Radial artery positioning method and system based on ultrasonic equipment and deep learning, and electronic equipment

Abstract

The invention discloses a radial artery positioning method and system based on ultrasonic equipment and deep learning and electronic equipment. The method comprises the following steps: converting an original ultrasonic signal into a wrist two-dimensional ultrasonic image by utilizing an ultrasonic beam synthesis algorithm in a training stage; marking the two-dimensional ultrasonic image of the wrist, and training a classification model based on deep learning; and converting the real-time ultrasonic signal into a real-time wrist two-dimensional ultrasonic image in a reasoning stage, inputting the real-time wrist two-dimensional ultrasonic image into the trained classification model to obtain a positioning result of the radial artery in the real-time wrist two-dimensional ultrasonic image, and outputting a corresponding voice prompt. According to the scheme, the radial artery positioning accuracy, the operation convenience and the cost controllability are improved.

Description

Technical Field

[0001] The present invention relates to the field of image recognition technology, and in particular to a radial artery positioning method, system and electronic equipment based on ultrasonic equipment and deep learning. Background Art

[0002] The location of the radial artery is of great significance in clinical medicine and health management, especially when taking blood samples, monitoring pulses, or performing other diagnostic and treatment operations. With the continuous development of medical imaging technology and sensor technology, some auxiliary devices and algorithms have been used to try to improve the accuracy and efficiency of radial artery location.

[0003] However, traditional radial artery positioning methods often rely on manual touch or experience-based skills, which have certain operating errors and uncertainties. Although existing technologies have made certain progress in improving the efficiency of radial artery positioning, they still do not fully meet actual needs in terms of positioning accuracy, user convenience, and controllability of equipment costs.

[0004] Due to individual differences in human skin thickness, blood pressure changes, etc., it may be difficult to accurately find the best position when automatically locating the radial artery and detecting pulse signals. Especially in solutions based on robotic arms and dexterous hands, the control requirements for multiple degrees of freedom are high, and a slight deviation will affect the accuracy of positioning. The equipment that collects pulse signals may be interfered by noise, especially when moving. The elasticity and pressure conduction characteristics of the skin can also affect the stability of the sensor, resulting in inconsistent pulse signals, which in turn affects the reliability of diagnosis. The auxiliary system uses a pressure roller to simulate the doctor's technique, but it may not be accurate enough in terms of strength and rhythm control, and cannot fully reproduce the Chinese medicine pulse diagnosis technique, which may lead to deviations in diagnostic results. In addition, the sensor needs to adapt to the accuracy of different pressures, and any slight error will affect the accuracy of the pulse data. The equipment includes components such as robotic arms, dexterous hands, and multi-degree-of-freedom control. The structure is complex, the cost is high, and the maintenance is difficult, which is not conducive to large-scale promotion. At the same time, the complex structure means high-frequency calibration and debugging requirements to ensure that the equipment is always in the best condition. Patients may feel uncomfortable wearing a fixed device or contacting a robotic arm for a long time, especially for the sensitive wrist area, and the discomfort may affect the stability of pulse measurement. In addition, errors or lags in mechanical movements can also increase patient discomfort. Summary of the invention

[0005] In order to solve the problems existing in the prior art, the present invention provides the following technical solutions.

[0006] The present invention first provides a radial artery positioning method based on ultrasound equipment and deep learning in a first aspect, comprising:

[0007] In the training phase, an ultrasonic device is used to detect original ultrasonic signals of the wrist for training, and an ultrasonic beam synthesis algorithm is used to convert the original ultrasonic signals into a two-dimensional ultrasonic image of the wrist for training;

[0008] Classifying and labeling the two-dimensional wrist ultrasound images used for training to form a labeling data set, wherein the labeling data set includes positioning results of the radial artery in the two-dimensional wrist ultrasound images, and using the labeling data set to train a classification model based on deep learning;

[0009] During the inference stage, the real-time wrist ultrasound signal detected by the ultrasound device is converted into a real-time two-dimensional wrist ultrasound image using an ultrasonic beam synthesis algorithm. The real-time two-dimensional wrist ultrasound image is input into the trained classification model to obtain the positioning result of the radial artery in the real-time two-dimensional wrist ultrasound image, and a voice prompt corresponding to the positioning result is output.

