Ultrasonic image recognition method and device

Abstract

The invention discloses an ultrasonic image recognition method and device, and the method comprises the steps: obtaining a first ultrasonic image at a preset initial collection point position, enabling the first ultrasonic image to recognize an anatomical structure, so as to obtain the position and distribution of the anatomical structure, and enabling the position and distribution of the anatomical structure to reflect a target region of an artery in the ultrasonic image. Therefore, the initial acquisition point position is adjusted according to the position and distribution of the target structure, the optimal target acquisition point position is determined, automatic optimization of the acquisition point position is realized, the image quality problem caused by manual adjustment is reduced, and the image definition and resolution are ensured. And the second ultrasonic image is acquired at the target acquisition point according to the specific identification task, so that the matching between the second ultrasonic image and the identification task is higher, and meanwhile, the optimal identification model is selected for different tasks to identify the second ultrasonic image, so that the overall efficiency of identification and the accuracy of the identification result are improved.

Description

Technical Field

[0001] The present invention relates to image processing technology, and in particular to an ultrasonic image recognition method and device. Background Art

[0002] B-mode ultrasound, also known as two-dimensional ultrasound, is a medical imaging technology that uses echo signals reflected by ultrasound in the internal tissues and organs of the human body to generate two-dimensional images. Ultrasound waves are transmitted into the human body through a probe, and the piezoelectric crystal in the probe can convert electrical signals into ultrasound waves. Ultrasound waves propagate in human tissues, and reflection, refraction, and absorption occur when encountering tissues of different densities. The reflected sound waves are received by the probe and converted into electrical signals. The electronic system processes these electrical signals and converts them into grayscale images, and different tissues are displayed with different brightness according to their reflection characteristics of ultrasound waves. Because B-mode ultrasound can provide real-time dynamic images without invasive operations, it is widely used in clinical medicine.

[0003] In the existing B-type ultrasound recognition methods, most of them collect B-type ultrasound images by technicians constantly verbally instructing patients to adjust their body positions. This is highly dependent, and the probe position and angle are adjusted only based on the technicians' experience, resulting in low recognition efficiency of the collected B-type ultrasound images. Summary of the invention

[0004] The invention provides an ultrasonic image recognition method to improve the efficiency and quality of B-type ultrasonic image acquisition and processing.

[0005] In order to solve the above technical problems, an embodiment of the present invention provides an ultrasonic image recognition method, comprising:

[0006] Acquire a first ultrasound image based on a preset initial acquisition point, identify an anatomical structure based on the first ultrasound image, and acquire a structural position and a structural distribution of the anatomical structure;

[0007] Adjusting the initial acquisition points based on the structural position and the structural distribution to determine target acquisition points;

[0008] A second ultrasonic image is acquired based on the recognition task and the target acquisition point, a recognition model is matched based on the recognition task, and the second ultrasonic image is recognized based on the recognition model to acquire an ultrasonic image recognition result.

[0009] The present invention obtains a first ultrasonic image at a preset initial acquisition point, so that the first ultrasonic image identifies the anatomical structure to obtain the position and distribution of the anatomical structure. Since the position and distribution of the anatomical structure can reflect the target area of ​​the artery in the ultrasonic image, the initial acquisition point is adjusted according to the position and distribution of the target structure, and the optimal target acquisition point is determined to achieve automatic optimization of the acquisition site, reduce image quality problems caused by manual adjustment, and ensure image clarity and resolution. In addition, at the target acquisition point, a second ultrasonic image is acquired according to a specific recognition task, so that the second ultrasonic image is more compatible with the recognition task, and at the same time, the optimal recognition model is selected for different tasks to recognize the second ultrasonic image, thereby improving the overall efficiency of recognition and the accuracy of the recognition results.

[0010] In a second aspect, the present invention provides an ultrasonic image recognition device, comprising: a preliminary recognition module, an adjustment module and a target recognition module; the target recognition module comprises a collection unit and a recognition unit;

[0011] The preliminary recognition module is used to acquire a first ultrasound image based on a preset initial acquisition point, recognize an anatomical structure based on the first ultrasound image, and acquire a structural position and a structural distribution of the anatomical structure;

[0012] The adjustment module is used to adjust the initial acquisition point based on the structure position and the structure distribution to determine the target acquisition point;

[0013] The acquisition unit is used to acquire a second ultrasonic image based on the recognition task and the target acquisition point;

[0014] The recognition unit is used to match the recognition model based on the recognition task, recognize the second ultrasound image based on the recognition model, and obtain an ultrasound image recognition result.

[0015] The present invention acquires a first ultrasonic image at a preset initial acquisition point through a preliminary recognition module, thereby recognizing an anatomical structure based on the first ultrasonic image to obtain the position and distribution of the anatomical structure. Since the position and distribution of the anatomical structure can reflect the target area of ​​the artery in the ultrasonic image, the initial acquisition point is adjusted according to the position and distribution of the anatomical structure through an adjustment module to determine the best target acquisition point, thereby realizing automatic optimization of the acquisition site, reducing image quality problems caused by manual adjustment, and ensuring image clarity and resolution. In addition, the acquisition module of the target recognition module acquires a second ultrasonic image at the target acquisition point according to a specific recognition task, so that the second ultrasonic image is more compatible with the recognition task, and at the same time, the optimal recognition model is selected according to the recognition unit for different tasks to recognize the second ultrasonic image, thereby improving the overall efficiency of recognition and the accuracy of the recognition result.

[0016] Furthermore, the preliminary identification module is used to:

[0017] Acquire a first ultrasound image based on a preset initial acquisition point, and perform standardization processing on the first ultrasound image to acquire a first image to be identified;

[0018] Inputting the first image to be identified into a pre-trained segmentation network model to obtain object structure data in the first ultrasound image; the segmentation network model is trained based on labeled ultrasound image data and a supervised learning algorithm;

[0019] The structural position and structural distribution of the anatomical structure are acquired based on the object structure data.

