution was a Gaussian curve with 10 pixels (11 mm) deviation standard in the beam propagation direction and 2.5 pixels (2.7 mm) deviation standard crosswise. The temporal dependence was linear, with a 1.7 C slope every second; the process was simulated for a duration of 27 s. These values were chosen to reproduce a real sonication in a clinical practice. A temperature map was simulated every 3.4 s and all the maps accumulated during the sonication time were used to calculate a map of the thermal dose accumulated. The number of pixels exceeding the threshold of 240 EM was stored as the measure of the sonicated spot (240 EM) size. At the end of the simulation, the average of the stored results was considered, along with its standard error. In this case a good compromise was reached with 5000 iterations, with a relative error lower then 2%. The heat distribution and temperature standard deviations simulated in ideal conditions are close enough to the real situation to be able to determine the behavior of the thermal dose bias and its impact on the determination of treated volumes.
428.7 71.7 290.0 6.21
2.0
0.4 0.2 0.2 0.07
715 162.0 303.9 11.3
2.5
1 0.4 0.3 0.1
1225 349 322.2 18.0
3.0
3 1 0.4 0.1
2377 771 346.9 27.1
12 5 0.5 0.2
Results PRF thermometry MRI thermal acquisitions were performed within a temperature range from 61.2 C to 37.6 C (close to the baseline human temperature). Data obtained in this temperature range show a very high linearity between phase and temperature (Fig. 3). The statistical regression evaluated (equation (1)) an a value of 0.0109 0.0002 ppm/ C (CL 68%, Gaussian distribution). The regression was statistically significant with a p < 0.03 (Chi squared test). Statistical uncertainty of temperature The temperature statistical uncertainty of our system resulted to be 1.9 0.3 C. The estimate was calculated by considering the average of seven different values drawn from DQAs performed over many days. Standard error was multiplied by the quantile to obtain a 70% confidence level in a Student distribution with six degrees of freedom. First Monte Carlo simulation, thermal dose bias Table 2 and Fig. 4A show the average thermal dose value resulting from the first simulation, a 1 s exposure at 57 C. Actual thermal dose accumulated would be 273 EM. The “Standard Deviation simulated” is the standard deviation of the Gaussian distribution for the temperature values sampled by the Monte Carlo. Since the temperature statistical uncertainty resulted to be 1.9 C from the DQA analysis, the standard deviation value of 2.0 C is the
Figure 4. A) Plot of the discrepancy factor vs standard deviation simulated calculated by the Monte Carlo simulation of the thermal dose accumulated in a sample. In black the calculations obtained by raw temperature maps, in white the calculations obtained with filtered temperature maps (see also Table 2). B) Plot of percentage discrepancy of the spot size vs standard deviation simulated in a treatment-like condition (see also Table 3).
F. Zucconi et al. / Physica Medica 30 (2014) 111e116
115
Table 3 Results of the second Monte Carlo simulation for different standard deviations of the temperature values distribution. The sonicated spot (240 EM) area in a treatment-like condition is compared to the actual value (138.4 px2). The unfiltered and filtered temperature map results are compared. Standard errors are reported. Standard deviation simulated ( C)
0.5
Unfiltered maps: spot area (px2) Unfiltered maps: discrepancy (%) Filtered maps: spot area (px2) Filtered maps: discrepancy (%)
138.69 0.20 137.78 0.46
1.0
0.05 0.04 0.04 0.03
139.84 1.03 137.87 0.39
1.5
0.05 0.04 0.04 0.03
one that best represents our instrumental conditions among all the value tested in the Monte Carlo simulations. “Unfiltered maps: TD measured” is the thermal dose measured by the simulations with unfiltered temperature maps. The “discrepancy factor” is the percentage discrepancy between the thermal dose measured and the effective value, 273 EM. The same analysis was applied in the case of low pass filtered temperature maps (“filtered maps”). Considering the 2.0 C column in the table, the thermal dose discrepancy factor with unfiltered temperature maps is 162% while it is 11% with the filtered maps. The discrepancy factors calculated by the Monte Carlo are in accordance with the ones calculated through a theoretical approach proposed by de Zwart [35]. Second Monte Carlo simulation, spot size discrepancy Table 3 and Fig. 4B show the results of the second Monte Carlo program. “Unfiltered maps: spot area” is the size of the simulated sonicated spot (240 EM) which comes from thermal dose calculations in the case of unfiltered temperature maps; “discrepancy” is the percentage discrepancy between this measured value and the actual a priori known value (in the implemented situation it is 138.4 square pixels). The same analysis was performed with low pass filtered temperature images (“filtered maps”). Considering the 2.0 C standard deviation simulated value (for the same reasons as above), discrepancy decreases from 4.18% without the filter to 0.19% with the filter. Discussion PRF thermometry In the thermometry calibration experiment the method consisted in heating the sample outside the MRI scanner and performing MRI thermal analysis in the cooling down process. This approach ensured that temperature inside the sample was almost uniform, and allowed phase measurements in a relatively large ROI, with low statistical uncertainty. It is probably for this reason that the a value resulting from the fit of the plot demonstrates a lower uncertainty compared to the ones obtained in other works where sharp heating sources like US or lasers [29] were used. The a value found is slightly different vs. other values in literature (Table 1). The reason might be a minor change of the a value during the heating process; in fact, as noticed by Peters [31], a few minutes preheating of the tissues led to an increase of a respect to the same sample when not preheated. Another considerable difference is the temperature range studied; in Table 2 it is shown that other calibrations were generally performed within lower temperature ranges, while our experiment studied the range of interest of MRgFUS treatment in vivo, namely from 37 C (baseline temperature) to approximately 60 C. Simulations The temperature statistical uncertainty evaluated in homogeneous gels was measured in several sessions and it resulted to be
141.56 2.27 137.92 0.36
2.0
0.05 0.04 0.05 0.03
144.20 4.18 138.15 0.19
2.5
0.05 0.04 0.05 0.04
147.42 6.50 138.39 0.02
3.0
0.05 0.04 0.05 0.04
151.45 9.45 138.80 0.28
0.05 0.04 0.05 0.04