[0010] Preferably, the detecting the original wrist ultrasonic signal for training using an ultrasonic device further comprises:

[0011] The ultrasound probe is connected to the main control device via wired or wireless means, and the sampling frequency and detection range of the ultrasound probe are selected according to the anatomical characteristics of the radial artery and the target application scenario.

[0012] Preferably, the converting the original ultrasonic signal into a two-dimensional ultrasonic image of the wrist for training by using an ultrasonic beam synthesis algorithm further comprises:

[0013] Receive the original ultrasonic signal, use a bandpass filter to remove background noise, and perform gain adjustment on the signal amplitude;

[0014] Delay correction is performed according to the scanning angle of the ultrasound probe and the speed of sound wave propagation, the delay time of each receiving channel is dynamically adjusted, and the delay-corrected signals are superimposed and synthesized to generate a two-dimensional ultrasound image of the wrist with depth information.

[0015] Preferably, the converting the original ultrasonic signal into a two-dimensional ultrasonic image of the wrist for training by using an ultrasonic beam synthesis algorithm further comprises:

[0016] The contrast of two-dimensional ultrasound images of the wrist was enhanced by adaptive histogram equalization algorithm, and the vascular edge was optimized by gradient filtering or Sobel operator.

[0017] Preferably, the step of classifying and labeling the two-dimensional wrist ultrasound images for training to form a labeling data set further comprises:

[0018] According to the position of the radial artery signal in the wrist 2D ultrasound image used for training, it is classified into one of four annotation types: offset to the left, offset to the right, exactly in the middle, and completely out of the picture, and the annotation results are output in .json or .xml format;

[0019] The labeled image data is randomly divided into training sets and test sets in proportion. Different usage scenarios are simulated through enhanced operations such as rotation, scaling, or flipping, and the image pixel values ​​are normalized.

[0020] Preferably, the deep learning-based classification model is a Vision Transformer model, a convolutional neural network model CNN or a MobileNet lightweight model.

[0021] Preferably, the outputting a voice prompt corresponding to the positioning result further comprises:

[0022] When the positioning result is a leftward deviation, a voice prompt "Please move the probe to the left" is output;

[0023] When the positioning result is a rightward deviation, a voice prompt "Please move the probe to the right" is output;

[0024] When the positioning result is exactly in the middle, a voice prompt "Please don't move" is output;

[0025] When the positioning result is completely out of the screen, a voice prompt "Please aim the tip of the probe at the base of the palm" is output.

[0026] Another aspect of the present invention provides a radial artery positioning system based on ultrasound equipment and deep learning, comprising:

[0027] An acquisition unit, used for detecting original ultrasonic signals of the wrist for training by using an ultrasonic device during a training phase, and converting the original ultrasonic signals into two-dimensional ultrasonic images of the wrist for training by using an ultrasonic beam synthesis algorithm;

[0028] a training unit, configured to classify and annotate the two-dimensional wrist ultrasound image for training to form an annotated data set, wherein the annotated data set includes a positioning result of the radial artery in the two-dimensional wrist ultrasound image, and to train a classification model based on deep learning using the annotated data set;

[0029] The inference unit is used to, during the inference stage, use the real-time wrist ultrasound signal detected by the ultrasound device, convert the real-time ultrasound signal into a real-time two-dimensional wrist ultrasound image using an ultrasound beam synthesis algorithm, input the real-time two-dimensional wrist ultrasound image into a trained classification model, obtain a positioning result of the radial artery in the real-time two-dimensional wrist ultrasound image, and output a voice prompt corresponding to the positioning result.

[0030] The third aspect of the present invention provides an electronic device, including a processor and a memory, wherein the memory stores multiple instructions, and the processor is used to read the instructions and execute the radial artery positioning method based on ultrasound equipment and deep learning as described in the first aspect above.

[0031] The fourth aspect of the present invention provides a computer-readable storage medium, which stores multiple instructions, and the multiple instructions can be read by a processor and executed by the radial artery positioning method based on ultrasound equipment and deep learning as described in the first aspect above.

[0032] The beneficial effects of the present invention are:

[0033] 1. Easy operation and improved user experience. Users only need to attach the ultrasound probe to their wrist, without having to manually adjust the position of the complex device. Voice prompts can guide users to complete the positioning operation, which reduces user learning and usage costs compared to traditional experience judgment methods;

[0034] 2. High-precision radial artery positioning. The ultrasonic beam synthesis algorithm generates high-resolution images and combines with the deep learning model for accurate classification, effectively solving the problem of inaccurate positioning due to individual differences in traditional methods. The classification accuracy of the four states of the radial artery is high, providing reliable guarantee for subsequent diagnosis or treatment operations.