[0020] The present invention collects a first ultrasound image through the preliminary recognition module, performs standardization processing on the first ultrasound image, eliminates inconsistencies in the first ultrasound image, and obtains a first image to be recognized. Since the segmentation network model is obtained through supervised learning training with a large amount of labeled ultrasound image data, it can accurately recognize and segment the structural data in the first image to be recognized. Since the structural data covers all structural information in the first image to be recognized, the position of the anatomical structure and its distribution range can be calculated based on the structural data.

[0021] Furthermore, the adjustment module is used to:

[0022] Calculating adjustment parameters based on a preset arterial structure range, the structure position and the structure distribution, the adjustment parameters including a probe moving distance and a probe deflection angle;

[0023] The initial acquisition point is adjusted based on the probe moving distance and the probe deflection angle to determine the target acquisition point.

[0024] The adjustment module in the present invention calculates adjustment parameters through a preset structural range. Since the preset structural range includes the target area of ​​the structure to be collected or detected, the adjustment parameters can be calculated according to the structural range and the structural position and structural distribution obtained by the current collection. The probe movement distance and the probe deflection angle are adjusted according to the adjustment parameters, and then the target collection site is determined, so that the target collection site better meets the collection requirements.

[0025] Furthermore, the recognition task includes a plaque recognition task, a blood flow analysis task and a spectrum analysis task; the acquisition unit is used to:

[0026] If the recognition task is a plaque recognition task, acquiring a two-dimensional ultrasound image based on the target acquisition point;

[0027] If the recognition task is a blood flow analysis task, acquiring a two-dimensional color ultrasound image based on a real-time imaging mode and the acquisition points;

[0028] If the recognition task is a spectrum analysis task, a spectrum diagram is obtained based on a pulse wave Doppler imaging mode and the target acquisition point.

[0029] The present invention adopts different imaging modes for different recognition tasks based on target acquisition points to generate different types of ultrasonic images, so that the ultrasonic images better meet the requirements of the recognition tasks and improve the recognition efficiency and recognition accuracy of the recognition tasks.

[0030] Furthermore, the identification unit includes a first identification subunit, and the first identification subunit is used to:

[0031] When the recognition task is a plaque recognition task, the recognition model is a first convolutional neural network model;

[0032] Preprocessing the second ultrasound image based on preset image adjustment parameters to obtain a second image to be identified;

[0033] Inputting the second image to be recognized into the first convolutional neural network model to obtain arterial structural parameters, wherein the arterial structural parameters include arterial inner diameter, arterial wall thickness, arterial plaque location, and arterial plaque size; the first convolutional neural network model is constructed based on a multi-layer convolutional neural network;

[0034] The second ultrasound image is annotated based on the arterial structure parameters to obtain the ultrasound image recognition result.

[0035] The present invention performs a preprocessing operation on the second ultrasound image through the first recognition subunit, optimizes the second ultrasound image, improves the quality of the second ultrasound image, and then matches the first convolutional neural network model to the first recognition task, so that the first convolutional neural network model recognizes the second image to be recognized. Since the first recognition task is a plaque recognition task, its main purpose is to recognize the location and size of the plaque on the artery, and the first convolutional neural network model is constructed based on a multi-layer convolutional neural network, which can quickly extract arterial structural features from the image, so the structural parameters of the target object can be quickly and accurately obtained based on the first convolutional neural network model, and then the arterial structural parameters can be marked on the picture, thereby improving the overall efficiency of image recognition and the accuracy of the recognition results.

[0036] Furthermore, the identification unit includes a second identification subunit, and the second identification subunit is used to:

[0037] When the recognition task is a blood flow analysis task, the recognition model is a long short-term memory network model;

[0038] Inputting the second ultrasound image into the long short-term memory network model to obtain blood flow characteristic parameters; the blood flow characteristic parameters include blood flow direction, blood flow velocity and blood flow turbulence;

[0039] The ultrasonic image recognition result is obtained based on a preset blood flow parameter threshold, the blood flow direction, the blood flow velocity and the blood flow turbulence condition.

[0040] In the present invention, the second recognition subunit recognizes the second ultrasound image through the long short-term memory network model. Since the long short-term memory network model can capture the long-term dependencies in the time series data, the real-time two-dimensional color ultrasound image can be recognized based on the long short-term memory network model, and the dynamic characteristics of the fluid movement in the ultrasound image sequence can be analyzed, the blood flow direction, blood flow velocity and blood flow turbulence can be extracted, and the image recognition results can be quickly obtained.

[0041] Furthermore, the identification unit includes a third identification subunit, and the third identification subunit is used to:

[0042] When the recognition task is a spectrum analysis task, the recognition model is a second convolutional neural network model;

[0043] Inputting the second ultrasound image into the second convolutional neural network model to obtain hemodynamic parameters; the second convolutional neural network model is constructed based on a convolutional neural network and a multi-head self-attention mechanism; the hemodynamic parameters include peak systolic velocity and end-diastolic velocity;

[0044] The ultrasound image recognition result is obtained based on a preset dynamic parameter threshold and the peak systolic velocity and the end-diastolic velocity.

[0045] In the present invention, the third recognition subunit recognizes the second ultrasound image through the second convolutional neural network model. Since the second convolutional neural network model is constructed based on the convolutional neural network and the multi-head self-attention mechanism, the second convolutional neural network model has an enhanced ability to extract complex fluid dynamic features. Therefore, the hemodynamic parameters can be quickly acquired and the image recognition efficiency can be improved. Since the preset hemodynamic parameter threshold defines the boundary between normal and abnormal hemodynamic parameters, the image recognition result can be automatically generated according to the hemodynamic parameter threshold, thereby improving the image recognition efficiency and accuracy.

[0046] Furthermore, the ultrasonic image recognition device further includes a self-checking module, and the self-checking module is used to:

[0047] Acquire the probe connection state and the display resolution, determine whether the probe connection state is abnormal based on a preset signal transmission threshold, and obtain a first determination result;

[0048] Determine whether the display resolution is abnormal based on a preset resolution threshold, and obtain a second determination result;

[0049] A detection report is generated based on the first judgment result and the second judgment result, and an early warning is issued based on the detection report.

[0050] The self-check module of the present invention performs component self-check through the probe connection status and the display resolution, and generates a test report, thereby improving the reliability of the device through regular automatic testing.

