AI enabling ultrasonic temperature measurement method for deep visceral organ HIFU treatment

Abstract

The invention discloses an AI enabling ultrasonic temperature measurement method for deep visceral organ HIFU treatment, and the method comprises the steps: collecting ultrasonic echo signals and temperature change data of deep visceral organs in a treatment process through an HIFU system, and providing a breath-guided multi-mode teacher-student model; a quasi-periodic respiratory phase curve is established through a normalized cross correlation coefficient of radio frequency data and a gating threshold strategy, and then six respiratory phases are extracted by using a phase analysis technology. And then, starting from the energy parameters of the high-intensity focused ultrasound probe, constructing multilayer perceptual mapping of acoustic parameters in a multi-modal teacher knowledge integration module based on intensity normalization and a stage feature fusion strategy. And then, the student knowledge module uses KL divergence as an optimization operator, establishes a probability relationship among the characteristic latent variables, and realizes temperature prediction through an entropy reduction strategy and a loss function minimization strategy. The temperature distribution of deep organs is effectively monitored in real time in the high-intensity focused ultrasound treatment process, and a foundation is laid for clinical high-intensity focused ultrasound application.

Description

Technical Field

[0001] The present invention relates to the field of medical diagnosis technology, and in particular to an AI-enabled ultrasonic temperature measurement method for HIFU treatment of deep organs. Background Art

[0002] High-intensity focused ultrasound tumor ablation technology is a non-invasive treatment method. Its basic principle is to use thermal effect and cavitation effect to rapidly increase the temperature of the focal area, thereby causing coagulative necrosis of tumor tissue, which is subsequently absorbed and cleared by the body. High-intensity focused ultrasound has the advantages of no radiation, non-invasiveness and real-time treatment, which makes it widely used in tumor treatment. Although high-intensity focused ultrasound technology has made significant progress in clinical practice, real-time and accurate monitoring of the distribution of acoustic energy absorption in the focal area, that is, the temperature field, remains a fundamental scientific challenge. During high-intensity focused ultrasound treatment, biological tissue absorbs acoustic energy and converts it into heat. Therefore, measuring and controlling the tissue temperature in the focal area of ​​high-intensity focused ultrasound is crucial to improving the therapeutic effect of high-intensity focused ultrasound.

[0003] Compared with traditional temperature measurement techniques such as thermocouples and magnetic resonance imaging, ultrasound has several advantages, including high temporal resolution, no radiation, low cost, strong resistance to electromagnetic interference, and easy integration with high-intensity focused ultrasound therapy systems. These advantages make ultrasound an ideal choice for monitoring the temperature field in the lesion area. In our previous study, we proposed a lightweight neural network algorithm, the deep multimodal teacher-student model, which established a nonlinear mapping relationship between ultrasound signals and temperature fields by utilizing ultrasound B-mode images, hemodynamics, and stress changes. This method successfully established a physical model of temperature distribution and achieved real-time monitoring of the two-dimensional temperature field of animal muscles in vivo (0.3 seconds per frame, maximum error less than 3°C, and average error less than 2.5°C). Although the deep multimodal teacher-student model performs well in detecting the temperature field of muscle tissue, it still faces challenges in reconstructing the temperature field of complex deep organs, which are susceptible to physiological movements (such as breathing and heartbeat) under ultrasound guidance.

[0004] Ultrasound artifacts and errors caused by physiological motion often interfere with the reconstruction of the temperature field in the focal area of ​​high-intensity focused ultrasound. To address this problem, researchers have proposed gating strategies that use electrocardiogram signals or image correlation coefficients to capture images at the same physiological motion stage and establish motion compensation strategies to correct deformed images. However, gating strategies require additional hardware and algorithm support, increase the complexity and cost of the system, and introduce time delays that affect the efficiency of real-time monitoring. Another research method is adaptive filters, such as least mean square, recursive least squares, and Kalman filters. Researchers have improved the robustness of these filters in complex environments by optimizing algorithms and improving filter design. However, adaptive filters require dynamic adjustment of filter parameters, which increases the computational burden and processing time. In addition, they are not effective in dealing with large-amplitude, rapidly changing noise. Therefore, researchers have explored fast artifact removal methods, such as regularization algorithms. These algorithms stabilize the calculation by introducing constraints in the optimization process and use global and local information to suppress local noise in displacement estimation. However, excessive regularization may lead to over-smoothing, resulting in the loss of important details and affecting measurement accuracy. In summary, these methods have limitations and cannot fully meet the requirements of end-to-end high-intensity focused ultrasound temperature field detection. Typically, motion correction is first achieved through a motion compensation algorithm, and then the temperature field is reconstructed using a temperature inversion algorithm. This two-step approach hinders rapid clinical temperature detection. Summary of the invention