[0035] 3. Strong real-time performance and rapid feedback. The model inference process is controlled at the millisecond level, and voice prompts are generated in real time. Users can quickly adjust the probe position, significantly improving positioning efficiency. The real-time feedback mechanism reduces the trial and error cost of traditional methods.

[0036] 4. Reduce equipment complexity and cost. Abandoning complex hardware structures such as robotic arms and multi-degree-of-freedom control, relying only on ultrasound equipment and deep learning models, the system architecture is simplified and manufacturing and maintenance costs are reduced. The system's miniaturized design is easy to carry and promote, and is suitable for different scenarios (such as clinical and home health monitoring).

[0037] 5. Enhance the scalability and iteration capabilities of the system. The deep learning model is used as the core algorithm to make the system have good scalability. By continuously collecting new ultrasound data and optimizing the model, it can quickly adapt to the individual characteristics of different users or the needs of new scenarios. Through model upgrades, higher-precision classification and adaptation to more complex scenarios can be achieved, supporting intelligent positioning of other blood vessels or tissues.

[0038] 6. The system has strong adaptability and reliability. It can adapt to individual differences of different patients (such as skin thickness, blood pressure, etc.) and maintain good performance in various usage environments. It provides abnormal state processing mechanism (such as probe position detachment prompt) to effectively enhance the robustness of the system.

[0039] 7. Humanized design, convenient for special groups of people. Voice prompts provide great convenience for people with visual impairments. Users can get real-time feedback on radial artery positioning without relying on the screen. The voice prompts are clear and concise, further expanding the scope of the system's applicable population. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 The present invention is a flowchart of a radial artery positioning method based on ultrasound equipment and deep learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0041] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0042] The method provided by the present invention can be implemented in the following terminal environment, and the terminal may include one or more of the following components: a processor, a memory, and a display screen. The memory stores at least one instruction, and the instruction is loaded and executed by the processor to implement the method described in the following embodiment.

[0043] The processor may include one or more processing cores. The processor uses various interfaces and lines to connect various parts of the entire terminal, and executes various functions of the terminal and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory, and calling data stored in the memory.

[0044] The memory may include random access memory (RAM) or read-only memory (ROM). The memory may be used to store instructions, programs, codes, code sets or instructions.

[0045] The display screen is used to display the user interface of each application.

[0046] In addition, those skilled in the art can understand that the structure of the above terminal does not constitute a limitation on the terminal, and the terminal may include more or fewer components, or combine certain components, or arrange the components differently. For example, the terminal also includes components such as a radio frequency circuit, an input unit, a sensor, an audio circuit, and a power supply, which will not be described in detail here.

[0047] Embodiment 1

[0048] like Figure 1 As shown, the embodiment of the present invention first provides a radial artery positioning method based on ultrasound equipment and deep learning, which is characterized by comprising:

[0049] S1. In a training phase, an ultrasonic device is used to detect original ultrasonic signals of a wrist for training, and an ultrasonic beam synthesis algorithm is used to convert the original ultrasonic signals into two-dimensional ultrasonic images of the wrist for training.

[0050] The ultrasound device is attached to the patient's wrist, and data is collected by detecting the ultrasound signal of the radial artery. To simplify the operation, the original detection results are not directly visualized, but the data is imaged and verified on the computer based on the ultrasound beam synthesis algorithm. The ultrasound beam synthesis algorithm is a technology that generates high-resolution two-dimensional ultrasound images of the wrist by delay correction and signal superposition of multi-channel ultrasound signals. There are four possible detection results: offset to the left, offset to the right, exactly in the middle, and completely out of the picture.

[0051] S2. Classify and annotate the two-dimensional wrist ultrasound images used for training to form an annotated data set, wherein the annotated data set includes positioning results of the radial artery in the two-dimensional wrist ultrasound images, and use the annotated data set to train a classification model based on deep learning.

[0052] The collected ultrasound image data is manually annotated, and the images are classified according to the above four classification criteria to form an annotated data set. Based on the annotated data set, the model is trained offline using deep learning technology (such as the Vision Transformer network) so that it can automatically identify which classification the current image belongs to. Vision Transformer is a deep learning model used to extract global features in images.