slightly variable. This quantity is expected to be even larger and even more unstable in a measurement inside a biological in vivo tissue; one reason might be the lack of homogeneity in real biological tissue, but the major issue while measuring temperature in vivo is the movement of the body. The PRF method is highly sensitive to motion and the slightest movement from the baseline acquisition to the end of the measurement can affect the precision of this technique. As shown in Table 2, the uncertainty on temperature measured generates a considerable overestimation on thermal dose calculated (up to 770%), so it is crucial to take into account this issue and to correct it. Moreover the overestimation factor is highly dependant on the temperature uncertainty level, for this reason we decided not to apply a unique correction factor on the thermal dose values, as proposed by other authors [34,35]. The 3 3 low pass filtering of temperature maps is a method to decrease the statistical variability of temperature measurements, thus decreasing the thermal dose bias. As the Monte Carlo simulation showed, this filter strongly reduces overestimations in thermal dose calculations. On the other hand, spot area measurement in a treatment-like condition revealed a more complicated situation. For the unfiltered maps, the discrepancy always enlarges the spot area because it is caused by the observed thermal dose overestimation. Instead, in the case of low pass filtering, two opposite factors affect the area value, a residual overestimation makes the area expand (this effect increases with the uncertainty level), while the loss of spatial resolution caused by the filter makes the area value decrease (this effect only depends on the type of filter kernel structure). As shown in Fig. 4B, the two factors tend to cancel out (discrepancy turns out to be 0) when the characteristic temperature statistical uncertainty level is around 0.8 C (for the geometry implemented). In the case of higher values, the effect of the loss in spatial resolution can be neglected and the low pass filtering reduces the sonicated spot size discrepancies. For lower values the overestimation factor is very low compared to the loss of spatial resolution factor and the filtering would cause a slight underestimation of the spot size. In the range of temperature statistical uncertainty measured with our system, however, filtering always seems to be effective in decreasing systematic discrepancies in spot size. Conclusions In this paper the MRI PRF thermometry was studied and tested in ex vivo tissue to investigate the reliability of this method for temperature monitoring in MRgFUS treatments. Monte Carlo simulations were implemented to study the effect of the thermal dose bias in a clinical geometry typical of a uterine fibroid sonication. The PRF thermometry method turned out to be a reliable way to measure temperature during the MRgFUS treatment with a high linearity between the phase measured and the temperature. The low pass filtering of temperature maps was proved to be a valid method to decrease systematic errors dealing with thermal dose for the specific geometry implemented. Applications of the method with sonication geometry different from the one studied for uterine fibroid should be studied with ad hoc implementation of the proposed simulations.
116
F. Zucconi et al. / Physica Medica 30 (2014) 111e116
Acknowledgments Thanks to Valeria Clementi and Amit Sokolov for useful advice and fruitful discussions on the performed work. This study was supported in part by Promedica for instruments. References [1] Hynynen K, Freund WR, Cline HE, Chung AH, Watkins RD, Vetro JP, et al. A clinical, noninvasive, MR imaging-monitored ultrasound surgery method. Radiographics 1996;16:185e95. [2] Huber PE, Jenne JW, Rastert R, Simiantonakis I, Sinn HP, Strittmatter HJ, et al. A new noninvasive approach in breast cancer therapy using magnetic resonance imaging-guided focused ultrasound surgery. Cancer Res 2001;61: 8441e7. [3] Daum DR, Hynynen K. A 256 element ultrasonic phased array system for treatment of large volumes of deep seated tissue. IEEE Trans Ultrason Ferroelectr Freq Contr 1999;46:1254e68. [4] LeBlang SD, Hoctor K, Steinberg FL. Leiomyoma shrinkage after MRI-guided focused ultrasound treatment: report of 80 patients. Am J Roentgenol 2010;1:274e80. [5] Vàrallyay C, Balàzs G, Lénàrd ZM, Bérczi V, Belics Z, Bajzik G, et al. MR imaging FOLLOW UP after MR-guided focused ultrasound surgery for uterine leiomyomas e early and mid term results. Int Med Appl Sc 2009;1:46e51. [6] Fennessy FM, Tempany CM, McDannold NJ, So MJ, Hesley G, Gostout B, et al. Uterine leiomyomas: MR imagingeguided focused ultrasound surgeryd results of different treatment protocols. Radiology 2007;243:885e93. [7] Lénàrd ZM, McDannold NJ, Fennessy FM, Stewart EA, Jolesz FA, Hynynen K, et al. Uterine leiomyomas: MR imaginge guided focused ultrasound surgeryd imaging predictors of success. Radiology 2008;249:187e94. [8] Okada A, Murakami T, Mikami K, Onishi H, Tanigawa N, Marukawa T, et al. A case of hepatocellular carcinoma treated by MR-guided focused ultrasound ablation with respiratory gating. Magn Reson Med Sci 2006;5:167e71. [9] Grissom WA, Rieke V, Holbrook AB, Medan Y, Lustig M, Santos J, et al. Hybrid referenceless and multibaseline subtraction MR thermometry for monitoring thermal therapies in moving organs. Med Phys 2010;37:5014e26. [10] Quesson B, Merle M, Roujol S, de Senneville BD, Moonen CT, Kohler MO, et al. A method for MRI guidance of intercostal high intensity focused ultrasound ablation in the liver. Med Phys 2010;37:2533e40. [11] Quesson B, Laurent C, Maclair G, de Senneville BD, Mougenot C, Ries M, et al. Real-time volumetric MRI thermometry of focused ultrasound ablation in vivo: a feasibility study in pig liver and kidney. NMR Biomed 2011;24: 145e53. [12] Wharton IP, Rivens IH, ter Haar GR, Gilderdale DJ, Collins DJ, Hand JW, et al. Design and development of a prototype endocavitary probe for high-intensity focused ultra-sound delivery with integrated magnetic resonance imaging. J Magn Reson Imaging 2007;25:548e56. [13] Salomir R, Rata M, Cadis D, Petrusca L, Auboiroux V, Cotton F. Endocavitary thermal therapy by MRI-guided phased-array contact ultrasound: experimental and numerical studies on the multi-input single-output PID temperature controller’s convergence and stability. Med Phys 2009;36: 4726e41. [14] Diederich CJ, Stafford RJ, Nau WH, Burdette EC, Price RE, Hazle JD. Transurethral ultrasound applicators with directional heating patterns for prostate thermal therapy: in vivo evaluation using magnetic resonance thermometry. Med Phys 2004;31:405e13. [15] Hutchinson EB, Hynynen K. Intracavitary ultrasound phased arrays for prostate thermal therapies: MRI compatibility and in vivo testing. Med Phys 1998;25:2392e9.
