An AI-assisted focused ultrasound focal zone temperature measurement method and system

Abstract

This invention discloses an AI-assisted focused ultrasound focal zone temperature measurement method and system, belonging to the field of non-destructive ultrasound temperature measurement technology. The method acquires medical ultrasound RF data during high-intensity focused ultrasound (HIFU) treatment, generates B-mode ultrasound images, and combines this with thermocouple temperature data to construct training and testing sets. An AI regression model is obtained by training a pre-defined spatial-frequency insight network (SFInsightNet). SFInsightNet includes a spatial feature analysis network and a frequency domain feature analysis network. The former extracts the horizontal and vertical spatial information of the temperature measurement area, while the latter learns frequency domain features from a global perspective. The two modules complement each other, improving temperature measurement accuracy. Finally, the testing set is optimized to improve the AI ​​regression model, achieving accurate single-point ultrasound temperature measurement of body fat. This invention is not only innovative but also improves the real-time performance and accuracy of clinical ultrasound temperature measurement technology, representing a valuable exploration in the field of medical ultrasound technology.

Description

Technical Field

[0001] This invention belongs to the field of ultrasonic non-destructive temperature measurement technology, specifically relating to an AI-assisted focused ultrasonic focal zone temperature measurement method and system. Background Technology

[0002] High-intensity focused ultrasound (HIFU) is a novel non-invasive cancer treatment technology, and it is highly favored as a non-invasive treatment method compared to traditional treatments. Monitoring the focal temperature field is crucial during HIFU treatment to ensure the ablation of necrotic cells and minimize collateral damage to healthy tissues. However, achieving non-invasive, accurate, and real-time temperature monitoring during current HIFU treatments remains a significant challenge in the field of medical engineering.

[0003] Since the last century, various temperature monitoring methods have emerged for HIFU treatment, but all have certain limitations: the most common temperature measurement techniques are thermocouples and magnetic resonance imaging (MRI), but thermocouples contradict the "non-invasive" concept of HIFU and can easily cause cancer cell metastasis; MRI equipment is bulky, expensive, and difficult to popularize, and the frame rate of the temperature field images it generates is low (0.1-1Hz), which cannot provide real-time guidance for HIFU treatment. In summary, neither thermocouple temperature measurement nor MRI temperature measurement can achieve real-time and accurate monitoring of the tissue focal temperature field during HIFU treatment.

[0004] Ultrasonic non-destructive temperature measurement is a classic non-destructive temperature measurement technology. This technology is characterized by low instrument cost, wide applicability, and ease of adoption. Furthermore, the probes used to acquire ultrasonic data can be directly integrated with HIFU treatment probes, achieving a unified solution. It is currently the most promising temperature monitoring technology for clinical HIFU treatment. However, ultrasonic non-destructive temperature measurement still suffers from insufficient accuracy and a low frame rate in temperature field images.

[0005] With the development of artificial intelligence (AI) technology and its integration with ultrasonic non-destructive thermometry, many existing AI-based ultrasonic non-destructive thermometry techniques have made significant progress in improving accuracy, eliminating motion artifacts, and increasing measurement speed. However, current ultrasonic non-destructive thermometry technology still has many areas that urgently need improvement. For example, regarding the objects of measurement, ultrasonic thermometry research mainly focuses on muscle tissue, with less research on adipose tissue thermometry. Liposarcoma, or liposarcoma, is a common cancer, accounting for 20% of tissue tumors and causing immense physical and psychological suffering to countless patients. Considering the differences in structure and acoustic properties between muscle and adipose tissue, which prevent the two thermometry techniques from being fully interchangeable, further research on adipose tissue thermometry is urgently needed. Summary of the Invention

[0006] In order to solve the technical problems existing in the background art, the present invention aims to provide an AI-assisted focused ultrasound focal zone temperature measurement method and system, to provide more technical support for medical staff in the HIFU treatment process, and to benefit a large number of patients.

