Low intensity pulsed ultrasound (LIPUS) has shown great utility and promise in medical therapeutic treatments, including bone and soft tissue healing, tooth root resorption [1
], stem cell proliferation and differentiation [5
], and antibody production [7
]. In these applications, 1.5 MHz has been validated to be an effective stimulation frequency, and the applied intensities are within a range between 30 mW/cm2
and 100 mW/cm2
. Accurate calibrations of ultrasound intensities have become important because treatment outcomes are highly dependent on the intensities and duration of LIPUS. If the ultrasound exposure level is too low, no biomedical or clinical effect will be observed, while too high a dose can cause adverse or damaging effects to the target tissues or cells [10
]. Radiation force balance is a benchmark technique used to measure ultrasound power and intensity. However, a radiation force balance system is costly, difficult to operate, and compromised by measurement noise [12
]. In an environment with noise vibration, the error of a radiation force balance for low intensity ultrasound measurement can easily surpass 20%. In a working environment such as in a biology laboratory, the measurement is likely to be affected by background vibrations produced from other lab equipment. In addition, the application of the radiation force balance is restricted in some situations. For example, an ultrasound holder with an array of transducers is commonly used in biological experiments to stimulate cells for various therapeutic applications. In this situation, it is difficult for the radiation force balance to directly measure the intensity of each individual transducer that has been fixed in the ultrasound holder, since it uses a large target to collect the ultrasonic beam which in turn would also collect the ultrasound beams of several other transducers simultaneously. Although the measurement may be realized by an absorber with a hole to isolate the each individual transducer by masking the other transducers, undoubtedly, it would also add to the complexity of the operation. Therefore, the development of a vibration-resistant, low-cost and easy-to-operate alternative measurement device is indispensable for rapid calibration of ultrasound intensity.
A vibration-resistant and cost-effective thermoacoustic sensor presents a suitable alternative to the radiation force balance. This type of sensor determines the ultrasound intensity according to the temperature rise caused by the heat produced from incident acoustic energy [13
]. A sensor using the thermal method was developed to directly measure spatial-peak temporal-average intensity (Ispta
) or determine intensity beam profiles of ultrasonic fields [13
], as it is commonly performed, for instance, in acoustic output characterization of diagnostic ultrasonic equipment for safety considerations. The thermoacoustic sensor is based on a conventional beam-plotting set-up in a water tank, which in principle can be used to calibrate a range of transducers. Different from spatial-peak temporal-average intensity (Ispta
), which represents the maximum intensity in the acoustic field, spatial-average temporal-average intensity (Isata
) describes the average intensity of the acoustic field [16
]. Therapeutic ultrasound applications require accurate measurements of spatial-average temporal-average (Isata
) which cannot be directly measured by hydrophones or the above mentioned sensor. To this end, we designed a thermoacoustic sensor to measure Isata
of a commercial ultrasound generator, SonaCell (IntelligentNano Inc., Edmonton, AB, Canada), which generates low intensity pulsed ultrasound at a frequency of 1.5 MHz for therapeutic applications. We adopted a thermal method based on a transient temperature model in order to quickly measure ultrasound intensities [17
]. To make the sensor easy to operate and maintain consistency during each measurement, we simplified the sensor set-up procedure using a concept of close-proximity [19
] where the designed sensor is directly coupled to the transducer through the ultrasound medium (ultrasound gel or degassed water) to perform the measurements. In applying the close-proximity thermoacoustic sensor design for ultrasound calibration, an enhancement in response time, accuracy, and consistency of sensor measurements becomes critical. Therefore, further studies are still needed. This paper focuses on how to improve the performance of the sensor design through a new structural design and artificial neural network algorithm.
The performance of the thermal sensor heavily depends on the conversion efficiency from low intensity ultrasonic energy into heat. In our current study, we proposed a two-layer structure to increase the absorption efficiency of ultrasound energy. The first layer is a cylindrical plexiglass absorber, the same as our previous design [19
]. A second layer is made of polyurethane rubber with a high attenuation coefficient to absorb extra ultrasound energy. This design provides higher conversion efficiency than our previous one-layer design. The detailed configuration is discussed in section two.
The thermoacoustic sensor measures the temperature increase caused by incident ultrasound energy to determine the ultrasound intensity. However, the sensor characteristics are not only dependent on applied ultrasound intensity, but also on ambient temperature and the slightly changing acoustic properties of absorber materials as the absorber heats up, which create a complex problem in sensor design. To obtain an accurate and consistent measurement, these effects should be considered and compensated for in the thermal sensor design. The traditional computational method usually identifies a deterministic mathematical relationship through data interpolation; however, this method is inadequate for solving the problem, since it is extremely difficult to resolve the mathematical formula, if not impossible, among multiple confounding variables such as the temperature change of the sensor, applied ultrasound intensity and ambient temperature from measured data. We previously showed a solution using extrapolation and interpolation based on calibration values [19
], but the method is still unsatisfactory due to the requirement of complex calibration, calculation procedures, and limited improvement in accuracy. In this paper, we propose the implementation of an artificial neural network to identify the relationship and solve the problem. An artificial neural network can map the implicit relationship of inputs and outputs through the training and testing of measured data, which has been applied to compensate for the various nonlinear errors in system designs [20
]. Through proper training, the artificial neural network can compensate for the nonlinear errors, enabling a direct read-out of the applied ultrasound intensity.