[16] Wootton JH, Hsu I-CJ, Diederich CJ. Endocervical ultrasound applicator for integrated hyperthermia and HDR brachytherapy in the treatment of locally advanced cervical carcinoma. Med Phys 2011;38:598e611. [17] Ram Z, Cohen ZR, Harnof S, Tal S, Faibel M, Nass D, et al. Magnetic resonance imaging- guided, high-intensity focused ultrasound for brain tumor therapy. Neurosurgery 2006;59:949e55. [18] Hynynen K. MRI-guided focused ultrasound treatments. Ultrasonics 2010;50: 221e9. [19] Larrat B, Pernot M, Aubry J-F, Dervishi E, Sinkus R, Seilhean D, et al. MRguided transcranial brain HIFU in small animal models. Phys Med Biol 2010;55:365e88. [20] Pinton G, Aubry JF, Fink M, Tanter M. Effects of nonlinear ultrasound propagation on high intensity brain therapy. Med Phys 2011;38:1207e16. [21] Gianfelice D, Gupta C, Kucharczyk W. Palliative treatment of painful bone metastases with MR imaginge guided focused ultrasound. Radiology 2008;249:355e63. [22] Liberman B, Gianfelice D, Inbar Y, Beck A, Rabin T, Shabshin N, et al. Pain palliation in patients with bone metastases using MR-guided focused ultrasound surgery: a multicenter study. Ann Surg Oncol 2009;16:140e6. [23] Chopra R, Curiel L, Staruch R, Morrison L, Hynynen K. An MRI-compatible system for focused ultrasound experiments in small animal models. Med Phys 2009;36:1867e74. [24] Hynynen K, McDannold NJ, Vykhodtseva N, Jolesz FA. Noninvasive MR imaging-guided focal opening of the blood-brain barrier in rabbits. Radiology 2001;220:640e6. [25] O’Reilly MA, Huang Y, Hynynen K. The impact of standing wave effects on transcranial focused ultrasound disruption of the bloodebrain barrier in a rat model. Phys Med Biol 2010;55:5251e67. [26] Rieke V, Pauly KB. MR thermometry. J Magn Reson Imaging 2008;27:376e90. [27] Samset E. Temperature mapping of thermal ablation using MRI. Minim Invasiv Ther 2006;15:36e41. [28] Quesson B, de Zwart JA, Moonen CTW. Magnetic resonance temperature imaging for guidance of thermotherapy. J Magn Reson Imaging 2000;12:525e33. [29] McDannold NJ. Quantitative MRI-based temperature mapping based on the proton resonant frequency shift: review of validation studies. Int J Hyperthermia 2005;21:533e46. [30] Hindman JC. Proton resonance shift of water in the gas and liquid states. J Chem Phys 1966;44:4582e92. [31] Peters RD, Hinks RS, Henkelman RM. Ex vivo tissue-type independence in proton-resonance frequency shift MR thermometry. Magn Reson Med 1998;40:454e9. [32] Sapareto SA, Dewey WC. Thermal dose determination in cancer therapy. Int J Radiat Oncol Biol Phys 1984;10:787e800. [33] McDannold NJ, Tempany CM, Fennessy FM, So MJ, Hynynen K, Rybicki FJ, et al. Uterine leiomyomas: MR imaginge based thermometry and thermal dosimetry during focused ultrasound thermal ablation. Radiology 2006;240:263e72. [34] Kohler MO, Mougenot C, Quesson B, Enholm J, Le Bail B, Laurent C, et al. Volumetric HIFU ablation under 3D guidance of rapid MRI thermometry. Med Phys 2009;36:3521e35. [35] de Zwart JA. Fast magnetic resonance temperature imaging for control of localized hyperthermia in medicine. Thesis. Universiteit Utrecht; 2000. [36] Muller N, Reiter RC. Temperature dependence of chemical shifts of protons in hydrogen bonds. J Chem Phys 1965;42:3265e9. [37] Schneider WG, Bernstein HJ, Poplet JA. Proton magnetic resonance chemical shift of free (Gaseous) and associated (Liquid) hydride molecules. J Chem Phys 1958;28:601e7. [38] Ishihara Y, Calderon A, Watanabe H, Okamoto K, Suzuki Y, Kuroda K, et al. A precise and fast mapping using water proton chemical shift. Magn Reson Med 1995;34:814e23. [39] McDannold NJ, Hynynen K. Quality assurance and system stability of a clinical MRI-guided focused ultrasound system: four-year experience. Med Phys 2006;33:4307e13.
Contents lists available at SciVerse ScienceDirect
Physica Medica journal homepage: http://www.physicamedica.com
Original paper
Analysis and reduction of thermal dose errors in MRgFUS treatment Fabio Zucconi a, *, Paola E. Colombo a, Stefano Pasetto a, Alessandro Lascialfari b, Cristiana Ticca c, Alberto Torresin a a
Department of Medical Physics, Ospedale Niguarda Ca’ Granda, Piazza Ospedale Maggiore 3, 20162 Milano, Italy Department of Physics, Università degli studi di Milano, via Celoria 16, 20133 Milano, Italy c Department of Radiology, Ospedale Niguarda Ca’ Granda, Piazza Ospedale Maggiore 3, 20162 Milano, Italy b
a r t i c l e i n f o
a b s t r a c t
Article history: Received 22 April 2012 Received in revised form 27 February 2013 Accepted 25 April 2013 Available online 5 June 2013
ExAblate 2000 MRgFUS system (InSightec) installed in Ospedale Maggiore Niguarda Ca’ Granda (Milano, Italy) is currently used to treat uterine fibroids. Through the magnetic resonance thermometry (PRF method), it is possible to monitor the temperature in the target in real-time and compute the treated region calculating the thermal dose. The purpose of this work is to investigate the errors in the temperature measurements and their effect on thermal dose. A low pass filtering of temperature maps is proposed to reduce the errors and therefore to improve the reliability of the treated regions calculated. The PRF method was studied through a calibration experiment on ex vivo pig muscle. The outcome resulted to be a very good linearity (p value 0.03) between phase and temperature in the range of interest, and an a value of 0.0109 0.0002 ppm/ C. Temperature statistical uncertainty was evaluated by analyzing the temperature readout variability in specific gel provided by InSightec for daily quality assurance control. It resulted to be 1.89 0.32 C. A Monte Carlo simulation of the MRI temperature measurement and thermal dose calculations in our specific conditions of geometry and statistical uncertainty revealed that a low-pass filtering process on each temperature map can strongly reduce systematic errors in thermal dose evaluations (1.11 overestimation factor instead of 2.62 without filter); consequently the systematic errors on the size of the predicted ablated area are reduced as well. Ó 2013 Associazione Italiana di Fisica Medica. Published by Elsevier Ltd. All rights reserved.