[0007] To solve the technical problem, the technical solution of the present invention is as follows:

[0008] An AI-assisted focused ultrasound focal zone temperature measurement method, the method comprising:

[0009] Medical ultrasound RF data in the focal zone during HIFU treatment is acquired and processed to obtain B-mode ultrasound images;

[0010] Training and testing sets were constructed using the B-mode ultrasound images and the corresponding thermocouple-recorded temperature data obtained experimentally, and an AI regression model based on a space-frequency domain intelligent network model was trained.

[0011] The AI ​​regression model is optimized based on the test set, and the spatial feature analysis network and the frequency domain feature analysis network are coupled to make full use of the information of the tissue around the measured point and the gray-scale change trend of the measured point to achieve single-point ultrasound temperature measurement of body fat. Among them, the spatial feature analysis module extracts the horizontal and vertical spatial information of the temperature measurement area, while the frequency domain feature analysis module learns the frequency domain features from a global perspective.

[0012] Furthermore, the spatial feature analysis network in the space-frequency domain intelligent network model includes three stages, each stage including multiple trans blocks, a convolutional block, and an average pooling layer;

[0013] The trans block in the spatial feature analysis network model includes four channels: H auxiliary channel, used to extract inter-layer spatial correlation; V auxiliary channel, used to extract intra-layer spatial correlation; H2V main channel, used to fuse the features of the two auxiliary channels; and V2H main channel, used to fuse the features of the two auxiliary channels.

[0014] Furthermore, the input to the spatial feature analysis network model is an ultrasound B-mode image; the ultrasound B-mode image is subjected to convolution processing to achieve downsampling; the downsampled ultrasound B-mode image is subjected to three types of word embedding: (1) the image is divided into multiple 16×16 blocks to form single visual words; (2) the image is divided into multiple layers along the depth H direction to form inter-layer visual words; (3) the image is divided into multiple layers along the horizontal V direction to form intra-layer visual words; the single visual words, inter-layer visual words and intra-layer visual words are flattened to obtain z0, z h and z v .

[0015] Furthermore, the four channel inputs of the l-th trans block in the s-th stage of the spatial feature analysis network model are respectively and The output of the previous layer becomes the input of the next layer, as shown in the following formula:

[0016]

[0017] The final output of the (s-1)th stage of the spatial feature analysis network model is the input of the first trans block of the s-th stage, as shown in the following formula:

[0018]

[0019] The initial inputs of the two auxiliary channels of the trans block are z. h z v The initial input for both main channels is z0;

[0020] The input of the auxiliary channel of the trans block is processed by a transformer neural network to obtain the output, as shown in the following formula:

[0021]

[0022] The main channel input of the trans block is passed through a transformer neural network, added to the output of the auxiliary channel, and then input into the transformer neural network again to obtain the output, as shown in the following formula:

[0023]

[0024] The final output of the trans block in each stage. and Input convolutional block: First, and The model is constructed by reconstructing a two-dimensional tensor from a one-dimensional vector and concatenating the ends of the vector. Then, three convolutions are performed, with batch normalization applied after each convolution to accelerate model convergence. All three convolutions use ReLU as the activation function, and the output z of the convolutional block is... s+1 Then, the average pooling layer is input to obtain the output of each stage.

[0025] Furthermore, the frequency domain feature analysis neural network includes: a frequency domain channel learning block and a frequency domain time learning block;

[0026] The frequency domain channel learning block consists of three parts: the first part is a domain converter based on Fourier transform, which is used to convert the new time domain signal into a frequency domain signal; the second part is a frequency domain multilayer perceptron, which is used to share learnable weights among different channels and capture the interaction and correlation between different variables; the third part is a domain inverter based on inverse Fourier transform, which is used to convert the new frequency domain signal into a time domain signal.

