Ultrasonic signal denoising method and device, model training method and device and storage medium

Abstract

The embodiment of the invention discloses an ultrasonic signal denoising method and device, a model training method and device and a storage medium, and the method comprises the steps: receiving a plurality of pieces of first ultrasonic channel data returned by a target tissue through an ultrasonic probe, inputting each piece of first ultrasonic channel data into a machine learning model, and carrying out the denoising processing, and then beam forming processing is performed on the multiple pieces of second ultrasonic channel data after denoising processing to obtain synthesized radio frequency data, and an image of the target tissue is obtained based on the synthesized radio frequency data and is displayed, so that noise in the ultrasonic echo signals is effectively removed. Wherein the machine learning model can remove noise data from the ultrasonic signal, so that the signal-to-noise ratio of the ultrasonic signal is improved, and the imaging quality of the ultrasonic image is improved. Based on this, the ultrasonic signal denoising method provided by the embodiment of the invention can effectively remove noise in the ultrasonic echo signal, improve the signal-to-noise ratio of the ultrasonic signal, and improve the imaging quality of the ultrasonic image.

Description

Technical Field

[0001] This application relates to the field of ultrasonic signal denoising technology, and in particular to a method for denoising ultrasonic signals, a model training method, a model training device, and a storage medium. Background Technology

[0002] Medical ultrasound imaging diagnostic equipment utilizes the propagation of ultrasound waves within the human body to obtain ultrasound characteristic information of human tissues and organs. Currently, medical ultrasound imaging diagnostic equipment is widely used in medical diagnostic scenarios. During transmission, the ultrasound echo signal passes through hardware modules such as ultrasound transducers, amplifiers, AD sampling devices, and digital demodulators. Various types of noise are generated in the ultrasound echo signal during this process, interfering with the final ultrasound imaging and thus affecting the quality of the ultrasound image. Currently, filters are used in the ultrasound receiving circuit to remove noise from the ultrasound signal. However, since different filters are often only effective against certain types of noise, the overall noise reduction capability of existing ultrasound signal denoising methods is not ideal. As the requirements for ultrasound image accuracy increase, current ultrasound denoising effects cannot meet clinical needs. Summary of the Invention

[0003] The following is an overview of the subject matter described in detail herein. This overview is not intended to limit the scope of the claims.

[0004] This application provides a method for denoising ultrasonic signals, a model training method, a model training device, and a storage medium, which can effectively remove noise from ultrasonic echo signals, improve the signal-to-noise ratio of ultrasonic signals, and improve the imaging quality of ultrasonic images.

[0005] In a first aspect, embodiments of this application provide a method for denoising ultrasonic signals, comprising:

[0006] Control the ultrasound probe to emit the first ultrasound wave toward the target tissue;

[0007] The ultrasound probe is controlled to receive the echo channel signal of the first ultrasound returned by the target tissue. The echo channel signal includes multiple first ultrasound channel data, and the multiple first ultrasound channel data correspond one-to-one with multiple receiving array elements on the ultrasound probe.

[0008] Multiple first ultrasound channel data are input into a machine learning model for noise reduction to obtain multiple second ultrasound channel data.

[0009] Beamforming processing is performed on multiple second ultrasound channel data to obtain synthesized radio frequency data;

[0010] Image data of the target tissue is obtained by performing image processing based on the synthesized radio frequency data;

[0011] An image of the target tissue is obtained based on the image data, and the image of the target tissue is displayed.

[0012] Secondly, embodiments of this application provide a method for denoising ultrasonic signals, comprising:

[0013] Control the ultrasound probe to emit the first ultrasound wave toward the target tissue;

[0014] The ultrasound probe is controlled to receive the echo channel signal of the first ultrasound returned by the target tissue. The echo channel signal includes multiple first ultrasound channel data, and the multiple first ultrasound channel data correspond one-to-one with multiple receiving array elements on the ultrasound probe.

[0015] Multiple first ultrasound channel data are input into a machine learning model for noise reduction to obtain multiple second ultrasound channel data.

[0016] The synthesized radio frequency data is obtained by weighted summation of multiple second ultrasound channel data.

[0017] Image data of the target tissue is obtained by performing image processing based on the synthesized radio frequency data;

[0018] An image of the target tissue is obtained based on the image data, and the image of the target tissue is displayed.

[0019] Thirdly, embodiments of this application provide a method for denoising ultrasonic signals, comprising:

[0020] The first ultrasonic wave is emitted into the target tissue through an ultrasonic probe;

[0021] The ultrasound probe receives the echo channel signal of the first ultrasound wave returned by the target tissue, and the echo channel signal includes multiple first ultrasound channel data corresponding to multiple receiving array elements on the ultrasound probe.

[0022] The data from each of the first ultrasound channels are merged to obtain the first combined channel data.

[0023] The first combined channel data is input into a machine learning model for noise reduction to obtain the second combined channel data.

[0024] Image data of the target tissue is obtained by performing image processing based on the second combined channel data;

[0025] An image of the target tissue is obtained based on the image data, and the image of the target tissue is displayed.

[0026] Fourthly, embodiments of this application provide a model training method, including:

[0027] Obtain a training sample dataset and a validation dataset, wherein the training sample dataset includes training sample data of multiple first ultrasound channels corresponding to multiple receiving array elements on an ultrasound probe, and the validation dataset includes valid data from the training sample data of the first ultrasound channels;

[0028] The first ultrasound channel training sample data is input into the machine learning model for noise reduction to obtain ultrasound channel prediction data.

[0029] The parameters of the machine learning model are corrected based on the ultrasound channel prediction data and the valid data.

[0030] Fifthly, embodiments of this application provide a model training method, including:

[0031] Acquire a training sample dataset and a validation dataset, wherein the training sample dataset includes training sample data of multiple first ultrasound channels corresponding to multiple receiving array elements on an ultrasound probe, and the validation dataset includes target noise data in the training sample data of the first ultrasound channels;

[0032] The training sample data of the first ultrasound channel is input into the machine learning model for noise extraction to obtain noise prediction data.

[0033] The parameters of the machine learning model are corrected based on the noise prediction data and the target noise data.

[0034] Sixthly, embodiments of this application provide an ultrasound imaging device, comprising:

[0035] Ultrasonic probe;

[0036] A transmitting / receiving circuit is used to control the ultrasound probe to transmit ultrasound waves to the target tissue and receive the echoes of the ultrasound waves.

[0037] A processor, configured to denoise the echo of the ultrasound to obtain an image of the target tissue;

[0038] A display for displaying the tissue image;

[0039] The processor is also used to execute the ultrasonic signal denoising method of the embodiments of the first, second and third aspects described above.

[0040] Seventhly, embodiments of this application provide a model training apparatus, comprising:

[0041] At least one memory for storing programs;

[0042] At least one processor is configured to load the program to execute the model training method described in the fourth and fifth aspect embodiments.

[0043] Eighthly, embodiments of this application provide a computer storage medium storing a computer program applied to an ultrasonic imaging device. When the computer program is executed by a processor, it implements the ultrasonic signal denoising method as described in the first, second, and third aspects of the above embodiments, or the model training method as described in the fourth and fifth aspects of the embodiments.

[0044] Ninthly, embodiments of this application provide a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the ultrasonic signal denoising method as described in the first, second, and third aspects of the embodiments above, or the model training method as described in the fourth and fifth aspects of the embodiments.

[0045] The embodiments of this application include at least the following beneficial effects:

[0046] On one hand, an embodiment of this application provides a method for denoising ultrasonic signals. This method utilizes an ultrasonic probe to receive multiple first ultrasonic channel data returned by a target tissue. Each first ultrasonic channel data point is input into a machine learning model for denoising processing. Then, beamforming is performed on the denoised multiple second ultrasonic channel data to obtain synthesized radio frequency (RF) data. An image of the target tissue is obtained and displayed based on the synthesized RF data, thereby effectively removing noise from the ultrasonic echo signal. The machine learning model can remove noise data from the ultrasonic signal, thereby improving the signal-to-noise ratio (SNR) of the ultrasonic signal and enhancing the imaging quality of the ultrasonic image. Based on this, the ultrasonic signal denoising method of this application can effectively remove noise from the ultrasonic echo signal, improve the SNR of the ultrasonic signal, and enhance the imaging quality of the ultrasonic image.

[0047] On the other hand, one embodiment of this application provides a method for denoising ultrasound signals. This method utilizes an ultrasound probe to receive multiple first ultrasound channel data returned by a target tissue. Each first ultrasound channel data point is input into a machine learning model for denoising processing. Then, the denoised multiple second ultrasound channel data points are weighted and summed to obtain synthetic radio frequency (RF) data. An image of the target tissue is obtained and displayed based on the synthetic RF data, thereby effectively removing noise from the ultrasound echo signal. The machine learning model can remove noise data from the ultrasound signal, thereby improving the signal-to-noise ratio (SNR) of the ultrasound signal and enhancing the imaging quality of the ultrasound image. Therefore, the ultrasound signal denoising method of this application can effectively remove noise from the ultrasound echo signal, improve the SNR of the ultrasound signal, and enhance the imaging quality of the ultrasound image.

[0048] On the other hand, one embodiment of this application provides a method for denoising ultrasound signals. This method utilizes an ultrasound probe to receive multiple first ultrasound channel data returned by the target tissue. The data from each first ultrasound channel is merged to obtain first combined channel data. This first combined channel data is then input into a machine learning model for denoising, achieving multi-channel combined data processing and thus improving processing efficiency. The denoised second combined channel data is then processed to obtain image data. Based on this image data, an image of the target tissue is obtained and displayed, effectively removing noise from the ultrasound echo signal. The machine learning model can remove noise data from the ultrasound signal, thereby improving the signal-to-noise ratio of the ultrasound signal and enhancing the imaging quality of the ultrasound image. Therefore, the ultrasound signal denoising method of this application can effectively remove noise from the ultrasound echo signal, improve the signal-to-noise ratio of the ultrasound signal, enhance the imaging quality of the ultrasound image, and improve denoising processing efficiency.

[0049] On the other hand, one embodiment of this application provides a model training method that inputs the first ultrasound channel training sample data into a machine learning model for denoising processing to obtain ultrasound channel prediction data. Then, the ultrasound channel prediction data is compared with the effective data to correct the parameters of the machine learning model. This allows the trained machine learning model to learn to extract noise features from the ultrasound signal and continuously correct the ultrasound signal prediction data, thereby improving the signal-to-noise ratio of the ultrasound signal and enhancing the imaging quality of the ultrasound image.

[0050] On the other hand, one embodiment of this application provides a model training method that inputs the training sample data of the first ultrasound channel into a machine learning model to extract noise, obtains noise prediction data, and then compares the noise prediction data with the target noise data to correct the parameters of the machine learning model. This allows the trained machine learning model to learn the noise features in the ultrasound signal and continuously correct the noise in the ultrasound signal, so as to subtract the noise data from the original ultrasound data, thereby improving the signal-to-noise ratio of the ultrasound signal and improving the imaging quality of the ultrasound image. Attached Figure Description

[0051] The accompanying drawings are used to provide a further understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application.

