|Home | About | Journals | Submit | Contact Us | Français|
It was previously demonstrated that it is feasible to simultaneously perform ultrasound therapy and imaging of a coagulated lesion during treatment with an integrated transducer that is capable of high intensity focused ultrasound (HIFU) and B-mode ultrasound imaging. It was found that coded excitation and fixed notch filtering upon reception could significantly reduce interference caused by the therapeutic transducer. During HIFU sonication, the imaging signal generated with coded excitation and fixed notch filtering had a range side-lobe level of less than −40 dB, while traditional short-pulse excitation and fixed notch filtering produced a range side-lobe level of −20 dB. The shortcoming is, however, that relatively complicated electronics may be needed to utilize coded excitation in an array imaging system. It is for this reason that in this paper an adaptive noise canceling technique is proposed to improve image quality by minimizing not only the therapeutic interference, but also the remnant side-lobe ‘ripples’ when using the traditional short-pulse excitation. The performance of this technique was verified through simulation and experiments using a prototype integrated HIFU/imaging transducer. Although it is known that the remnant ripples are related to the notch attenuation value of the fixed notch filter, in reality, it is difficult to find the optimal notch attenuation value due to the change in targets or the media resulted from motion or different acoustic properties even during one sonication pulse. In contrast, the proposed adaptive noise canceling technique is capable of optimally minimizing both the therapeutic interference and residual ripples without such constraints. The prototype integrated HIFU/imaging transducer is composed of three rectangular elements. The 6 MHz center element is used for imaging and the outer two identical 4 MHz elements work together to transmit the HIFU beam. Two HIFU elements of 14.4 mm × 20.0 mm dimensions could increase the temperature of the soft biological tissue from 55 °C to 71 °C within 60 s. Two types of experiments for simultaneous therapy and imaging were conducted to acquire a single scan-line and B-mode image with an aluminum plate and a slice of porcine muscle, respectively. The B-mode image was obtained using the single element imaging system during HIFU beam transmission. The experimental results proved that the combination of the traditional short-pulse excitation and the adaptive noise canceling method could significantly reduce therapeutic interference and remnant ripples and thus may be a better way to implement real-time simultaneous therapy and imaging.
During noninvasive high intensity focused ultrasound (HIFU) surgery, imaging modalities such as magnetic resonance imaging (MRI) or ultrasound is often used to allow a physician to have a live view of the target being treated (Cline et al 1995, Ebbini et al 2006, Hynynen et al 1996, Vaezy et al 2001a). Among them, ultrasound guidance has been widely used for the treatment of benign and malignant prostate tumors due to its fast real-time imaging capability and low cost (Sanghvi et al 1999, Blana et al 2004, Azzouz and Rosette 2006). Typically, a HIFU surgical process is composed of pre-, mid- and post- treatments. The physician should identify the location of the target precisely and determine the proper dose in pre-treatment. In mid-treatment, monitoring the response of the treated target and tracking the motion of the target or patient are required to ensure delivering a proper ultrasound dose to the target without damaging normal tissues. An analysis of the ablated target may be conducted in post-treatment (Chinn 2005). Especially, by employing B-mode (Brightness mode) ultrasound imaging for monitoring of the thermal ablation, echogenic regions induced by cavitation of the coagulated lesion can be monitored during HIFU sonication (Chan et al 2002, Melodelima et al 2007, Owen et al 2006, Vaezy et al 2001b).
Although ultrasound guidance with an integrated HIFU/imaging transducer can provide real-time images during pre- and post-treatments, it has had limited success in realizing real-time B-mode imaging in mid-treatment due to the significant therapeutic interference received by the imaging transducer (Owen et al 2006, Vaezy et al 2001b, Wu et al 2008). One of the potential methods to reduce this interference is designing the imaging transducer at a higher center frequency than that of the therapeutic transducer (Kluiwstra et al 1997, Azzouz and Rosette 2006). Another approach is to transmit mixed-pulse signals by controlling pulse repetition frequency (PRF) of synchronized imaging and therapeutic signals (Seip et al 2002). More recently it was suggested that coded excitation with the fixed notch filtering (Jeong et al 2009) might be an alternative for live imaging during treatment. In this method, the coded signal for imaging and the continuous wave (CW) signal for therapy were transmitted to the target simultaneously, and the reflected-mixed signal was received by the imaging transducer. After pulse compression and fixed notch filtering, the final imaging signal had range side-lobe levels lower than −40 dB, while the traditional short-pulse excitation showed higher side-lobe levels due to the remnant ripples after the main-lobe.