[0027] The frequency domain time learning block comprises three parts: the first part is a domain converter based on Fourier transform, used to convert the new time domain signal into a frequency domain signal; the second part is a frequency domain multilayer perceptron, used to study the frequency domain characteristics within the same channel; and the third part is a domain inverter based on inverse Fourier transform, used to convert the new frequency domain signal into a time domain signal.

[0028] Furthermore, the training of the AI ​​regression model based on the spatial feature analysis network model and the frequency domain feature analysis network model includes:

[0029] The frequency domain feature analysis network converts time domain information into the frequency domain to learn global features from a global perspective, and combines it with the spatial features learned by the spatial feature analysis network to perform temperature regression calculation.

[0030] The input to the frequency domain feature analysis network is a B-mode ultrasound image. From the acquired B-mode ultrasound images, after selecting 10 consecutive frames preceding time t, a 16×16 small image block {p} centered on the measured point is cropped. t-9 ,p t-8, ...,p t After straightening the image patch, the tensor {x} is obtained. t-9 ,x t-8, ...,x t |x i ∈R N}, X t =[x t-9 ,x t-8, ...,x t ]∈R N×L Here, L=10, and the input frequency domain feature analysis network is used to infer the temperature Y corresponding to time t. t ;

[0031] For the tensor X t Expand the dimension by multiplying it by a learnable vector. The formula used to obtain more hidden information is as follows:

[0032]

[0033] The tensor H t ∈R N×L×dThis serves as the input to the frequency domain channel learning block; it includes Fourier transform to convert time-domain information to frequency-domain information, and inverse Fourier transform to convert frequency-domain information to time-domain information, with the formulas as follows:

[0034]

[0035] Where f is the frequency, v is the integral variable, j is the imaginary number, H(v)cos(2πfv) is the real part, and H(v)sin(2πfv) is the imaginary part;

[0036]

[0037] Where f is the integration variable, Re(h(f)) is the real part, and Im(h(f)) is the imaginary part;

[0038] When considering time series forecasting, channel dependence can provide insights into the interactions and correlations between different variables, thereby improving prediction accuracy. Therefore, frequency domain channel learning blocks share weights across L timestamps to understand channel relationships: for input H... t ∈R N×L×d Consider the timestamp corresponding to the l-th timestamp Input it into the frequency domain channel learning block:

[0039]

[0040] Where, fourier() and Fourier() represent the Fourier transform and inverse Fourier transform, respectively. express The frequency domain components, and the weights and biases satisfy:

[0041]

[0042] The output Z∈R of the frequency domain channel learning block N×L×d As input to the frequency domain time learning block, consider the nth channel corresponding to Input it into the frequency domain time learning block:

[0043]

[0044] S t ∈R N×L×d After being straightened into a one-dimensional vector, it is connected to the output features of the spatial feature analysis network and then input into a fully connected layer for regression calculation, which is used to perform AI regression calculation of the temperature at time t.

[0045] Furthermore, the features output from the spatial feature analysis network model and the frequency domain feature analysis network model are concatenated and then input into a fully connected layer for regression calculation to infer the temperature at the measured point. The spatial feature analysis network model and the frequency domain feature analysis network model calculate the loss function based on the mean square error, and the formula for the loss function is as follows:

[0046]

[0047] Where n represents the batch size of the model; Indicates the temperature recorded by the corresponding thermocouple; y i =f(x) i ), representing the predicted temperature result, x i For input data.

[0048] An AI-assisted focused ultrasonic focal zone temperature measurement system, characterized in that the system is used to perform any of the methods described above, the system comprising:

[0049] The data acquisition module is used to acquire and process medical ultrasound RF data in the focal zone during HIFU treatment to obtain B-mode ultrasound images.

[0050] The model training module is used to construct training and testing sets using the B-mode ultrasound images and the corresponding thermocouple temperature data obtained from the experiment, and to train an AI regression model based on a space-frequency domain intelligent network model.