[0052] Figure 1 This is a structural block diagram of an ultrasound imaging device provided in one embodiment of this application;

[0053] Figure 2 This is a flowchart of a method for denoising ultrasonic signals according to an embodiment of this application;

[0054] Figure 3A This is a schematic diagram of the internal structure of a signal processing module provided in one embodiment of this application;

[0055] Figure 3B This is a schematic diagram of the internal structure of a signal processing module provided in another embodiment of this application;

[0056] Figure 4 This is a flowchart of steps 410 to 420 provided in one embodiment of this application;

[0057] Figure 5 This is a flowchart of steps 510 to 530 provided in one embodiment of this application;

[0058] Figure 6 This is a flowchart of a method for denoising ultrasonic signals provided in another embodiment of this application;

[0059] Figure 7 This is a flowchart of steps 710 to 730 provided in one embodiment of this application;

[0060] Figure 8 This is a flowchart of a method for denoising ultrasonic signals provided in another embodiment of this application;

[0061] Figure 9 This is a flowchart of steps 910 to 930 provided in one embodiment of this application;

[0062] Figure 10 This is a flowchart of steps 1010 to 1020 provided in one embodiment of this application;

[0063] Figure 11 This is a flowchart of a model training method provided in one embodiment of this application;

[0064] Figure 12 This is a schematic diagram of a model training process provided in one embodiment of this application;

[0065] Figure 13 This is a flowchart of steps 1310 to 1330 provided in one embodiment of this application;

[0066] Figure 14 This is a flowchart of steps 1410 to 1420 provided in one embodiment of this application;

[0067] Figure 15 This is a flowchart of steps 1510 to 1520 provided in one embodiment of this application;

[0068] Figure 16 This is a flowchart of steps 1610 to 1620 provided in one embodiment of this application;

[0069] Figure 17 This is a flowchart of steps 1710 to 1730 provided in one embodiment of this application;

[0070] Figure 18 This is a flowchart of a model training method provided in another embodiment of this application;

[0071] Figure 19 This is a flowchart of steps 1910 to 1930 provided in one embodiment of this application;

[0072] Figure 20 This is a flowchart of steps 2010 to 2020 provided in one embodiment of this application;

[0073] Figure 21 This is a flowchart of steps 2110 to 2130 provided in one embodiment of this application;

[0074] Figure 22 This is a flowchart of steps 2210 to 2230 provided in one embodiment of this application;

[0075] Figure 23 This is a flowchart of steps 2310 to 2340 provided in one embodiment of this application. Detailed Implementation

[0076] The present application will be further described below with reference to the accompanying drawings and specific embodiments. The described embodiments should not be considered as limitations on the present application, and all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of the present application.

[0077] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0078] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0079] Medical ultrasound imaging diagnostic equipment utilizes the propagation of ultrasound waves through the human body to obtain ultrasound images of human tissues and organs. These images include tissue images of target tissues (such as the heart, lungs, and ovaries). Doctors can use their experience to identify target tissue structures within these images (e.g., blood vessels in the heart, alveoli in the lungs, follicles in the ovaries), observe and / or measure the state of these structures to obtain a diagnostic result. However, diagnosis based solely on visual observation and experience is prone to errors.

[0080] Currently, medical ultrasound imaging diagnostic equipment is widely used in medical diagnostic scenarios. During the transmission of ultrasound signals, the echo signals pass through hardware modules such as ultrasound transducers, amplifiers, AD sampling devices, and digital demodulators. Various types of noise are generated in the ultrasound echo signals during this process, interfering with the final ultrasound imaging and thus affecting the quality of the ultrasound images. Currently, filters are used in the ultrasound receiving circuit to remove noise from the ultrasound signals. However, since different filters are often only effective against certain types of noise, the overall noise reduction capability of existing ultrasound signal denoising methods is not ideal. As the requirements for ultrasound image accuracy increase, current ultrasound denoising effects cannot meet clinical needs.

[0081] Based on this, the present invention proposes a noise reduction method, a model training method, a model training device, and a storage medium for ultrasonic signals, which effectively removes noise from ultrasonic echo signals, improves the signal-to-noise ratio of ultrasonic signals, and improves the imaging quality of ultrasonic images.

[0082] This invention provides an ultrasound imaging device that improves the ease of operation and efficiency of editing ultrasound recognition images. For example... Figure 1The diagram shows a structural block diagram of an ultrasound imaging device. The ultrasound imaging device 10 may include a probe 100, a transmitting circuit 101, a transmitting / receiving selection switch 102, a receiving circuit 103, a beamforming circuit 104, a processor 105, a display 106, a memory 107, and a signal processing module 108.

[0083] The ultrasound probe 100 includes a transducer (not shown) composed of multiple array elements arranged in an array. These elements can be arranged in a row to form a linear array, or in a two-dimensional matrix to form a planar array; they can also form a convex array. Each element is used to emit an ultrasonic beam based on an excitation electrical signal, or to convert a received ultrasonic beam into an electrical signal. Therefore, each element can be used to achieve the mutual conversion between electrical pulse signals and ultrasonic beams, thereby enabling the emission of ultrasonic waves to target tissues in the human body (e.g., the heart, lungs, ovaries, etc.), and can also be used to receive echoes of ultrasonic waves reflected back from the tissue. During ultrasound detection, the transmit / receive selection switch 102 can control which elements are used to emit ultrasonic beams, which elements are used to receive ultrasonic beams, or control the elements to be used in time-slotted manner for emitting ultrasonic beams or receiving echoes of ultrasonic beams. Elements participating in ultrasonic wave emission can be simultaneously excited by electrical signals to emit ultrasonic waves simultaneously; or elements participating in ultrasonic wave emission can be excited by several electrical signals with a certain time interval to continuously emit ultrasonic waves with a certain time interval.

[0084] The transmitting circuit 101 generates a transmission sequence under the control of the processor 105. This transmission sequence controls some or all of the multiple array elements to transmit ultrasound waves towards the target tissue. The transmission sequence parameters include the array element positions, the number of array elements, and the ultrasound beam transmission parameters (e.g., amplitude, frequency, number of transmissions, transmission interval, transmission angle, waveform, focusing position, etc.). In some cases, the transmitting circuit 101 also performs phase delay on the transmitted beam, allowing different transmitting array elements to transmit ultrasound waves at different times, so that each transmitted ultrasound beam can be focused in a predetermined region of interest. Different operating modes, such as B-image mode, C-image mode, and D-image mode (Doppler mode), may have different transmission sequence parameters. After the echo signal is received by the receiving circuit 320 and processed by subsequent modules and corresponding algorithms, a B-image reflecting the tissue anatomy, a C-image reflecting the tissue anatomy and blood flow information, and a D-image reflecting the Doppler spectrum can be generated.

[0085] The receiving circuit 103 receives the electrical signal of the ultrasonic echo from the ultrasonic probe 100 and processes the electrical signal of the ultrasonic echo. The receiving circuit 103 may include one or more amplifiers, analog-to-digital converters (ADCs), etc. The amplifier amplifies the received ultrasonic echo electrical signal after appropriate gain compensation, and the ADC samples the analog echo signal at predetermined time intervals to convert it into a digitized signal. The digitized echo signal still retains amplitude, frequency, and phase information. The data output by the receiving circuit 103 can be sent to the beamforming circuit 104 for processing, or to the memory 107 for storage.

[0086] The signal processing module 108 is used to denoise the signal output by the receiving circuit 103. The signal processing module 108 includes an averaging module, a channel filtering module, an interpolation filtering module, and a machine learning model. The averaging module smooths the signal and reduces the impact of noise by averaging continuous data points in the signal to reduce short-term fluctuations and noise. The channel filtering module processes different frequency components of the signal by selectively allowing certain frequencies to pass through (passband) while blocking other frequencies (stopband), thereby altering the frequency characteristics of the signal. The interpolation filtering module 1083 inserts new data points between signal sampling points to improve signal resolution or reduce the sampling rate by interpolating known sample points to estimate the values ​​between sample points. The machine learning model learns at least one of the averaging module, channel filtering module, and interpolation filtering module. The machine learning model can be composed of a neural network trained on a constructed ultrasound signal dataset. The neural network includes, but is not limited to, convolutional neural networks, recurrent neural networks, adversarial networks, and neural networks composed of attention mechanisms. It improves the quality of the ultrasound signal by learning the noise characteristics in the ultrasound signal and continuously correcting the noise in the ultrasound signal. It should be noted that the embodiments of this application do not specifically limit the position of the neural network in the signal processing module 108. For example, the averaging module, the channel filtering module, and the interpolation filtering module are each connected to a neural network, and each neural network independently learns the denoising capability of its corresponding module; or, for another example, the averaging module, the channel filtering module, and the interpolation filtering module are all connected to a single neural network, and this shared neural network simultaneously learns the denoising capabilities of the averaging module, the channel filtering module, and the interpolation filtering module. It should be pointed out that the signal processing module 108, in addition to being able to... Figure 1 The device shown can be independently installed in the ultrasound imaging device 10, or it can be installed in the processor 105, or located in the cloud server.

[0087] The beamforming circuit 104 is signal-connected to the signal processing module 108 and is used to perform beamforming processing such as delay and weighted summation on the signal output by the signal processing module 108. Because the distance from the ultrasonic receiving point in the tested tissue to the receiving array element varies, the channel data of the same receiving point output by different receiving array elements has delay differences, requiring delay processing to align the phases and perform weighted summation on the different channel data of the same receiving point to obtain the beamformed ultrasonic image data. The ultrasonic image data output by the beamforming circuit 104 is also called radio frequency (RF) data. The beamforming circuit 104 can output the RF data to the IQ demodulation circuit. In some embodiments, the beamforming circuit 104 can also output the RF data to the memory 107 for caching or storage, or directly output the RF data to the image processing module of the processor 105 for image processing.

[0088] The beamforming circuit 104 can perform the above functions in hardware, firmware or software. For example, the beamforming circuit 104 may include a central controller circuit (CPU), one or more microprocessor chips or any other electronic components capable of processing input data according to specific logic instructions. When the beamforming circuit 104 is implemented in software, it can execute instructions stored on a tangible and non-transitory computer-readable medium (e.g., memory 107) to perform beamforming calculations using any appropriate beamforming method.

[0089] The processor 105 is configured to process input data according to specific logic instructions. It is a central controller circuit (CPU), one or more microprocessors, a graphics controller circuit (GPU), or any other electronic component. It can control peripheral electronic components according to input instructions or predetermined instructions, or perform data reading and / or saving on the memory 107. It can also process input data by executing programs in the memory 107. For example, it can perform one or more processing operations on the acquired ultrasound data according to one or more operating modes. The processing operations include, but are not limited to, adjusting or limiting the form of ultrasound waves emitted by the ultrasound probe 100, generating various image frames for display on the display 106 of the subsequent human-machine interaction device, or adjusting or limiting the content and form displayed on the display 106, or adjusting one or more image display settings displayed on the display 106 (e.g., ultrasound images, interface components, locating regions of interest, ultrasound recognition images, etc.).

[0090] The image processing module of processor 105 processes the data output from beamforming circuit 104 or IQ demodulation circuit to generate a grayscale image reflecting the signal strength variations within the scanning range. This grayscale image reflects the internal anatomical structure of the tissue and is called a B-image. The image processing module can output the B-image to the display 106 of the human-computer interaction device for display. The human-computer interaction device is used for human-computer interaction, i.e., receiving user input and outputting visual information; it can receive user input via keyboard, operation buttons, mouse, trackball, etc., or via a touch screen integrated with the display; it outputs visual information via display 106.

[0091] The memory 107 may be a tangible and non-transitory computer-readable medium, such as a flash memory card, solid-state memory, hard disk, etc., for storing data or programs. For example, the memory 107 may be used to store acquired ultrasound data or image frames generated by the processor 105 that are not immediately displayed, or the memory 107 may store a graphical user interface, one or more default image display settings, or programming instructions for the processor, beamforming circuit, or IQ demodulation circuit.