[0051] Furthermore, the ultrasonic image recognition device further includes a visualization module, and the visualization module is used to:

[0052] The ultrasonic image recognition result is obtained, and the ultrasonic image recognition result is visualized. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 A schematic diagram of a flow chart of an ultrasonic image recognition method provided in an embodiment of the present invention;

[0054] Figure 2 A schematic structural diagram of an ultrasonic image recognition device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0055] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0056] The terms "first" and "second" and the like in the specification, claims and drawings of this application are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device comprising a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products or devices.

[0057] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0058] Example 1

[0059] See also Figure 1, Figure 1 The following is a flow chart of an ultrasonic image recognition method provided in an embodiment of the present invention. The present invention provides an ultrasonic image recognition method, including steps 101 to 103, which are as follows:

[0060] Step 101: acquiring a first ultrasound image based on a preset initial acquisition point, identifying an anatomical structure based on the first ultrasound image, and acquiring a structural position and a structural distribution of the anatomical structure;

[0061] Step 102: adjusting the initial acquisition points based on the structural position and the structural distribution to determine target acquisition points;

[0062] Step 103: Acquire a second ultrasonic image based on the recognition task and the target acquisition point, match a recognition model based on the recognition task, recognize the second ultrasonic image based on the recognition model, and obtain an ultrasonic image recognition result.

[0063] The present invention obtains a first ultrasonic image at a preset initial acquisition point, so that the first ultrasonic image identifies the anatomical structure to obtain the position and distribution of the anatomical structure. Since the position and distribution of the anatomical structure can reflect the target area of ​​the artery in the ultrasonic image, the initial acquisition point is adjusted according to the position and distribution of the target structure, and the optimal target acquisition point is determined to achieve automatic optimization of the acquisition site, reduce image quality problems caused by manual adjustment, and ensure image clarity and resolution. In addition, at the target acquisition point, a second ultrasonic image is acquired according to a specific recognition task, so that the second ultrasonic image is more compatible with the recognition task, and at the same time, the optimal recognition model is selected for different tasks to recognize the second ultrasonic image, thereby improving the overall efficiency of recognition and the accuracy of the recognition results.

[0064] In this embodiment, acquiring a first ultrasound image based on a preset initial acquisition point, identifying an anatomical structure based on the first ultrasound image, and acquiring a structural position and a structural distribution of the anatomical structure include:

[0065] Acquire a first ultrasound image based on a preset initial acquisition point, and perform standardization processing on the first ultrasound image to acquire a first image to be identified;

[0066] Inputting the first image to be identified into a pre-trained segmentation network model to obtain object structure data in the first ultrasound image; the segmentation network model is trained based on labeled ultrasound image data and a supervised learning algorithm;

[0067] The structural position and structural distribution of the anatomical structure are acquired based on the object structure data.

[0068] The present invention collects a first ultrasound image and performs standardization processing on it to eliminate inconsistencies in the first ultrasound image and obtain a first image to be identified. Since the segmentation network model is obtained by supervised learning training through a large amount of labeled ultrasound image data, it can accurately identify and segment the structural data in the first image to be identified. Since the structural data covers all structural information in the first image to be identified, the position of the anatomical structure and its distribution range can be calculated based on the structural data.

[0069] In this embodiment, the adjusting the initial acquisition point based on the structure position and the structure distribution to determine the target acquisition point includes:

[0070] Calculating adjustment parameters based on a preset arterial structure range, the structure position and the structure distribution, the adjustment parameters including a probe moving distance and a probe deflection angle;

[0071] The initial acquisition point is adjusted based on the probe moving distance and the probe deflection angle to determine the target acquisition point.

[0072] The present invention calculates adjustment parameters through a preset structure range. Since the preset structure range includes the target area of ​​the structure to be collected or detected, the adjustment parameters can be calculated according to the structure range and the structure position and structure distribution obtained by the current collection. The probe movement distance and the probe deflection angle are adjusted according to the adjustment parameters, and then the target collection site is determined, so that the target collection site better meets the collection requirements.

[0073] In this embodiment, the recognition task includes a plaque recognition task, a blood flow analysis task, and a spectrum analysis task; and acquiring a second ultrasound image based on the recognition task and the target acquisition point includes:

[0074] If the recognition task is a plaque recognition task, acquiring a two-dimensional ultrasound image based on the target acquisition point;

[0075] If the recognition task is a blood flow analysis task, acquiring a two-dimensional color ultrasound image based on a real-time imaging mode and the acquisition points;

[0076] If the recognition task is a spectrum analysis task, a spectrum diagram is obtained based on a pulse wave Doppler imaging mode and the target acquisition point.

[0077] The present invention adopts different imaging modes for different recognition tasks based on target acquisition points to generate different types of ultrasonic images, so that the ultrasonic images better meet the requirements of the recognition tasks and improve the recognition efficiency and recognition accuracy of the recognition tasks.

[0078] In this embodiment, matching the recognition model based on the recognition task, recognizing the second ultrasound image based on the recognition model, and obtaining the ultrasound image recognition result includes:

[0079] When the recognition task is a plaque recognition task, the recognition model is a first convolutional neural network model;

[0080] Preprocessing the second ultrasound image based on preset image adjustment parameters to obtain a second image to be identified;

[0081] Inputting the second image to be recognized into the first convolutional neural network model to obtain arterial structural parameters, wherein the arterial structural parameters include arterial inner diameter, arterial wall thickness, arterial plaque location, and arterial plaque size; the first convolutional neural network model is constructed based on a multi-layer convolutional neural network;

[0082] The second ultrasound image is annotated based on the arterial structure parameters to obtain the ultrasound image recognition result.

[0083] The present invention performs a preprocessing operation on the second ultrasound image through the first recognition subunit, optimizes the second ultrasound image, improves the quality of the second ultrasound image, and then matches the first convolutional neural network model to the first recognition task, so that the first convolutional neural network model recognizes the second image to be recognized. Since the first recognition task is a plaque recognition task, its main purpose is to recognize the location and size of the plaque on the artery, and the first convolutional neural network model is constructed based on a multi-layer convolutional neural network, which can quickly extract arterial structural features from the image, so the structural parameters of the target object can be quickly and accurately obtained based on the first convolutional neural network model, and then the arterial structural parameters can be marked on the picture, thereby improving the overall efficiency of image recognition and the accuracy of the recognition results.