Keywords: HIFU Thermal dose MRgFUS
Introduction MRgFUS (Magnetic Resonance guided Focused Ultrasound Surgery) is a technique that combines the power of a focused ultrasound beam to locally heat biological tissue over a necrotic level and an MRI performed at the same time. Although not necessarily required, MR imaging equipment provides several advantages, such as the guidance of the ultrasound beam in order to bypass critical anatomical structures and makes it possible to evaluate the necrotic volume after the treatment (contrast-enhanced imaging). However, the main potential of a MRgFUS system is the real-time monitoring of temperature. The quantification of heating generated by the ultrasound wave and the determination of its exact location allows the control of clinical effects, even before exceeding the necrotic threshold. After early applications of MRgFUS performed on breast tumors [1,2], the technique was improved by the introduction of phased
* Corresponding author. Tel.: þ39 3388360321. E-mail address: [email protected] (F. Zucconi).
array transducers [3]. MRgFUS was then applied to the treatment of uterine leiomyoma (fibroids), which is currently the major application for MRgFUS treatments (Insightec MRgFUS system ExAblate 2000 received FDA approval for this application) [4e7]. Other applications of MRgFUS, especially in the treatment of liver, prostate and cervical carcinoma [8e16], brain lesions [17e20], bone metastasis [21,22] and the most recent which is the targeted drug delivery methodology [23e25] are under investigation and seem very promising. It has been widely demonstrated that different methods can successfully be used to measure temperature using an MRI [26e 28], but the most relevant for MRgFUS treatments is the Proton Resonance Frequency (PRF) method, based on a dependence of the Larmor frequency on temperature. Several calibrations in vivo of the PRF method in different tissues have been performed [29] resulting in a good linearity between the Larmor frequency and the temperature in the range of interest for ultrasound applications, with a proportionality (Table 1) close to that of pure water found by Hindman [30]. Peters et al. [31] investigated the relationship in ex vivo tissues and reported a substantial independence from the tissue type.
1120-1797/$ e see front matter Ó 2013 Associazione Italiana di Fisica Medica. Published by Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.ejmp.2013.04.003
112
F. Zucconi et al. / Physica Medica 30 (2014) 111e116
Table 1 a Values obtained during PRF thermometry calibrations [4]. Tissue
Temperature range ( C)
a (ppm/ C)
Rabbit muscle Rabbit muscle Rabbit muscle Pig muscle Rabbit muscle
0e20 0e50 0e40 22e40 5e35
0.007 0.00876 0.00909 0.009 0.0104
0.001 0.00069 0.00068 0.001 0.0002
length of the hydrogen bonds between water molecules [30,36], resulting in a change of the chemical shift of the MRI signal. This model can explain the linear behavior observed experimentally between the water proton chemical shift and temperature [37]. The PRF method can be applied in MRI by measuring the phase accumulated by magnetization during the echo time of a gradient echo sequence and comparing it to the phase accumulated in a situation in which temperature is known a-priori [38].
D4 ¼ a$gB0 TE$DT Thermal heating measured while delivering US sonication allows the evaluation of the thermal dose (TD) introduced by Sapareto [32]. This empirical parameter is an indicator of the effects of biological tissue heating. Thermal dose is expressed as equivalent minutes (EM) at 43 C; for a given time-dependent temperature profile, this parameter quantifies how many minutes are necessary at a constant temperature of 43 C to reach the same biological effect. When analyzing hyperthermia treatment data, Sapareto found that a thermal dose threshold of 240 EM is required to obtain a total coagulative necrosis in biological tissue. In recent years, when comparing MRI thermal dose maps of MRgFUS treatments to post-treatment contrast-enhanced MRI images, several studies suggest that a lower thermal dose threshold can be assumed to be a threshold for total necrosis [33,34]. Evaluation of the thermal dose threshold to predict necrosis can be highly dependent on the method used to calculate the thermal dose, since it is well known [34,35] that the statistical uncertainty of temperature measurements generates a systematic error in thermal dose values, because of the strong non-linearity in the relation between temperature and thermal dose. This error will be called “thermal dose bias”. In this work, a quantitatively evaluation of the thermal dose bias was performed in simulated sonications of a uterine fibroid treatment. A method to reduce the bias is proposed and tested in the same simulated environment. Materials and methods MRgFUS system The system used in this work is ExAblate 2000, developed by Insightec (Israel). This system consists of a circular (12 cm in diameter) phased array transducer with 208 active independent piezoelectric elements capable of generating an ultrasound beam with a frequency ranging from 0.95 to 1.35 MHz (in the uterine fibroid mode). The transducer of ExAblate 2000 is placed inside a modified MRI table and the entire system is fully integrated into an MRI scanner (SIGNA HD EXCITE 1.5 T, General Electric Medical Systems). During a uterine fibroid treatment the patient is positioned prone inside the MRI scanner with the lesion facing the transducer. The space between the patient’s skin and the transducer is filled with water and a specific gel pad in order to assure a good acoustic coupling and minimize the loss in energy. Each US sonication has a duration of between 20 s and 30 s and causes a cylindrical-shaped thermal lesion in the focal point having a typical length between 2 and 4 cm, approximately 0.5 cm in diameter and a depth of 12e14 cm (the focal length of the beam can be electronically set from 6 to 20 cm).