[0092] It should be noted that, Figure 1 The structure shown is for illustrative purposes only and may include more... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown. Figure 1 The components shown can be implemented in hardware and / or software.

[0093] In one embodiment of this application, the display 106 of the aforementioned ultrasonic imaging device 10 may be a touch screen, a liquid crystal display, or an independent display device such as a liquid crystal display or a television set, separate from the ultrasonic imaging device 10, or a display screen on an electronic device such as a mobile phone or tablet computer.

[0094] In one embodiment of this application, the memory 107 of the aforementioned ultrasound imaging device 10 may be a flash memory card, a solid-state memory, a hard disk, etc.

[0095] In one embodiment of this application, a computer-readable storage medium is also provided, which stores a plurality of program instructions. After being called and executed by the processor 105, the plurality of program instructions can execute some or all of the steps or any combination of the steps in the ultrasonic signal denoising method in various embodiments of this application.

[0096] In one embodiment, the computer-readable storage medium may be a memory 107, which may be a non-volatile storage medium such as a flash memory card, a solid-state memory, or a hard disk.

[0097] In one embodiment of this application, the processor 105 of the aforementioned ultrasonic imaging device 10 can be implemented by software, hardware, firmware, or a combination thereof. It can use at least one of the following: application-specific integrated circuit (ASIC), digital signal processor (DSP), digital signal processing device (DSPD), programmable logic device (PLD), field-programmable gate array (FPGA), central processing unit (CPU), controller, microcontroller, and microprocessor, so that the processor 105 can execute the corresponding steps of the ultrasonic signal denoising method in the various embodiments of this application.

[0098] The embodiments of the ultrasonic imaging device provided in this application can perform the following ultrasonic signal denoising methods. The specific embodiments are described below using ultrasonic signal denoising methods.

[0099] like Figure 2 As shown in the embodiment of this application, a method for denoising ultrasonic signals is provided. This method for denoising ultrasonic signals can be applied to... Figure 1 The ultrasound imaging device shown includes:

[0100] Step 110: Control the ultrasound probe to emit the first ultrasound wave toward the target tissue.

[0101] In this step, the ultrasound imaging device 10 generally supports multiple ultrasound examination modes, such as B-mode, color Doppler mode, ultrasound elastography mode, power Doppler mode, and vector flow mode. In this step 110, the ultrasound imaging device 10 can use B-image mode (reflecting tissue anatomy), C-image mode (reflecting tissue anatomy and blood flow information), E-image mode (elastography), or M-image mode (also known as echocardiography), which can display tissue structure images. The target tissue can be set according to the needs of the ultrasound examination; for example, ultrasound examination can be performed on organs such as the heart, lungs, and ovaries. For example, it can be used through... Figure 1The ultrasound probe 100 shown emits a first ultrasound wave toward the target tissue. By controlling the emission sequence parameters of the emission circuit 101, the first ultrasound wave emitted by the ultrasound probe 100 toward the target tissue can meet the working requirements for tissue structure image examination, such as B-image mode, C-image mode, E-image mode, or M-image mode.

[0102] In one embodiment, a linear array ultrasonic probe can be used to emit the first ultrasonic wave described above for two-dimensional ultrasonic detection (e.g., to detect two-dimensional images of target tissue). In another embodiment, an area array probe can be used to emit the first ultrasonic wave described above for three-dimensional ultrasonic detection (e.g., to detect three-dimensional images of target tissue).

[0103] Step 120: Control the ultrasound probe to receive the echo channel signal of the first ultrasound returned from the target tissue. The echo channel signal includes multiple first ultrasound channel data, and the multiple first ultrasound channel data correspond one-to-one with multiple receiving array elements on the ultrasound probe.

[0104] In this step, after receiving the first ultrasound wave, the target tissue returns the echo channel signal of the first ultrasound wave. This echo channel signal is received by the ultrasound probe 100. The echo channel signal includes multiple first ultrasound channel data, each corresponding one-to-one with a different receiving element on the ultrasound probe 100. The echo channel signal is then sent to the receiving circuit 103 to be converted into an electrical signal, awaiting subsequent noise reduction processing.

[0105] Step 130: Input multiple first ultrasound channel data into a machine learning model for noise reduction processing to obtain multiple second ultrasound channel data.

[0106] In this step, multiple first ultrasound channel data are input into the machine learning model 1084 for denoising. Since the machine learning model 1084 can remove noise data from ultrasound signals, the noise data in the multiple first ultrasound channel data will be removed by the machine learning model 1084, resulting in multiple second ultrasound channel data. This effectively removes noise from the ultrasound echo signal, improves the signal-to-noise ratio of the ultrasound signal, and enhances the imaging quality of the ultrasound image. It should be noted that the input first ultrasound channel data includes, but is not limited to, RF data and IQ data. Ultrasonic noise includes noise signals determined by the characteristics of ultrasound imaging, such as side lobes, grating lobes, and reverberation, which are common in array ultrasound imaging.

[0107] It should be noted that the machine learning model can learn at least one of the averaging module, channel filtering module, and interpolation filtering module. The machine learning model can be constructed from a neural network trained on a built ultrasound signal dataset. This neural network includes, but is not limited to, convolutional neural networks, recurrent neural networks, adversarial networks, and neural networks composed of attention mechanisms. It improves the quality of the ultrasound signal by learning the noise features in the ultrasound signal and continuously correcting the noise. It should also be noted that the embodiments of this application do not specifically limit the position of the neural network in the signal processing module 108; it can be as follows: Figure 3A As shown, the averaging module 1081, the channel filtering module 1082, and the interpolation filtering module 1083 are each connected to a neural network. Each neural network independently learns the denoising capability of its corresponding module. Alternatively, as shown... Figure 3B As shown, the averaging module 1081, channel filtering module 1082, and interpolation filtering module 1083 are all connected to a shared neural network. This shared neural network simultaneously learns the denoising capabilities of the averaging module 1081, channel filtering module 1082, and interpolation filtering module 1083. It should be noted that the signal processing module 108, in addition to... Figure 1 The device shown can be independently installed in the ultrasound imaging device 10, or it can be installed in the processor 105, or located in the cloud server.

[0108] Step 140: Beamforming the data from multiple second ultrasound channels to obtain synthesized radio frequency data.

[0109] In this step, after each first ultrasound channel data is denoised by the machine learning model 1084, the corresponding second ultrasound channel data is obtained. Then, the beamforming circuit 104 performs beamforming processing on the multiple second ultrasound channel data to obtain the synthesized radio frequency data. For example, the synthesized radio frequency data after beamforming can be obtained by delaying and weighting the data of each second ultrasound channel. It should be noted that the beamforming processing includes, but is not limited to, delay processing and weighted summation processing.

[0110] Step 150: Image data of the target tissue is obtained by performing image processing based on the synthesized radio frequency data.

[0111] In this step, the synthesized radio frequency data can be processed by the processor 105 to convert it into image data of the target tissue. It should be noted that the image data of the target tissue includes, but is not limited to, B-image data, C-image data, E-image data, or M-image data. Based on the image data of the target tissue, a visual image of the target tissue can be generated.

[0112] Step 160: Obtain an image of the target tissue based on the image data, and display the image of the target tissue.

[0113] In this step, the processor 105 can generate a visualized image of the target tissue based on image data. For example, it can generate a B image of the target tissue based on B image data, a C image of the target tissue based on C image data, an E image of the target tissue based on E image data, and an M image of the target tissue based on M image data. The image of the target tissue is then displayed on the display 106. Therefore, in this embodiment, the tissue image of the target tissue displayed on the display 106 includes, but is not limited to, B images, C images, E images, or M images. Furthermore, the generated visualized image of the target tissue can be stored in the memory 107 for easy saving or retrieval.

[0114] The ultrasonic signal denoising method provided in this application embodiment utilizes an ultrasonic probe to receive multiple first ultrasonic channel data returned by a target tissue. Each first ultrasonic channel data point is input into a machine learning model for denoising processing. Then, beamforming is performed on the denoised multiple second ultrasonic channel data to obtain synthesized radio frequency (RF) data. An image of the target tissue is obtained and displayed based on the synthesized RF data, thereby effectively removing noise from the ultrasonic echo signal. The machine learning model can remove noise data from the ultrasonic signal, thereby improving the signal-to-noise ratio (SNR) and enhancing the imaging quality of the ultrasonic image. Therefore, the ultrasonic signal denoising method of this application embodiment can effectively remove noise from ultrasonic echo signals, improve the SNR, and enhance the imaging quality of ultrasonic images.

[0115] In one embodiment, such as Figure 4 As shown, step 140 includes:

[0116] Step 410: After delaying the data from multiple second ultrasound channels, multiple delayed channel data are obtained.

[0117] Step 420: Perform weighted summation on the data from multiple delay channels to obtain the synthesized radio frequency data.

[0118] Because the distances from the ultrasonic receiving point to the receiving array element in the target tissue vary, the channel data of the same receiving point output by different receiving array elements have delay differences. Delay processing is required to align the phases and then perform weighted summation of the different channel data from the same receiving point to obtain the beam-synthesized ultrasonic image data. The ultrasonic image data output by the beam-synthesizing circuit 104 is also called radio frequency (RF) data. In this step, after each second ultrasonic channel data undergoes denoising processing by the machine learning model 1084, delay processing is performed to obtain multiple delayed channel data. Phase alignment is then performed, and the weighted summation of the different channel data from the same receiving point yields the beam-synthesized RF data.

[0119] In one embodiment, step 130 includes:

[0120] Multiple first ultrasound channel data are input into the machine learning model corresponding to each receiving array element for denoising processing to obtain multiple second ultrasound channel data.

[0121] Alternatively, multiple first ultrasound channel data can be sequentially input into a machine learning model for denoising processing to obtain multiple second ultrasound channel data.

[0122] Alternatively, it can include multiple parallel machine learning models, each corresponding to an array element of an ultrasound probe. Each array element includes multiple receiving array elements. The multiple first channel data corresponding to each array element are sequentially input into the corresponding machine learning model for noise reduction processing to obtain multiple second ultrasound channel data.

[0123] In this step, the data from each first ultrasound channel can be input into the machine learning model 1084 corresponding to each receiving array element for denoising processing. That is, one ultrasound channel data corresponds to one machine learning model 1084 for denoising processing, thus obtaining multiple second ultrasound channel data. This method allows the first ultrasound channel data to be processed in parallel, thereby improving denoising efficiency. Alternatively, the data from each first ultrasound channel can be input into the machine learning model 1084 sequentially for denoising processing. That is, one machine learning model can sequentially denoise the ultrasound channel data, thus obtaining multiple second ultrasound channel data. This method eliminates the need to configure multiple machine learning models 1084, reducing denoising processing costs. Alternatively, it can include multiple parallel machine learning models 1084, each corresponding to an array element group of the ultrasound probe. Each array element group includes multiple receiving array elements. The multiple first-channel data corresponding to each array element group are sequentially input into the corresponding machine learning model 1084 for denoising processing. This is equivalent to grouping the multiple first-channel data, and then inputting the grouped multiple first-channel data into the corresponding machine learning model 1084 for denoising processing to obtain multiple second-channel ultrasound data, thereby achieving more targeted noise processing for different groups of first-channel data. It should be noted that the multiple array element groups can be arranged continuously or non-contiguously.