[0053] S3. In the inference stage, the real-time wrist ultrasound signal detected by the ultrasound device is converted into a real-time two-dimensional wrist ultrasound image using an ultrasound beam synthesis algorithm. The real-time two-dimensional wrist ultrasound image is input into the trained classification model to obtain the positioning result of the radial artery in the real-time two-dimensional wrist ultrasound image, and a voice prompt corresponding to the positioning result is output.

[0054] In actual use by users, the real-time acquired ultrasound image is input into the trained neural network model, and the model automatically predicts the offset of the current image. Based on the results of the model output, the system provides operation prompts through voice feedback. For example: if the radial artery signal is offset to the left, the prompt "Please move to the left"; if the radial artery signal is offset to the right, the prompt "Please move to the right"; if the radial artery signal is exactly in the middle, the prompt "Please do not move"; if the radial artery signal is completely out of the picture, the prompt "Please aim the tip of the probe at the base of the palm" is given.

[0055] The above-mentioned radial artery positioning method based on ultrasound equipment and deep learning abandons the complex mechanical structure and multi-degree-of-freedom control, relies only on ultrasound equipment and deep learning models, and combines voice prompts to achieve accurate positioning of the radial artery. It is easy to operate, has real-time feedback, low equipment cost, and is easy to promote.

[0056] The solution of the present invention is divided into two processes, the data labeling and training process, and the user actual use process. The main purpose of the data labeling and training process is to build a data set and a training model. The user actual use process mainly uses the model trained in the previous process to provide prompts for users to find the radial artery.

[0057] In a specific embodiment, the above step S1 can be further implemented by a data acquisition module and a data processing module.

[0058] During the device connection stage, the ultrasound probe is connected to the main control device (such as a computer or embedded system) by wire or wireless means. Ensure that the data transmission channel is stable to avoid acquisition interruption due to signal loss. In a preferred embodiment, the sampling frequency and detection range of the ultrasound probe are selected according to the anatomical characteristics of the radial artery and the target application scenario to ensure the optimal balance between signal resolution, radial artery focusing position and processing efficiency. For example, a 128-element ultrasound linear array probe is used with a center frequency of 10Mhz and a sampling frequency of 40Mhz. The detection range is 6 cm long and 2 cm deep below the wrist crease.

[0059] The data acquisition module is used to obtain the signal of the patient's radial artery in real time through the ultrasound probe as the basis for subsequent processing. The collected raw data is preliminarily denoised to remove environmental interference signals and improve the quality of the raw data. The raw signal is stored and transmitted to the subsequent data processing module via wired means.

[0060] The data processing module is used to restore the original ultrasonic signal generated by the data acquisition module into a two-dimensional ultrasonic image of the wrist using an ultrasonic beam synthesis algorithm. This image provides a data basis for subsequent model training and reasoning. The specific process is as follows:

[0061] 1. Signal input and preprocessing:

[0062] Receive the original ultrasonic signal output by the data acquisition module, use a bandpass filter to remove background noise (such as equipment environment interference) to improve signal quality, and adjust the signal amplitude to enhance the visibility of weak signals.

[0063] 2. Ultrasonic beam synthesis:

[0064] The ultrasonic beam synthesis algorithm (such as DAS (Delay-and-Sum) algorithm) is used to restore the ultrasonic signals received by multiple channels into a two-dimensional grayscale image. Delay correction is performed according to the scanning angle of the ultrasonic probe and the propagation speed of the sound wave, and the delay time of each receiving channel is dynamically adjusted. The delay-corrected signals are then superimposed and synthesized to generate a two-dimensional ultrasonic image of the wrist with depth information.

[0065] 3. Image Optimization:

[0066] The image contrast is enhanced by adaptive histogram equalization (CLAHE) algorithm to make the radial artery features clearer. The gradient filter or Sobel operator is used to optimize the blood vessel edge.

[0067] 4. Output data format:

[0068] The optimized image is exported to a standard format (such as .png or .tiff) for easy reading by subsequent training and inference modules. The image contains annotation meta-information, such as acquisition time, device parameters, etc., for easy traceability and verification.

[0069] In a preferred embodiment, the above step S2 is implemented by a training module. The training module manually annotates the ultrasound images generated by the data processing module and constructs a data set, and uses a deep learning algorithm to perform offline training on the model to generate a neural network model capable of classifying radial artery deviation states.

