Apparatus and method for generating real-time monitoring image by using high-intensity focused ultrasound transducer
The method addresses limitations in HIFU monitoring by generating high-resolution images using only an HIFU transducer through optimized transfer functions and deep learning, enhancing precision and applicability in constrained environments.
Patent Information
- Application Number
- PCT/KR2026/096195
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2025-05-15
- Filing Date
- 2026-04-23
- Publication Date
- 2026-11-19
AI Technical Summary
Conventional HIFU monitoring technologies face limitations such as low accuracy in cavitation monitoring, difficulty in determining treatment location and size, and increased complexity due to separate imaging and HIFU transducers, especially in array transducers, which are not suitable for environments with size constraints.
A real-time monitoring image generation method using a high-intensity focused ultrasound (HIFU) transducer that generates high-resolution images without an imaging transducer by optimizing the transmit/receive transfer function and applying coded excitation and deep learning-based deconvolution techniques, utilizing a deep learning model to process low-resolution ultrasound image data.
Enables high-resolution real-time treatment monitoring using only an array HIFU transducer, simplifying the ultrasound transceiver structure, reducing manufacturing difficulty and costs, and expanding applicability to environments with size limitations, while improving precision and reliability of treatment site analysis.
Smart Images

Figure KR2026096195_19112026_PF_FP_ABST
Abstract
Description
Real-time monitoring image generation device and method using high-intensity focused ultrasound transducer
[0001] The present invention relates to a real-time monitoring image generation device and method using a high-intensity focused ultrasound transducer, and more specifically, to a real-time monitoring image generation device and method using a high-intensity focused ultrasound transducer capable of generating real-time high-resolution monitoring images using only a high-intensity focused ultrasound (HIFU) transducer without an imaging transducer.
[0002]
[0003] High-intensity focused ultrasound (HIFU) is a treatment technique that focuses high-intensity ultrasound energy onto a localized (specific) area to increase temperature and damage (heat or destroy) lesions. Recently, compared to conventional surgical methods, HIFU treatment is becoming increasingly preferred by patients and medical professionals due to the fact that it can treat specific lesions in internal organs non-invasively without skin incision and has a short recovery period.
[0004] The basic principle of HIFU is that ultrasound is a sound wave with a high frequency (generally 0.8 - 3.5 MHz), and HIFU delivers high energy by focusing this ultrasound onto a specific focal point. The heat generated at the focal point (approximately 60 - 90 ℃) causes tissue necrosis or destruction, thereby inducing a therapeutic effect. Since it does not significantly affect surrounding normal tissue, it is used in various fields such as tumor removal, treatment of uterine fibroids, treatment of neurological diseases, and cosmetic and obesity treatments.
[0005] In particular, the fact that treatment is possible without tissue damage in the treatment of cancer and neurological diseases serves as a major advantage.
[0006] Array HIFU transducers, which are widely used recently, are transducers composed of multiple ultrasound transducer elements and are a technology that controls the focus by precisely controlling the ultrasound signals of individual elements. Compared to conventional single-element HIFU transducers, they have the advantage of enabling more precise ultrasound focus control and treatment.
[0007] Array HIFU transducers can typically be composed of piezoelectric elements, a matching layer, a backing layer, and phase control circuitry. Piezoelectric elements are the core components that generate ultrasound and consist of multiple small elements, while the matching layer is a layer that regulates the effective transmission of ultrasound. Additionally, the backing layer contributes to improving image quality by reducing the length of ultrasound pulses delivered to biological tissues through the prevention of reflection of ultrasound propagating to the back of the piezoelectric elements; however, it may not be suitable for HIFU transducers due to the resulting loss of ultrasound energy. Nevertheless, it is sometimes applied considering the ease of manufacturing the transducer. Furthermore, the phase control circuitry enables focus shifting by adjusting the signals of each element.
[0008] The success of such HIFU treatment depends largely on the advancement of real-time monitoring technology, and accurate real-time monitoring technology is essential to obtain effective treatment results.
[0009] Conventional HIFU monitoring technologies include Active Cavitation Mapping, which estimates the size and location of treatment by comparing signals before and after ultrasound transmission for HIFU treatment, and Passive Acoustic Mapping, which determines whether treatment is being performed by receiving ultrasound emitted from cavitation in the treatment area. However, each has limitations, such as low accuracy in cavitation monitoring or difficulty in determining the location and size.
[0010] Another prior art, Korean registered patent No. 10-1625646, proposes a HIFU treatment probe having a structure in which an imaging transducer and a HIFU transducer are combined.
[0011] More specifically, using a real-time cavitation mapping method, a short-pulse HIFU signal for monitoring, distinct from a long-pulse therapeutic HIFU signal, is transmitted, and a reflected signal is received by an imaging transducer, thereby enabling the visualization of the cavitation location and treatment progress in real time during the treatment process. This improves the accuracy of HIFU treatment and allows treatment to proceed while verifying the focal point location in real time.
[0012] However, the problem of increased complexity arises as separate systems are added for each transducer (imaging transducer and HIFU transducer) to operate independently. Furthermore, due to the structure of the HIFU transducer including the imaging transducer, it requires high manufacturing difficulty and reduces treatment efficacy when expanded into an array HIFU transducer or when there are size constraints on the transducer, such as in endoscopic ultrasound and transrectal ultrasound. In addition, due to the characteristics of HIFU transducers, which have a narrow frequency bandwidth for high energy delivery, when transmitting and receiving ultrasound for image monitoring using a HIFU transducer, the received signal is observed as a long signal in the axial direction, which limits the ability to accurately measure the location and size of the treatment site.