[0124] In one embodiment, such as Figure 5 As shown, step 130 includes:

[0125] Step 510: The data from multiple first ultrasound channels are merged to obtain the first combined channel data;

[0126] Step 520: Input the first combined channel data into the machine learning model for noise reduction processing to obtain the second combined channel data;

[0127] Step 530: Extract and process the data from the second combined channel to obtain multiple second ultrasound channel data.

[0128] In one embodiment, the data from each first ultrasound channel are merged to obtain first combined channel data, i.e., the first combined channel data is multi-channel data. The first combined channel data is input into a machine learning model 1084 for denoising processing. This is equivalent to simultaneously denoising the multi-channel data input into the machine learning model 1084, resulting in denoised second combined channel data, thereby improving denoising efficiency. The second combined channel data is then subjected to data extraction processing to obtain multiple denoised second ultrasound channel data corresponding to the first ultrasound channel data. In one embodiment, after simultaneously denoising the multi-channel data, the second combined channel data can be delayed to obtain multiple delayed channel data, and then beamforming processing can be performed on the multiple delayed channel data to obtain synthesized radio frequency data.

[0129] In one embodiment, the system may further include multiple parallel machine learning models 1084, each corresponding to an array element group of an ultrasound probe. Each array element group includes multiple receiving array elements. Multiple first ultrasound channel data corresponding to the array element group are merged to obtain first combined channel data. The first combined channel data corresponding to each array element group are then input into the corresponding machine learning model 1084 for denoising. This is equivalent to grouping the multiple first combined channel data. The grouped first combined channel data are then input into the corresponding machine learning model 1084 for denoising to obtain multiple second combined channel data, thereby achieving more targeted noise processing for different groups of first combined channel data. Furthermore, data extraction processing is performed on the second combined channel data to obtain multiple second ultrasound channel data. It should be noted that the multiple array element groups can be arranged continuously or non-contiguously.

[0130] In one embodiment, the machine learning model includes multiple denoising sub-models, with different denoising sub-models used to remove different types of noise. Step 130 includes:

[0131] Multiple first ultrasound channel data are input into a machine learning model, and noise in the first ultrasound channel data is removed using each denoising sub-model to obtain multiple second ultrasound channel data.

[0132] In this step, multiple first ultrasound channel data are input into the machine learning model 1084. Since the machine learning model 1084 includes multiple denoising sub-models, noise in the first ultrasound channel data can be removed using each denoising sub-model to obtain multiple second ultrasound channel data. It should be noted that the denoising sub-model can include, but is not limited to, at least one of an averaging filter sub-model, a channel filter sub-model, and an interpolation filter sub-model. Specifically, the averaging filter sub-model is used to perform average filtering on the input first ultrasound channel data, the channel filter sub-model is used to perform channel filtering on the input first ultrasound channel data, and the interpolation filter sub-model is used to perform interpolation filtering on the input first ultrasound channel data.

[0133] In one embodiment, step 130 includes:

[0134] Multiple first ultrasound channel data are input into a machine learning model to extract effective data, resulting in multiple second ultrasound channel data.

[0135] Alternatively, multiple first ultrasound channel data can be input into a machine learning model for noise extraction to obtain multiple noise data. Based on each noise data, the corresponding first ultrasound channel data can be denoised to obtain multiple second ultrasound channel data.

[0136] In this step, the machine learning model can be considered a feature extractor, capable of directly extracting valid data from the original ultrasound data, or extracting and removing noise data from the original ultrasound data. Based on this, multiple first ultrasound channel data can be input into the machine learning model 1084 for valid data extraction, resulting in multiple second ultrasound channel data that are denoised valid ultrasound data. Alternatively, multiple first combined channel data can be input into the machine learning model 1084 for noise extraction, obtaining noise data, which is then subtracted from the original data to obtain multiple second ultrasound channel data, which are also denoised valid ultrasound data. This effectively removes noise from the ultrasound echo signal, improves the signal-to-noise ratio of the ultrasound signal, and enhances the imaging quality of the ultrasound image.

[0137] like Figure 6 As shown in the embodiments of this application, another method for denoising ultrasonic signals is also provided. This method for denoising ultrasonic signals can be applied to... Figure 1 The ultrasound imaging device shown includes:

[0138] Step 210: Control the ultrasound probe to emit the first ultrasound wave toward the target tissue.

[0139] In this step, the ultrasound imaging device 10 generally supports multiple ultrasound examination modes, such as B-mode, color Doppler mode, ultrasound elastography mode, power Doppler mode, and vector flow mode. In this step 210, the ultrasound imaging device 10 can use B-image mode (reflecting tissue anatomy), C-image mode (reflecting tissue anatomy and blood flow information), E-image mode (elastography), or M-image mode (also known as echocardiography), which can display tissue structure images. The target tissue can be set according to the needs of the ultrasound examination; for example, ultrasound examination can be performed on organs such as the heart, lungs, and ovaries. For example, it can be used through... Figure 1 The ultrasound probe 100 shown emits a first ultrasound wave toward the target tissue. By controlling the emission sequence parameters of the emission circuit 101, the first ultrasound wave emitted by the ultrasound probe 100 toward the target tissue can meet the working requirements for tissue structure image examination, such as B-image mode, C-image mode, E-image mode, or M-image mode.

[0140] Step 220: Control the ultrasound probe to receive the echo channel signal of the first ultrasound returned from the target tissue. The echo channel signal includes multiple first ultrasound channel data, and the multiple first ultrasound channel data correspond one-to-one with multiple receiving array elements on the ultrasound probe.

[0141] In this step, after receiving the first ultrasonic wave, the target tissue returns the echo channel signal of the first ultrasonic wave. This echo channel signal is received by the ultrasonic probe 100. The echo channel signal includes multiple first ultrasonic channel data, each corresponding one-to-one with a different receiving element on the ultrasonic probe 100. The echo channel signal is then sent to the receiving circuit 103 to be converted into an electrical signal, awaiting subsequent noise reduction processing. It should be noted that the first ultrasonic channel data includes, but is not limited to, RF data, IQ data, and beamforming data.

[0142] Step 230: Input multiple first ultrasound channel data into a machine learning model for noise reduction processing to obtain multiple second ultrasound channel data.

[0143] In this step, multiple first ultrasound channel data are input into the machine learning model 1084 for denoising. Since the machine learning model 1084 can remove noise data from ultrasound signals, the noise data in the multiple first ultrasound channel data will be removed by the machine learning model 1084, resulting in multiple second ultrasound channel data. This effectively removes noise from the ultrasound echo signal, improves the signal-to-noise ratio of the ultrasound signal, and enhances the imaging quality of the ultrasound image. It should be noted that the input first ultrasound channel data includes, but is not limited to, RF data and IQ data. Ultrasonic noise includes noise signals determined by the characteristics of ultrasound imaging, such as side lobes, grating lobes, and reverberation, which are common in array ultrasound imaging.

[0144] It should be noted that the machine learning model can learn at least one of the averaging module, channel filtering module, and interpolation filtering module. The machine learning model can be constructed from a neural network trained on a built ultrasound signal dataset. This neural network includes, but is not limited to, convolutional neural networks, recurrent neural networks, adversarial networks, and neural networks composed of attention mechanisms. It improves the quality of the ultrasound signal by learning the noise features in the ultrasound signal and continuously correcting the noise. It should also be noted that the embodiments of this application do not specifically limit the position of the neural network in the signal processing module 108; it can be as follows: Figure 3A As shown, the averaging module 1081, the channel filtering module 1082, and the interpolation filtering module 1083 are each connected to a neural network. Each neural network independently learns the denoising capability of its corresponding module. Alternatively, as shown... Figure 3B As shown, the averaging module 1081, channel filtering module 1082, and interpolation filtering module 1083 are all connected to a shared neural network. This shared neural network simultaneously learns the denoising capabilities of the averaging module 1081, channel filtering module 1082, and interpolation filtering module 1083. It should be noted that the signal processing module 108, in addition to... Figure 1 The device shown can be independently installed in the ultrasound imaging device 10, or it can be installed in the processor 105, or located in the cloud server.

[0145] Step 240: Perform weighted summation on multiple second ultrasound channel data to obtain synthesized radio frequency data.

[0146] In this step, after each first ultrasound channel data is denoised by the machine learning model 1084, the corresponding second ultrasound channel data is obtained. Then, the beamforming circuit 104 performs a weighted summation process on the multiple second ultrasound channel data to obtain the synthesized radio frequency data. It should be noted that the multiple second ultrasound channel data can be delayed before or after the weighted summation process.

[0147] Step 250: Image data of the target tissue is obtained by performing image processing based on the synthesized radio frequency data.

[0148] In this step, the synthesized radio frequency data can be processed by the processor 105 to convert it into image data of the target tissue. It should be noted that the image data of the target tissue includes, but is not limited to, B-image data, C-image data, E-image data, or M-image data. Based on the image data of the target tissue, a visual image of the target tissue can be generated.

[0149] Step 260: Obtain an image of the target tissue based on the image data, and display the image of the target tissue.

[0150] In this step, the processor 105 can generate a visualized image of the target tissue based on image data. For example, it can generate a B image of the target tissue based on B image data, a C image of the target tissue based on C image data, an E image of the target tissue based on E image data, and an M image of the target tissue based on M image data. The image of the target tissue is then displayed on the display 106. Therefore, in this embodiment, the tissue image of the target tissue displayed on the display 106 includes, but is not limited to, B images, C images, E images, or M images. Furthermore, the generated visualized image of the target tissue can be stored in the memory 107 for easy saving or retrieval.

[0151] The ultrasonic signal denoising method provided in this application embodiment utilizes an ultrasonic probe to receive multiple first ultrasonic channel data returned by the target tissue. Each first ultrasonic channel data point is input into a machine learning model for denoising processing. Then, the denoised multiple second ultrasonic channel data points are weighted and summed to obtain synthetic radio frequency (RF) data. An image of the target tissue is obtained and displayed based on the synthetic RF data, thereby effectively removing noise from the ultrasonic echo signal. The machine learning model can remove noise data from the ultrasonic signal, thereby improving the signal-to-noise ratio (SNR) and enhancing the imaging quality of the ultrasonic image. Therefore, the ultrasonic signal denoising method of this application embodiment can effectively remove noise from the ultrasonic echo signal, improve the SNR, and enhance the imaging quality of the ultrasonic image.

[0152] In one embodiment, step 140 includes:

[0153] After delaying the data from each of the second ultrasound channels, the delayed data from each second ultrasound channel are weighted and summed to obtain the synthesized radio frequency data.

[0154] Because the distances from the ultrasonic receiving points to the receiving array elements in the target tissue vary, the channel data from the same receiving point output by different receiving array elements have delay differences. Delay processing is required to align the phases, and the different channel data from the same receiving point are then weighted and summed to obtain the beam-synthesized ultrasonic image data. The ultrasonic image data output by the beam-synthesizing circuit 104 is also called radio frequency (RF) data. In this step, before weighted summation, the data from each second ultrasonic channel is first delayed, and then the delayed data from each second ultrasonic channel is weighted and summed to obtain the synthesized RF data. That is, the data from each second ultrasonic channel is first delayed and then weighted and summed.

[0155] In one embodiment, step 130 includes:

[0156] After delaying the data from each of the first ultrasound channels, the delayed data from the multiple first ultrasound channels is input into a machine learning model for denoising to obtain multiple second ultrasound channel data.