[0070] First, image annotation is performed. Use data annotation software (such as LabelImg or a custom annotation system) to manually classify and annotate the radial artery position in the image. The annotation types are offset to the left, offset to the right, exactly in the middle, and completely out of the picture. According to the position of the radial artery signal in the image, it is classified into the above four situations. The annotation results are output in .json or .xml format and stored in association with the image file.

[0071] Next, we build the dataset. We randomly divide the labeled image data into training and test sets in a ratio of 8:2. We simulate different usage scenarios and improve the generalization ability of the model through enhancement operations such as rotation, scaling, or flipping. We normalize the image pixel values ​​to ensure the consistency of the model input.

[0072] The training module uses the Vision Transformer (ViT) model. The Vision Transformer (ViT) model is used to capture global features and is suitable for processing complex patterns of ultrasound images. The Vision Transformer model includes an input layer, an encoder layer, and a classification head. The input layer receives a normalized two-dimensional ultrasound image of the wrist. The encoder layer uses a multi-layer Transformer encoder to extract global features. The classification head is used to output four classification labels (left deviation, right deviation, center, and detached).

[0073] In an optional embodiment, as an alternative to the Vision Transformer model, other multimodal models or visual models that support image input, such as CLIP, convolutional neural networks (CNN), or lightweight models (such as MobileNet) can be used for reasoning.

[0074] During the training process, standard deep learning optimization methods (such as Adam optimizer) are used to evaluate the model classification performance through the cross entropy loss function (weighted combination of focal loss + dice loss). The model is trained on a high-performance computing environment (such as GPU) to ensure that the classification accuracy meets the clinical application requirements.

[0075] After training is completed, use the test set to evaluate the model performance; adjust the model hyperparameters or structure based on the verification results to optimize the performance.

[0076] Save the trained model in a standard format (such as .pt or .onnx) so that it can be loaded and used in testing and real-time inference.

[0077] In a preferred embodiment, the above step S3 is further implemented by a data acquisition module, a data processing module, a model reasoning module and a voice prompt module.

[0078] The model inference module is used to use the trained neural network model in actual use to infer the real-time acquired ultrasound images, determine the deviation state of the radial artery, and provide real-time operation guidance to the user.

[0079] First, the real-time acquired ultrasound images are used as the input for model reasoning. Noise suppression and normalization operations are performed on the input images to ensure that the data format is consistent with the training phase. The preprocessing results are directly passed to the model for reasoning.

[0080] During the inference process, the neural network model classifies and infers the input image to determine the current state type of the radial artery (shifted to the left; shifted to the right; exactly in the middle; completely out of the picture).

[0081] In a preferred embodiment, the reasoning time is controlled at the millisecond level to ensure that the user can receive feedback information quickly.

[0082] The inference results are output in the form of state labels and passed to the voice prompt module.

[0083] The voice prompt module receives the classification results output by the model inference module and provides real-time voice feedback to the user, guiding the user to adjust the position of the ultrasound device to ensure accurate positioning of the radial artery. The classification results include the following classification labels: offset to the left, offset to the right, exactly in the middle, and completely out of the picture.

[0084] The preset voice prompt can be a pre-recorded voice file. The content is as follows:

[0085] Leftward deviation: prompt "Please move the probe to the left";

[0086] Offset to the right: prompt "Please move the probe to the right";

[0087] Right in the middle: the prompt "Please don't move";

[0088] Completely out of the picture: prompt "Please aim the tip of the probe at the base of your palm."

[0089] Select the corresponding voice prompt content according to the category label.

[0090] The prompt tone is played through the built-in speaker or a connected audio output device. The voice playback volume and frequency are adjustable for clarity.

[0091] The above voice prompt module is tightly coupled with the model inference module to ensure the real-time nature of the prompt, and the delay is controlled within 100 milliseconds. In a further embodiment, if the user does not adjust the device position in time, the prompt is repeated every 2 seconds until the classification state changes.

[0092] It can be seen that the radial artery positioning method based on ultrasound equipment and deep learning of the present invention has the following advantages compared with the prior art methods:

[0093] 1. Easy operation and improved user experience. Users only need to attach the ultrasound probe to their wrist, without having to manually adjust the position of the complex device. Voice prompts can guide users to complete the positioning operation, which reduces user learning and usage costs compared to traditional experience judgment methods;

[0094] 2. High-precision radial artery positioning. The ultrasonic beam synthesis algorithm generates high-resolution images and combines with the deep learning model for accurate classification, effectively solving the problem of inaccurate positioning due to individual differences in traditional methods. The classification accuracy of the four states of the radial artery is high, providing reliable guarantee for subsequent diagnosis or treatment operations.