[0013]
[0014] Accordingly, the present invention has been devised to solve the problems of the prior art as described above. The objective of the present invention is to provide a real-time monitoring image generation device and method using a high-intensity focused ultrasound transducer that can generate real-time high-resolution monitoring images, enabling real-time treatment monitoring even when using only an array HIFU transducer without an imaging transducer.
[0015]
[0016] A real-time monitoring image generation device using a high-intensity focused ultrasound transducer according to an embodiment of the present invention for achieving the above-mentioned purpose preferably comprises a first generation unit that uses a high-intensity focused ultrasound (HIFU) transducer to transmit a monitoring ultrasound signal to an object and receives a signal reflected from the object to generate low-resolution ultrasound image data, and a second generation unit that inputs the low-resolution ultrasound image data generated by a stored deep learning model and outputs high-resolution ultrasound image data.
[0017] Furthermore, it is preferable that the first generation unit transmits the monitoring ultrasonic signal to the target object using a coded excitation technique, acquires frame data for a region of interest using the reflected signal, and performs pulse compression on the frame data to generate low-resolution ultrasonic image data as input to the deep learning model.
[0018] Furthermore, the real-time monitoring image generation device using the high-intensity focused ultrasound transducer described above further comprises: a learning data generation unit that uses a high-intensity focused ultrasound (HIFU) transducer to transmit a monitoring ultrasound signal to an object, receives a signal reflected from the object, measures a transmit / receive transfer function, and generates at least one ultrasound image data; a learning processing unit that inputs the generated ultrasound image data into a pre-stored deep learning algorithm to perform learning processing; a first post-processing unit that inputs the low-resolution ultrasound image data generated by the learning data generation unit into a deep learning model according to the learning processing result to receive high-resolution ultrasound image data output, and generates simulated low-resolution ultrasound image data by performing blurring using the transmit / receive transfer function input to the high-resolution ultrasound image data; and a model update unit that calculates the error between the original low-resolution ultrasound image data generated by the learning data generation unit and the simulated low-resolution ultrasound image data generated by the first post-processing unit, sets the calculated error value as a correction value, and performs an update of the deep learning model, wherein the model update unit It is preferable that the final updated deep learning model is stored in the second generation unit.
[0019] Furthermore, it is preferable that the above-described learning data generation unit transmits the monitoring ultrasonic signal to the target object using a coded excitation technique, acquires frame data for a region of interest using the reflected signal, and performs pulse compression on the frame data to generate low-resolution ultrasonic image data as input to a deep learning model.
[0020] Furthermore, the real-time monitoring image generation device using the high-intensity focused ultrasound transducer described above further comprises a second post-processing unit that generates simulated low-resolution ultrasound image data by performing blurring using the transmission-reception transfer function measured on the high-resolution ultrasound image data output by the deep learning model, and a function update unit that calculates the error between the input low-resolution ultrasound image data corresponding to the input of the deep learning model generated by the learning data generation unit and the simulated low-resolution ultrasound image data generated by the second post-processing unit, sets the calculated error value as a correction value, and performs an update of the transmission-reception transfer function, and it is preferable that the transmission-reception transfer function updated by the function update unit is input to the first post-processing unit.
[0021] Furthermore, it is preferable that the learning processing unit, the first post-processing unit, the model update unit, the second post-processing unit, and the function update unit are performed through a feedback loop to optimize the parameters of the deep learning model and the transmission / reception transfer function in each iteration.
[0022]
[0023] A method for generating a real-time monitoring image using a high-intensity focused ultrasound transducer, wherein each step is performed by a computational processing means according to another embodiment of the present invention, comprising: a first generation step (S100) of transmitting a monitoring ultrasound signal to an object using a high-intensity focused ultrasound (HIFU) transducer and receiving a signal reflected from the object to generate low-resolution ultrasound image data; and a second generation step (S200) of inputting the generated low-resolution ultrasound image data into a stored deep learning model to receive high-resolution ultrasound image data as output. Preferably, the first generation step (S100) uses a coded excitation technique to transmit the monitoring ultrasound signal to the object, acquires frame data for a region of interest using the reflected signal, and performs pulse compression on the frame data to generate low-resolution ultrasound image data as input to the deep learning model.
[0024] Furthermore, to store the deep learning model, the method for generating a real-time monitoring image using the high-intensity focused ultrasound transducer comprises: a learning data generation step (S10) in which the ultrasonic image data generated is input into a pre-stored deep learning algorithm to perform learning processing by inputting the generated ultrasonic image data into the deep learning model according to the learning processing result to receive high-resolution ultrasonic image data, and performing blurring using the transmission-reception transfer function input to the high-resolution ultrasonic image data to generate simulated low-resolution ultrasonic image data; and the low-resolution ultrasonic image data generated by the learning data generation step (S10) and the first post-processing The method further includes a model update step (S40) for performing an update of the deep learning model by calculating the error between the simulated low-resolution ultrasound image data generated by the step (S30) and setting the calculated error value as a correction value, and it is preferable that the deep learning model finally updated by the model update step (S40) is applied to the second generation step (S200).