Although coded excitation showed high robustness against residual ripples, relatively complicated hardware components are required for its implementation, i.e. transmission/reception of long pulses per channel (O’Donnell 1992). Thus, the objective of this research is to reduce both reflected therapeutic interference and residual ripples with traditional short-pulse excitation, accomplished by adaptive noise canceling and subsequently to determine the feasibility of using such a technique for simultaneous therapy and imaging. Because the proposed algorithm can be implemented with the same hardware for fixed notch filtering, no additional imaging hardware components will be needed to produce an acceptable image. A prototype integrated HIFU/imaging transducer was fabricated for demonstrating the feasibility of the proposed algorithm. All specifications for the prototype transducer such as the total dimension, center frequencies for therapy/imaging and focal depth were designed for satisfying the needs of treating benign or malignant prostate tumors. Single scan-line and B-mode imaging experiments were conducted with an aluminum plate as a strong reflector and a slice of porcine muscle, respectively. B-mode image data were collected with a mechanical imaging scanner during HIFU beam transmission.
In general, the ultrasonic diagnostic image quality during HIFU treatment is affected by the spurious therapeutic interference. Fortunately, the notch filtering technique may be used for suppressing this interference that is sinusoidal with a known center frequency. A notch filter has been widely used for suppression of sinusoidal interference in such diverse fields as sonar, speech processing and physiological signal processing (Hirano et al 1974, Ferdjallah and Barr 1994, Dragosevic and Stankovic 1995). The notch filter can remove specific frequency components without affecting other frequency components, and it can be implemented by relatively simplified hardware. Typically, the transfer function of the notch filter can be designed by finite impulse response (FIR) or infinite impulse response (IIR). An IIR notch filter can provide a relatively narrower bandwidth than a FIR notch filter and thus produces less effect on frequencies near the notch frequency. The main challenge however for designing a fixed notch filter is to determine the proper notch attenuation value against the interference signals generated by various targets. If the notch attenuation value is insufficient, high side-lobe level and ripples following the main-lobe of the imaging signal resulting from the remaining interference components will be produced. Since notch attenuation and bandwidth are directly related, an increase in notch attenuation may add to the distortion of the filtered imaging signal.
A simulation was carried out to demonstrate this deleterious effect. Here, 4 MHz and 6 MHz frequencies were used for therapy and imaging, respectively, considering that the location of the prostate tissue is approximately 4–5 cm from the rectal surface (Sanghvi et al 1999). Thus, 4 MHz fundamental and 8 MHz harmonic signals generated by the 4 MHz HIFU transducer severely degrade the image quality of the 6 MHz imaging transducer. Previously, it was shown that two fixed notch filters with 4 MHz and 8 MHz notch frequencies were required to minimize these interference signals (Jeong et al 2009). Figure 1 shows the frequency response of the fixed notch filters designed by a second-order IIR type based on a biquadratic filter with 4 MHz and 8 MHz notch frequencies. When the quality factor Q is 2 (at 4 MHz) and 4 (at 8 MHz), the notch attenuation values (solid line) are −48 dB and −42 dB, respectively. When Q is 8 (at 4 MHz) and 16 (at 8 MHz), the notch attenuation values (dashed line) are −36 dB and −30 dB, respectively. In this simulation, the amplitudes of the fundamental and the second harmonic interference signals are 40 dB and 34 dB, respectively. Subsequently, the two sets of notch filters so designed are used to eliminate these interference signals. Thus, one set of the notch filters has 4 dB shallower notch depth and the other set has 8 dB deeper notch depth than the amplitude of the interference signal. The 6 dB notch depth difference between 4 MHz and 8 MHz notch filter responses may be compensated by the amplitude difference between the fundamental and the second harmonic components of the therapeutic interference signal.