[0005] In order to solve the technical problems existing in the background technology, the present invention aims to provide an AI-enabled ultrasonic temperature measurement method for HIFU treatment of deep organs, which can effectively monitor the temperature distribution of deep organs in real time during high-intensity focused ultrasound treatment, lay a foundation for the clinical application of high-intensity focused ultrasound, provide a scientific and accurate dose planning solution, and provide effective guidance for non-invasive cancer treatment.

[0006] In order to solve the technical problem, the technical solution of the present invention is:

[0007] An AI-enabled ultrasound temperature measurement method for deep-layer organ HIFU treatment, the method comprising: collecting ultrasound echo signals and temperature data, constructing a breathing-guided multimodal teacher-student model, and reconstructing a deep-layer organ temperature field;

[0008] The multimodal teacher-student model of breathing guidance includes: a breathing prediction module, a multimodal teacher knowledge integration module and a student knowledge permission module.

[0009] Furthermore, the multimodal teacher-student model of breathing guidance is constructed, including:

[0010] A quasi-periodic respiratory phase curve is established through the normalized cross-correlation coefficient of RF data and the gating threshold strategy, and then the phase analysis technology is used to extract the six respiratory phases. Subsequently, starting from the energy parameters of the high-intensity focused ultrasound probe, a multi-layer perceptual mapping of acoustic parameters is constructed in the multimodal teacher knowledge integration module based on intensity normalization and stage feature fusion strategy; then the student knowledge licensing module uses KL divergence as the optimization operator to establish a probabilistic relationship between feature latent variables, and realizes temperature prediction through entropy reduction strategy and loss function minimization strategy.

[0011] Furthermore, a high-intensity focused ultrasound system was used to conduct in vivo experiments to collect ultrasound echo signals and temperature data.

[0012] Further, the breathing prediction module is used to demodulate the radio frequency data into a quasi-periodic breathing curve, establish the relationship between the ultrasonic signal and the breathing phase through phase analysis, and thus capture the displacement caused by the physiological movement of the deep organs;

[0013] The multimodal teacher knowledge integration module is used to reconstruct the ultrasound signal into a temperature representation by combining high-intensity focused ultrasound acoustic parameters and an intensity normalization strategy and a phase feature fusion strategy;

[0014] The student knowledge licensing module is used to fit the probability distribution matrix in the latent space of the multimodal teacher knowledge integration module, reduce the matrix calculation load in the neural network and improve the reasoning speed.

[0015] Furthermore, the respiratory prediction module specifically includes: firstly calculating the normalized cross-correlation coefficient of the RF data to extract the respiratory phase of the animal; then setting the normalized cross-correlation coefficient threshold based on the amplitude gating principle to preliminarily screen the RF data; finally, establishing a small area around the thermocouple, and using the gradient update strategy to map the relationship between the small area and the respiratory phase, wherein the normalized cross-correlation coefficient formula of the RF data is as follows:

[0016]

[0017] Among them, γ(m)∈R 1×M ,COV(X,X m ) is the RF reference data X and the i-th frame data X m The covariance of X and σ X,m X and X respectively m The standard deviation of

[0018] Taking the six breathing stages as labels, multi-class cross entropy is used to update the weights, and the formula is as follows:

[0019]

[0020] Where S is the total number of samples, C = 6 is the breathing stage category, represents the actual label of the i-th sample, It represents the probability distribution of the model predicting that the i-th sample belongs to each breathing stage category;

[0021] Furthermore, the multimodal teacher knowledge integration module incorporates high-intensity focused ultrasound acoustic parameters and treatment duration as prior knowledge into the neural network. represents the high-intensity focused ultrasound acoustic parameters and treatment duration, where H f is the high-intensity focused ultrasound frequency, H p represents the high-intensity focused ultrasound power, The duration of the high-intensity focused ultrasound pulse is H t Duration of high-intensity focused ultrasound therapy; The intensity normalization strategy introduced by the multimodal teacher knowledge integration module is to h ∈R S×m Normalize, which represents the feature map output of H, and then use the affine transformation of the trainable parameters, that is, rescaling parameter α h and the bias parameter β h , the formula is as follows:

[0022]

[0023] in is the output of the intensity normalization strategy, μ h and σ h They are x h The mean and standard deviation of , ε is used for stability.