[0157] Because the distances from the ultrasonic receiving points to the receiving array elements in the target tissue vary, the channel data of the same receiving point output by different receiving array elements have delay differences. Delay processing is required to align the phases, and the different channel data of the same receiving point are weighted and summed to obtain the beamformed ultrasonic image data. The ultrasonic image data output by the beamformer circuit 104 is also called radio frequency (RF) data. In this step, the data of each first ultrasonic channel is first delayed to obtain multiple corresponding first ultrasonic channel data. Then, a machine learning model 1084 is used to denoise the input multiple first ultrasonic channel data to obtain multiple denoised second ultrasonic channel data.

[0158] In one embodiment, such as Figure 7 As shown, step 130 includes:

[0159] Step 710: The data from each first ultrasound channel are merged to obtain the first combined channel data;

[0160] Step 720: Input the first combined channel data into the machine learning model for noise reduction processing to obtain the second combined channel data;

[0161] Step 730: The second combined channel data is processed to obtain multiple second ultrasound channel data.

[0162] In one embodiment, the data from each first ultrasound channel are merged to obtain first combined channel data, i.e., the first combined channel data is multi-channel data. The first combined channel data is then input into a machine learning model 1084 for denoising processing. This is equivalent to simultaneously denoising the multi-channel data input into the machine learning model 1084, resulting in denoised second combined channel data, thereby improving denoising efficiency. The second combined channel data is then subjected to data extraction processing to obtain multiple denoised second ultrasound channel data corresponding to the first ultrasound channel data.

[0163] In one embodiment, after denoising the multi-channel data simultaneously, the second combined channel data is then delayed to obtain multiple delayed channel data, and the multiple delayed channel data are then weighted and summed to obtain synthesized radio frequency data.

[0164] In one embodiment, to obtain the first combined channel data after merging the data of each first ultrasound channel, specifically, each first ultrasound channel data can be delayed to obtain multiple delayed channel data, and then the multiple delayed channel data can be merged to obtain the merged first combined channel data. That is, the data of each first ultrasound channel is delayed first and then merged.

[0165] like Figure 8 As shown in the embodiments of this application, another method for denoising ultrasonic signals is also provided. This method for denoising ultrasonic signals can be applied to... Figure 1 The ultrasound imaging device shown includes:

[0166] Step 310: Control the ultrasound probe to emit the first ultrasound wave toward the target tissue.

[0167] In this step, the ultrasound imaging device 10 generally supports multiple ultrasound examination modes, such as B-mode, color Doppler mode, ultrasound elastography mode, power Doppler mode, and vector flow mode. In this step 210, the ultrasound imaging device 10 can use B-image mode (reflecting tissue anatomy), C-image mode (reflecting tissue anatomy and blood flow information), E-image mode (elastography), or M-image mode (also known as echocardiography), which can display tissue structure images. The target tissue can be set according to the needs of the ultrasound examination; for example, ultrasound examination can be performed on organs such as the heart, lungs, and ovaries. For example, it can be used through... Figure 1 The ultrasound probe 100 shown emits a first ultrasound wave toward the target tissue. By controlling the emission sequence parameters of the emission circuit 101, the first ultrasound wave emitted by the ultrasound probe 100 toward the target tissue can meet the working requirements for tissue structure image examination, such as B-image mode, C-image mode, E-image mode, or M-image mode.

[0168] In one embodiment, a linear array ultrasonic probe can be used to emit the first ultrasonic wave described above for two-dimensional ultrasonic detection (e.g., to detect two-dimensional images of target tissue). In another embodiment, an area array probe can be used to emit the first ultrasonic wave described above for three-dimensional ultrasonic detection (e.g., to detect three-dimensional images of target tissue).

[0169] Step 320: Control the ultrasound probe to receive the echo channel signal of the first ultrasound returned from the target tissue. The echo channel signal includes multiple first ultrasound channel data, and the multiple first ultrasound channel data correspond one-to-one with multiple receiving array elements on the ultrasound probe.

[0170] In this step, after receiving the first ultrasonic wave, the target tissue returns the echo channel signal of the first ultrasonic wave. This echo channel signal is received by the ultrasonic probe 100. The echo channel signal includes multiple first ultrasonic channel data, each corresponding one-to-one with a different receiving element on the ultrasonic probe 100. The echo channel signal is then sent to the receiving circuit 103 to be converted into an electrical signal, awaiting subsequent noise reduction processing. It should be noted that the first ultrasonic channel data includes, but is not limited to, RF data, IQ data, and beamforming data.

[0171] Step 330: Merge multiple first ultrasound channel data to obtain first combined channel data.

[0172] In this step, before each first ultrasound channel data is input into the machine learning model 1084 for denoising, the data from each first ultrasound channel can be merged to obtain a first combined channel data. This allows all the first ultrasound channel data to be simultaneously input into the machine learning model for denoising, instead of inputting each first ultrasound channel data into the machine learning model one by one, thereby achieving multi-channel combined data processing and improving processing efficiency. It should be noted that the merging process can be data merging, such as data stitching, or beamforming. The data merging process is executed by the processor 105, and the beamforming process is executed by the beamforming circuit 104.

[0173] Step 340: Input the first combined channel data into the machine learning model for noise reduction processing to obtain the second combined channel data.

[0174] In this step, the merged first combined channel data is input into the machine learning model 1084 for denoising. Since the machine learning model 1084 can remove noise data from the ultrasound signal, the noise data in the first combined channel data is removed by the machine learning model 1084, resulting in the second combined channel data. This effectively removes noise from the ultrasound echo signal, improves the signal-to-noise ratio of the ultrasound signal, and enhances the imaging quality of the ultrasound image. It should be noted that ultrasound noise includes noise signals commonly found in array ultrasound imaging, such as side lobes, grating lobes, and reverberation, which are determined by the characteristics of ultrasound imaging.

[0175] It should be noted that the machine learning model can learn at least one of the averaging module, channel filtering module, and interpolation filtering module. The machine learning model can be constructed from a neural network trained on a built ultrasound signal dataset. This neural network includes, but is not limited to, convolutional neural networks, recurrent neural networks, adversarial networks, and neural networks composed of attention mechanisms. It improves the quality of the ultrasound signal by learning the noise features in the ultrasound signal and continuously correcting the noise. It should also be noted that the embodiments of this application do not specifically limit the position of the neural network in the signal processing module 108; it can be as follows: Figure 3A As shown, the averaging module 1081, the channel filtering module 1082, and the interpolation filtering module 1083 are each connected to a neural network. Each neural network independently learns the denoising capability of its corresponding module. Alternatively, as shown... Figure 3B As shown, the averaging module 1081, channel filtering module 1082, and interpolation filtering module 1083 are all connected to a shared neural network. This shared neural network simultaneously learns the denoising capabilities of the averaging module 1081, channel filtering module 1082, and interpolation filtering module 1083. It should be noted that the signal processing module 108, in addition to... Figure 1 The device shown can be independently installed in the ultrasound imaging device 10, or it can be installed in the processor 105, or located in the cloud server.

[0176] Step 350: Perform image processing based on the second combined channel data to obtain image data of the target tissue.

[0177] In this step, the processor 105 can perform image processing on the second combined channel data to convert it into image data of the target tissue. It should be noted that the image data of the target tissue includes, but is not limited to, B-image data, C-image data, E-image data, or M-image data. A visualization image of the target tissue can be generated based on the image data of the target tissue.

[0178] Step 360: Obtain an image of the target tissue based on the image data and display the image of the target tissue.

[0179] In this step, the processor 105 can generate a visualized image of the target tissue based on image data. For example, it can generate a B image of the target tissue based on B image data, a C image of the target tissue based on C image data, an E image of the target tissue based on E image data, and an M image of the target tissue based on M image data. The image of the target tissue is then displayed on the display 106. Therefore, in this embodiment, the tissue image of the target tissue displayed on the display 106 includes, but is not limited to, B images, C images, E images, or M images. Furthermore, the generated visualized image of the target tissue can be stored in the memory 107 for easy saving or retrieval.

[0180] The ultrasonic signal denoising method provided in this application embodiment utilizes an ultrasonic probe to receive multiple first ultrasonic channel data returned by the target tissue. The data from each first ultrasonic channel is merged to obtain first combined channel data. This first combined channel data is then input into a machine learning model for denoising processing, achieving multi-channel combined data processing and thus improving processing efficiency. The denoised second combined channel data is then processed to obtain image data. Based on the image data, an image of the target tissue is obtained and displayed, effectively removing noise from the ultrasonic echo signal. The machine learning model can remove noise data from the ultrasonic signal, thereby improving the signal-to-noise ratio of the ultrasonic signal and enhancing the imaging quality of the ultrasonic image. Therefore, the ultrasonic signal denoising method of this application embodiment can effectively remove noise from the ultrasonic echo signal, improve the signal-to-noise ratio of the ultrasonic signal, enhance the imaging quality of the ultrasonic image, and improve denoising processing efficiency.

[0181] In one embodiment, such as Figure 9 As shown, step 250 includes:

[0182] Step 910: The second combined channel data is split to obtain multiple second ultrasound channel data.

[0183] Step 920: Beamforming the data from multiple second ultrasound channels to obtain synthesized radio frequency data;

[0184] Step 930: Image data of the target tissue is obtained by performing image processing based on the synthesized radio frequency data.

[0185] In this step, the second combined channel data is merged data after denoising processing. Therefore, it is necessary to split the denoised second combined channel data into multiple second ultrasound channel data, and then perform beamforming on the multiple second ultrasound channel data to obtain synthesized radio frequency data. The processor 105 then performs image processing on the synthesized radio frequency data to obtain image data of the target tissue. This embodiment denoises the second combined channel data before beamforming, enabling beamforming of the effective ultrasound data to obtain denoised synthesized radio frequency data, thereby effectively removing noise from the ultrasound echo signal.

[0186] In one embodiment, step 230 includes:

[0187] Beamforming processing is performed on multiple first ultrasound channel data to obtain first combined channel data.

[0188] In this step, before the multiple first ultrasound channel data are input into the machine learning model 1084 for denoising, beamforming can be performed on the multiple first ultrasound channel data to obtain first combined channel data. This allows all the beamformed first ultrasound channel data to be simultaneously input into the machine learning model for denoising. In other words, this embodiment denoises the beamformed first combined channel data, effectively removing noise from the ultrasound echo signal.

[0189] In one embodiment, such as Figure 10 As shown, step 230 includes:

[0190] Step 1010: Delay the data from multiple first ultrasound channels to obtain multiple delayed data from the first ultrasound channels.

[0191] Step 1020: Merge the multiple delayed first ultrasound channel data to obtain the first combined channel data.

[0192] Because the distances from the ultrasonic receiving points to the receiving array elements in the target tissue vary, the channel data from the same receiving point output by different receiving array elements have delay differences. Delay processing is required to align the phases and perform weighted summation of the different channel data from the same receiving point to obtain the beamformed ultrasonic image data. The ultrasonic image data output by the beamformer circuit 104 is also called radio frequency (RF) data. In this step, before denoising, multiple first ultrasonic channel data are delayed to obtain multiple delayed first ultrasonic channel data. Then, these multiple delayed first ultrasonic channel data are merged, for example, by aligning the phases and performing weighted summation of the different channel data from the same receiving point to obtain the merged first combined channel data.

[0193] In one embodiment, the machine learning model includes multiple denoising sub-models, with different denoising sub-models used to remove different types of noise. Step 240 includes:

[0194] The first combined channel data is input into the machine learning model, and noise in the first combined channel data is removed using each denoising sub-model to obtain the second combined channel data.