[0084] In this embodiment, matching the recognition model based on the recognition task, recognizing the second ultrasound image based on the recognition model, and obtaining the ultrasound image recognition result includes:

[0085] When the recognition task is a blood flow analysis task, the recognition model is a long short-term memory network model;

[0086] Inputting the second ultrasound image into the long short-term memory network model to obtain blood flow characteristic parameters; the blood flow characteristic parameters include blood flow direction, blood flow velocity and blood flow turbulence;

[0087] The ultrasonic image recognition result is obtained based on a preset blood flow parameter threshold, the blood flow direction, the blood flow velocity and the blood flow turbulence condition.

[0088] In the present invention, the second ultrasound image is identified by a long short-term memory network model. Since the long short-term memory network model can capture the long-term dependencies in time series data, the real-time two-dimensional color ultrasound image can be identified based on the long short-term memory network model, and the dynamic characteristics of the fluid movement in the ultrasound image sequence can be analyzed, the blood flow direction, blood flow velocity and blood flow turbulence can be extracted, and the image recognition result can be quickly obtained.

[0089] In this embodiment, matching the recognition model based on the recognition task, recognizing the second ultrasound image based on the recognition model, and obtaining the ultrasound image recognition result includes:

[0090] When the recognition task is the third recognition task, the recognition model is the second convolutional neural network model,

[0091] Inputting the second ultrasound image into the second convolutional neural network model to obtain hemodynamic parameters; the second convolutional neural network model is constructed based on a convolutional neural network and a multi-head self-attention mechanism; the hemodynamic parameters include peak systolic velocity and end-diastolic velocity;

[0092] The ultrasound image recognition result is obtained based on a preset dynamic parameter threshold and the peak systolic velocity and the end-diastolic velocity.

[0093] The present invention identifies the second ultrasound image through a second convolutional neural network model. Since the second convolutional neural network model is constructed based on the convolutional neural network and the multi-head self-attention mechanism, the ability of the second convolutional neural network model to extract complex fluid dynamics features is enhanced. Therefore, hemodynamic parameters can be quickly acquired and image recognition efficiency can be improved. Since the preset hemodynamic parameter threshold defines the boundary between normal and abnormal hemodynamic parameters, image recognition results can be automatically generated according to the hemodynamic parameter threshold, thereby improving image recognition efficiency and accuracy.

[0094] An embodiment of the present invention further provides an ultrasonic image recognition device, which performs ultrasonic image recognition based on the above ultrasonic image recognition method.

[0095] Please refer to Figure 2 , Figure 2 A schematic diagram of the structure of an ultrasonic image recognition device provided by an embodiment of the present invention includes: a preliminary recognition module 201, an adjustment module 202 and a target recognition module 203; the target recognition module 203 includes a collection unit 2031 and a recognition unit 2032;

[0096] The preliminary recognition module 201 is used to acquire a first ultrasound image based on a preset initial acquisition point, recognize an anatomical structure based on the first ultrasound image, and acquire a structural position and a structural distribution of the anatomical structure;

[0097] The adjustment module 202 is used to adjust the initial acquisition point based on the structure position and the structure distribution to determine the target acquisition point;

[0098] The acquisition unit 2031 is used to acquire a second ultrasonic image based on the recognition task and the target acquisition point;

[0099] The recognition unit 2032 is used to match the recognition model based on the recognition task, recognize the second ultrasound image based on the recognition model, and obtain an ultrasound image recognition result.

[0100] In this embodiment, the preliminary identification module 201 is used to:

[0101] Acquire a first ultrasound image based on a preset initial acquisition point, and perform standardization processing on the first ultrasound image to acquire a first image to be identified;

[0102] Inputting the first image to be identified into a pre-trained segmentation network model to obtain object structure data in the first ultrasound image; the segmentation network model is trained based on labeled ultrasound image data and a supervised learning algorithm;

[0103] The structural position and structural distribution of the anatomical structure are acquired based on the object structure data.

[0104] In this embodiment, the first ultrasound image is collected by the preliminary recognition module and is standardized to eliminate inconsistencies in the first ultrasound image, so as to obtain a first image to be recognized. Since the segmentation network model is obtained by supervised learning training with a large amount of labeled ultrasound image data, it can accurately recognize and segment the structural data in the first image to be recognized. Since the structural data covers all structural information in the first image to be recognized, the position of the anatomical structure and its distribution range can be calculated based on the structural data.

[0105] In this embodiment, the preliminary recognition module receives the image data sent back by the probe in real time, and performs preprocessing operations such as image standardization, data enhancement and data annotation on the image data. Subsequently, the preprocessed image is input into a pre-trained segmentation network model to identify the anatomical structure in the image, including the position and morphology of the artery.

[0106] In this embodiment, the segmentation network model is obtained by supervised learning training through a large amount of labeled ultrasound image data, and can accurately identify and segment the anatomical structure in the image. The segmentation network model includes U-Net, DeepLab, Mask R-CNN, etc. The preprocessed image is input into the segmentation network model. The model will classify each pixel in the image according to the learned features and patterns, and generate a segmentation map of the anatomical structure. Each pixel in the segmentation map is marked as a specific anatomical structure category. These marking information is the anatomical data. Then, by statistically analyzing the marking information in the segmentation map, the spatial distribution of the anatomical structure in the image can be obtained, and the density, shape, size and other characteristics of the anatomical structure can be analyzed to obtain the structural position and distribution of the anatomical structure.

[0107] As a specific example of an embodiment of the present invention, the segmentation network model is constructed based on the U-Net model, which includes an encoder and a decoder. The decoder gradually reduces the spatial dimension and increases the number of channels to extract high-level semantic features, and then the encoder restores the spatial dimension and fuses the features of the corresponding layer of the encoder to achieve accurate segmentation.

[0108] In this embodiment, the segmentation network model is trained using the cross entropy loss function and the Adam optimizer, and the segmentation results obtained by the model are post-processed, such as morphological operations (opening and closing operations) to remove noise and smooth the boundaries. At the same time, the model output is further analyzed according to actual needs, such as calculating parameters such as the inner diameter and area of ​​the artery.