[0051] The model optimization module is used to optimize the AI ​​regression model based on the test set. It couples the spatial feature analysis network with the frequency domain feature analysis network to make full use of the information of the tissue around the test point and the gray-scale change trend of the test point to achieve single-point ultrasound temperature measurement of body fat. The spatial feature analysis module extracts the horizontal and vertical spatial information of the temperature measurement area, while the frequency domain feature analysis module learns the frequency domain features from a global perspective.

[0052] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement an AI-assisted focused ultrasonic focal zone temperature measurement method as described above.

[0053] A computer-readable storage medium storing a computer program that, when executed by a processor, implements an AI-assisted focused ultrasonic focal zone temperature measurement method as described above.

[0054] Compared with the prior art, the advantages of the present invention are as follows:

[0055] This invention collects RF data from within a living organism, processes the data to obtain B-mode ultrasound images, constructs training and testing sets based on the ultrasound-acquired data and the temperatures recorded by hotspots in experiments, and trains an optimal AI regression model based on a deep learning model to achieve single-point ultrasound thermometry of body fat. This effectively saves manpower, shortens measurement time, and improves measurement accuracy. Attached Figure Description

[0056] Figure 1 This is a schematic diagram of the structure of the AI-assisted focused ultrasound focal zone temperature measurement system disclosed in this invention;

[0057] Figure 2 This is a schematic diagram of the spatial-frequency domain intelligent network model 200 of the AI-assisted focused ultrasound focal domain temperature measurement method disclosed in this invention.

[0058] Figure 3 This is a schematic diagram of the spatial feature analysis network and frequency domain feature analysis network model 300 disclosed in this invention. Detailed Implementation

[0059] The specific implementation of the present invention is described below with reference to embodiments:

[0060] It should be noted that the structures, proportions, sizes, etc. shown in this specification are only used to complement the content disclosed in the specification for those skilled in the art to understand and read, and are not intended to limit the conditions under which the present invention can be implemented. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0061] Furthermore, the terms such as "upper," "lower," "left," "right," "middle," and "one" used in this specification are merely for clarity of description and are not intended to limit the scope of the invention. Any changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention.

[0062] Example 1:

[0063] Existing non-destructive ultrasound thermometry lacks technology for measuring body fat during HIFU treatment. Therefore, this embodiment proposes an AI-assisted focused ultrasound focal-area thermometry system and method.

[0064] Figure 1 This is a schematic diagram of the structure of an AI-assisted focused ultrasound focal zone temperature measurement system 100 according to the first embodiment of this disclosure. This AI-assisted single-point ultrasound temperature measurement system for body fat during HIFU treatment includes the following functional modules: a data acquisition module 101, a model training module 102, and a model optimization module 103.

[0065] The data acquisition module 101 is used to acquire ultrasound RF data of the HIFU treatment area and process the data to obtain B-mode ultrasound images. In addition, the data acquisition module 101 also uses thermocouples to acquire temperature data of the focal zone.

[0066] The raw RF data is first filtered to remove noise and artifacts, improving signal quality. Then, envelope detection (e.g., Hilbert transform) is used to extract the envelope signal from the RF signal, i.e., the echo intensity of the tissue. The extracted envelope data is then converted into grayscale values ​​for imaging processing to obtain a B-mode ultrasound image.

[0067] The model training module 102 is used to build a training set based on a deep learning model, using B-mode ultrasound images and corresponding thermocouple temperature data, and train an AI regression model.

[0068] In this embodiment, the B-mode ultrasound image obtained by the data acquisition module 101 and the temperature data recorded by the corresponding thermocouple are divided into two parts, a training set and a test set, in a ratio of 4:1. The test set is used to train the AI ​​regression model. Based on the calculation result of the mean squared error loss function, the backpropagation algorithm is used to update the parameters of the AI ​​regression model, thereby realizing the training process of the AI ​​regression model.

[0069] The model optimization module 103 is used to optimize the AI ​​regression model from the test set constructed from the B-mode ultrasound images and the corresponding thermocouple recording data, so that the model training module trains the best AI regression model.