[0195] In this step, the first combined channel data is input into the machine learning model 1084. Since the machine learning model 1084 includes multiple denoising sub-models, noise in the first combined channel data can be removed using each denoising sub-model to obtain the second combined channel data. It should be noted that the denoising sub-models may include, but are not limited to, at least one of an averaging filter sub-model, a channel filter sub-model, and an interpolation filter sub-model. Specifically, the averaging filter sub-model is used to perform average filtering on the input first combined channel data, the channel filter sub-model is used to perform channel filtering on the input first combined channel data, and the interpolation filter sub-model is used to perform interpolation filtering on the input first combined channel data.

[0196] In one embodiment, step 240 includes:

[0197] The first combination channel data is input into the machine learning model to extract effective data, thus obtaining the second combination channel data.

[0198] Alternatively, the first combined channel data can be input into a machine learning model for noise extraction to obtain noise data. Based on the noise data, the first combined channel data can be denoised to obtain the second combined channel data.

[0199] In this step, the machine learning model can be considered a feature extractor, capable of directly extracting valid data from the original ultrasound data, or extracting and removing noise data from the original ultrasound data. Based on this, the first combined channel data can be input into the machine learning model 1084 for valid data extraction, resulting in the second combined channel data, which is the denoised valid ultrasound data. Alternatively, the first combined channel data can be input into the machine learning model 1084 for noise extraction, obtaining noise data, which is then subtracted from the original data to obtain the second combined channel data, which is the denoised valid ultrasound data. This effectively removes noise from the ultrasound echo signal, improves the signal-to-noise ratio of the ultrasound signal, and enhances the imaging quality of the ultrasound image.

[0200] like Figure 11 As shown in the embodiments of this application, a model training method is also provided, including:

[0201] Step 1110: Obtain the training sample dataset and the validation dataset. The training sample dataset includes training sample data of multiple first ultrasound channels corresponding to multiple receiving array elements on the ultrasound probe. The validation dataset includes valid data from the training sample data of the first ultrasound channels.

[0202] In this step, the construction of the training dataset involves using noisy ultrasonic signals as the first dataset and ultrasonic signals with no noise or high signal-to-noise ratios as the second dataset. Based on this, this embodiment uses the first dataset as the training sample dataset and the second dataset as the validation dataset. It should be noted that the dataset can contain ultrasonic signals actually acquired by the ultrasonic imaging device or ultrasonic signals constructed through a simulation program. Regarding circuit noise, if the first dataset contains acquired ultrasonic system signals, high-quality ultrasonic signals are constructed using methods such as averaging, Butterworth filtering, and wavelet transform filtering as the second dataset. If it contains simulation data, the original simulation data is used as the second dataset, and the original simulation data is supplemented with different proportions of white noise, salt-and-pepper noise, and other circuit noise interference to form the first dataset. Ultrasonic noise is primarily constructed from simulation data. In ultrasonic imaging, the transducer chip size is closely related to sensitivity; larger chips result in higher sensitivity but also more pronounced sidelobe phenomena. Conversely, smaller chips result in lower sensitivity but less pronounced sidelobe phenomena. The grating lobe is related to the spacing between the ultrasonic transducer chip array. Generally, when strong reflective interfaces exist within a tissue, multiple reflections of ultrasonic signals between these interfaces produce reverberation. The aforementioned ultrasonic noise can be simulated to construct a first dataset and a second dataset. For example, by keeping the wafer center position constant and varying the wafer size, ultrasonic signals with and without sidelobes can be constructed, corresponding to the first and second datasets, respectively. By setting different acoustic parameters of the tissue, the reflection phenomenon of the ultrasonic signal at the interface can be controlled. The ultrasonic data without multiple reflections constitutes the first dataset, while the data with multiple reflections constitutes the second dataset.

[0203] Step 1120: Input the first ultrasound channel training sample data into the machine learning model for noise reduction to obtain ultrasound channel prediction data.

[0204] In this step, during the training phase, the training sample data of the first ultrasound channel is input into the machine learning model for denoising processing to obtain ultrasound channel prediction data. This ultrasound channel prediction data is the prediction data output by the machine learning model, and it may differ from the effective data. The machine learning model is a neural network, including but not limited to convolutional neural networks, recurrent neural networks, adversarial neural networks, and attention neural networks. It should be noted that the location of the neural network in the signal processing module 108 is not specifically limited, and can be as follows: Figure 3A As shown, the averaging module 1081, the channel filtering module 1082, and the interpolation filtering module 1083 are each connected to a neural network. Each neural network independently learns the denoising capability of its corresponding module. Alternatively, as shown... Figure 3B As shown, the averaging module 1081, the channel filtering module 1082, and the interpolation filtering module 1083 are all connected to a neural network. This shared neural network learns the denoising capabilities of the averaging module 1081, the channel filtering module 1082, and the interpolation filtering module 1083 simultaneously.

[0205] Step 1130: Based on the ultrasound channel prediction data and valid data, the parameters of the machine learning model are corrected.

[0206] In this step, because there is a certain difference between the ultrasound channel prediction data and the effective data, the machine learning model needs to be trained multiple times to correct its parameters until the ultrasound channel prediction data output by the machine learning model is very close to the effective data. For example, when the error between the ultrasound channel prediction data and the effective data is less than a set threshold, the training is complete. At this point, the target parameters of the machine learning model can be determined, resulting in a well-trained machine learning model. The well-trained machine learning model can improve the imaging quality of ultrasound images and can be installed in an ultrasound imaging device.

[0207] In one embodiment, the model training process is as follows: Figure 12As shown, the first dataset is input into the generator of the machine learning model for denoising, outputting ultrasound channel prediction data. This ultrasound channel prediction data, along with the second dataset, is then input into the machine learning model's judge to determine if the error between the two datasets is less than a set threshold. The judgment result is then fed back to the generator. If the judgment result indicates that the error between the ultrasound channel prediction data and the second dataset is greater than the set threshold, the machine learning model needs to be trained further to refine its parameters. If the judgment result indicates that the error between the ultrasound channel prediction data and the second dataset is less than the set threshold, the machine learning model training is complete. At this point, the target parameters of the machine learning model can be determined, resulting in a trained machine learning model. The trained machine learning model can improve the imaging quality of ultrasound images and can be integrated into an ultrasound imaging device.

[0208] The model training method provided in this application involves inputting the first ultrasound channel training sample data into a machine learning model for denoising to obtain ultrasound channel prediction data. The ultrasound channel prediction data is then compared with the effective data to correct the parameters of the machine learning model. This allows the trained machine learning model to learn to extract noise features from the ultrasound signal and continuously correct the ultrasound signal prediction data, thereby improving the signal-to-noise ratio of the ultrasound signal and enhancing the imaging quality of the ultrasound image.

[0209] It should be noted that, in the model training method provided in this application embodiment, the simulated noise data includes, but is not limited to, at least one of circuit noise, ultrasonic noise, white noise, Gaussian noise, Rayleigh noise, triangular noise, additive noise, multiplicative noise, thermal noise of amplifying devices, sidelobe noise, grating lobe noise, and reverberation noise. Circuit noise refers to noise signals such as white noise and salt-and-pepper noise generated in ultrasonic circuits; while ultrasonic noise refers to noise signals such as sidelobes, grating lobes, and reverberation, which are common in array ultrasonic imaging and determined by the characteristics of ultrasonic imaging.

[0210] It should be noted that in the model training method provided in this application embodiment, the input data to the machine learning model can be single-channel data (N×1) or multiple-channel data (N*M) simultaneously as input data. Generally, circuit noise is relatively independent, and the influence between each channel is small, so single-channel data is sufficient as input. Of course, the advantage of using multi-channel data as input in terms of effect and efficiency cannot be ruled out. For ultrasonic noise, there is often a certain correlation between channels; therefore, multi-channel data can be considered as input to the machine learning model. Finally, the machine learning model outputs high-quality ultrasonic signals of the same dimension.

[0211] In one embodiment, such as Figure 13As shown, step 1120 includes:

[0212] Step 1310: The training sample data of each first ultrasound channel are merged to obtain combined channel data;

[0213] Step 1320: Input the combined channel data into the machine learning model for noise reduction to obtain the combined channel prediction data;

[0214] Step 1330: Perform data extraction and processing on the combined channel prediction data to obtain multiple ultrasound channel prediction data.

[0215] For ultrasonic noise, there is often a certain correlation between different ultrasonic channels. Therefore, multi-channel data can be used as input into a machine learning model. Based on this, the training sample data of each first ultrasonic channel can be merged to obtain combined channel data. This combined channel data is then input into the machine learning model for noise reduction to obtain combined channel prediction data. Further data extraction processing is performed on the combined channel prediction data to separate it into individual ultrasonic channel prediction data.

[0216] In one embodiment, such as Figure 14 As shown, step 1120 includes:

[0217] Step 1410: The training sample data of each first ultrasound channel are merged to obtain combined channel data;

[0218] Step 1420: Input the combined channel data into the machine learning model for noise reduction to obtain ultrasound channel prediction data.

[0219] For ultrasonic noise, there is often a certain correlation between different ultrasonic channels. Therefore, multi-channel data can be used as input to a machine learning model. Based on this, the training sample data of each first ultrasonic channel can be merged to obtain combined channel data. This combined channel data is then input as multi-channel data into the machine learning model for noise reduction, resulting in ultrasonic channel prediction data. This embodiment does not split the ultrasonic channel prediction data output by the machine learning model.

[0220] In one embodiment, such as Figure 15 As shown, step 1110 includes:

[0221] Step 1510: Emit the first ultrasound wave towards the target tissue through the ultrasound probe;

[0222] Step 1520: Receive the echo channel signal of the first ultrasound returned by the target tissue through the ultrasound probe to obtain the training sample dataset;

[0223] Step 1530: The noise in the training sample data of the first ultrasound channel is filtered by the filtering module to obtain the ultrasound channel prediction data, and a validation dataset is constructed based on the ultrasound channel prediction data.

[0224] In this step, a first ultrasonic wave is emitted towards the target tissue using an ultrasound probe, and the echo channel signal of the first ultrasonic wave returned by the target tissue is received. This actually acquired echo channel signal is used as the training sample dataset. The training sample dataset includes training sample data for multiple first ultrasonic channels corresponding to multiple receiving array elements on the ultrasound probe. A filtering module can be used to filter noise in the first ultrasonic channel training sample data to obtain ultrasonic channel prediction data. It should be noted that the filtering module includes, but is not limited to, at least one of an averaging filtering module, a channel filtering module, and an interpolation filtering module. After filtering, the obtained ultrasonic channel prediction data is the valid data in the first ultrasonic channel training sample data, and a validation dataset is then constructed based on this valid data.

[0225] In one embodiment, such as Figure 16 As shown, step 1110 includes:

[0226] Step 1610: Obtain the simulation echo channel data and construct a verification dataset based on the simulation echo channel data. The simulation echo channel data includes simulation data of multiple ultrasound channels corresponding to multiple receiving array elements on the ultrasound probe.

[0227] Step 1620: Obtain simulated noise data and overlay it onto the ultrasonic channel simulation data to obtain the training sample dataset.

[0228] In this step, an effective ultrasound signal is constructed using a simulation program to obtain simulated echo channel data. This simulated echo channel data includes simulated ultrasound data corresponding to multiple receiving array elements on the ultrasound probe. The simulated echo channel data is effective ultrasound data without noise; therefore, a validation dataset can be constructed based on this data. Furthermore, an ultrasound noise signal can be simulated using a simulation program to obtain simulated noise data. This simulated noise data is then superimposed on the simulated ultrasound channel data to obtain multiple simulated ultrasound channel data containing noise. Therefore, a training sample dataset can be constructed based on this multiple simulated ultrasound channel data containing noise.