[0025] Furthermore, the above-described training data generation step (S10) preferably uses a coded excitation technique to transmit the monitoring ultrasonic signal to the target object, uses the reflected signal to acquire frame data for the region of interest, and performs pulse compression on the frame data to generate low-resolution ultrasonic image data as input to the deep learning model.
[0026] Furthermore, the method for generating a real-time monitoring image using the high-intensity focused ultrasound transducer described above further includes a second post-processing step (S50) for generating simulated low-resolution ultrasound image data by performing blurring using the transmission-reception transfer function measured on the high-resolution ultrasound image data output by the deep learning model, and a function update step (S60) for performing an update of the transmission-reception transfer function by calculating the error between the input low-resolution ultrasound image data generated by the learning data generation step (S10) and the simulated low-resolution ultrasound image data generated by the second post-processing step (S50), and setting the calculated error value as a correction value. It is preferable that the transmission-reception transfer function updated by the function update step (S60) is input to the first post-processing step (S30).
[0027] Furthermore, it is preferable that the above learning processing step (S20), first post-processing step (S30), model update step (S40), second post-processing step (S50), and function update step (S60) be performed through a feedback loop to optimize the parameters of the deep learning model and the transmission / reception transfer function in each iteration.
[0028]
[0029] According to the present invention, through a real-time monitoring image generation device and method using a high-intensity focused ultrasound transducer, there is an advantage that high-resolution HIFU treatment monitoring is possible using only array HIFU without an imaging transducer.
[0030] In particular, since it does not use an imaging transducer, the structure of the ultrasound transceiver is simplified, the difficulty of manufacturing medical devices including probes is reduced, and production costs can be reduced. It can also be applied in environments where there are size limitations on the transducer, such as intravascular ultrasound, which can expand the scope of application for HIFU treatment.
[0031] In addition, by progressively improving the transmit and receive transfer function of the HIFU transducer through an iterative algorithm, there is an advantage in that more precise analysis of the treatment site is possible.
[0032] In addition, unlike conventional frequency domain inverse operation-based deconvolution methods, high-resolution monitoring images are inferred through convolution and non-linear deep learning models in the time domain, which has the advantage of solving ringing and computational instability problems that may occur in conventional technology.
[0033] In addition, by optimizing the transfer functions of different ultrasound elements within the array HIFU transducer to improve the performance of the deconvolution deep learning model, the precision of HIFU treatment can be increased. Furthermore, by training the deep learning model including the transmitted pulse signal to enable the application of coded excitation to compensate for the low reception efficiency of the HIFU transducer, the reliability of HIFU treatment monitoring can be improved through precise inference based on the transmitted pulse during the inference process.
[0034]
[0035] FIG. 1 is a diagram illustrating the configuration of a real-time monitoring image generation device using a high-intensity focused ultrasound transducer according to one embodiment of the present invention.
[0036] FIG. 2 is a flowchart illustrating a method for generating real-time monitoring images using a high-intensity focused ultrasound transducer according to an embodiment of the present invention.
[0037]
[0038] Hereinafter, a real-time monitoring image generation device and method using a high-intensity focused ultrasound transducer according to the present invention having the configuration as described above will be explained in detail with reference to the attached drawings. The drawings presented below are provided as examples to ensure that the concept of the present invention is sufficiently conveyed to those skilled in the art. Accordingly, the present invention is not limited to the drawings presented below and may be embodied in other forms. In addition, throughout the specification, the same reference numerals indicate the same components.
[0039] Unless otherwise defined, technical and scientific terms used herein have the meaning commonly understood by those skilled in the art to which this invention pertains, and descriptions of known functions and configurations that could unnecessarily obscure the essence of the invention are omitted in the following description and accompanying drawings.
[0040] Furthermore, a system refers to a set of components, including devices, mechanisms, and means, that are organized and interact regularly to perform necessary functions.
[0041]
[0042] Previously, the treatment area was imaged using an imaging transducer, but the real-time monitoring image generation device and method using a high-intensity focused ultrasound transducer according to one embodiment of the present invention has the advantage of enabling image generation that allows real-time monitoring of the treatment area without a separate imaging transducer by optimizing the receiving function of the array HIFU transducer. In particular, it has the advantage of being able to generate high-resolution images from reflected signals by optimizing the transmit / receive transfer function and applying a non-linear deep learning-based signal processing technique.
[0043] In addition, since HIFU transducers typically have low reception efficiency in order to achieve high transmission efficiency, to overcome this, the real-time monitoring image generation device and method using a high-intensity focused ultrasound transducer according to one embodiment of the present invention preferably include technical features for reinforcing low reception signals by applying coded excitation technology such as Barker Code and Golay Code. Coded excitation technology is a method of transmitting a long coded signal along the time axis, and this technology can increase transmission energy by up to 15-20 dB without increasing the transmission peak voltage, and consequently, it brings about the effect of improving the signal-to-noise ratio (SNR) by that amount, which can be used to increase resolution or penetration.
[0044] In addition, by applying a non-linear deep learning-based deconvolution technique to solve the problem of the narrow frequency bandwidth of the HIFU converter, the deconvolution performed in the frequency domain previously caused ringing and instability, but by applying a time domain-based deep learning model, it has the advantage of improving image quality and generating high-resolution image data.