[0070] In this example, a test set constructed from B-mode ultrasound images and corresponding thermocouple-recorded temperature data is used to perform performance testing on the AI ​​regression model. The main evaluation method is the mean squared error loss function, which assesses the difference between the AI ​​regression model's predicted results and the thermocouple-recorded results. The closer the difference in the loss function is to zero, the stronger the performance of the AI ​​regression training model. The test results can guide model training and adjust the hyperparameters of the neural network model to achieve optimal performance, enabling the AI ​​regression model to predict temperatures as close as possible to the thermocouple-recorded temperatures at time t, given the input of B-mode ultrasound images at time t and the preceding nine frames.

[0071] Figure 2 This is a schematic diagram of the spatial-frequency domain intelligent network model 200 of the AI-assisted focused ultrasound focal domain temperature measurement method according to an embodiment of the present disclosure.

[0072] In this example, the space-frequency domain intelligent neural network model 200 comprises two main parts: a spatial feature analysis network 301, used to extract spatial information from B-mode ultrasound images; and a frequency domain feature analysis network 302, which converts time-domain signals into frequency-domain signals, enabling the neural network to learn features from a global perspective. Finally, the outputs of the two parallel neural networks are combined and input into a fully connected layer to enable the AI ​​regression model to predict temperature.

[0073] Figure 3 This is a schematic diagram of the spatial feature analysis network model 301 and the frequency domain feature analysis network 302 of the AI-assisted focused ultrasound focal domain temperature measurement method according to an embodiment of the present disclosure.

[0074] In this embodiment, a spatial feature analysis network and a frequency domain feature analysis network are combined. The spatial feature analysis network starts from spatial information. Considering that a B-mode ultrasound image centered on adipose tissue may contain information from the superficial epidermis and deeper muscles, and that acoustic parameters such as sound attenuation coefficients and sound velocities, as well as their rates of change with temperature, differ between different tissues, temperature calculation can be attempted based on the correlation information between layers. Therefore, this spatial feature analysis network not only includes traditional single-layer visual terms, but also incorporates intra-layer and inter-layer visual terms that are divided horizontally and vertically. Intra-layer visual terms can be used to extract spatial information at the same depth, while inter-layer visual terms are used to extract correlation information between different tissues at different depths. The single-layer, inter-layer, and intra-layer visual terms are flattened to obtain z0, z... h and z v The initial inputs of the two auxiliary channels of the trans block are z. h z v The initial inputs for both main channels are z0. Considering the spatial feature analysis network model, the four channel inputs of the l-th trans block in the s-th stage are respectively... and The output of the previous layer becomes the input of the next layer, as shown in the following formula:

[0075]

[0076] Where f() represents the function of the trans block. In this spatial feature analysis network model, the final output of the (s-1)th stage is the input of the first trans block of the s-th stage, as shown in the following formula:

[0077]

[0078] To address the limitation of spatial feature analysis networks in not considering the changing trends of the measured points, a frequency domain feature analysis network is introduced. This network converts time-domain information to the frequency domain, enabling the learning of global features from a global perspective. These features are then combined with the spatial features learned by the spatial feature analysis network to perform temperature regression calculations. The input to the frequency domain feature analysis network is 10 consecutive frames of processed B-mode ultrasound images. A 16×16 B-mode ultrasound image centered on the measured temperature point is cropped to obtain {p t-9 ,p t-8, ...,p t Then straighten it to get 10 one-dimensional vectors {x}. t-9 ,x t-8, ...,x t |x i ∈R N}, X t =[x t-9 ,x t-8, ...,x t ]∈R N×L (Here, L = 10) serves as the input to the frequency domain feature analysis network. For tensor X... t Expand the dimension by multiplying it by a learnable vector. The formula used to obtain more hidden information is as follows:

[0079]

[0080] Obtain tensor H t ∈R N×L×d This serves as the input to the frequency domain channel learning block. It includes Fourier transform to convert time-domain information to frequency-domain information, and inverse Fourier transform to convert frequency-domain information back to time-domain information, with the formulas as follows:

[0081]

[0082] Where f is the frequency, v is the integral variable, j is the imaginary number, H(v)cos(2πfv) is the real part, and H(v)sin(2πfv) is the imaginary part.