[0024] Furthermore, the student knowledge licensing module makes full use of multimodal data to explore the temperature prediction of the focal area of ​​HIFU based on ultrasound signals. Specifically, a lightweight ResNet18 is determined as E s , where x and H are E s The input, z s ∈R S×S YesE s The extracted feature map, the feature map in the latent space represents the probability distribution of the neural network output, so KL divergence is used to establish z s and x hifu The probability relationship between them is as follows:

[0025]

[0026] By minimizing the KL divergence loss function, z can be reduced s and Bx hifu The relative entropy between them makes z in the latent space sThe distribution gradually approaches x hifu ;

[0027] Both the multimodal teacher knowledge integration module and the student knowledge licensing module use the mean square error loss function, and the formula is as follows:

[0028] where y h is the prediction result, y t For label.

[0029] Furthermore, the multimodal data includes: acoustic parameters, treatment duration, respiratory stage category and ultrasonic signal.

[0030] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any one of the above-mentioned AI-enabled ultrasonic temperature measurement methods for deep organ HIFU treatment is implemented.

[0031] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-mentioned AI-enabled ultrasonic temperature measurement methods for HIFU treatment of deep organs.

[0032] Compared with the prior art, the advantages of the present invention are:

[0033] Firstly, the ultrasound echo signals and temperature change data of deep organs during treatment were collected through the high-intensity focused ultrasound system, and an artificial intelligence algorithm, namely the breathing-guided multimodal teacher-student model, was proposed. The model established a quasi-periodic respiratory phase curve through the normalized cross-correlation coefficient of radio frequency data and the gated threshold strategy, and then used the phase analysis technology to extract six respiratory phases. Subsequently, starting from the energy parameters of the high-intensity focused ultrasound probe, a multi-layer perceptual mapping of acoustic parameters was constructed in the multimodal teacher knowledge integration module based on intensity normalization and stage feature fusion strategy. Then, the student knowledge module used KL divergence as an optimization operator to establish a probabilistic relationship between feature latent variables, and achieved temperature prediction through entropy reduction strategy and loss function minimization strategy. This method effectively monitors the temperature distribution of deep organs in real time during high-intensity focused ultrasound treatment, lays the foundation for the clinical application of high-intensity focused ultrasound, provides a scientific and accurate dose planning solution, and provides effective guidance for non-invasive cancer treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 , the main flow chart of an AI-enabled ultrasonic temperature measurement method for HIFU treatment of deep organs in the present invention. DETAILED DESCRIPTION

[0035] The specific implementation mode of the present invention is described below in conjunction with embodiments:

[0036] It should be noted that the structures, proportions, sizes, etc. shown in this specification are only used to match the contents disclosed in the specification so that people familiar with this technology can understand and read them, and are not used to limit the conditions under which the present invention can be implemented. Any structural modification, change in proportional relationship or adjustment of size should still fall within the scope of the technical content disclosed in the present invention without affecting the effects and purposes that can be achieved by the present invention.

[0037] At the same time, the terms such as "upper", "lower", "left", "right", "middle" and "one" cited in this specification are only for the convenience of description and are not used to limit the scope of implementation of the present invention. Changes or adjustments to their relative relationships should be regarded as the scope of implementation of the present invention without substantially changing the technical content.

[0038] HIFU: High-intensity focused ultrasound.

[0039] Embodiment 1:

[0040] Power ultrasound transmits energy through high-intensity sound waves and has great potential in clinical applications. A classic example is high-intensity focused ultrasound (HIFU), which is widely used for non-invasive tumor ablation in deep organs. During HIFU treatment, biological tissues absorb sound wave energy and convert it into heat. Then, the temperature in the HIFU focal area rises rapidly, leading to coagulative necrosis of tumor tissue. Therefore, controlling tissue temperature is crucial to improve the treatment effect. Although HIFU technology has made significant progress in clinical practice, non-invasive and accurate real-time temperature monitoring within the HIFU focal area remains a fundamental scientific challenge.