[0229] In one embodiment, such as Figure 17 As shown, step 1110 includes:

[0230] Step 1710: Obtain the simulated echo channel data and construct a verification dataset based on the simulated echo channel data. The simulated echo channel data includes simulated data of multiple ultrasound channels corresponding to multiple receiving array elements on the ultrasound probe.

[0231] Step 1720: The simulation data of each ultrasonic channel is merged to obtain the combined channel simulation data;

[0232] Step 1730: Obtain simulated noise data and overlay it onto the combined channel simulation data to obtain the training sample dataset.

[0233] In this step, an effective ultrasound signal is constructed using a simulation program to obtain simulated echo channel data. This simulated echo channel data includes simulated ultrasound channel data corresponding to multiple receiving array elements on the ultrasound probe. The simulated echo channel data is effective ultrasound data without noise; therefore, a validation dataset can be constructed based on this data. The simulated ultrasound channel data are then merged to obtain combined channel simulated data. Furthermore, an ultrasound noise signal can be simulated using a simulation program to obtain simulated noise data. This simulated noise data is then superimposed onto the entire combined channel simulated data to obtain combined channel simulated data containing noise. Therefore, a training sample dataset can be constructed based on this combined channel simulated data containing noise. Based on this, this embodiment can superimpose simulated noise data onto the merged combined channel simulated data, improving noise superposition efficiency and thus accelerating the generation of the training sample dataset.

[0234] like Figure 18 As shown in the embodiments of this application, a model training method is also provided, including:

[0235] Step 1810: Obtain the training sample dataset and the validation dataset. The training sample dataset includes training sample data of multiple first ultrasound channels corresponding to multiple receiving array elements on the ultrasound probe. The validation dataset includes target noise data in the training sample data of the first ultrasound channels.

[0236] In this step, the training dataset is constructed by using noisy ultrasound signals as the first dataset and target noise data from the first dataset as the second dataset. Based on this, this embodiment uses the first dataset as the training sample dataset and the second dataset as the validation dataset. It should be noted that the dataset can be ultrasound signals actually acquired by the ultrasound imaging device or ultrasound signals constructed through a simulation program.

[0237] Step 1820: Input the training sample data of the first ultrasound channel into the machine learning model for noise extraction to obtain noise prediction data;

[0238] In this step, during the training phase, the training sample data of the first ultrasound channel is input into the machine learning model for noise extraction to obtain ultrasound channel prediction data. This ultrasound channel prediction data is the noise prediction data output by the machine learning model, and there is a certain difference between the noise prediction data and the target noise data. The machine learning model is a neural network, including but not limited to convolutional neural networks, recurrent neural networks, adversarial neural networks, and attention neural networks. It should be noted that the location of the neural network in the signal processing module 108 is not specifically limited, and can be as follows: Figure 3A As shown, the averaging module 1081, the channel filtering module 1082, and the interpolation filtering module 1083 are each connected to a neural network. Each neural network independently learns the denoising capability of its corresponding module. Alternatively, as shown... Figure 3B As shown, the averaging module 1081, channel filtering module 1082, and interpolation filtering module 1083 are all connected to a neural network. This shared neural network simultaneously learns the denoising capabilities of the averaging module 1081, channel filtering module 1082, and interpolation filtering module 1083. It should be noted that the purpose of extracting the noise data is to facilitate subsequent subtraction of the noise data from the noisy ultrasound signal, thus obtaining the effective ultrasound data.

[0239] Step 1830: Adjust the parameters of the machine learning model based on the noise prediction data and the target noise data.

[0240] In this step, because there is a certain difference between the noise prediction data and the target noise data, the machine learning model needs to be trained multiple times to correct its parameters until the noise prediction data extracted by the machine learning model is very close to the target noise data. For example, when the error between the noise prediction data and the target noise data is less than a set threshold, the training is complete. At this point, the target parameters of the machine learning model can be determined, resulting in a well-trained machine learning model. The well-trained machine learning model can be used to accurately extract noise data from the original ultrasound data, so that subsequent subtraction of the noise data from the noisy ultrasound signal can yield effective ultrasound data, thereby improving the imaging quality of the ultrasound image. Furthermore, the well-trained machine learning model can be integrated into an ultrasound imaging device.

[0241] In one embodiment, the model training process is as follows: Figure 12As shown, the first dataset is input into the generator of the machine learning model for denoising, outputting ultrasound channel prediction data. This ultrasound channel prediction data, along with the second dataset, is then input into the machine learning model's judge to determine if the error between the two datasets is less than a set threshold. The judgment result is then fed back to the generator. If the judgment result indicates that the error between the ultrasound channel prediction data and the second dataset is greater than the set threshold, the machine learning model needs to be trained further to refine its parameters. If the judgment result indicates that the error between the ultrasound channel prediction data and the second dataset is less than the set threshold, the machine learning model training is complete. At this point, the target parameters of the machine learning model can be determined, resulting in a trained machine learning model. The trained machine learning model can improve the imaging quality of ultrasound images and can be integrated into an ultrasound imaging device.

[0242] The model training method provided in this application involves inputting the training sample data of the first ultrasound channel into a machine learning model to extract noise and obtain noise prediction data. The noise prediction data is then compared with the target noise data to correct the parameters of the machine learning model. This allows the trained machine learning model to learn the noise features in the ultrasound signal and continuously correct the noise in the ultrasound signal, thereby subtracting the noise data from the original ultrasound data, thus improving the signal-to-noise ratio of the ultrasound signal and enhancing the imaging quality of the ultrasound image.

[0243] In one embodiment, such as Figure 19 As shown, step 1820 includes:

[0244] Step 1910: The training sample data of each first ultrasound channel are merged to obtain combined channel data;

[0245] Step 1920: Input the combined channel data into the machine learning model for noise extraction to obtain the combined channel prediction data;

[0246] Step 1930: Perform data extraction processing on the combined channel prediction data to obtain multiple noise prediction data.

[0247] For ultrasonic noise, there is often a certain correlation between different ultrasonic channels. Therefore, multi-channel data can be used as input into a machine learning model. Based on this, the training sample data of each first ultrasonic channel can be merged to obtain combined channel data. This combined channel data is then input into the machine learning model as multi-channel data for noise extraction, yielding combined channel prediction data. Further data extraction processing is performed on the combined channel prediction data to separate it into noise prediction data for each individual channel.

[0248] In one embodiment, such as Figure 20 As shown, step 1820 includes:

[0249] Step 2010: The training sample data of each first ultrasound channel are merged to obtain combined channel data;

[0250] Step 2020: Input the combined channel data into the machine learning model for noise extraction to obtain noise prediction data.

[0251] For ultrasonic noise, there is often a certain correlation between the various ultrasonic channels. Therefore, multi-channel data can be used as input to a machine learning model. Based on this, the training sample data of each first ultrasonic channel can be merged to obtain combined channel data. This combined channel data is then input as multi-channel data into the machine learning model for denoising to obtain noise prediction data. This embodiment does not split the noise prediction data output by the machine learning model.

[0252] In one embodiment, such as Figure 21 As shown, step 1810 includes:

[0253] Step 2110: Emit the first ultrasound wave towards the target tissue using an ultrasound probe;

[0254] Step 2120: Receive the echo channel signal of the first ultrasound returned by the target tissue through the ultrasound probe to obtain the training sample dataset;

[0255] Step 2130: The noise in the training sample data of the first ultrasonic channel is extracted by the filtering module to obtain the target noise data, and a verification dataset is constructed based on the target noise data.

[0256] In this step, a first ultrasonic wave is emitted towards the target tissue using an ultrasonic probe, and the echo channel signal of the first ultrasonic wave returned by the target tissue is received. This actually acquired echo channel signal is used as the training sample dataset. The training sample dataset includes training sample data for multiple first ultrasonic channels corresponding to multiple receiving array elements on the ultrasonic probe. A filtering module can be used to extract noise from the first ultrasonic channel training sample data to obtain target noise data. It should be noted that the filtering module includes, but is not limited to, at least one of an averaging filtering module, a channel filtering module, and an interpolation filtering module. After noise extraction, the obtained target noise data is the noise data in the first ultrasonic channel training sample data, and a validation dataset is then constructed based on this target noise data.

[0257] In one embodiment, such as Figure 22 As shown, step 1810 includes:

[0258] Step 2210: Obtain simulation echo channel data, which includes simulation data of multiple ultrasonic channels corresponding to multiple receiving array elements on the ultrasonic probe.

[0259] Step 2220: Obtain simulated noise data and construct a validation dataset based on the simulated noise data;

[0260] Step 2230: Superimpose simulated noise data onto the ultrasonic channel simulation data to obtain the training sample dataset.

[0261] In this step, an effective ultrasound signal is constructed using a simulation program to obtain simulated echo channel data. This simulated echo channel data includes simulated ultrasound channel data corresponding to multiple receiving array elements on the ultrasound probe. The simulated echo channel data is effective ultrasound data without noise. Furthermore, an ultrasound noise signal can be simulated using a simulation program to obtain simulated noise data. Therefore, a validation dataset can be constructed based on the simulated noise data. Simulated noise data is then superimposed on the simulated ultrasound channel data to obtain simulated ultrasound channel data containing noise. Therefore, a training sample dataset can be constructed based on this simulated ultrasound channel data containing noise.

[0262] In one embodiment, such as Figure 23 As shown, step 1810 includes:

[0263] Step 2310: Obtain simulation echo channel data, which includes simulation data of multiple ultrasonic channels corresponding to multiple receiving array elements on the ultrasonic probe.

[0264] Step 2320: The simulation data of each ultrasonic channel is merged to obtain the combined channel simulation data;

[0265] Step 2330: Obtain simulated noise data and construct a validation dataset based on the simulated noise data;

[0266] Step 2340: Superimpose simulated noise data onto the combined channel simulation data to obtain the training sample dataset.

[0267] In this step, an effective ultrasound signal is constructed using a simulation program to obtain simulated echo channel data. This simulated echo channel data includes simulated ultrasound data corresponding to multiple receiving array elements on the ultrasound probe. The simulated echo channel data consists of effective ultrasound data without noise. The simulated ultrasound channel data are then merged to obtain combined channel simulated data. Furthermore, an ultrasound noise signal can be simulated using a simulation program to obtain simulated noise data. Therefore, a validation dataset can be constructed based on the simulated echo channel data. Simulated noise data is superimposed on the entire combined channel simulated data to obtain combined channel simulated data containing noise. Therefore, a training sample dataset can be constructed based on the combined channel simulated data containing noise. Based on this, this embodiment can superimpose simulated noise data onto the merged entire combined channel simulated data, which can improve noise superposition efficiency and thus accelerate the generation of the training sample dataset.

[0268] It should be noted that, in the model training method provided in this application embodiment, the simulated noise data includes, but is not limited to, at least one of circuit noise, ultrasonic noise, white noise, Gaussian noise, Rayleigh noise, triangular noise, additive noise, multiplicative noise, thermal noise of amplifying devices, sidelobe noise, grating lobe noise, and reverberation noise. Circuit noise refers to noise signals such as white noise and salt-and-pepper noise generated in ultrasonic circuits; while ultrasonic noise refers to noise signals such as sidelobes, grating lobes, and reverberation, which are common in array ultrasonic imaging and determined by the characteristics of ultrasonic imaging.