[0045] As such, conventionally, real-time treatment monitoring was performed by adding a separate imaging transducer, but the present invention improves the ability to perform real-time treatment monitoring using only a HIFU transducer. To achieve this, it includes technical features for optimizing the transmit and receive transfer function of the HIFU transducer and improving the quality of the ultrasound reflection signal by applying coded excitation and deep learning-based deconvolution techniques. Since an imaging transducer is not required, the device is simplified, the size of the ultrasound probe is reduced, and manufacturing costs are lowered.
[0046] Through these technological advancements, there is an advantage in that high-resolution real-time treatment monitoring is possible even in environments where it was previously difficult to apply, such as intravascular ultrasound therapy.
[0047]
[0048] FIG. 1 is a diagram illustrating the configuration of a real-time monitoring image generation device using a high-intensity focused ultrasound transducer according to one embodiment of the present invention.
[0049] As illustrated in FIG. 1, a real-time monitoring image generation device using a high-intensity focused ultrasound transducer according to one embodiment of the present invention includes a first generation unit (100) and a second generation unit (200), and each component is operated in a computational processing means including a computer. That is, a real-time monitoring image generation device using a high-intensity focused ultrasound transducer according to one embodiment of the present invention includes a memory that stores one or more commands and a processor that executes the commands stored in the memory, and the processor performs the operation of the components by executing the commands.
[0050] Let's take a closer look at each component.
[0051] It is preferable that the first generating unit (100) uses a High Intensity Focused Ultrasound (HIFU) transducer to transmit a monitoring ultrasound signal to an object and receive a signal reflected from the object to generate low-resolution ultrasound image data for input to a deep learning model.
[0052] At this time, it is desirable to include all types of array structure converters as the HIFU converter. Typically, a high-intensity therapeutic HIFU signal having a long pulse length in the time axis and a relatively low-intensity monitoring signal having a short pulse length in the time axis can be transmitted. However, in the present invention, a coded excitation technique is applied to the relatively low-intensity monitoring signal to perform modulation into a signal having a long pulse length in the time axis, and this is transmitted to the target.
[0053] The coded signal used in the coded excitation technique is a signal capable of pulse compression and having a large time-bandwidth product (TBP), and may use any one of the Barker code, Chirp code, and Golay code, but is not limited thereto. In addition, the ultrasonic signal can be modulated to the center frequency of the HIFU transducer through the coded excitation technique.
[0054] It is preferable that the first generating unit (100) uses the reflected signal to acquire frame data for a region of interest and performs pulse compression on the frame data to generate the low-resolution ultrasound image data.
[0055] It is preferable that the second generation unit (200) inputs the low-resolution ultrasound image data generated by the first generation unit (100) into a stored deep learning model and outputs high-resolution ultrasound image data.
[0056] Considering that applying deconvolution to each channel received data results in a problem where real-time performance is difficult to guarantee, and that numerical analysis methods result in ringing or divergence of calculation results while performing inverse inversion in the frequency domain, the present invention further includes a technical feature of training a deep learning model capable of generating / inferring high-resolution image data based on supervised training or unsupervised training.
[0057] To this end, a real-time monitoring image generation device using a high-intensity focused ultrasound transducer according to one embodiment of the present invention preferably further includes a learning data generation unit (10), a learning processing unit (20), a first post-processing unit (30), a model update unit (40), a second post-processing unit (40), and a function update unit (50) as shown in FIG. 1 to generate the deep learning model stored in the second generation unit (200).
[0058] It is preferable that the above learning data generation unit (10) uses the HIFU converter to transmit a monitoring ultrasonic signal to an object, receives a signal reflected from the object, measures a transmission and reception transfer function, and generates at least one ultrasonic image data.
[0059] The above-mentioned learning data generation unit (10), similar to the above-mentioned first generation unit (100), uses a coded excitation technique to modulate a signal for relatively low-intensity monitoring having a short pulse length in the time axis into a signal having a long pulse length in the time axis, and transmits this to the target object.
[0060] The above-mentioned learning data generation unit (10) can also pulse-compress the coded signal used in the coded excitation technique and, as a signal with a large time-bandwidth product (TBP), can use any one of Wacker code, Chirp code, and Golay code, but is not limited thereto. In addition, the ultrasonic signal can be modulated to the center frequency of the HIFU transducer through the coded excitation technique.
[0061] At this time, it is desirable to use the reflected signal to acquire frame data for the region of interest and to perform pulse compression on the frame data to generate low-resolution ultrasound image data as input to the deep learning model.
[0062] Furthermore, the problem of a narrow frequency bandwidth can be overcome by subjecting the received signal to a deconvolution process using a transfer function. This numerical deconvolution method is a mathematical technique that estimates the original form by removing blur effects or distortions during the process of restoring signals or images; it is not limited to any numerical method, such as Wiener deconvolution, ForWaRd, and Richardson-Lucy deconvolution, as it can be applied.
[0063] Thus, for effective deconvolution, accurate measurement of the transmit / receive transfer function is required. To this end, the learning data generation unit (10) applies short pulses to all array elements to focus the array elements (multi-channel transmission), and it is preferable to store the transmit / receive transfer function of the corresponding element only from the received signal of the element whose transfer function is to be measured (single-channel reception). In other words, while the transmit / receive transfer function of a single-element ultrasonic transducer can be measured via pulse-echo, the transmit / receive transfer function of an array-type transducer must consider the influence of lateral direction mode and cross talk caused by surrounding channel transmission. Accordingly, in the present invention, it is preferable to store the transmit / receive transfer function of the corresponding element by using only the received signal of the element whose transfer function is to be measured among the echo signals reflected after multi-channel focused transmission. Compared to a measurement method through single-channel transmission and single-channel reception, this measurement method has the advantage of effectively measuring the operation transfer function of the array transducer, including lateral direction mode and cross talk.