(1)
where D4 is the difference in the accumulated phase in the voxels between the image acquired at an unknown temperature and the baseline image at a known temperature, DT is the difference in temperature in the sample between the two acquisitions, g is the gyromagnetic ratio of the proton, B0 is the static magnetic field of the MRI scanner and TE is the echo time of the gradient echo sequence. Parameter a is the proportionality constant obtained through a calibration process. To experimentally find the value of a in Equation (1), a pig dorsal muscle sample (nearly a cube of 8 cm per side) was used. This sample was heated in a water bath and then placed inside the MRI scanner during the cooling process. Inside the scanner, the temperature of the sample was monitored by two fiber optic temperature sensors (OTG-A fibers connected to TempSens signal conditioner, developed by Opsens Inc.) with an accuracy of 0.8 C and a resolution of 0.1 C. During the cooling down of the sample 9 scans were performed with a Fast Spoiled Gradient Echo sequence (FSGR) (Echo Time: 12.3 ms, Repetition Time: 25.1 ms, Flip Angle: 30 , Bandwidth: 44 kHz, matrix size: 256 128, FOV: 28 cm 28 cm). For each scan, real and imaginary components of the signal were used to build the phase image. An 8 8 pixels ROI (Fig. 1) was chosen in a position between the temperature sensor terminations. Fatty regions, where the PRF method might face issues, were carefully avoided by visual analysis. The mean and standard deviation of phase values in the ROI were considered and plotted in a phase vs. temperature graph to fit the data and obtain the calibration parameter. Thermal dose evaluation During a uterine fibroid treatment several sonications are performed with different beam orientations and transducer positions to cover the entire target volume, the aim is to create a coagulative
PRF thermometry The MRI thermometry method used in this work is the Proton Resonance Frequency shift method, called PRF method. This method is based on the dependence of the water proton chemical shift on temperature; the physical reason of this dependence is related to modifications, driven by temperature, occurring in the average
Figure 1. MRI scan of the muscle sample. Calibration 8 8 pixels ROI is highlighted.
F. Zucconi et al. / Physica Medica 30 (2014) 111e116
113
necrosis region (as large as possible) inside the fibroid. In the months following the treatment this region will gradually be absorbed and the fibroid volume will decrease, thus reducing the patient’s symptoms. During each sonication, 9 to 12 thermal MRI acquisitions are performed in order to build a temperature vs. time profile of the tissue exposed to ultrasound using the same FSGR protocol as in the calibration experiment. MRI scans are usually sagittally oriented and the position of the plane scanned is in the middle of the focal spot of the ultrasound beam. On the basis of these data thermal dose (TD) is calculated e pixel by pixel e from the Sapareto’s original definition:
Z TD ¼
R43TðtÞ dt
(2)
where R is 0.5 for temperature values above 43 C or 0.25 for temperatures below 43 C, t is the time measured in minutes, T(t) is the temperature at time t. Thermal dose TD is expressed in EM (43 C equivalent minutes). A linear interpolation of the temperature versus time profile has been made for the calculation. Equation (2) is used to transform the temperature maps of a sonication, pixel by pixel, into a single thermal dose map. If x is a generic value of thermal dose, in a thermal dose map we defined a quantity called sonicated spot (x) as the region overcoming the threshold x of thermal dose (generally 240 EM is assumed as the threshold to reach coagulative necrosis). Evaluation of temperature statistical uncertainty Because of the exponential relation between temperature and thermal dose, a statistical uncertainty in MRI temperature measurement will cause the presence of the thermal dose bias [34,35]. A homogeneous gel was scanned and temperature in sagittal orientation was used to measure the temperature statistical uncertainty of the system. The gel was provided by Insightec for Daily Quality Assurance (DQA) and mimics biological tissue in the acoustic impedance and a value (equation (1)). The statistical uncertainty of temperature measure requires to consider a ROI in each temperature map, far enough from the heated region so that a homogeneous temperature can be assumed, and to calculate the standard deviation of the pixels inside (Fig. 2). This value will be considered the measure of temperature uncertainty of our system [39]. Simulation To study the bias in the thermal dose maps caused by the temperature uncertainty, two Monte Carlo programs were developed using MatLab software in order to simulate MRI temperature images and calculate thermal dose maps under ideal conditions known a priori. A Monte Carlo simulation is a statistical nonparametric method useful to determine numerical outcomes related to a complex physical process that can be decomposed in several elementary variables, each with a known probability distribution. It consists in iterative random sampling from the distribution of all the variables to build the distribution of outcomes of the physical process. In this specific case, the aim was to evaluate the difference in thermal dose values caused by the bias (first Monte Carlo simulation) and to measure the discrepancy in the sonicated spot (240 EM) size (second Monte Carlo simulation). The elementary variables were temperature values collected in 256 256 simulated temperature maps and randomly sampled from a gaussian distribution. The mean of this distribution is the real temperature (known a priori as a simulation parameter) while the standard deviation represents the statistical uncertainty
Figure 2. MRI temperature map (sagittal plane) of the gel phantom during a sonication. The 11 11 pixels ROI chosen for the statistical uncertainty test is located at a certain distance from the focus of the ultrasound beam in order to assume that temperature is uniform.
affecting temperature measurement. Both Monte Carlo simulations were performed with six different values of standard deviation: 0.5 C, 1 C, 1.5 C, 2 C, 2.5 C and 3 C. Since the thermal dose bias is expected to depend strongly on the temperature statistical uncertainty, the method proposed to reduce the bias effect is a low pass filter applied to the temperature images, with the aim of reducing the noise in the temperature values. The temperature images produced by the simulations were therefore processed through a uniform 3 3 low pass convolutional filter. The kernel size 3 3 was a priori chosen as a compromise between variance reduction and conservation of spatial resolution; larger kernel was not considered because of the higher loss in spatial resolution. Outcomes of thermal dose bias and spot size discrepancy performed with filtered images were compared to the same analysis with unfiltered images. First Monte Carlo simulation, thermal dose bias In the first simulation the aim was to evaluate the difference in thermal dose values caused by the bias. In the implementation a homogeneous sample was exposed to a uniform and constant temperature of 57 C for a duration of 1 s. In this situation the thermal dose accumulated was known a priori to be 273 EM. At each iteration a temperature map was created and the thermal dose
Figure 3. Phase vs temperature plot for PRF thermometry calibration. aValues results to be 0.0109 0.0002 ppm/ C.