[0041] Unlike other temperature measurement techniques such as thermocouples and magnetic resonance imaging, ultrasound has the advantages of real-time, non-invasiveness, and easy integration with HIFU treatment systems. Therefore, these advantages make ultrasound an ideal choice for monitoring the temperature field in the lesion area. Traditional ultrasound monitoring of the temperature field in the HIFU focal area mainly relies on echo signals, because the acoustic characteristic parameters of tissues, such as ultrasound attenuation, sound velocity, density, and thermal expansion, usually appear as a linear function of temperature in the range of 37°C to 50°C. However, when the temperature exceeds 50°C, a nonlinear relationship will occur, limiting its application. In fact, during HIFU treatment, controlling the temperature of the focal area to exceed 60°C is crucial to ensure effective coagulative necrosis. Therefore, in order to resolve the contradiction between existing technologies and clinical needs, it is essential to explore new ultrasound temperature measurement technologies and expand the measurement range.

[0042] Recently, artificial intelligence (AI) has shown significant potential in solving clinical medical problems, ranging from image recognition and segmentation to treatment planning. Therefore, it is worthwhile to investigate whether AI-driven ultrasound technology can address the challenge of monitoring hyperthermia (T>50°C) in the HIFU focal area. However, to the best of our knowledge, such studies are still relatively scarce.

[0043] Here, an AI-based ultrasound temperature measurement technique is proposed, which adopts an end-to-end deep neural network model, called breathing-guided multimodal teacher-student network (BMTS). The BMTS model is shown to be able to elucidate the interaction between HIFU and complex heterogeneous biological media. After eliminating physiological noise, the two-dimensional temperature distribution of the HIFU focal area in the rabbit liver can be accurately reconstructed with an average error of 0.8℃ and a frame rate of 0.37 seconds. Most importantly, the temperature measurement range of the ultrasound technique has been successfully extended from 37℃-50℃ to 37℃-67℃. This breakthrough shows that AI-based ultrasound temperature measurement technology is a promising solution for monitoring temperature changes in deep organs during HIFU ablation.

[0044] like Figure 1 As shown, the present invention provides a technical solution: a method for real-time reconstruction of the focal temperature field of deep organs under high-intensity focused ultrasound. In this invention, the ultrasonic echo signal and temperature change data of the deep organs during the treatment process are first collected by the high-intensity focused ultrasound system. In addition, a breathing-guided multimodal teacher-student model is proposed. The model establishes a quasi-periodic respiratory phase curve through the normalized mutual correlation coefficient of radio frequency data and the gated threshold strategy, and then uses the phase analysis technology to extract six respiratory phases. Subsequently, starting from the energy parameters of the high-intensity focused ultrasound probe, a multi-layer perceptual mapping of acoustic parameters is constructed in the multimodal teacher knowledge integration module based on the intensity normalization and stage feature fusion strategy. Then the student knowledge module uses KL divergence as an optimization operator to establish a probabilistic relationship between feature latent variables, and realizes temperature prediction through the entropy reduction strategy and the loss function minimization strategy. This method effectively monitors the temperature distribution of deep organs in real time during high-intensity focused ultrasound treatment, lays a foundation for the clinical application of high-intensity focused ultrasound, provides a scientific and accurate dose planning solution, and provides effective guidance for non-invasive cancer treatment.

[0045] A method for real-time reconstruction of the focal temperature field of deep organs under high-intensity focused ultrasound, the steps of which are as follows: Step 1, conduct animal in vivo experiments to collect ultrasound echo signals and temperature change data. Step 2, process the data and send them to a breathing-guided multimodal teacher-student model for training. Step 3, fix the model parameters. Step 4, use the trained model to reconstruct the temperature field of deep organs.

[0046] For the in vivo animal experiment, five New Zealand rabbits (weight: 4.5±1kg) were selected. Anesthesia was induced by intramuscular injection of 1.3ml of ketamine. After establishing intravenous access in the rabbit's marginal ear vein, 2-3ml of vecuronium bromide was injected intravenously to achieve muscle relaxation. After anesthesia induction, the rabbit was intubated and connected to a ventilator to replace spontaneous breathing. Anesthesia was maintained using isoflurane at a rate of 0.6ml / min. The inspiratory to expiratory time ratio was set to approximately 0.5, and the respiratory rate was maintained at 30 times per minute. Both parameters were adjusted to keep the oxygen saturation above 95%. The rabbit was placed supine on a board with the forelimbs crossed. The hair on the chest and upper abdomen was shaved using a trimmer and depilatory cream to expose the skin for high-intensity focused ultrasound treatment.