[0064] It is preferable that the learning processing unit (20) inputs at least one ultrasound image data generated through the learning data generation unit (10) into a pre-stored deep learning algorithm to perform learning processing. The pre-stored deep learning algorithm includes all types of generally used Image-to-Image algorithms such as U-net, GAN, and Diffusion net, but is not limited thereto.
[0065] It is preferable that the first post-processing unit (30) inputs the low-resolution ultrasound image data of the deep learning model generated by the learning data generation unit (10) into the deep learning model according to the learning processing result by the learning processing unit (20), and outputs high-resolution ultrasound image data.
[0066] Afterwards, it is desirable to generate simulated low-resolution ultrasound image data by performing blurring using a transmit / receive transfer function input to the outputted high-resolution ultrasound image data. That is, low-resolution ultrasound image data (simulated low-resolution ultrasound image data) is generated again using the high-resolution ultrasound image data output by the deep learning model.
[0067] Here, using the above-mentioned transmission and reception transfer function means using a mathematical model based on the transmission and reception characteristics of the array HIFU converter according to the present invention.
[0068] It is preferable that the model update unit (40) calculates the error between the low-resolution ultrasound image data generated by the learning data generation unit (10) and the simulated low-resolution ultrasound image data generated by the first post-processing unit (30), sets the calculated error value as a correction value, and performs an update of the deep learning model.
[0069] It is preferable to calculate the error by comparing the low-resolution ultrasound image data generated by the learning data generation unit (10) and the simulated low-resolution ultrasound image data generated by the first post-processing unit (30) at the pixel level, and to set the calculated error value as a correction value to perform weight updates of the deep learning model.
[0070] In the present invention, the deep learning model finally updated by the model update unit (40) is stored in the second generation unit (200) and performs operations.
[0071] That is, a high-resolution echogenicity image representing the shape and location of the tissue can be output through the deep learning model finally updated by the model update unit (40).
[0072] In addition, in order to improve the update results of the deep learning model, it is preferable for the present invention to also update the transmission and reception transfer function input to the first post-processing unit (30).
[0073] It is preferable that the second post-processing unit (50) generates simulated low-resolution ultrasound image data by performing blurring using the transmission and reception transfer function measured on the high-resolution ultrasound image data output by the deep learning model.
[0074] Afterwards, it is preferable for the function update unit (60) to calculate the error between the low-resolution ultrasound image data generated by the learning data generation unit (10) and the simulated low-resolution ultrasound image data generated by the second post-processing unit (50), set the calculated error value as a correction value, and perform an update of the transmission and reception transfer function.
[0075] At this time, the transmission and reception transfer function updated by the function update unit (60) is input to the first post-processing unit (30), and based on this, blurring of the high-resolution ultrasound image data output by the deep learning model is performed.
[0076] As such, it is preferable that the learning processing unit (20), the first post-processing unit (30), the model update unit (40), the second post-processing unit (50), and the function update unit (60) are performed through a feedback loop to optimize the parameters of the deep learning model and the transmission / reception transfer function in each iteration.
[0077]
[0078] FIG. 2 is a flowchart illustrating a method for generating real-time monitoring images using a high-intensity focused ultrasound transducer according to an embodiment of the present invention.
[0079] As illustrated in FIG. 2, a method for generating a real-time monitoring image using a high-intensity focused ultrasound transducer according to one embodiment of the present invention includes a first generation step (S100) and a second generation step (S200), and each step is performed by a real-time monitoring image generating device using a high-intensity focused ultrasound transducer in which each step is performed by a computational processing means.
[0080] Let's take a closer look at each step.
[0081] The first generation step (S100) above involves the first generation unit (100) using a high-intensity focused ultrasound (HIFU) transducer to transmit a monitoring ultrasound signal to an object and receive a signal reflected from the object to generate low-resolution ultrasound image data.
[0082] At this time, it is desirable to include all types of array structure converters as the HIFU converter. Typically, a high-intensity therapeutic HIFU signal having a long pulse length in the time axis and a relatively low-intensity monitoring signal having a short pulse length in the time axis can be transmitted. However, in the present invention, a coded excitation technique is applied to the relatively low-intensity monitoring signal to perform modulation into a signal having a long pulse length in the time axis, and this is transmitted to the target.
[0083] The coded signal used in the coded excitation technique is a signal capable of pulse compression and having a large time-bandwidth product (TBP), and may use any one of the Barker code, Chirp code, and Golay code, but is not limited thereto. In addition, the ultrasonic signal can be modulated to the center frequency of the HIFU transducer through the coded excitation technique.
[0084] The first generation step (S100) uses the reflected signal to acquire frame data for the region of interest and performs pulse compression on the frame data to generate low-resolution ultrasound image data as input to the deep learning model.
[0085] In the second generation step (S200), the low-resolution ultrasound image data generated by the first generation step (S100) is input into the stored deep learning model in the second generation unit (200), and high-resolution ultrasound image data is output.