114
F. Zucconi et al. / Physica Medica 30 (2014) 111e116
Table 2 Results of the first Monte Carlo simulation for different standard deviations of the temperature values distribution. The thermal dose (TD) measurement is compared to the actual thermal dose accumulated (273 EM) and the discrepancy factor is calculated. Unfiltered and filtered temperature maps results are compared. Standard errors are reported. Standard deviation simulated ( C)
0.5
Unfiltered maps: TD measured (EM) Unfiltered maps: discrepancy factor (%) Filtered maps: TD measured (EM) Filtered maps: discrepancy factor (%)
289.85 6.17 274.82 0.67
1.0
0.06 0.02 0.06 0.02
347.1 27.16 280.3 2.66
1.5
0.2 0.06 0.1 0.05
map was computed. The mean thermal dose was calculated among the pixels and the result was stored. At the end of the simulation, the average of the stored results was considered, along with its standard error. As a compromise between precision of the simulation and computing time, we performed 1000 iterations, with a relative error lower then 1%. Second Monte Carlo simulation, spot size discrepancy In the second simulation the aim was to measure the discrepancy in the sonicated spot (240 EM) size caused by the thermal dose bias. In the implementation a homogeneous sample was exposed (at each iteration) to a simulated MRgFUS sonication; the model used for the temperature distribution was a Gaussian curve with 10 pixels (11 mm) deviation standard in the beam propagation direction and 2.5 pixels (2.7 mm) deviation standard crosswise. The temporal dependence was linear, with a 1.7 C slope every second; the process was simulated for a duration of 27 s. These values were chosen to reproduce a real sonication in a clinical practice. A temperature map was simulated every 3.4 s and all the maps accumulated during the sonication time were used to calculate a map of the thermal dose accumulated. The number of pixels exceeding the threshold of 240 EM was stored as the measure of the sonicated spot (240 EM) size. At the end of the simulation, the average of the stored results was considered, along with its standard error. In this case a good compromise was reached with 5000 iterations, with a relative error lower then 2%. The heat distribution and temperature standard deviations simulated in ideal conditions are close enough to the real situation to be able to determine the behavior of the thermal dose bias and its impact on the determination of treated volumes.
428.7 71.7 290.0 6.21
2.0
0.4 0.2 0.2 0.07
715 162.0 303.9 11.3
2.5
1 0.4 0.3 0.1
1225 349 322.2 18.0
3.0
3 1 0.4 0.1
2377 771 346.9 27.1
12 5 0.5 0.2
Results PRF thermometry MRI thermal acquisitions were performed within a temperature range from 61.2 C to 37.6 C (close to the baseline human temperature). Data obtained in this temperature range show a very high linearity between phase and temperature (Fig. 3). The statistical regression evaluated (equation (1)) an a value of 0.0109 0.0002 ppm/ C (CL 68%, Gaussian distribution). The regression was statistically significant with a p < 0.03 (Chi squared test). Statistical uncertainty of temperature The temperature statistical uncertainty of our system resulted to be 1.9 0.3 C. The estimate was calculated by considering the average of seven different values drawn from DQAs performed over many days. Standard error was multiplied by the quantile to obtain a 70% confidence level in a Student distribution with six degrees of freedom. First Monte Carlo simulation, thermal dose bias Table 2 and Fig. 4A show the average thermal dose value resulting from the first simulation, a 1 s exposure at 57 C. Actual thermal dose accumulated would be 273 EM. The “Standard Deviation simulated” is the standard deviation of the Gaussian distribution for the temperature values sampled by the Monte Carlo. Since the temperature statistical uncertainty resulted to be 1.9 C from the DQA analysis, the standard deviation value of 2.0 C is the
Figure 4. A) Plot of the discrepancy factor vs standard deviation simulated calculated by the Monte Carlo simulation of the thermal dose accumulated in a sample. In black the calculations obtained by raw temperature maps, in white the calculations obtained with filtered temperature maps (see also Table 2). B) Plot of percentage discrepancy of the spot size vs standard deviation simulated in a treatment-like condition (see also Table 3).
F. Zucconi et al. / Physica Medica 30 (2014) 111e116
115
Table 3 Results of the second Monte Carlo simulation for different standard deviations of the temperature values distribution. The sonicated spot (240 EM) area in a treatment-like condition is compared to the actual value (138.4 px2). The unfiltered and filtered temperature map results are compared. Standard errors are reported. Standard deviation simulated ( C)
0.5
Unfiltered maps: spot area (px2) Unfiltered maps: discrepancy (%) Filtered maps: spot area (px2) Filtered maps: discrepancy (%)
138.69 0.20 137.78 0.46
1.0
0.05 0.04 0.04 0.03
139.84 1.03 137.87 0.39
1.5
0.05 0.04 0.04 0.03
one that best represents our instrumental conditions among all the value tested in the Monte Carlo simulations. “Unfiltered maps: TD measured” is the thermal dose measured by the simulations with unfiltered temperature maps. The “discrepancy factor” is the percentage discrepancy between the thermal dose measured and the effective value, 273 EM. The same analysis was applied in the case of low pass filtered temperature maps (“filtered maps”). Considering the 2.0 C column in the table, the thermal dose discrepancy factor with unfiltered temperature maps is 162% while it is 11% with the filtered maps. The discrepancy factors calculated by the Monte Carlo are in accordance with the ones calculated through a theoretical approach proposed by de Zwart [35]. Second Monte Carlo simulation, spot size discrepancy Table 3 and Fig. 4B show the results of the second Monte Carlo program. “Unfiltered maps: spot area” is the size of the simulated sonicated spot (240 EM) which comes from thermal dose calculations in the case of unfiltered temperature maps; “discrepancy” is the percentage discrepancy between this measured value and the actual a priori known value (in the implemented situation it is 138.4 square pixels). The same analysis was performed with low pass filtered temperature images (“filtered maps”). Considering the 2.0 C standard deviation simulated value (for the same reasons as above), discrepancy decreases from 4.18% without the filter to 0.19% with the filter. Discussion PRF thermometry In the thermometry calibration experiment the method consisted in heating the sample outside the MRI scanner and performing MRI thermal analysis in the cooling down process. This approach ensured that temperature inside the sample was almost uniform, and allowed phase measurements in a relatively large ROI, with low statistical uncertainty. It is probably for this reason that the a value resulting from the fit of the plot demonstrates a lower uncertainty compared to the ones obtained in other works where sharp heating sources like US or lasers [29] were used. The a value found is slightly different vs. other values in literature (Table 1). The reason might be a minor change of the a value during the heating process; in fact, as noticed by Peters [31], a few minutes preheating of the tissues led to an increase of a respect to the same sample when not preheated. Another considerable difference is the temperature range studied; in Table 2 it is shown that other calibrations were generally performed within lower temperature ranges, while our experiment studied the range of interest of MRgFUS treatment in vivo, namely from 37 C (baseline temperature) to approximately 60 C. Simulations The temperature statistical uncertainty evaluated in homogeneous gels was measured in several sessions and it resulted to be