A novel two-layer thermoacoustic sensor based on an artificial neural network as a means to adapt to temperature drifts was proposed, implemented, and investigated in order to measure low ultrasound intensities. Compared to our previous sensor design where only one layer of plexiglass absorber is used, the new sensor design employed two absorber layers with a plexiglass layer in the front and a rubber layer at the back. The current design has improved the sensor’s absorption efficiency, as demonstrated by simulation, and thus resulting in a more rapid and reliable estimation of ultrasound intensity. The two-layer sensor design demonstrates that sensor performance can be improved by optimizing its physical structure.
The temperature increase of the sensor depends not only on the ultrasound intensity, but also ambient temperature and the slightly changing acoustic properties of absorber materials that change with temperature, which in turn makes sensor measurements a complicated inverse problem. It is extremely difficult to resolve the exact mathematical relation. To overcome this difficulty, the method of artificial neural networks is proposed and applied into the thermoacoustic sensor design. Application of the artificial neural network method requires training the data in a proper manner. The data are divided into two groups. Most of the data are used to directly train the network, while the other data are set aside for validation and testing purposes. If all the data are used to train the network, the problem of over-training appears, in which the trained network works well for the training data with minimum error, but cannot be generalized to the untrained data well. Therefore, some data are set aside to check for the presence of over-training and to decide when to stop training in order to obtain an optimal network that can minimize the generalization error. Through proper training, the artificial neural network can compensate for measurement errors caused by temperature drifts and enable the sensor to directly measure the applied ultrasound intensity. The experimental results demonstrate that the trained network not only validates on the data sets with training, but also is adaptive to the untrained data sets in the training range. The experimental results show that measurement error is reduced from greater than 15% without network compensation, to 1% with network compensation. The current design trained the network to adapt to room temperatures ranging from 20 °C to 26 °C, however the adaptive temperature range of the sensor design can easily be extended by training more data sets in a wider temperature range.
Thermocaoustic sensors, due to their relatively simple structure, have the advantage of low cost and simplicity over the other techniques used to measure ultrasound intensities. Our design is based on the close-proximity sensor concept as a means to measure spatial-average temporal-average intensity (Isata
), which not only simplifies the set-up, but also guarantees the consistency of each measurement. For the Isata
measurement, radiation force balance is a benchmark technique and has a minimal measurement uncertainty (3%) without vibration noises. The radiation force balance is used to calibrate the ultrasound generator for the sensor design, which would link the thermoacoustic sensor’s measurement error to the radiation force balance’s uncertainty. In addition, the uncertainty of the beam area used to derive ultrasound intensity also contributes to the absolute calibration uncertainty of the thermoacoustic sensor, and therefore, our sensor design cannot provide more accurate measurements than the radiation force balance in a vibration-free environment. However, in an environment with vibration noises, such as in a biology laboratory, the error of a radiation force balance can easily surpass 20% for low ultrasound intensity measurements, whereas our sensor design only has an overall measurement uncertainty of 5%. For thermoacoustic sensors, it is better to let the sensor cool down before beginning the next measurement in order to maintain measurement accuracy. However, the cool-down time between measurements should not be considered as a disadvantage for the thermoacoustic sensor design, since the radiation force balance, as the benchmark technique, also needs a short period of time between measurements while it waits for a force balance. It takes around 10~15 s for the radiation force balance to take a measurement and 10~20 s between measurements to wait for a force balance to settle down, while it takes 12 s for our sensor to take a measurement and around 13 s for the sensor to cool down between measurements. Table 6
shows advantages and disadvantages of both techniques for further comparison.
Advantages and disadvantages of radiation force balance and the thermocaoustic sensor.
Advantages and disadvantages of radiation force balance and the thermocaoustic sensor.
|Radiation Force Balance|
The designed thermoacoustic sensor can provide a real-time measurement and process the sensor’s temperature change, since a microcontroller is used in the sensor’s design to form an embedded system. By measuring and processing temperature data, the value of ultrasound intensity is obtained in 12 s. The microcontroller is implemented in a printed circuit board outside the sensor to process the temperature data sent by the thermistor and to perform compensations based on the artificial neural network. In the future design, the microcontroller can be integrated into the ultrasound generator system in order to measure and control the output intensity simultaneously, which makes the ultrasound auto-calibration system become a possibility.
A novel two-layer thermoacoustic sensor based on an artificial neural network is described and investigated in this paper. The structure design improvement enables the sensor to reduce its measurement time from 20 s to 12 s. The artificial neural network algorithm is integrated in order to compensate for the influence of the temperature drifts, adapting the sensor for a range of ambient temperatures, and providing an accurate and consistent measurement of the ultrasound intensities. The experimental result show that the compensation provided by the artificial neural network reduced the temperature drift errors from more than 15% to 1%. The final results show that the new sensor achieves an average error of 1.31 mW/cm2 over 18 measurement samples.
The new sensor design is a low-cost alternative method that can provide rapid ultrasound intensity measurements without any complex set-up procedure. Although the radiation force balance is the golden standard method with minimal measurement error, the application of the technique is limited by the requirements of experience in regards to equipment set-up and operation. In addition, measurement accuracy is affected by background vibrations, which in turn limits its application in biology laboratories. The new thermoacoustic sensor design is an easy-to-operate alternative method for rapidly measuring low ultrasound intensity (30 mW/cm2 to 120 mW/cm2) with high accuracy, especially in a practical environment like a biology laboratory.