[0086] Considering that applying deconvolution to each channel received data results in a problem where real-time performance is difficult to guarantee, and that numerical analysis methods result in ringing or divergence of calculation results while performing inverse inversion in the frequency domain, the present invention further includes a technical feature of training a deep learning model capable of generating / inferring high-resolution image data based on supervised training or unsupervised training.
[0087] To this end, a method for generating a real-time monitoring image using a high-intensity focused ultrasound transducer according to one embodiment of the present invention further performs a learning data generation step (S10), a learning processing step (S20), a first post-processing step (S30), a model update step (S40), a second post-processing step (S40), and a function update step (S50), as illustrated in FIG. 2.
[0088] The above learning data generation step (S10) involves the learning data generation unit (10) using the HIFU converter to transmit a monitoring ultrasonic signal to an object, receiving a signal reflected from the object, measuring a transmission and reception transfer function, and generating at least one ultrasonic image data.
[0089] The above training data generation step (S10) also uses a coded excitation technique to modulate a signal for relatively low-intensity monitoring, which has a short pulse length in the time axis, into a signal having a long pulse length in the time axis, and transmits this to the target object. In addition, the coded signal used in the coded excitation technique is a signal capable of pulse compression and has a large time-bandwidth product (TBP), and any one of Wacker code, Chirp code, and Golay code may be used, but is not limited thereto. Furthermore, the ultrasonic signal may be modulated to the center frequency of the HIFU transducer through the coded excitation technique.
[0090] At this time, frame data for the region of interest is obtained using the reflected signal, and pulse compression is performed on the frame data to generate low-resolution ultrasound image data as input to the deep learning model.
[0091] Furthermore, the problem of a narrow frequency bandwidth can be overcome by subjecting the received signal to a deconvolution process using a transfer function. This numerical deconvolution method is a mathematical technique that estimates the original form by removing blur effects or distortions during the process of restoring signals or images; it is not limited to any numerical method, such as Wiener deconvolution, ForWaRd, and Richardson-Lucy deconvolution, as it can be applied.
[0092] Thus, for effective deconvolution, accurate measurement of the transmit / receive transfer function is required. To this end, the learning data generation unit (10) applies short pulses to all array elements to focus the array elements (multi-channel transmission), and it is preferable to store the transmit / receive transfer function of the corresponding element only from the received signal of the element whose transfer function is to be measured (single-channel reception). In other words, while the transmit / receive transfer function of a single-element ultrasonic transducer can be measured via pulse-echo, the transmit / receive transfer function of an array-type transducer must consider the influence of lateral direction mode and cross talk caused by surrounding channel transmission. Accordingly, in the present invention, it is preferable to store the transmit / receive transfer function of the corresponding element by using only the received signal of the element whose transfer function is to be measured among the echo signals reflected after multi-channel focused transmission. Compared to a measurement method through single-channel transmission and single-channel reception, this measurement method has the advantage of effectively measuring the operation transfer function of the array transducer, including lateral direction mode and cross talk.
[0093] The above learning processing step (S20) performs learning processing by inputting at least one ultrasound image data generated through the learning data generation step (S10) into a deep learning algorithm that is stored in advance in the above learning processing unit (20).
[0094] At this time, the aforementioned pre-stored deep learning algorithms include all types of commonly used Image-to-Image algorithms such as U-net, GAN, and Diffusion net, but are not limited to these.
[0095] In the first post-processing step (S30), the low-resolution ultrasound image data generated by the learning data generation step (S10) is input into the deep learning model according to the learning processing result of the learning processing step (S20) in the first post-processing unit (30), and high-resolution ultrasound image data is output.
[0096] Afterwards, the first post-processing step (S30) preferably generates simulated low-resolution ultrasound image data by performing blurring using a transmission and reception transfer function input to the outputted high-resolution ultrasound image data. That is, low-resolution ultrasound image data (simulated low-resolution ultrasound image data) is generated again using the high-resolution ultrasound image data output by the deep learning model.
[0097] Here, using the above-mentioned transmission and reception transfer function means using a mathematical model based on the transmission and reception characteristics of the array HIFU converter according to the present invention.
[0098] The above model update step (S40) calculates the error between the low-resolution ultrasound image data generated by the learning data generation step (S10) and the simulated low-resolution ultrasound image data generated by the first post-processing step (S30) in the above model update unit (40), sets the calculated error value as a correction value, and performs an update of the deep learning model.
[0099] In the model update unit (40), it is preferable to compare the low-resolution ultrasound image data generated by the learning data generation step (S10) and the simulated low-resolution ultrasound image data generated by the first post-processing step (S30) at the pixel level for error calculation, and the calculated error value is set as a correction value to perform weight updates of the deep learning model.
[0100] Through this, the deep learning model finally updated by the model update step (S40) is applied to the second generation step (S200), and a high-resolution echogenicity image representing the shape and location of the tissue can be output.
[0101] In order to improve the update result of the deep learning model, it is preferable for the present invention to also update the transmission and reception transfer function input to the first post-processing step (S30).
[0102] The second post-processing step (S50) generates simulated low-resolution ultrasound image data by performing blurring using the transmission and reception transfer function measured on the high-resolution ultrasound image data output by the deep learning model in the second post-processing unit (50).
[0103] The function update step (S60) calculates the error between the low-resolution ultrasound image data generated by the learning data generation step (S10) and the simulated low-resolution ultrasound image data generated by the second post-processing step (S50) in the function update unit (60), sets the calculated error value as a correction value, and performs an update of the transmission and reception transfer function.