[0083]

[0084] Where f is the integration variable, Re(h(f)) is the real part, and Im(h(f)) is the imaginary part.

[0085] When considering time series forecasting, channel dependence can provide insights into the interactions and correlations between different variables, thereby improving prediction accuracy. Therefore, frequency domain channel learning blocks share weights across L timestamps to understand channel relationships: for input H... t ∈R N×L×d Consider the timestamp corresponding to the l-th timestamp Input it into the frequency domain channel learning block:

[0086]

[0087] Where fourier() and Fourier() represent the Fourier transform and inverse Fourier transform, respectively. express The frequency domain components, and the weights and biases satisfy:

[0088]

[0089] The output Z∈R of the frequency domain channel learning block N×L×d As input to the frequency domain time learning block, consider the nth channel corresponding to Input it into the frequency domain time learning block:

[0090]

[0091] S t ∈R N×L×d After being straightened into a one-dimensional vector, it is connected to the output features of the spatial feature analysis network and then input into a fully connected layer for regression calculation, which is used to perform AI regression calculation of the temperature at time t.

[0092] According to a preferred embodiment, the neural network model combining spatial feature analysis and frequency domain feature analysis calculates a loss function based on mean square error, and the formula for the loss function is as follows:

[0093]

[0094] Where n represents the batch size of the model; Indicates the temperature recorded by the corresponding thermocouple; y i =f(x) i ), representing the predicted temperature result, x i For input data.

[0095] The AI-assisted focused ultrasound focal zone temperature measurement system and method disclosed herein have good temperature monitoring capabilities, can provide technical reference for the field of ultrasound non-destructive temperature measurement technology, and contribute to the wider application of HIFU treatment technology.

[0096] Example 2:

[0097] This embodiment discloses an AI-assisted focused ultrasound focal zone temperature measurement system. The system is used to execute the aforementioned AI-assisted focused ultrasound focal zone temperature measurement method. Specifically, the system includes:

[0098] The data acquisition module is used to acquire and process medical ultrasound RF data in the focal zone during HIFU treatment to obtain B-mode ultrasound images.

[0099] The model training module is used to construct training and testing sets using the B-mode ultrasound images and the corresponding thermocouple temperature data obtained from the experiment, and to train an AI regression model based on a preset spatial feature analysis network model and a frequency domain feature analysis network model.

[0100] The model optimization module is used to optimize the AI ​​regression model based on the test set to obtain the optimal AI regression model, which enables single-point ultrasound temperature measurement of body fat.

[0101] Example 3:

[0102] This embodiment provides a terminal device, which includes a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve a corresponding method flow or function. The processor described in this embodiment can be used in the operation of an AI-assisted focused ultrasonic focal zone temperature measurement method, including the following steps:

[0103] Medical ultrasound RF data in the focal zone during HIFU treatment is acquired and processed to obtain B-mode ultrasound images;

[0104] Training and testing sets were constructed using the B-mode ultrasound images and the corresponding thermocouple temperature data obtained experimentally, and an AI regression model based on a spatial feature analysis network model and a frequency domain feature analysis network model was trained.

[0105] The AI ​​regression model is optimized based on the test set to achieve single-point ultrasound temperature measurement of body fat.

[0106] Example 4:

[0107] This embodiment provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a terminal device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and extended storage media supported by the terminal device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device.