Figure 2 shows the performance of the fixed notch filtering technique. The side-lobe level of the original two-cycle short-pulse excitation shown in figure 2(a) was increased to approximately −4 dB after being mixed with sinusoidal interference (in figure 2(b)). Two types of notch filters (in figure 1) with different Qs were used to remove the therapeutic interference signal as shown in figures 2(c) and (d). One (Q = 8/16) of them shows approximately a −54 dB side-lobe level and high ripples immediately following the main-lobe. The other one (Q = 2/4) shows less than −60 dB side-lobe level and lower ripples than low Q results, but the output signal is distorted significantly when compared to original signal (figure 2(a)). Therefore, the notch attenuation value must be properly determined considering the amplitude of the interference signal and the bandwidth of the imaging signal. However, it is often difficult to select a proper notch attenuation value using the fixed notch filter as shown in figures 2(c) and (d). As a potential solution for this problem, an adaptive noise canceling method to achieve optimized notch attenuation values is proposed.
In HIFU treatment, the signal type and the frequency of the therapeutic interference are known. It was suggested that the adaptive noise canceling technique may provide the desired signal with optimally canceled noise components (Glover 1977). Figure 3 shows a block diagram of the proposed adaptive noise canceling algorithm that may be applied to solving the problem on hand. The primary input signal p(k) is composed of the original imaging signal s(k) mixed with therapeutic interference n(k). The reference signal r(k) is the sinusoidal signal with the same frequency of the therapeutic interference signals. This reference signal r(k) will be adaptively matched to the real interference signal n(k) through an iterative algorithm, and thus the final system output e(k) will be as close to original signal s(k) as possible. In essence, the adaptive system algorithm involves an iterative gradient-descent process to find the optimal filter coefficient vectors at the minimum mean-square error position (Widrow 1996). This least-mean-square (LMS) algorithm can be described by several relative equations:
where y(k) is the adaptive filter output, WT(k) is the N-dimensional transversal filter coefficient vector for adaptation process, e(k) is the error signal between the primary input signal and adaptive filter output signal and β is the adaptation step size. The βe(k)r(k) is the negative gradient to find the optimal WT(k) through the LMS process. As the adaptive filter output y(k) approaches n(k), the e(k) converges to s(k). Figure 4 shows the simulation results of the adaptive noise canceling method when β is 0.000 512 based on the same conditions shown in figure 2. The results of this test are promising as the final output signal shape is very close to original signal. Although, the −6 dB beamwidth is 11% broader, the signal pattern as a whole closely matches the original signal in comparison to the results from the fixed notch filters in figures 2(c) and (d).
A prototype integrated HIFU/imaging transducer with cylindrically focusing was fabricated in order to demonstrate the performance of the proposed algorithm experimentally. This transducer was composed of three single elements. Inner element was used for imaging and outer two identical elements were used for therapy. Considering the treatment of malignant prostate tissues located at 4–5 cm depth, the dimensions of the transducer (14.4 mm × 28 mm) and the center frequency of HIFU (4 MHz) and imaging transducer (6 MHz) were chosen. As a piezoelectric material for the HIFU transducer, PZT4 (840, APC Company, Mackeyville, PA) with a high Curie temperature and a low dielectric/mechanical loss was selected, and PZT-5H was used for the imaging transducer (Zipparo 2003, Zhang et al 2005). PZT4 was diced to make 1–3 composite structure by using a 25 μm width blade at a 250 μm pitch. After cleaning and drying, a high temperature epoxy (EPO-TEK 314, Epoxy Technology, Billerica, MA) was used to fill in the kerfs and cured at 120 °C for 3 h. The final thickness of 450 μm was obtained by lapping, and a 0.5 μm thick gold/chrome electrode was sputtered on it. The cylindrical curved shape with 4 cm radius of curvature was obtained through pressing at 130 °C using a stainless steel tube and a rubber mold. In order to minimize heat absorption and to maximize transmit intensity, the prototype transducer did not have a matching or backing layer. The Ispta (spatial-peak temporal-average) at a 4 cm focal point was 350 W cm−2 and the transducer efficiency was 54%.