[0104] At this time, the transmission and reception transfer function updated by the function update step (S60) is input to the first post-processing step (S30), and based on this, blurring of the high-resolution ultrasound image data output by the deep learning model is performed.
[0105] As such, the learning processing step (S20), the first post-processing step (S30), the model update step (S40), the second post-processing step (S50), and the function update step (S60) are performed through a feedback loop to optimize the parameters of the deep learning model and the transmission / reception transfer function in each iteration.
[0106]
[0107] Through these technical features, the real-time monitoring image generation device and method using a high-intensity focused ultrasound transducer according to one embodiment of the present invention has the advantage of enabling high-resolution monitoring that can detect and analyze changes in the treatment area in real time using only the array HIFU transducer without an imaging transducer by optimizing the transmit / receive transfer function of the array HIFU transducer and applying a deep learning model to visualize the treatment area in high resolution.
[0108] In addition, HIFU converters typically use piezoelectric elements, which have excellent transmission efficiency despite low reception efficiency, to maximize transmission efficiency for high energy transfer. This results in a problem where the magnitude of the received reflected signal is very small. To overcome this, the problem of low reception efficiency is compensated for through coded excitation techniques using Wacker codes or Golay codes. However, since code compression does not operate normally in a narrow frequency bandwidth, a deconvolution process is added to improve code compression performance.
[0109] In other words, the images obtained from the ultrasound device are represented as echogenicity images indicating the shape and location of the tissue. These ultrasound image data are acquired through a transmission-reception transfer function and a convolution process. In this invention, after optimizing the transmission-reception transfer function, linear / iterative deconvolution operations and deep learning-based deconvolution operations are applied, which has the advantage of allowing the treatment area to be visualized more clearly. In particular, by performing deconvolution using convolution operations in the time domain and a non-linear deep learning model, it has the advantage of overcoming ringing phenomena or computational instability issues that occur in existing frequency domain-based deconvolution (Wiener Deconvolution, Richardson-Lucy Deconvolution).
[0110] Finally, considering that the transmit and receive transfer functions of ultrasonic transducers may vary depending on the transducer, optimization is performed by applying an iterative algorithm to train the deconvolution deep learning model with higher accuracy.
[0111]
[0112] Meanwhile, a real-time monitoring image generation device and method using a high-intensity focused ultrasound transducer according to one embodiment of the present invention may be implemented in the form of program instructions that can be executed through various means of electronically processing information and recorded on a storage medium. The storage medium may include program instructions, data files, data structures, etc., either individually or in combination.
[0113] Program instructions recorded on a storage medium may be those specifically designed and configured for the present invention, or they may be those known and available to those skilled in the art of software. Examples of storage media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, and flash memory. Examples of program instructions include machine code, such as that generated by a compiler, as well as high-level language code that can be executed by a device that processes information electronically using an interpreter, such as a computer.
[0114] As described above, the present invention has been explained with specific details such as specific constituent elements and limited exemplary drawings; however, this is provided merely to aid in a more comprehensive understanding of the invention, and the invention is not limited to the above-mentioned embodiment. A person skilled in the art can make various modifications and variations from this description.
[0115] Accordingly, the scope of the present invention should not be limited to the described embodiments, and all things equivalent to or having equivalent variations to the claims set forth below, as well as the claims set forth below, shall be considered to fall within the scope of the concept of the present invention.
[0116]
[0117] [Explanation of the symbol]
[0118] 100 : First generation unit
[0119] 200 : Second generation unit
[0120] 10: Training Data Generation Unit
[0121] 20 : Learning processing unit
[0122] 30: 1st Post-processing Unit
[0123] 40: Model Update Section
[0124] 50 : 2nd Post-processing Unit
[0125] 60 : Function date section
Claims
1. A first generating unit that uses a High Intensity Focused Ultrasound (HIFU) transducer to transmit a monitoring ultrasound signal to an object and receives a signal reflected from the object to generate low-resolution ultrasound image data; and A second generation unit that inputs the low-resolution ultrasound image data generated by the stored deep learning model and outputs high-resolution ultrasound image data; A real-time monitoring image generation device using a high-intensity focused ultrasound transducer, comprising 2. In Paragraph 1, The above first generating part Using the coded excitation technique, A real-time monitoring image generation device using a high-intensity focused ultrasound transducer, which transmits the above-mentioned monitoring ultrasound signal to the above-mentioned target object, acquires frame data for a region of interest using the reflected signal, and generates the above-mentioned low-resolution ultrasound image data by performing pulse compression on the frame data.
3. In Paragraph 1, The real-time monitoring image generation device using the above high-intensity focused ultrasound transducer A learning data generation unit that uses a High Intensity Focused Ultrasound (HIFU) transducer to transmit a monitoring ultrasound signal to a target, receives a signal reflected from the target, measures a transmission-reception transfer function, and generates at least one ultrasound image data; A learning processing unit that inputs generated ultrasound image data into a stored deep learning algorithm to perform learning processing; A first post-processing unit that inputs low-resolution ultrasound image data generated by the training data generation unit into a deep learning model based on the training processing result, outputs high-resolution ultrasound image data, and generates simulated low-resolution ultrasound image data by performing blurring using a transmit / receive transfer function input to the high-resolution ultrasound image data; and A model update unit that calculates the error between the low-resolution ultrasound image data generated by the training data generation unit and the simulated low-resolution ultrasound image data generated by the first post-processing unit, sets the calculated error value as a correction value, and performs an update of the deep learning model; Includes more, A real-time monitoring image generation device using a high-intensity focused ultrasound transducer, wherein the deep learning model finally updated by the model update unit is stored in the second generation unit.