[0109] In this embodiment, it is also necessary to use indicators such as the Dice coefficient and IoU (Intersection over Union) to evaluate the performance of the model in anatomical structure recognition in order to optimize the model.

[0110] In this embodiment, the adjustment module 202 is used to:

[0111] Calculating adjustment parameters based on a preset arterial structure range, the structure position and the structure distribution, the adjustment parameters including a probe moving distance and a probe deflection angle;

[0112] The initial acquisition point is adjusted based on the probe moving distance and the probe deflection angle to determine the target acquisition point.

[0113] In this embodiment, in order to ensure that high-quality ultrasound images are collected and the structural information of the target area is accurately obtained, the collection points of the probe need to be adjusted.

[0114] In this embodiment, the normal range of the target arterial structure (such as diameter, length, etc.) is predefined, and the specific position of the target arterial structure and the distribution characteristics of the target arterial structure (such as direction, branching, etc.) are obtained through ultrasonic image recognition. Therefore, according to the deviation between the position of the target structure and the current position of the probe, the distance that the probe needs to move is calculated; according to the distribution characteristics of the target structure (such as angle, curvature, etc.), the deflection angle that the probe needs to adjust is calculated.

[0115] In this embodiment, the parameters are calculated and adjusted to ensure that the probe can be accurately moved to the actual position of the target arterial structure, and the probe angle is adjusted according to the characteristics of the target structure to ensure that the ultrasound can optimally cover the target area and avoid occlusion or signal attenuation.

[0116] In this embodiment, the distance that the probe moves in a specific direction (such as front and back, left and right, up and down) is controlled based on the probe movement distance and the probe deflection angle, or the rotation angle of the probe relative to the initial direction (such as up and down tilt angle, left and right tilt angle, etc.) is controlled. This reduces the time and error of manual adjustment and improves the acquisition efficiency. It ensures that the acquired ultrasound image can accurately reflect the morphology and dynamic characteristics of the target artery structure.

[0117] In this embodiment, the adjustment module calculates the adjustment parameters through a preset structure range. Since the preset structure range includes the target area of ​​the structure to be collected or detected, the adjustment parameters can be calculated according to the structure range and the structure position and structure distribution obtained by the current collection. The probe movement distance and the probe deflection angle are adjusted according to the adjustment parameters, and then the target collection site is determined, so that the target collection site better meets the collection requirements.

[0118] In this embodiment, the recognition task includes a plaque recognition task, a blood flow analysis task and a spectrum analysis task; the acquisition unit 2031 is used to:

[0119] If the recognition task is a plaque recognition task, acquiring a two-dimensional ultrasound image based on the target acquisition point;

[0120] If the recognition task is a blood flow analysis task, acquiring a two-dimensional color ultrasound image based on a real-time imaging mode and the acquisition points;

[0121] If the recognition task is a spectrum analysis task, a spectrum diagram is obtained based on a pulse wave Doppler imaging mode and the target acquisition point.

[0122] The present invention adopts different imaging modes for different recognition tasks based on target acquisition points to generate different types of ultrasonic images, so that the ultrasonic images better meet the requirements of the recognition tasks and improve the recognition efficiency and recognition accuracy of the recognition tasks.

[0123] In this embodiment, if the recognition task is a plaque recognition task, the imaging mode is two-dimensional ultrasound imaging, so as to obtain a two-dimensional ultrasound image, wherein the two-dimensional ultrasound imaging can clearly display the structure of the blood vessel wall, including the location, size and shape of the plaque. By analyzing the two-dimensional image, the stability of the plaque can be evaluated, such as whether there is a vulnerable plaque.

[0124] In this embodiment, if the recognition task is a spectrum analysis task, the imaging mode is two-dimensional color ultrasound imaging in real-time imaging mode, so that a two-dimensional color ultrasound image is acquired in real time. Color Doppler ultrasound can display the direction and speed of blood flow in real time, expressed in different colors, and evaluate the blood flow characteristics in the blood vessels by measuring parameters such as blood flow velocity and blood flow.

[0125] In this embodiment, if the identification task is a spectrum analysis task, the imaging mode is pulse wave Doppler imaging mode, so as to obtain a spectrum diagram, and pulse wave Doppler can accurately measure the speed and direction of blood flow. Through the spectrum diagram, the spectrum characteristics of blood flow, such as peak velocity, spectrum width, etc., can be analyzed to evaluate the degree of vascular stenosis and judge the blood flow state.

[0126] In this embodiment, the most suitable imaging mode and acquisition point are selected according to different recognition tasks to obtain the most relevant ultrasound data. By selecting the appropriate imaging technology, the target features can be displayed and measured more accurately, thereby improving the accuracy of diagnosis. And using different imaging modes for different tasks can more effectively utilize equipment resources and improve work efficiency. Combining multiple imaging modes provides comprehensive vascular structure and function information to support comprehensive analysis and diagnosis.

[0127] In this embodiment, the identification unit 2032 includes a first identification subunit, and the first identification subunit is used to:

[0128] When the recognition task is a plaque recognition task, the recognition model is a first convolutional neural network model;

[0129] Preprocessing the second ultrasound image based on preset image adjustment parameters to obtain a second image to be identified;

[0130] Inputting the second image to be recognized into the first convolutional neural network model to obtain arterial structural parameters, wherein the arterial structural parameters include arterial inner diameter, arterial wall thickness, arterial plaque location, and arterial plaque size; the first convolutional neural network model is constructed based on a multi-layer convolutional neural network;

[0131] The second ultrasound image is annotated based on the arterial structure parameters to obtain the ultrasound image recognition result.

[0132] The present invention performs a preprocessing operation on the second ultrasound image through the first recognition subunit, optimizes the second ultrasound image, improves the quality of the second ultrasound image, and then matches the first convolutional neural network model to the first recognition task, so that the first convolutional neural network model recognizes the second image to be recognized. Since the first recognition task is a plaque recognition task, its main purpose is to recognize the location and size of the plaque on the artery, and the first convolutional neural network model is constructed based on a multi-layer convolutional neural network, which can quickly extract arterial structural features from the image, so the structural parameters of the target object can be quickly and accurately obtained based on the first convolutional neural network model, and then the arterial structural parameters can be marked on the picture, thereby improving the overall efficiency of image recognition and the accuracy of the recognition results.