PZT-5H (TFT L-145N, TFT Corporation, Tokyo, Japan) and unloaded epoxy (EPO-TEK 301, Epoxy Technology, Billerica, MA) were used to make 1–3 composite structure for the imaging transducer. The fabrication procedure was the same as described for the therapy transducer except that a matching and a backing layer were used. A 110 μm thick layer of unloaded epoxy (EPO-TEK 301, Epoxy Technology, Billerica, MA) was used as the 3 MRayl matching layer, and loaded epoxy (EPO-TEK 301) mixed with tungsten oxide particles was used for the 4.4 MRayl backing layer. Finally, all transducers were assembled in an aluminum housing. The total acoustic power (TAP) of the imaging transducer with two-cycle pulsed wave was 8.5 mW, mechanical index (MI) was 0.7, and Ispta was 144 mW cm−2 which satisfy Food and Drug Administration (FDA) guideline (Barnett et al 2000) and will not contribute to any heating on the target. The designed transducer has a confocal point at 4 cm to increase detect ability. Because the −6 dB bandwidth was 52%, the 12 MHz third harmonic signal could be significantly reduced. Figure 5 shows a photograph of the prototype integrated HIFU/imaging transducer and its cross-sectional view. Sheets of copper and plastic along with coaxial cables were used inside the housing for radio frequency (RF) shielding between the HIFU and imaging transducer.
The performance of the prototype transducer was tested first to verify whether it can deliver enough ultrasound energy to heat a target to more than 70 °C. Figure 6(a) shows the setup for this experiment. The tip of a thermocouple (TMTSS-020G-6 and HH806AU, OMEGA Engineering Inc., Stamford, CT) was placed inside the porcine muscle slice. A high power amplifier (A300, ENI Co., Santa Clara, CA) driven by a function generator (33250A, Agilent, Santa Clara, CA) was connected to the transducer. The temperature profile was recorded for 60 s. Figure 6(b) shows the measured temperature profile, and the temperature of the tissue was increased from 55 °C to 71 °C in less than 60 s.
A strong sinusoidal interference signal was collected from a polished aluminum plate reflector as shown in figure 7 to verify the effect of the adaptive noise canceling method precisely. An arbitrary waveform generator (AFG3021, Tektronix, Beaverton, OR) was connected to the power amplifier (325LA, ENI Co., Santa Clara, CA) with a 50 dB gain. The reflected signal was amplified by a receiver (5900PR, Panametrics Inc., Waltham, MA) and recorded by the digital oscilloscope (LC534, LeCroy, Chestnut Ridge, NY) with 8-bit analog to digital converter (ADC). A diode expander (DEX-3, Matec, Northborough, MA) and a diode limiter (DL-1, Matec, Northborough, MA) were used to protect the electrical equipment; the same equipment described in figure 6(a) was used to activate the HIFU transducer in this experiment. Note that all measured signals were obtained using an average mode of the oscilloscope, i.e. averaging 15 frames per signal to minimize white noise interference.
Figure 8(a) shows envelope signals of the original two-cycle short-pulse excitation when the 8-bit ADC was used. Averaging resulted in a side-lobe level of −60 dB. When the HIFU transducer was activated, the high amplitude interference was added onto the imaging signal (in figure 8(b)) resulting in a −1.3 dB side-lobe level. After notch filtering with Q = 8/16, as in figure 1, the pedestal level was −50 dB, but the highest side-lobe which came from the ripples immediately following the main-lobe was nearly −28 dB, as shown in figure 8(c). With adaptive noise canceling, the peak side-lobe level was reduced to −48 dB and the ripples were eliminated (figure 8(d)). A 512-tap FIR band pass filter was also used in order to remove the direct current signal and the third harmonic signal which was relatively high.
The B-mode imaging experiment was conducted using a 5 cm thick slice of porcine muscle (in vitro) as the target. Linear mechanical scanning was used to generate an image because the prototype transducer was composed of three single elements. Therefore, the HIFU transducer was also scanned along with the imaging transducer resulting in generation of high amplitude interference but no coagulated lesions.