[0047] In order to accurately locate the temperature measurement position and install the temperature measurement device, the following procedures were performed: First, under ultrasound guidance, the middle lobe of the liver was located, and a flat area with significant tissue thickness and no major blood vessels was selected as the location for the focus and thermocouple placement. The shaved area was disinfected with 75% alcohol, and three acupuncture points were marked on the skin with a marking pen, each with a depth of 1-3 cm and in the same plane. The skin at the marked point was incised with the tip of a scalpel (the incision was approximately 0.3 cm). Under ultrasound guidance, a 15G puncture needle and cannula were inserted at the marked point. After ensuring that the puncture depth was 3-5 cm, the needle was removed and the thermocouple was inserted through the cannula. During the experiment, three thermocouples were used to record the spatiotemporal temperature distribution perpendicular to the axis of the heating beam. The treatment head was adjusted so that the focus and the tips of the three thermocouples were in the same plane and visible on the ultrasound image at the same time. The thermocouples were connected to the temperature detection system and numbered TC1, TC2, and TC3, respectively, and each thermocouple was gradually moved away from the heating area. TC1 was located at the focus, and the adjacent thermocouples were separated by 2 mm. At the end of the experiment, the rabbits were euthanized by injecting air (20–40 ml) through the marginal ear vein. All animal experiments were performed according to protocols approved by the Animal Welfare and Ethics Group of the Department of Laboratory Animal Science, Fudan University (202310034S).

[0048] During the experiment, HIFU treatment waves were used to heat the focal area. At the beginning of the treatment, the temperature was recorded in real time using thermocouples, and RF data was collected and exported. The sampling rates of thermocouple data and RF data were 10 Hz and 24 MHz, respectively, and ultrasound images were captured at a fixed frame rate of 9 FPS. Each ultrasound image covered a depth range of 0-12 cm, and the frame size was 128 (horizontal) × 2078 (vertical) pixels.

[0049] The breathing prediction module first calculates the normalized cross-correlation coefficient of the RF data to extract the animal's breathing phase. Then, the normalized cross-correlation coefficient threshold is set based on the amplitude gating principle to preliminarily filter the RF data. Finally, small areas are established around the thermocouples, and the relationship between these small areas and the breathing phase is mapped using a gradient update strategy. The formula for the normalized cross-correlation coefficient of the RF data is as follows:

[0050]

[0051] Among them, γ(m)∈R 1×M ,COV(X,X m ) is the RF reference data X and the i-th frame data X m The covariance of X and σ X,m X and X respectively m The standard deviation of .

[0052] The breathing prediction module uses six breathing stages as labels and uses multi-class cross entropy to update the weights. The formula is as follows:

[0053]

[0054] Where S is the total number of samples, C = 6 is the breathing stage category, represents the actual label of the i-th sample, It represents the probability distribution that the model predicts that the i-th sample belongs to each breathing stage category.

[0055] The multimodal teacher knowledge integration module incorporates high-intensity focused ultrasound acoustic parameters and treatment duration as prior knowledge into the neural network. represents the high-intensity focused ultrasound acoustic parameters and treatment duration, where H f is the high-intensity focused ultrasound frequency, H p represents the high-intensity focused ultrasound power, The duration of the high-intensity focused ultrasound pulse is H t Duration of high-intensity focused ultrasound treatment. The intensity normalization strategy introduced by the multimodal teacher knowledge integration module is to h ∈R S×m Normalize, which represents the feature map output of H, and then use the affine transformation of the trainable parameters, that is, rescaling parameter α h and the bias parameter β h , the formula is as follows:

[0056]

[0057] in is the output of the intensity normalization strategy. h and σh They are x h The mean and standard deviation of , ε is used for stability.

[0058] The student knowledge license module makes full use of multimodal data (acoustic parameters, treatment duration, respiratory stage category and ultrasound signal) to explore the temperature prediction of the focal area of ​​HIFU based on ultrasound signal. Specifically, a lightweight ResNet18 is defined as E s , where x and H are E s The input, z s ∈R S×S YesE s Extracted feature maps. Since the feature maps in the latent space represent the probability distribution of the neural network output, KL divergence can be used to establish z s and x hifu The probability relationship between them is as follows:

[0059]

[0060] By minimizing the KL divergence loss function, z can be reduced s and Bx hifu The relative entropy between them makes z in the latent space s The distribution gradually approaches x hifu .