[0108] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the AI-assisted focused ultrasonic focal zone temperature measurement method in the above embodiments; one or more instructions in the computer-readable storage medium are loaded and executed by the processor to perform the following steps:

[0109] Medical ultrasound RF data in the focal zone during HIFU treatment is acquired and processed to obtain B-mode ultrasound images;

[0110] Training and testing sets were constructed using the B-mode ultrasound images and the corresponding thermocouple temperature data obtained experimentally, and an AI regression model based on a spatial feature analysis network model and a frequency domain feature analysis network model was trained.

[0111] The AI ​​regression model is optimized based on the test set to achieve single-point ultrasound temperature measurement of body fat.

[0112] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied 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.

[0113] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0114] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0115] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0116] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

[0117] Many other changes and modifications can be made without departing from the concept and scope of this invention. It should be understood that this invention is not limited to the specific embodiments, and the scope of this invention is defined by the appended claims.

Claims

1. An AI-assisted focused ultrasonic focal zone temperature measurement method, characterized in that, The method includes: Medical ultrasound RF data in the focal zone during HIFU treatment is acquired and processed to obtain B-mode ultrasound images; Training and testing sets were constructed using the B-mode ultrasound images and the corresponding thermocouple temperature data obtained experimentally, and an AI regression model based on spatial feature analysis network and frequency domain feature analysis network was trained. The AI ​​regression model is optimized based on the test set. The frequency domain feature analysis network converts time domain information into frequency domain to learn global features from a global perspective, and combines it with the spatial features learned by the spatial feature analysis network to perform temperature regression calculation. The spatial feature analysis network consists of three stages, each stage including multiple trans blocks, a convolutional block, and an average pooling layer; The trans block in the spatial feature analysis network includes four channels: H auxiliary channel, used to extract inter-layer spatial correlation; V auxiliary channel, used to extract intra-layer spatial correlation; H2V main channel, used to fuse the features of the two auxiliary channels; and V2H main channel, used to fuse the features of the two auxiliary channels. The input to the spatial feature analysis network is an ultrasound B-mode image; the ultrasound B-mode image is subjected to convolution processing to achieve downsampling; the downsampled ultrasound B-mode image is subjected to three types of word embedding: (1) the image is divided into multiple 16×16 blocks to form single visual words; (2) the image is divided into multiple layers along the depth V direction to form inter-layer visual words; (3) the image is divided into multiple layers along the horizontal H direction to form intra-layer visual words; the single visual words, inter-layer visual words and intra-layer visual words are flattened to obtain , and ; The four channel inputs of the l-th trans block in the s-th stage of the spatial feature analysis network are respectively , , and The output of the previous layer becomes the input of the next layer, as shown in the following formula: ， The final output of the (s-1)th stage of the spatial feature analysis network is the input of the first trans block of the s-th stage, as shown in the following formula: ， The initial inputs of the two auxiliary channels of the trans block are respectively , The initial inputs for both main channels are ; The input of the auxiliary channel of the trans block is processed by a transformer neural network to obtain the output, as shown in the following formula: ， ， The main channel input of the trans block is passed through a transformer neural network, added to the output of the auxiliary channel, and then input into the transformer neural network again to obtain the output, as shown in the following formula: ， ， The final output of the trans block in each stage. and Input convolutional block: First, and The model is converted from a one-dimensional vector into a two-dimensional tensor and concatenated end to end. Then, three convolutions are performed. After each convolution, batch normalization is performed to accelerate the convergence speed of the model. ReLU is used as the activation function for all three convolutions. After the output of the convolution block, it is input into the average pooling layer to obtain the output of each stage.

2. The AI-assisted focused ultrasound focal zone temperature measurement method according to claim 1, characterized in that, The frequency domain feature analysis network includes: a frequency domain channel learning block and a frequency domain time learning block; The frequency domain channel learning block comprises three parts: the first part is a domain converter based on Fourier transform, used to convert time domain signals into frequency domain signals; the second part is a frequency domain multilayer perceptron, used to share learnable weights among different channels and capture the interaction and correlation between different variables; and the third part is a domain inverter based on inverse Fourier transform, used to convert frequency domain signals into time domain signals. The frequency domain time learning block comprises three parts: the first part is a domain converter based on Fourier transform, used to convert time domain signals into frequency domain signals; the second part is a frequency domain multilayer perceptron, used to study frequency domain characteristics within the same channel; and the third part is a domain inverter based on inverse Fourier transform, used to convert frequency domain signals into time domain signals.