[0095] 3. Strong real-time performance and rapid feedback. The model inference process is controlled at the millisecond level, and voice prompts are generated in real time. Users can quickly adjust the probe position, significantly improving positioning efficiency. The real-time feedback mechanism reduces the trial and error cost of traditional methods.

[0096] 4. Reduce equipment complexity and cost. Abandoning complex hardware structures such as robotic arms and multi-degree-of-freedom control, relying only on ultrasound equipment and deep learning models, the system architecture is simplified and manufacturing and maintenance costs are reduced. The system's miniaturized design is easy to carry and promote, and is suitable for different scenarios (such as clinical and home health monitoring).

[0097] 5. Enhance the scalability and iteration capabilities of the system. The deep learning model is used as the core algorithm to make the system have good scalability. By continuously collecting new ultrasound data and optimizing the model, it can quickly adapt to the individual characteristics of different users or the needs of new scenarios. Through model upgrades, higher-precision classification, adaptation to more complex scenarios, and support for intelligent positioning of other blood vessels or tissues can be achieved.

[0098] 6. The system has strong adaptability and reliability. It can adapt to individual differences of different patients (such as skin thickness, blood pressure, etc.) and maintain good performance in various usage environments. It provides abnormal state processing mechanism (such as probe position detachment prompt) to effectively enhance the robustness of the system.

[0099] 7. Humanized design, convenient for special groups of people. Voice prompts provide great convenience for people with visual impairments. Users can get real-time feedback on radial artery positioning without relying on the screen. The voice prompts are clear and concise, further expanding the scope of the system's applicable population.

[0100] Embodiment 2

[0101] Another aspect of the present invention also includes a functional module architecture that is completely consistent with the aforementioned method flow, that is, an embodiment of the present invention also provides a radial artery positioning system based on ultrasound equipment and deep learning, including:

[0102] An acquisition unit, used for detecting original ultrasonic signals of the wrist for training by using an ultrasonic device during a training phase, and converting the original ultrasonic signals into two-dimensional ultrasonic images of the wrist for training by using an ultrasonic beam synthesis algorithm;

[0103] a training unit, configured to classify and annotate the two-dimensional wrist ultrasound image for training to form an annotated data set, wherein the annotated data set includes a positioning result of the radial artery in the two-dimensional wrist ultrasound image, and to train a classification model based on deep learning using the annotated data set;

[0104] The inference unit is used to, during the inference stage, use the real-time wrist ultrasound signal detected by the ultrasound device, convert the real-time ultrasound signal into a real-time two-dimensional wrist ultrasound image using an ultrasound beam synthesis algorithm, input the real-time two-dimensional wrist ultrasound image into a trained classification model, obtain a positioning result of the radial artery in the real-time two-dimensional wrist ultrasound image, and output a voice prompt corresponding to the positioning result.

[0105] The system can be implemented through the radial artery positioning method based on ultrasound equipment and deep learning provided in the above-mentioned embodiment 1. The specific implementation method can be found in the description in embodiment 1 and will not be repeated here.

[0106] The present invention also provides a computer-readable memory storing a plurality of instructions, wherein the instructions are used to implement the radial artery positioning method based on ultrasound equipment and deep learning as described in Example 1.

[0107] The present invention also provides an electronic device, comprising a processor and a memory connected to the processor, wherein the memory stores a plurality of instructions, and the instructions can be loaded and executed by the processor so that the processor can execute the method described in the first embodiment.

[0108] Although preferred embodiments of the present invention have been described, additional changes and modifications may be made to these embodiments by those skilled in the art once the basic inventive concepts are known. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention. Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A radial artery positioning method based on ultrasound equipment and deep learning, characterized in that: include: In the training phase, an ultrasonic device is used to detect original ultrasonic signals of the wrist for training, and an ultrasonic beam synthesis algorithm is used to convert the original ultrasonic signals into a two-dimensional ultrasonic image of the wrist for training; Classifying and labeling the two-dimensional wrist ultrasound images used for training to form a labeling data set, wherein the labeling data set includes positioning results of the radial artery in the two-dimensional wrist ultrasound images, and using the labeling data set to train a classification model based on deep learning; During the inference stage, the real-time wrist ultrasound signal detected by the ultrasound device is converted into a real-time two-dimensional wrist ultrasound image using an ultrasonic beam synthesis algorithm. The real-time two-dimensional wrist ultrasound image is input into the trained classification model to obtain the positioning result of the radial artery in the real-time two-dimensional wrist ultrasound image, and a voice prompt corresponding to the positioning result is output.