[0133] In this embodiment, in the plaque recognition task, the first convolutional neural network model (built based on a multi-layer convolutional neural network) is used to process the second ultrasound image, with the aim of extracting arterial structural parameters (such as arterial inner diameter, arterial wall thickness, arterial plaque location and size) from the ultrasound image, and annotating the image with these parameters to ultimately generate an ultrasound image recognition result.

[0134] In this embodiment, image adjustment parameters (such as contrast enhancement, noise removal, image size adjustment, etc.) of the preprocessing operation are predefined. Thus, according to the image adjustment parameters, the brightness and contrast of the image are improved, the arterial structure features are highlighted, the noise in the image is removed, the image quality is improved, and the image size is adjusted to meet the input requirements of the model.

[0135] In this embodiment, a first convolutional neural network model is constructed based on a multi-layer convolutional neural network, and high-level features (such as the edge, texture, structure, etc. of the artery) in the image are gradually extracted by using multi-layer convolution and pooling operations. Thus, the arterial structural parameters are automatically extracted through the convolutional neural network, reducing the workload of manual analysis.

[0136] As a specific example of an embodiment of the present invention, a large amount of high-quality medical imaging data of PW signals is collected, and experts accurately mark key features such as arterial inner diameter, IMT and plaque location, and perform normalization, denoising, enhancement and other operations on the image data to ensure data quality.

[0137] The first convolutional neural network model includes a multi-layer convolutional neural network, which extracts image features based on the multi-layer convolutional neural network and realizes accurate detection of arterial inner diameter, IMT and plaque location through multi-scale feature fusion.

[0138] In this embodiment, target detection algorithms such as Faster R-CNN or YOLOv5 are used to locate arterial structures and plaque areas, combined with U-Net or Mask R-CNN for accurate segmentation, and deep learning is used to detect key points to measure arterial diameter and IMT. The position of key points in the image is predicted by regression or classification methods, and then their distance is calculated. When plaques are detected, their area, volume and other features can be further calculated.

[0139] In this embodiment, the identification unit 2032 includes a second identification subunit, and the second identification subunit is used to:

[0140] When the recognition task is a blood flow analysis task, the recognition model is a long short-term memory network model;

[0141] Inputting the second ultrasound image into the long short-term memory network model to obtain blood flow characteristic parameters; the blood flow characteristic parameters include blood flow direction, blood flow velocity and blood flow turbulence;

[0142] The ultrasonic image recognition result is obtained based on a preset blood flow parameter threshold, the blood flow direction, the blood flow velocity and the blood flow turbulence condition.

[0143] In the present invention, the second recognition subunit recognizes the second ultrasound image through the long short-term memory network model. Since the long short-term memory network model can capture the long-term dependencies in the time series data, the real-time two-dimensional color ultrasound image can be recognized based on the long short-term memory network model, and the dynamic characteristics of the fluid movement in the ultrasound image sequence can be analyzed, the blood flow direction, blood flow velocity and blood flow turbulence can be extracted, and the image recognition results can be quickly obtained.

[0144] In this embodiment, in the blood flow analysis task, a long short-term memory network model (LSTM) is used to process the second ultrasound image, with the aim of extracting blood flow characteristic parameters (such as blood flow direction, blood flow velocity, and blood flow turbulence) from the image, and analyzing these parameters according to preset blood flow parameter thresholds, and finally generating an ultrasound image recognition result.

[0145] In this embodiment, LSTM is a deep learning model suitable for time series data, which can capture long-term dependencies in the data. In blood flow analysis, LSTM can extract dynamic characteristics of blood flow (such as blood flow direction, speed and turbulence) from image sequences or dynamic videos to ensure the reliability of parameters.

[0146] In this embodiment, according to clinical needs or pathological characteristics, thresholds for blood flow direction, velocity and turbulence (such as normal blood flow velocity range, upper limit of turbulence intensity, etc.) are pre-set, and recognition results are generated in sequence by judging whether the blood flow is heading toward or away from the target area; or comparing the blood flow velocity with the threshold, judging whether there is an abnormal velocity (such as high-speed or low-speed blood flow) or evaluating whether the turbulence intensity exceeds the normal range, judging whether there is an abnormal blood flow state (such as blood flow disorder). The recognition results include text reports (such as blood flow state assessment) or image annotation results.

[0147] In this embodiment, automatic analysis of blood flow status is achieved by comparing preset thresholds and blood flow characteristic parameters, and based on the recognition results, an accurate blood flow status assessment basis is provided for the doctor's clinical diagnosis.

[0148] In this embodiment, the identification unit includes a third identification subunit, and the third identification subunit is used to:

[0149] When the recognition task is a spectrum analysis task, the recognition model is a second convolutional neural network model;

[0150] Inputting the second ultrasound image into the second convolutional neural network model to obtain hemodynamic parameters; the second convolutional neural network model is constructed based on a convolutional neural network and a multi-head self-attention mechanism; the hemodynamic parameters include peak systolic velocity and end-diastolic velocity;

[0151] The ultrasound image recognition result is obtained based on a preset dynamic parameter threshold and the peak systolic velocity and the end-diastolic velocity.

[0152] In the present invention, the third recognition subunit recognizes the second ultrasound image through the second convolutional neural network model. Since the second convolutional neural network model is constructed based on the convolutional neural network and the multi-head self-attention mechanism, the second convolutional neural network model has an enhanced ability to extract complex fluid dynamic features. Therefore, the hemodynamic parameters can be quickly acquired and the image recognition efficiency can be improved. Since the preset hemodynamic parameter threshold defines the boundary between normal and abnormal hemodynamic parameters, the image recognition result can be automatically generated according to the hemodynamic parameter threshold, thereby improving the image recognition efficiency and accuracy.