4. In Paragraph 3, The above-mentioned training data generation unit Using the coded excitation technique, A real-time monitoring image generation device using a high-intensity focused ultrasound transducer, which transmits the above-mentioned monitoring ultrasound signal to the above-mentioned target object, acquires frame data for a region of interest using the reflected signal, and performs pulse compression on the frame data to generate low-resolution ultrasound image data as input to a deep learning model.
5. In Paragraph 3, The real-time monitoring image generation device using the above high-intensity focused ultrasound transducer A second post-processing unit that generates simulated low-resolution ultrasound image data by performing blurring using the measured transmit / receive transfer function on the high-resolution ultrasound image data output by the deep learning model; and A function update unit that calculates the error between the low-resolution ultrasound image data generated by the learning data generation unit and the simulated low-resolution ultrasound image data generated by the second post-processing unit, sets the calculated error value as a correction value, and performs an update of the transmission and reception transfer function; Includes more, A real-time monitoring image generation device using a high-intensity focused ultrasound transducer, wherein the transmission and reception transfer function updated by the above-mentioned function update unit is input to the above-mentioned first post-processing unit.
6. In Paragraph 5, The above learning processing unit, first post-processing unit, model update unit, second post-processing unit, and function update unit A real-time monitoring image generation device using a high-intensity focused ultrasound transducer, which is performed through a feedback loop to optimize the parameters of the deep learning model and the transmission / reception transfer function in each iteration.
7. A method for generating a real-time monitoring image using a high-intensity focused ultrasound transducer by a real-time monitoring image generating device using a high-intensity focused ultrasound transducer in which each step is performed by a computational processing means, A first generation step (S100) of transmitting a monitoring ultrasound signal to an object using a High Intensity Focused Ultrasound (HIFU) transducer and receiving a signal reflected from the object to generate low-resolution ultrasound image data; and A second generation step (S200) of inputting the low-resolution ultrasound image data generated in the stored deep learning model and receiving high-resolution ultrasound image data; Includes, The above first generation step (S100) is Using the coded excitation technique, A method for generating real-time monitoring images using a high-intensity focused ultrasound transducer, wherein the above-mentioned monitoring ultrasound signal is transmitted to the above-mentioned target object, frame data for a region of interest is acquired using the reflected signal, and pulse compression is performed on the frame data to generate low-resolution ultrasound image data as input to the deep learning model.
8. In Paragraph 7, To store the above deep learning model, the method for generating a real-time monitoring image using the above high-intensity focused ultrasound transducer is, A learning data generation step (S10) of using a High Intensity Focused Ultrasound (HIFU) transducer to transmit a monitoring ultrasound signal to a target, receiving a signal reflected from the target, measuring a transmission-reception transfer function, and generating at least one ultrasound image data; A learning processing step (S20) for performing learning processing by inputting generated ultrasound image data into a stored deep learning algorithm; A first post-processing step (S30) for inputting low-resolution ultrasound image data generated by the training data generation step (S10) into a deep learning model based on the training processing result, receiving high-resolution ultrasound image data, and generating simulated low-resolution ultrasound image data by performing blurring using a transmit / receive transfer function input to the high-resolution ultrasound image data; and A model update step (S40) for performing an update of the deep learning model by calculating the error between the low-resolution ultrasound image data generated by the above training data generation step (S10) and the simulated low-resolution ultrasound image data generated by the above first post-processing step (S30), and setting the calculated error value as a correction value; Includes more, A method for generating real-time monitoring images using a high-intensity focused ultrasound transducer, wherein the deep learning model finally updated by the above model update step (S40) is applied to the above second generation step (S200).
9. In Paragraph 8, The above training data generation step (S10) is Using the coded excitation technique, A method for generating real-time monitoring images using a high-intensity focused ultrasound transducer, wherein the above-mentioned monitoring ultrasound signal is transmitted to the above-mentioned target object, frame data for a region of interest is acquired using the reflected signal, and pulse compression is performed on the frame data to generate low-resolution ultrasound image data as input to a deep learning model.
10. In Paragraph 8, The method for generating real-time monitoring images using the above-described high-intensity focused ultrasound transducer is, A second post-processing step (S50) for generating simulated low-resolution ultrasound image data by performing blurring using the transmission and reception transfer function measured on the high-resolution ultrasound image data output by the deep learning model; and A function update step (S60) for calculating the error between the low-resolution ultrasound image data generated by the above learning data generation step (S10) and the simulated low-resolution ultrasound image data generated by the above second post-processing step (S50), setting the calculated error value as a correction value, and performing an update of the above transmission and reception transfer function; Includes more, A method for generating a real-time monitoring image using a high-intensity focused ultrasound transducer, wherein the transmission and reception transfer function updated by the above function update step (S60) is input to the above first post-processing step (S30).
11. In Paragraph 10, The above learning processing step (S20), first post-processing step (S30), model update step (S40), second post-processing step (S50) and function update step (S60) are A method for generating real-time monitoring images using a high-intensity focused ultrasound transducer, performed through a feedback loop to optimize the parameters of the deep learning model and the transmission / reception transfer function in each iteration.