2. The method according to claim 1, characterized in that The method of using an ultrasonic device to detect the original ultrasonic signal of the wrist for training further includes: The ultrasound probe is connected to the main control device via wired or wireless means, and the sampling frequency and detection range of the ultrasound probe are selected according to the anatomical characteristics of the radial artery and the target application scenario.

3. The method according to claim 1, characterized in that The method of converting the original ultrasonic signal into a two-dimensional ultrasonic image of the wrist for training by using an ultrasonic beam synthesis algorithm further comprises: Receive the original ultrasonic signal, use a bandpass filter to remove background noise, and perform gain adjustment on the signal amplitude; Delay correction is performed according to the scanning angle of the ultrasound probe and the speed of sound wave propagation, the delay time of each receiving channel is dynamically adjusted, and the delay-corrected signals are superimposed and synthesized to generate a two-dimensional ultrasound image of the wrist with depth information.

4. The method according to claim 3, characterized in that The method of converting the original ultrasonic signal into a two-dimensional ultrasonic image of the wrist for training by using an ultrasonic beam synthesis algorithm further comprises: The contrast of two-dimensional ultrasound images of the wrist was enhanced by adaptive histogram equalization algorithm, and the vascular edge was optimized by gradient filtering or Sobel operator.

5. The method according to claim 1, characterized in that The step of classifying and labeling the wrist two-dimensional ultrasound images for training to form a labeling data set further includes: According to the position of the radial artery signal in the wrist 2D ultrasound image used for training, it is classified into one of four annotation types: offset to the left, offset to the right, exactly in the middle, and completely out of the picture, and the annotation results are output in .json or .xml format; The labeled image data is randomly divided into training sets and test sets in proportion. Different usage scenarios are simulated through enhanced operations such as rotation, scaling, or flipping, and the image pixel values ​​are normalized.

6. The method according to claim 1, characterized in that The deep learning-based classification model is a multimodal model or a visual model that supports image input.

7. The method according to claim 1, characterized in that The outputting a voice prompt corresponding to the positioning result further comprises: When the positioning result is a leftward deviation, a voice prompt "Please move the probe to the left" is output; When the positioning result is a rightward deviation, a voice prompt "Please move the probe to the right" is output; When the positioning result is exactly in the middle, a voice prompt "Please don't move" is output; When the positioning result is completely out of the screen, a voice prompt "Please aim the tip of the probe at the base of the palm" is output.

8. A radial artery positioning system based on ultrasound equipment and deep learning, characterized in that: include: An acquisition unit, used for detecting original ultrasonic signals of the wrist for training by using an ultrasonic device during a training phase, and converting the original ultrasonic signals into two-dimensional ultrasonic images of the wrist for training by using an ultrasonic beam synthesis algorithm; A training unit, configured to classify and annotate the two-dimensional wrist ultrasound image for training to form an annotated data set, wherein the annotated data set includes a positioning result of the radial artery in the two-dimensional wrist ultrasound image, and use the annotated data set to train a classification model based on deep learning; The inference unit is used to, during the inference stage, use the real-time wrist ultrasound signal detected by the ultrasound device, convert the real-time ultrasound signal into a real-time two-dimensional wrist ultrasound image using an ultrasound beam synthesis algorithm, input the real-time two-dimensional wrist ultrasound image into a trained classification model, obtain a positioning result of the radial artery in the real-time two-dimensional wrist ultrasound image, and output a voice prompt corresponding to the positioning result.

9. An electronic device, characterized in that: It includes a processor and a memory, the memory stores multiple instructions, and the processor is used to read the instructions and execute the radial artery positioning method based on ultrasound equipment and deep learning as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a plurality of instructions, which can be read by a processor and executed according to any one of claims 1 to 7 as to the radial artery positioning method based on ultrasound equipment and deep learning.

Images