[0061] Both the multimodal teacher knowledge integration module and the student knowledge licensing module use the mean square error loss function, and the formula is as follows:

[0062] where y h is the prediction result, y t For label.

[0063] The method for real-time reconstruction of the focal temperature field of deep organs in high-intensity focused ultrasound of the present invention can be applied to the reconstruction of the temperature field of deep organs in high-intensity focused ultrasound treatment. The method can obtain a result with a maximum temperature error of about 2.6°C and a reconstruction speed of 0.37 seconds per frame. The method lays a foundation for the clinical application of high-intensity focused ultrasound, provides a scientific and accurate dose planning solution, and provides effective guidance for non-invasive cancer treatment.

[0064] Embodiment 2:

[0065] This embodiment provides a terminal device, which includes a processor and a memory, wherein the memory is used to store a computer program, wherein the computer program includes program instructions, and the processor is used to execute the program instructions stored in the computer storage medium. The processor may be a central processing unit (CPU), or may be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, and is specifically suitable for loading and executing one or more instructions to implement the corresponding method flow or corresponding function; the processor described in the embodiment of the present invention can be used for the operation of a method for real-time reconstruction of the focal temperature field of deep organs of high-intensity focused ultrasound, comprising the following steps:

[0066] A method for real-time reconstruction of the focal temperature field of deep organs in high-intensity focused ultrasound, the steps of which are as follows: Step 1, conduct an in vivo experiment, and use a high-intensity focused ultrasound system to collect the ultrasound echo signal and temperature change data of the deep organs during the treatment process. Step 2, process the data and send it to the breathing-guided multimodal teacher-student model for training. Step 3, fix the model parameters. Step 4, use the trained model to reconstruct the temperature field of the deep organs.

[0067] Embodiment 3:

[0068] This embodiment provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a terminal device for storing programs and data. It is understandable that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and the extended storage medium supported by the terminal device. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory.

[0069] One or more instructions stored in a computer-readable storage medium may be loaded and executed by a processor to implement corresponding steps of a method for real-time reconstruction of a focal temperature field of a deep organ in high-intensity focused ultrasound in the above embodiment; one or more instructions in the computer-readable storage medium are loaded and executed by a processor in the following steps:

[0070] Step 1: Conduct in vivo experiments and use a high-intensity focused ultrasound system to collect ultrasound echo signals and temperature change data of deep organs during treatment. Step 2: Process the data and send it to the breathing-guided multimodal teacher-student model for training. Step 3: Fix the model parameters. Step 4: Use the trained model to reconstruct the temperature field of deep organs.

[0071] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0072] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0073] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0074] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0075] The preferred embodiments of the present invention are described in detail above, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

[0076] Many other changes and modifications may be made without departing from the concept and scope of the present invention.It should be understood that the present invention is not limited to the specific embodiments, and the scope of the present invention is defined by the appended claims.

Claims

1. An AI-enabled ultrasound temperature measurement method for deep organ HIFU treatment, It is characterized in that The method comprises: collecting ultrasonic echo signals and temperature data, constructing a respiratory-guided multimodal teacher-student model, and reconstructing a deep-layer organ temperature field; The multimodal teacher-student model of breathing guidance includes: a breathing prediction module, a multimodal teacher knowledge integration module and a student knowledge permission module.

2. According to claim 1, an AI-enabled ultrasonic temperature measurement method for deep organ HIFU treatment, It is characterized in that The multimodal teacher-student model of breathing guidance is constructed, including: A quasi-periodic respiratory phase curve is established through the normalized cross-correlation coefficient of RF data and the gating threshold strategy, and then the phase analysis technology is used to extract the six respiratory phases. Subsequently, starting from the energy parameters of the high-intensity focused ultrasound probe, a multi-layer perceptual mapping of acoustic parameters is constructed in the multimodal teacher knowledge integration module based on intensity normalization and stage feature fusion strategy; then the student knowledge licensing module uses KL divergence as the optimization operator to establish a probabilistic relationship between feature latent variables, and realizes temperature prediction through entropy reduction strategy and loss function minimization strategy.