141.56 2.27 137.92 0.36
2.0
0.05 0.04 0.05 0.03
144.20 4.18 138.15 0.19
2.5
0.05 0.04 0.05 0.04
147.42 6.50 138.39 0.02
3.0
0.05 0.04 0.05 0.04
151.45 9.45 138.80 0.28
0.05 0.04 0.05 0.04
slightly variable. This quantity is expected to be even larger and even more unstable in a measurement inside a biological in vivo tissue; one reason might be the lack of homogeneity in real biological tissue, but the major issue while measuring temperature in vivo is the movement of the body. The PRF method is highly sensitive to motion and the slightest movement from the baseline acquisition to the end of the measurement can affect the precision of this technique. As shown in Table 2, the uncertainty on temperature measured generates a considerable overestimation on thermal dose calculated (up to 770%), so it is crucial to take into account this issue and to correct it. Moreover the overestimation factor is highly dependant on the temperature uncertainty level, for this reason we decided not to apply a unique correction factor on the thermal dose values, as proposed by other authors [34,35]. The 3 3 low pass filtering of temperature maps is a method to decrease the statistical variability of temperature measurements, thus decreasing the thermal dose bias. As the Monte Carlo simulation showed, this filter strongly reduces overestimations in thermal dose calculations. On the other hand, spot area measurement in a treatment-like condition revealed a more complicated situation. For the unfiltered maps, the discrepancy always enlarges the spot area because it is caused by the observed thermal dose overestimation. Instead, in the case of low pass filtering, two opposite factors affect the area value, a residual overestimation makes the area expand (this effect increases with the uncertainty level), while the loss of spatial resolution caused by the filter makes the area value decrease (this effect only depends on the type of filter kernel structure). As shown in Fig. 4B, the two factors tend to cancel out (discrepancy turns out to be 0) when the characteristic temperature statistical uncertainty level is around 0.8 C (for the geometry implemented). In the case of higher values, the effect of the loss in spatial resolution can be neglected and the low pass filtering reduces the sonicated spot size discrepancies. For lower values the overestimation factor is very low compared to the loss of spatial resolution factor and the filtering would cause a slight underestimation of the spot size. In the range of temperature statistical uncertainty measured with our system, however, filtering always seems to be effective in decreasing systematic discrepancies in spot size. Conclusions In this paper the MRI PRF thermometry was studied and tested in ex vivo tissue to investigate the reliability of this method for temperature monitoring in MRgFUS treatments. Monte Carlo simulations were implemented to study the effect of the thermal dose bias in a clinical geometry typical of a uterine fibroid sonication. The PRF thermometry method turned out to be a reliable way to measure temperature during the MRgFUS treatment with a high linearity between the phase measured and the temperature. The low pass filtering of temperature maps was proved to be a valid method to decrease systematic errors dealing with thermal dose for the specific geometry implemented. Applications of the method with sonication geometry different from the one studied for uterine fibroid should be studied with ad hoc implementation of the proposed simulations.
116
F. Zucconi et al. / Physica Medica 30 (2014) 111e116
Acknowledgments Thanks to Valeria Clementi and Amit Sokolov for useful advice and fruitful discussions on the performed work. This study was supported in part by Promedica for instruments. References [1] Hynynen K, Freund WR, Cline HE, Chung AH, Watkins RD, Vetro JP, et al. A clinical, noninvasive, MR imaging-monitored ultrasound surgery method. Radiographics 1996;16:185e95. [2] Huber PE, Jenne JW, Rastert R, Simiantonakis I, Sinn HP, Strittmatter HJ, et al. A new noninvasive approach in breast cancer therapy using magnetic resonance imaging-guided focused ultrasound surgery. Cancer Res 2001;61: 8441e7. [3] Daum DR, Hynynen K. A 256 element ultrasonic phased array system for treatment of large volumes of deep seated tissue. IEEE Trans Ultrason Ferroelectr Freq Contr 1999;46:1254e68. [4] LeBlang SD, Hoctor K, Steinberg FL. Leiomyoma shrinkage after MRI-guided focused ultrasound treatment: report of 80 patients. Am J Roentgenol 2010;1:274e80. [5] Vàrallyay C, Balàzs G, Lénàrd ZM, Bérczi V, Belics Z, Bajzik G, et al. MR imaging FOLLOW UP after MR-guided focused ultrasound surgery for uterine leiomyomas e early and mid term results. Int Med Appl Sc 2009;1:46e51. [6] Fennessy FM, Tempany CM, McDannold NJ, So MJ, Hesley G, Gostout B, et al. Uterine leiomyomas: MR imagingeguided focused ultrasound surgeryd results of different treatment protocols. Radiology 2007;243:885e93. [7] Lénàrd ZM, McDannold NJ, Fennessy FM, Stewart EA, Jolesz FA, Hynynen K, et al. Uterine leiomyomas: MR imaginge guided focused ultrasound surgeryd imaging predictors of success. Radiology 2008;249:187e94. [8] Okada A, Murakami T, Mikami K, Onishi H, Tanigawa N, Marukawa T, et al. A case of hepatocellular carcinoma treated by MR-guided focused ultrasound ablation with respiratory gating. Magn Reson Med Sci 2006;5:167e71. [9] Grissom WA, Rieke V, Holbrook AB, Medan Y, Lustig M, Santos J, et al. Hybrid referenceless and multibaseline subtraction MR thermometry for monitoring thermal therapies in moving organs. Med Phys 2010;37:5014e26. [10] Quesson B, Merle M, Roujol S, de Senneville BD, Moonen CT, Kohler MO, et al. A method for MRI guidance of intercostal high intensity focused ultrasound ablation in the liver. Med Phys 2010;37:2533e40. [11] Quesson B, Laurent C, Maclair G, de Senneville BD, Mougenot C, Ries M, et al. Real-time volumetric MRI thermometry of focused ultrasound ablation in vivo: a feasibility study in pig liver and kidney. NMR Biomed 2011;24: 145e53. [12] Wharton IP, Rivens IH, ter Haar GR, Gilderdale DJ, Collins DJ, Hand JW, et al. Design and development of a prototype endocavitary probe for high-intensity focused ultra-sound delivery with integrated magnetic resonance imaging. J Magn Reson Imaging 2007;25:548e56. [13] Salomir R, Rata M, Cadis D, Petrusca L, Auboiroux V, Cotton F. Endocavitary thermal therapy by MRI-guided phased-array contact ultrasound: experimental and numerical studies on the multi-input single-output PID temperature controller’s convergence and stability. Med Phys 2009;36: 4726e41. [14] Diederich CJ, Stafford RJ, Nau WH, Burdette EC, Price RE, Hazle JD. Transurethral ultrasound applicators with directional heating patterns for prostate thermal therapy: in vivo evaluation using magnetic resonance thermometry. Med Phys 2004;31:405e13. [15] Hutchinson EB, Hynynen K. Intracavitary ultrasound phased arrays for prostate thermal therapies: MRI compatibility and in vivo testing. Med Phys 1998;25:2392e9.