3. The AI-assisted focused ultrasound focal zone temperature measurement method according to claim 1, characterized in that, AI regression models based on spatial feature analysis networks and frequency domain feature analysis networks include: The input to the frequency domain feature analysis network is a B-mode ultrasound image. From the acquired B-mode ultrasound images, 10 consecutive frames preceding time t are selected, and a 16×16 image block centered on the measured point is cropped. After straightening the image patch, a tensor is obtained. ,Will Here, L=10, and the input frequency domain feature analysis network is used to infer the temperature at time t. ; For the tensor Expand the dimension by multiplying it by a learnable vector. This is used to obtain more hidden information, and the formula is as follows: ， The tensor This serves as the input to the frequency domain channel learning block; it includes Fourier transform to convert time-domain information to frequency-domain information, and inverse Fourier transform to convert frequency-domain information to time-domain information, with the formulas as follows: ， in, It's frequency. It is an integral variable. It is an imaginary number. For the real part, It is the imaginary part; ， in, It is an integral variable For the real part, It is the imaginary part; When considering time series forecasting, channel dependence can provide insights into the interactions and correlations between different variables, thereby improving prediction accuracy. Therefore, frequency domain channel learning blocks share weights across L timestamps to understand channel relationships: for input... Consider the timestamp corresponding to the l-th timestamp Input it into the frequency domain channel learning block: ， ， ， in, and These represent the Fourier transform and the inverse Fourier transform, respectively. express The frequency domain components, and the weights and biases satisfy: ， ， The output of the frequency domain channel learning block As input to the frequency domain time learning block, consider the nth channel corresponding to Input it into the frequency domain time learning block: ， ， ， Will After being straightened into a one-dimensional vector, it is connected to the output features of the spatial feature analysis network and then input into a fully connected layer for regression calculation, which is used to perform AI regression calculation of the temperature at time t.

4. The AI-assisted focused ultrasound focal zone temperature measurement method according to claim 1, characterized in that, The features output from the spatial feature analysis network and the frequency domain feature analysis network are concatenated and then input into a fully connected layer for regression calculation to infer the temperature of the measured point. The spatial feature analysis network and the frequency domain feature analysis network calculate the loss function based on the mean square error. The formula for the loss function is as follows: ， Where n represents the batch size of the model; This indicates the temperature recorded by the corresponding thermocouple; This indicates the predicted temperature result.

5. An AI-assisted focused ultrasonic focal zone temperature measurement system, characterized in that, The system is used to perform the method according to any one of claims 1-4, the system comprising: The data acquisition module is used to acquire and process medical ultrasound RF data in the focal zone during HIFU treatment to obtain B-mode ultrasound images. The model training module is used to construct training and testing sets using the B-mode ultrasound images and the corresponding thermocouple temperature data obtained from the experiment, and to train an AI regression model based on spatial feature analysis network and frequency domain feature analysis network. The model optimization module is used to optimize the AI ​​regression model based on the test set. It couples the spatial feature analysis network and the frequency domain feature analysis network to make full use of the information of the tissue around the test point and the gray-scale change trend of the test point to achieve single-point ultrasound temperature measurement of body fat. The spatial feature analysis network extracts the horizontal and vertical spatial information of the temperature measurement area, while the frequency domain feature analysis network learns the frequency domain features from a global perspective.

6. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement an AI-assisted focused ultrasonic focal zone temperature measurement method according to any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements an AI-assisted focused ultrasonic focal zone temperature measurement method according to any one of claims 1 to 4.

Images