3. According to claim 1, an AI-enabled ultrasonic temperature measurement method for deep organ HIFU treatment, It is characterized in that A high-intensity focused ultrasound system was used to conduct in vivo experiments and collect ultrasound echo signals and temperature data.

4. According to claim 1, an AI-enabled ultrasonic temperature measurement method for deep organ HIFU treatment, It is characterized in that The breathing prediction module is used to demodulate the radio frequency data into a quasi-periodic breathing curve, establish the relationship between the ultrasonic signal and the breathing phase through phase analysis, and thus capture the displacement caused by the physiological movement of the deep organs; The multimodal teacher knowledge integration module is used to reconstruct the ultrasound signal into a temperature representation by combining high-intensity focused ultrasound acoustic parameters and an intensity normalization strategy and a phase feature fusion strategy; The student knowledge licensing module is used to fit the probability distribution matrix in the latent space of the multimodal teacher knowledge integration module, reduce the matrix calculation load in the neural network and improve the reasoning speed.

5. According to claim 1, an AI-enabled ultrasonic temperature measurement method for deep organ HIFU treatment, It is characterized in that The respiratory prediction module specifically includes: firstly calculating the normalized cross-correlation coefficient of the RF data to extract the respiratory phase of the animal; then setting the normalized cross-correlation coefficient threshold based on the amplitude gating principle to preliminarily screen the RF data; finally, establishing a small area around the thermocouple and using the gradient update strategy to map the relationship between the small area and the respiratory phase, wherein the normalized cross-correlation coefficient formula of the RF data is as follows: Among them, γ(m)∈R 1×M ,COV(X,X m ) is the RF reference data X and the i-th frame data X m The covariance of X and σ X,m X and X respectively m The standard deviation of Taking the six breathing stages as labels, multi-class cross entropy is used to update the weights, and the formula is as follows: Where S is the total number of samples, C = 6 is the breathing stage category, represents the actual label of the i-th sample, It represents the probability distribution that the model predicts that the i-th sample belongs to each breathing stage category.

6. According to claim 1, an AI-enabled ultrasonic temperature measurement method for deep organ HIFU treatment, It is characterized in that The multimodal teacher knowledge integration module incorporates high-intensity focused ultrasound acoustic parameters and treatment duration into the neural network as prior knowledge. represents the high-intensity focused ultrasound acoustic parameters and treatment duration, where H f is the high-intensity focused ultrasound frequency, H p represents the high-intensity focused ultrasound power, The duration of the high-intensity focused ultrasound pulse is H t Duration of HIFU treatment; The strength normalization strategy introduced by the multimodal teacher knowledge integration module is to h ∈R S×m Normalize, which represents the feature map output of H, and then use the affine transformation of the trainable parameters, that is, rescaling parameter α h and the bias parameter β h , the formula is as follows: in is the output of the intensity normalization strategy, μ h and σ h They are x h The mean and standard deviation of , ε is used for stability.

7. According to claim 1, an AI-enabled ultrasonic temperature measurement method for deep organ HIFU treatment, It is characterized in that The student knowledge licensing module makes full use of multimodal data to explore the temperature prediction of the focal area of ​​HIFU based on ultrasound signals. Specifically, a lightweight ResNet18 is determined as E s , where x and H are E s The input, z s ∈R S ×S YesE s The extracted feature map, the feature map in the latent space represents the probability distribution of the neural network output, so KL divergence is used to establish z s and x hifu The probability relationship between them is as follows: By minimizing the KL divergence loss function, z can be reduced s and Bx hifu The relative entropy between them makes z in the latent space s The distribution gradually approaches x hifu ; Both the multimodal teacher knowledge integration module and the student knowledge licensing module use the mean square error loss function, and the formula is as follows: where y h is the prediction result, y t For label.

8. According to claim 7, an AI-enabled ultrasonic temperature measurement method for deep organ HIFU treatment, It is characterized in that The multimodal data includes: acoustic parameters, treatment duration, respiratory phase category and ultrasound signal.

9. A computer device, It is characterized in that The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, an AI-enabled ultrasonic temperature measurement method for deep organ HIFU treatment according to any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, implements an AI-enabled ultrasonic temperature measurement method for deep organ HIFU treatment according to any one of claims 1 to 8.

Images