[16] Wootton JH, Hsu I-CJ, Diederich CJ. Endocervical ultrasound applicator for integrated hyperthermia and HDR brachytherapy in the treatment of locally advanced cervical carcinoma. Med Phys 2011;38:598e611. [17] Ram Z, Cohen ZR, Harnof S, Tal S, Faibel M, Nass D, et al. Magnetic resonance imaging- guided, high-intensity focused ultrasound for brain tumor therapy. Neurosurgery 2006;59:949e55. [18] Hynynen K. MRI-guided focused ultrasound treatments. Ultrasonics 2010;50: 221e9. [19] Larrat B, Pernot M, Aubry J-F, Dervishi E, Sinkus R, Seilhean D, et al. MRguided transcranial brain HIFU in small animal models. Phys Med Biol 2010;55:365e88. [20] Pinton G, Aubry JF, Fink M, Tanter M. Effects of nonlinear ultrasound propagation on high intensity brain therapy. Med Phys 2011;38:1207e16. [21] Gianfelice D, Gupta C, Kucharczyk W. Palliative treatment of painful bone metastases with MR imaginge guided focused ultrasound. Radiology 2008;249:355e63. [22] Liberman B, Gianfelice D, Inbar Y, Beck A, Rabin T, Shabshin N, et al. Pain palliation in patients with bone metastases using MR-guided focused ultrasound surgery: a multicenter study. Ann Surg Oncol 2009;16:140e6. [23] Chopra R, Curiel L, Staruch R, Morrison L, Hynynen K. An MRI-compatible system for focused ultrasound experiments in small animal models. Med Phys 2009;36:1867e74. [24] Hynynen K, McDannold NJ, Vykhodtseva N, Jolesz FA. Noninvasive MR imaging-guided focal opening of the blood-brain barrier in rabbits. Radiology 2001;220:640e6. [25] O’Reilly MA, Huang Y, Hynynen K. The impact of standing wave effects on transcranial focused ultrasound disruption of the bloodebrain barrier in a rat model. Phys Med Biol 2010;55:5251e67. [26] Rieke V, Pauly KB. MR thermometry. J Magn Reson Imaging 2008;27:376e90. [27] Samset E. Temperature mapping of thermal ablation using MRI. Minim Invasiv Ther 2006;15:36e41. [28] Quesson B, de Zwart JA, Moonen CTW. Magnetic resonance temperature imaging for guidance of thermotherapy. J Magn Reson Imaging 2000;12:525e33. [29] McDannold NJ. Quantitative MRI-based temperature mapping based on the proton resonant frequency shift: review of validation studies. Int J Hyperthermia 2005;21:533e46. [30] Hindman JC. Proton resonance shift of water in the gas and liquid states. J Chem Phys 1966;44:4582e92. [31] Peters RD, Hinks RS, Henkelman RM. Ex vivo tissue-type independence in proton-resonance frequency shift MR thermometry. Magn Reson Med 1998;40:454e9. [32] Sapareto SA, Dewey WC. Thermal dose determination in cancer therapy. Int J Radiat Oncol Biol Phys 1984;10:787e800. [33] McDannold NJ, Tempany CM, Fennessy FM, So MJ, Hynynen K, Rybicki FJ, et al. Uterine leiomyomas: MR imaginge based thermometry and thermal dosimetry during focused ultrasound thermal ablation. Radiology 2006;240:263e72. [34] Kohler MO, Mougenot C, Quesson B, Enholm J, Le Bail B, Laurent C, et al. Volumetric HIFU ablation under 3D guidance of rapid MRI thermometry. Med Phys 2009;36:3521e35. [35] de Zwart JA. Fast magnetic resonance temperature imaging for control of localized hyperthermia in medicine. Thesis. Universiteit Utrecht; 2000. [36] Muller N, Reiter RC. Temperature dependence of chemical shifts of protons in hydrogen bonds. J Chem Phys 1965;42:3265e9. [37] Schneider WG, Bernstein HJ, Poplet JA. Proton magnetic resonance chemical shift of free (Gaseous) and associated (Liquid) hydride molecules. J Chem Phys 1958;28:601e7. [38] Ishihara Y, Calderon A, Watanabe H, Okamoto K, Suzuki Y, Kuroda K, et al. A precise and fast mapping using water proton chemical shift. Magn Reson Med 1995;34:814e23. [39] McDannold NJ, Hynynen K. Quality assurance and system stability of a clinical MRI-guided focused ultrasound system: four-year experience. Med Phys 2006;33:4307e13.