[0153] In this embodiment, in the spectrum analysis task, a second convolutional neural network model (built based on a convolutional neural network and a multi-head self-attention mechanism) is used to process the second ultrasound image, with the aim of extracting hemodynamic parameters (such as peak systolic velocity and end-diastolic velocity) from the image, and analyzing these parameters according to preset dynamic parameter thresholds, and finally generating an ultrasound image recognition result.

[0154] In this embodiment, the spatial features of the image (such as the structure and morphology of blood flow) are extracted based on a convolutional neural network (CNN), and the key areas and features in the image (such as the areas with high dynamic changes in blood flow) are captured based on a multi-head self-attention mechanism. Hemodynamic parameters (such as peak systolic velocity and end-diastolic velocity) are extracted from ultrasound images.

[0155] In this embodiment, according to clinical needs or pathological characteristics, thresholds of hemodynamic parameters (such as normal systolic velocity range, upper limit of end-diastolic velocity, etc.) are pre-set, so as to compare the peak systolic velocity with the threshold to determine whether there is an abnormality (such as velocity is too high or too low); or compare the end-diastolic velocity with the threshold to determine whether there is an abnormality (such as velocity is too high or too low), and then generate recognition results based on the comparison results, and the recognition results include text reports (such as blood flow status assessment) or image annotation results.

[0156] In this embodiment, the automatic analysis of blood flow status is realized by comparing the preset threshold and the hemodynamic parameters. Based on the recognition results, an accurate blood flow status evaluation basis is provided for the doctor's clinical diagnosis.

[0157] In this embodiment, the ultrasonic image recognition device further includes a self-checking module, and the self-checking module is used to:

[0158] Acquire the probe connection state and the display resolution, determine whether the probe connection state is abnormal based on a preset signal transmission threshold, and obtain a first determination result;

[0159] Determine whether the display resolution is abnormal based on a preset resolution threshold, and obtain a second determination result;

[0160] A detection report is generated based on the first judgment result and the second judgment result, and an early warning is issued based on the detection report.

[0161] In this embodiment, the self-test module of the present invention performs component self-test through the probe connection status and display resolution, and generates a test report, thereby improving the reliability of the device through regular automated testing.

[0162] In this embodiment, the ultrasonic image recognition device further includes a visualization module, and the visualization module is used to:

[0163] The ultrasonic image recognition result is obtained, and the ultrasonic image recognition result is visualized.

[0164] In this embodiment, when the ultrasonic image recognition device performs recognition, the technician starts the device, and the device self-check module automatically runs the self-check program. The system detects the probe connection status, display resolution, computing unit performance, etc., and generates a self-check report. If the probe connection is found to be unstable, the system issues an alarm to remind the technician to adjust the probe connection.

[0165] In this embodiment, when the patient enters the examination room and lies on the examination bed, the visualization module of the device can provide the patient with graphic or voice prompts to assist the patient in adjusting the body position. For example, the patient is prompted to lie on his back, tilt his head slightly back and turn to one side. The feedback system based on image recognition monitors the patient's body position in real time and adjusts the instructions according to the anatomical model to ensure the best imaging posture.

[0166] In this embodiment, the technician starts scanning the carotid artery, and the preliminary recognition module receives and analyzes the image data in real time. The adjustment module calculates the probe adjustment parameters, and then displays the adjustment parameters according to the visualization module, for example, prompting "Please move the probe 2 cm to the left and adjust the angle downward by 5 degrees."

[0167] In this embodiment, the imaging mode is switched for the recognition task by the acquisition unit.

[0168] In this embodiment, when the recognition task is a plaque recognition task, after the initial image is acquired by the first recognition subunit system, the contrast, brightness and other parameters are automatically adjusted to improve the image quality, and the deep learning algorithm is used to identify the inner diameter of the artery, IMT and plaque location, and automatically calculate related parameters to generate a standardized measurement report containing all key data and their location marks.

[0169] In this embodiment, when the recognition task is a blood flow analysis task, real-time blood flow analysis and abnormal prompts are performed, the imaging mode is switched to the color Doppler mode, and the second recognition subunit starts to analyze the blood flow data in real time. For example, an abnormal increase in the blood flow velocity of the right carotid artery is detected, indicating that there may be stenosis, and the area of ​​interest is marked on the image.

[0170] In this embodiment, when the recognition task is a spectrum analysis task, the imaging mode is switched to the pulse wave Doppler mode, and the third recognition subunit automatically captures the PW signal and displays the spectrum diagram, calculates important parameters such as PSV and EDV, and generates a detailed blood flow spectrum report. The report marks the abnormal value area and provides an explanation to help doctors make subsequent diagnoses.

[0171] In this embodiment, the first ultrasound image is acquired at a preset initial acquisition point, so that the first ultrasound image recognizes the anatomical structure to obtain the position and distribution of the anatomical structure. Since the position and distribution of the anatomical structure can reflect the target area of ​​the artery in the ultrasound image, the initial acquisition point is adjusted according to the position and distribution of the target structure, and the best target acquisition point is determined to achieve automatic optimization of the acquisition site, reduce image quality problems caused by manual adjustment, and ensure image clarity and resolution. In addition, at the target acquisition point, the second ultrasound image is acquired according to the specific recognition task, so that the second ultrasound image is more compatible with the recognition task, and the optimal recognition model is selected for different tasks to recognize the second ultrasound image, thereby improving the overall efficiency of recognition and the accuracy of the recognition result.

[0172] In an embodiment of the present invention, an ultrasound image recognition device is also provided, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and the processor implements the above-mentioned ultrasound image recognition method when executing the computer program.

[0173] In an embodiment of the present invention, a computer-readable storage medium is further provided. The computer-readable storage medium includes a stored computer program. When the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the above-mentioned ultrasound image recognition method.

[0174] Exemplarily, the computer program can be divided into one or more modules, one or more modules are stored in the memory and executed by the processor to complete the present invention. One or more modules can be a series of computer program instruction segments that can complete specific functions, and the instruction segments are used to describe the execution process of the computer program in the ultrasound image recognition device.