As shown in figure 9, the 4 MHz HIFU activation equipment was the same as in figure 6(a) and an imaging system was connected to the 6 MHz imaging transducer. The imaging system was composed of a linear motor (LAR37, SMAC Inc., Carlsbad, CA, USA) and a 12-bit ADC (CS12400, Gage Applied Technologies Inc., Lachine, QC, Canada) controlled by the LabVIEW program (LabVIEW, National instruments Co., Austin, TX). This system was connected to the same equipment as described in figure 7 to transmit the two-cycle short pulse and receive the pulse echo signals. After obtaining data, post-processing such as envelope detection and logarithmical compression was done using Matlab (The MathWorks Inc., Natick, MA).
Figure 10(a) shows the original porcine muscle image before activating the HIFU transducer, and figure 10(b) shows the image after transmitting the HIFU beam to the target. A fixed notch filter with a high Q value (20/40) was used to consider the average amplitude and bandwidth of the interference signals in one frame of image. After notch filtering, the image in figure 10(c) shows remnant interference components. Conversely, the adaptive notch filtered image, figure 10(d), shows clearly that the interference was removed. Note that the speckle pattern of the image generated with the proposed adaptive filtering algorithm is very similar to the original image.
We proposed an adaptive noise canceling algorithm to effectively suppress therapeutic interference of varying amplitude while retaining the original signal form as closely as possible. This algorithm could minimize the reflected therapeutic interference generated by the HIFU transducer and significantly decrease side-lobe ripples immediately following the main-lobe. The simulation results show that the recovered imaging signal after applying the proposed algorithm was nearly identical to the original signal. However, the fixed notch filter with too shallow or too deep a notch attenuation value shows high remnant ripples or signal distortion, respectively. The remnant ripple was one of the main factors that increase the side-lobe level of normal short-pulse excitation which is relatively weaker when compared to the noise signal. Although this artifact may be reduced with a deeper notch attenuation value, significant frequency distortion around notch frequency will also result in a distorted signal compared to the original signal.
A prototype integrated HIFU/imaging transducer for the treatment of malignant prostate tissues was fabricated. Although it had a cylindrically focused aperture, it could increase the target temperature from 55 °C to 71 °C after 60 s of operation. This transducer provided an interference/ripple-free single scan-line and B-mode image after applying adaptive noise canceling. In the presence of a strong reflector, the side-lobe level after adaptive canceling was −48 dB, whereas the fixed notch filter had a −28 dB side-lobe level. Also, the adaptive noise canceling method could provide a high quality B-mode image with minimized interference and ripples. The stripe-patterned artifact in the B-mode image after fixed notch filtering might come from both remnant interference and ripples as a result of the fixed notch filtering. This means that fixed notch filtering may not be able to eliminate the variable interference signal completely while maintaining the original signal shape. Further, the amplitude of the reflected therapeutic interference may vary depending on the target property during treatment. The adaptive noise canceling technique is shown to be capable of overcoming this limitation. The proposed algorithm may be implemented either in the software or in the hardware with a high speed digital signal processor (DSP).
It is well known that the B-mode image can be used for observing lesion formation. However, most ultrasound image-guided HIFU systems that implement simultaneous therapy and imaging suffer from HIFU interference received by the imaging transducer. The proposed method provides a potential solution to minimize this HIFU interference by allowing real-time B-mode imaging during treatment. Although our current experimental arrangement did not allow monitoring of lesion formation in real time, the experimental results demonstrated that the proposed algorithm could minimize HIFU interference. If B-mode imaging is possible during treatment, observation of lesion formation may be possible based on the hyperechoic response due to cavitation or different acoustic impedance between a coagulated lesion and a normal tissue.
In this paper, it is demonstrated that in traditional short-pulse excitation an adaptive noise canceling technique could be implemented to reduce therapeutic interference signals and remnant ripples to provide high-resolution B-mode imaging during treatment. Therefore, it may be a promising approach to achieve real-time simultaneous therapy and imaging in ultrasound image-guided HIFU.
The authors acknowledge the support of NIH grant #P41-EB2182, Mr Jay Williams for help with fabrication of the housing, Mr Jin Hyoung Park for help with the soft biological tissue imaging experiments and Mr Hyung Ham Kim for help with the hydrophone measurements.
Online at stacks.iop.org/PMB/55/1889