[0269] It should be noted that in the model training method provided in this application embodiment, the input data to the machine learning model can be single-channel data (N×1) or multiple-channel data (N*M) simultaneously as input data. Generally, circuit noise is relatively independent, and the influence between each channel is small, so single-channel data is sufficient as input. Of course, the advantage of using multi-channel data as input in terms of effect and efficiency cannot be ruled out. For ultrasonic noise, there is often a certain correlation between channels; therefore, multi-channel data can be considered as input to the machine learning model. Finally, the machine learning model outputs high-quality ultrasonic signals of the same dimension.

[0270] This application provides an ultrasound imaging device, including:

[0271] Ultrasonic probe;

[0272] Transmitter / receiver circuit: The transmitter / receiver circuit is used to control the ultrasound probe to transmit ultrasound waves to the target tissue and receive ultrasound echoes.

[0273] A processor is used to denoise the echo of ultrasound waves to obtain an image of the target tissue;

[0274] A monitor, used to display tissue images

[0275] The processor is also used to execute the ultrasonic signal denoising method provided in any of the above embodiments.

[0276] This application provides a model training apparatus, including:

[0277] At least one memory for storing programs;

[0278] At least one processor is used to load a program to execute the model training method provided in any of the above embodiments.

[0279] This application provides a computer storage medium storing a computer program applied to an ultrasonic imaging device. When executed by a processor, the computer program implements the ultrasonic emission method as described in the above embodiments, or it is also used to execute the image processing method described in the above embodiments.

[0280] This application provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the ultrasonic emission method as described in the above embodiments, or further to perform the image processing method as described in the above embodiments.

[0281] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0282] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0283] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0284] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0285] It should also be understood that the various implementation methods provided in this application can be combined arbitrarily to achieve different technical effects.

[0286] The above provides a detailed description of the preferred embodiments of this application. However, this application is not limited to the above-described embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application. All such equivalent modifications or substitutions are included within the scope defined by the claims of this application.

Claims

1. A method for denoising ultrasonic data, characterized in that, The method includes: Control the ultrasound probe to emit the first ultrasound wave toward the target tissue; The ultrasound probe is controlled to receive the echo channel signal of the first ultrasound returned by the target tissue. The echo channel signal includes multiple first ultrasound channel data, and the multiple first ultrasound channel data correspond one-to-one with multiple receiving array elements on the ultrasound probe. Multiple first ultrasound channel data are input into a machine learning model for noise reduction to obtain multiple second ultrasound channel data. Beamforming processing is performed on multiple second ultrasound channel data to obtain synthesized radio frequency data; Image data of the target tissue is obtained by performing image processing based on the synthesized radio frequency data; An image of the target tissue is obtained based on the image data, and the image of the target tissue is displayed.

2. The method for denoising ultrasonic data according to claim 1, characterized in that, The process of beamforming multiple second ultrasound channel data to obtain synthesized radio frequency data includes: After delaying the data from multiple second ultrasound channels, multiple delayed channel data are obtained. The synthesized radio frequency data is obtained by weighted summation of the data from multiple delay channels.

3. The method for denoising ultrasonic data according to claim 1, characterized in that, The step of inputting multiple first ultrasound channel data into a machine learning model for noise reduction processing to obtain multiple second ultrasound channel data includes: Multiple first ultrasound channel data are respectively input into the machine learning model corresponding to each of the receiving array elements for noise reduction processing to obtain multiple second ultrasound channel data. Alternatively, multiple sets of first ultrasound channel data can be sequentially input into a machine learning model for denoising processing to obtain multiple sets of second ultrasound channel data. Alternatively, it may include multiple parallel machine learning models, each machine learning model corresponding to an array element group of the ultrasound probe, each array element group including multiple receiving array elements, and the multiple first channel data corresponding to each array element group are sequentially input into the corresponding machine learning model for noise reduction processing to obtain multiple second ultrasound channel data.

4. The method for denoising ultrasonic data according to claim 1 or 2, characterized in that, The process of inputting multiple first ultrasound channel data into a machine learning model for noise reduction to obtain multiple second ultrasound channel data includes one of the following: Multiple first ultrasound channel data are merged to obtain first combined channel data. The first combined channel data is then input into a machine learning model for noise reduction to obtain second combined channel data. The second combined channel data is then subjected to data extraction processing to obtain multiple second ultrasound channel data. Alternatively, it may include multiple parallel machine learning models, each machine learning model corresponding to an array element group of the ultrasound probe, each array element group including multiple receiving array elements, merging multiple first ultrasound channel data corresponding to the array element group to obtain first combined channel data, distributing the first combined channel data corresponding to each array element group into the corresponding machine learning model for noise reduction processing to obtain multiple second combined channel data, and performing data extraction processing on the second combined channel data to obtain multiple second ultrasound channel data.

5. The method for denoising ultrasonic data according to claim 1, characterized in that, The machine learning model includes multiple denoising sub-models, each used to remove different types of noise; the process of inputting multiple first ultrasound channel data into the machine learning model for denoising processing to obtain multiple second ultrasound channel data includes: Multiple sets of first ultrasound channel data are input into a machine learning model, and noise in the first ultrasound channel data is removed using each of the denoising sub-models to obtain multiple sets of second ultrasound channel data.

6. The method for denoising ultrasonic data according to claim 5, characterized in that, The plurality of denoising sub-models in the machine learning model include at least one of the following: The average filter sub-model used for average filtering; Alternatively, a channel filtering sub-model for channel filtering; Alternatively, an interpolation filtering sub-model used for interpolation filtering.

7. The method for denoising ultrasonic data according to claim 1, characterized in that, The process of inputting multiple first ultrasound channel data into a machine learning model for noise reduction to obtain multiple second ultrasound channel data includes one of the following: Multiple first ultrasound channel data are input into a machine learning model to extract effective data, thereby obtaining multiple second ultrasound channel data. Alternatively, multiple sets of first ultrasound channel data can be input into a machine learning model for noise extraction to obtain multiple sets of noise data. Then, based on each set of noise data, the corresponding first ultrasound channel data can be denoised to obtain multiple sets of second ultrasound channel data.

8. A method for denoising ultrasonic data, characterized in that, The method includes: Control the ultrasound probe to emit the first ultrasound wave toward the target tissue; The ultrasound probe is controlled to receive the echo channel signal of the first ultrasound returned by the target tissue. The echo channel signal includes multiple first ultrasound channel data, and the multiple first ultrasound channel data correspond one-to-one with multiple receiving array elements on the ultrasound probe. Multiple first ultrasound channel data are input into a machine learning model for noise reduction to obtain multiple second ultrasound channel data. The synthesized radio frequency data is obtained by weighted summation of multiple second ultrasound channel data. Image data of the target tissue is obtained by performing image processing based on the synthesized radio frequency data; An image of the target tissue is obtained based on the image data, and the image of the target tissue is displayed.

9. The method for denoising ultrasonic data according to claim 8, characterized in that, The process of weighted summation of multiple second ultrasound channel data to obtain synthesized radio frequency data includes: After delaying the data from each of the second ultrasound channels, the delayed data from each of the second ultrasound channels are weighted and summed to obtain the synthesized radio frequency data.

10. The method for denoising ultrasonic data according to claim 8, characterized in that, The step involves inputting multiple first ultrasound channel data into a pre-trained target machine learning model for denoising processing to obtain multiple second ultrasound channel data, including: After delaying each of the first ultrasound channel data, the delayed first ultrasound channel data is input into a machine learning model for denoising to obtain multiple second ultrasound channel data.

11. The method for denoising ultrasonic data according to claim 10, characterized in that, The step involves inputting multiple first ultrasound channel data into a pre-trained target machine learning model for denoising processing to obtain multiple second ultrasound channel data, including: The data from each of the first ultrasound channels are merged to obtain the first combined channel data. The first combined channel data is input into a machine learning model for noise reduction to obtain the second combined channel data. The second combined channel data is processed to obtain multiple second ultrasound channel data.

12. The method for denoising ultrasonic data according to claim 11, characterized in that, The process of merging the data from each of the first ultrasound channels to obtain the first combined channel data includes: After delaying the data from each of the first ultrasonic channels, multiple delayed channel data are obtained. The first combined channel data is obtained by merging the data from each of the delay channels.

13. A method for denoising ultrasonic data, characterized in that, The method includes: Control the ultrasound probe to emit the first ultrasound wave toward the target tissue; The ultrasound probe is controlled to receive the echo channel signal of the first ultrasound returned by the target tissue. The echo channel signal includes multiple first ultrasound channel data, and the multiple first ultrasound channel data correspond one-to-one with multiple receiving array elements on the ultrasound probe. Multiple first ultrasound channel data are merged to obtain first combined channel data; The first combined channel data is input into a machine learning model for noise reduction to obtain the second combined channel data. Image data of the target tissue is obtained by performing image processing based on the second combined channel data; An image of the target tissue is obtained based on the image data, and the image of the target tissue is displayed.

14. The method for denoising ultrasonic data according to claim 13, characterized in that, The step of obtaining image data of the target tissue by image processing based on the second combined channel data includes: The second combined channel data is split to obtain multiple second ultrasound channel data; Beamforming processing is performed on multiple second ultrasound channel data to obtain synthesized radio frequency data; Image data of the target tissue is obtained by performing image processing on the synthesized radio frequency data.

15. The method for denoising ultrasonic data according to claim 13, characterized in that, The step of merging multiple sets of first ultrasound channel data to obtain first combined channel data includes: Beamforming processing is performed on multiple first ultrasound channel data to obtain first combined channel data.

16. The method for denoising ultrasonic data according to claim 13, characterized in that, The step of merging multiple sets of first ultrasound channel data to obtain first combined channel data includes: Multiple sets of the first ultrasound channel data are delayed, and the delayed sets of the first ultrasound channel data are merged to obtain the first combined channel data.

17. The method for denoising ultrasonic data according to claim 13, characterized in that, The machine learning model includes multiple denoising sub-models, each used to remove different types of noise; the step of inputting the first combined channel data into the machine learning model for denoising processing to obtain the second combined channel data includes: The first combined channel data is input into the machine learning model, and noise in the first combined channel data is removed using each of the denoising sub-models to obtain the second combined channel data.

18. The method for denoising ultrasonic data according to claim 17, characterized in that, The plurality of denoising sub-models in the machine learning model include at least one of the following: The average filter sub-model used for average filtering; Alternatively, a channel filtering sub-model for channel filtering; Alternatively, an interpolation filtering sub-model used for interpolation filtering.

19. The method for denoising ultrasonic data according to claim 13, characterized in that, The step of inputting the first combined channel data into a machine learning model for noise reduction processing to obtain the second combined channel data includes one of the following: The first combined channel data is input into a machine learning model to extract effective data, resulting in the second combined channel data. Alternatively, the first combined channel data can be input into a machine learning model for noise extraction to obtain noise data, and the first combined channel data can be denoised based on the noise data to obtain the second combined channel data.

20. An ultrasonic imaging device, characterized in that, include: Ultrasonic probe; A transmitting / receiving circuit is used to control the ultrasound probe to transmit ultrasound waves to the target tissue and receive the echoes of the ultrasound waves. A processor, configured to denoise the echo of the ultrasound to obtain an image of the target tissue; A display for displaying the tissue image; The processor is also used to perform the ultrasonic data denoising method according to any one of claims 1 to 19.

21. A computer-readable storage medium storing computer-executable instructions, characterized in that: The computer-executable instructions are used to execute the ultrasonic data denoising method according to any one of claims 1 to 19.

22. A computer program product, comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the ultrasonic data denoising method according to any one of claims 1 to 19.

Images