[0175] The ultrasonic image recognition device may be a computing device such as a desktop computer, a notebook, a PDA, and a cloud server. The ultrasonic image recognition device may include, but is not limited to, a processor, a memory, and a display. Those skilled in the art may understand that the above components are merely examples of ultrasonic image recognition devices and do not constitute a limitation on the ultrasonic image recognition device. The ultrasonic image recognition device may include more or fewer components than the components, or a combination of certain components, or different components. For example, the ultrasonic image recognition device may also include input and output devices, network access devices, buses, etc.

[0176] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the ultrasonic image recognition device, and uses various interfaces and lines to connect various parts of the entire ultrasonic image recognition device.

[0177] The memory can be used to store computer programs and / or modules. The processor realizes various functions of the ultrasonic image recognition device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function (such as a sound playback function, a text conversion function, etc.); the data storage area can store data created according to the use of the mobile phone (such as audio data, text message data, etc.). In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0178] Wherein, if the module based on ultrasonic image recognition is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0179] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. An ultrasonic image recognition method, characterized in that: include: Acquire a first ultrasound image based on a preset initial acquisition point, identify an anatomical structure based on the first ultrasound image, and acquire a structural position and a structural distribution of the anatomical structure; Adjusting the initial acquisition points based on the structural position and the structural distribution to determine target acquisition points; A second ultrasonic image is acquired based on the recognition task and the target acquisition point, a recognition model is matched based on the recognition task, and the second ultrasonic image is recognized based on the recognition model to acquire an ultrasonic image recognition result.

2. An ultrasonic image recognition device, characterized in that: include: Preliminary recognition module, adjustment module and target recognition module; The target recognition module includes a collection unit and a recognition unit; The preliminary recognition module is used to acquire a first ultrasound image based on a preset initial acquisition point, recognize an anatomical structure based on the first ultrasound image, and acquire a structural position and a structural distribution of the anatomical structure; The adjustment module is used to adjust the initial acquisition point based on the structure position and the structure distribution to determine the target acquisition point; The acquisition unit is used to acquire a second ultrasonic image based on the recognition task and the target acquisition point; The recognition unit is used to match the recognition model based on the recognition task, recognize the second ultrasound image based on the recognition model, and obtain an ultrasound image recognition result.

3. An ultrasonic image recognition device as claimed in claim 2, characterized in that: The preliminary identification module is used to: Acquire a first ultrasound image based on a preset initial acquisition point, and perform standardization processing on the first ultrasound image to acquire a first image to be identified; Inputting the first image to be identified into a pre-trained segmentation network model to obtain object structure data in the first ultrasound image; the segmentation network model is trained based on labeled ultrasound image data and a supervised learning algorithm; The structural position and structural distribution of the anatomical structure are acquired based on the object structure data.

4. An ultrasonic image recognition device as claimed in claim 3, characterized in that: The adjustment module is used for: Calculating adjustment parameters based on a preset arterial structure range, the structure position and the structure distribution, the adjustment parameters including a probe moving distance and a probe deflection angle; The initial acquisition point is adjusted based on the probe moving distance and the probe deflection angle to determine the target acquisition point.

5. An ultrasonic image recognition device as claimed in claim 4, characterized in that: The recognition tasks include plaque recognition tasks, blood flow analysis tasks and spectrum analysis tasks; the acquisition unit is used to: If the recognition task is a plaque recognition task, acquiring a two-dimensional ultrasound image based on the target acquisition point; If the recognition task is a blood flow analysis task, acquiring a two-dimensional color ultrasound image based on a real-time imaging mode and the acquisition points; If the recognition task is a spectrum analysis task, a spectrum diagram is obtained based on a pulse wave Doppler imaging mode and the target acquisition point.

6. An ultrasonic image recognition device as claimed in claim 5, characterized in that: The identification unit comprises a first identification subunit, wherein the first identification subunit is used to: When the recognition task is a plaque recognition task, the recognition model is a first convolutional neural network model; Preprocessing the second ultrasound image based on preset image adjustment parameters to obtain a second image to be identified; Inputting the second image to be recognized into the first convolutional neural network model to obtain arterial structural parameters, wherein the arterial structural parameters include arterial inner diameter, arterial wall thickness, arterial plaque location, and arterial plaque size; the first convolutional neural network model is constructed based on a multi-layer convolutional neural network; The second ultrasound image is annotated based on the arterial structure parameters to obtain the ultrasound image recognition result.

7. An ultrasonic image recognition device as claimed in claim 6, characterized in that: The identification unit comprises a second identification subunit, wherein the second identification subunit is used to: When the recognition task is a blood flow analysis task, the recognition model is a long short-term memory network model; Inputting the second ultrasound image into the long short-term memory network model to obtain blood flow characteristic parameters; the blood flow characteristic parameters include blood flow direction, blood flow velocity and blood flow turbulence; The ultrasonic image recognition result is obtained based on a preset blood flow parameter threshold, the blood flow direction, the blood flow velocity and the blood flow turbulence condition.

8. An ultrasonic image recognition device as claimed in claim 7, characterized in that: The identification unit comprises a third identification subunit, and the third identification subunit is used to: When the recognition task is a spectrum analysis task, the recognition model is a second convolutional neural network model; Inputting the second ultrasound image into the second convolutional neural network model to obtain hemodynamic parameters; the second convolutional neural network model is constructed based on a convolutional neural network and a multi-head self-attention mechanism; the hemodynamic parameters include peak systolic velocity and end-diastolic velocity; The ultrasound image recognition result is obtained based on a preset dynamic parameter threshold and the peak systolic velocity and the end-diastolic velocity.

9. The ultrasonic image recognition device according to claim 7, characterized in that: The ultrasonic image recognition device further includes a self-checking module, wherein the self-checking module is used to: Acquire the probe connection state and the display resolution, determine whether the probe connection state is abnormal based on a preset signal transmission threshold, and obtain a first determination result; Determine whether the display resolution is abnormal based on a preset resolution threshold, and obtain a second determination result; A detection report is generated based on the first judgment result and the second judgment result, and an early warning is issued based on the detection report.

10. The ultrasonic image recognition device according to claim 7, characterized in that: The ultrasonic image recognition device further comprises a visualization module, wherein the visualization module is used to: The ultrasonic image recognition result is obtained, and the ultrasonic image recognition result is visualized.

Images