Metacarpal super intima middle layer thickness measuring system and method

Abstract

The invention provides a palm hyperintima middle layer thickness measuring system and method, and relates to the technical field of palm ultrasound, and the system comprises an image collection module which is used for obtaining a carotid artery image of a patient collected by a palm ultrasound device; the boundary identification module is connected with the carotid artery image acquisition module and is used for identifying the boundary between the inner membrane and the middle layer in the carotid artery image; and the thickness measurement module is connected with the boundary identification module and used for calculating the boundary thickness to obtain an inner film middle layer thickness measurement result. The beneficial effects are that the system can realize automatic identification of the boundary between the inner membrane and the middle layer in the carotid artery, after the boundary is identified, an inner membrane and middle layer thickness measurement result is obtained by calculating the thickness of the boundary, and compared with high-end ultrasonic equipment in the prior art, the cost is lower, and the accuracy is higher. Compared with an experienced ultrasonic doctor for thickness detection, automatic detection can be realized, the labor cost can be saved, and the efficiency can be improved.

Description

Technical Field

[0001] The invention relates to the technical field of palm ultrasound, and in particular to a palm ultrasound intima-media thickness measurement system and method. Background Art

[0002] In modern medical imaging, the measurement of carotid intima-media thickness (IMT) is an important indicator for assessing the degree of atherosclerosis and the risk of cardiovascular disease. Traditional IMT measurement usually relies on high-end ultrasound equipment and experienced ultrasound physicians. However, with the popularity of portable handheld ultrasound equipment, how to achieve accurate IMT measurement on such portable equipment has become a research hotspot.

[0003] The existing IMT measurement technology mainly relies on high-end ultrasound equipment and experienced doctors. Although it has high accuracy, it is complicated to operate and the equipment is expensive. The emergence of handheld ultrasound equipment has improved portability and ease of use, but due to equipment performance limitations, the measurement accuracy is low. In addition, although handheld ultrasound equipment can acquire images of the patient's carotid artery, it cannot automatically identify the boundary between the intima and the middle layer, nor can it automatically calculate the thickness of the boundary, which is very inconvenient for doctors to diagnose and reduces efficiency. Summary of the invention

[0004] In view of the problems existing in the prior art, the present invention provides a palm ultrasound intima-media thickness measurement system, which is characterized by comprising:

[0005] An image acquisition module, used to acquire a carotid artery image of a patient acquired by a handheld ultrasound device;

[0006] a boundary recognition module, connected to the carotid artery image acquisition module, and used to recognize the boundary between the intima and the middle layer in the carotid artery image;

[0007] The thickness measurement module is connected to the boundary recognition module and is used to calculate the thickness of the boundary to obtain the intima-media thickness measurement result.

[0008] Preferably, it further comprises an image preprocessing module, connected to the carotid artery image acquisition module and the boundary recognition module, comprising:

[0009] a contrast enhancement unit, configured to perform contrast adjustment on the carotid artery image using an adaptive contrast network model, and then update the carotid artery image to the carotid artery image after contrast adjustment;

[0010] a resolution reconstruction unit, connected to the contrast enhancement unit, configured to perform super-resolution reconstruction on the carotid artery image by using an adversarial network model, and then update the carotid artery image to the carotid artery image after resolution reconstruction;

[0011] an image denoising unit, connected to the resolution reconstruction unit, and configured to perform denoising processing on the carotid artery image using a convolutional neural network, and then update the carotid artery image to the denoised carotid artery image;

[0012] An edge enhancement unit is connected to the image denoising unit and is used to perform edge enhancement on the carotid artery image using an edge enhancement network model, and then update the carotid artery image to the edge-enhanced carotid artery image.

[0013] Preferably, a hardware acceleration module is further included, which is connected to the boundary recognition module and is used to construct a recognition task for identifying the boundary between the intima and the middle layer in the carotid artery image when acquiring the carotid artery image, and then call a pre-configured hardware acceleration processor to assign the established recognition task to the hardware acceleration processor for execution, so as to identify the boundary in the carotid artery image.

[0014] Preferably, it further comprises a model building module connected to the boundary recognition module, wherein the boundary recognition module inputs the carotid artery image into the boundary recognition model constructed by the model building module to recognize the boundary between the intima and the middle layer in the carotid artery image;

[0015] The model building module includes:

[0016] A multi-scale feature extraction unit is used to add a multi-scale feature extraction module between the encoder and the decoder of the pre-built standard U-NET network model to extract features of different scales on the carotid artery image to obtain multi-scale features;

[0017] An attention mechanism adding unit, connected to the multi-scale feature extraction unit, for enhancing the expression of multi-scale features by connecting the attention module to the skip layer of the standard U-NET network model after adding the multi-scale feature extraction module;

[0018] A pre-training unit is connected to the attention mechanism adding unit and is used to perform self-supervised pre-training on the standard U-NET network model after adding the multi-scale feature extraction module and the attention module by using a contrastive learning method, and perform formal training after the pre-training to obtain the boundary recognition model.

[0019] Preferably, the thickness measurement module comprises:

[0020] A coordinate establishing unit, used to establish a reference coordinate system in the carotid artery image, and then obtain the coordinates of each pixel point on the boundary;

[0021] The thickness calculation unit is connected to the coordinate establishment unit and is used to traverse each pixel point on the boundary. For each pixel point, the difference between the longitudinal coordinates of the inner membrane and the middle layer corresponding to the pixel point is calculated, and then the sum of all the differences is calculated and divided by the total number of pixel points on the boundary to obtain the measurement result of the inner membrane and the middle layer thickness.

[0022] The present invention also provides a palm ultrasound intima-media thickness measurement method, which uses the palm ultrasound intima-media thickness measurement system as described above, comprising:

[0023] Step S1, the palm ultrasound intima-media thickness measurement system obtains a carotid artery image of a patient collected by a palm ultrasound device;

[0024] Step S2, the palmar ultrasound intima-media thickness measurement system identifies the boundary between the intima and the media in the carotid artery image;

[0025] Step S3, the palm ultrasound intima-media thickness measurement system calculates the thickness of the boundary to obtain an intima-media thickness measurement result.

[0026] Preferably, after step S1, an image preprocessing process is also included, including:

[0027] Step A1, the palmar ultrasound intima-media thickness measurement system uses an adaptive contrast network model to perform contrast adjustment on the carotid artery image, and then updates the carotid artery image to the carotid artery image after contrast adjustment;

[0028] Step A2, the palmar ultrasound intima-media thickness measurement system uses an adversarial network model to perform super-resolution reconstruction on the carotid artery image, and then updates the carotid artery image to the carotid artery image after super-resolution reconstruction;

[0029] Step A3, the palmar ultrasound intima-media thickness measurement system uses a convolutional neural network to perform denoising on the carotid artery image, and then updates the carotid artery image to the denoised carotid artery image;

[0030] Step A4: the palmar ultrasound intima-media thickness measurement system uses an edge enhancement network model to perform edge enhancement on the carotid artery image, and then updates the carotid artery image to the edge-enhanced carotid artery image.

[0031] Preferably, a hardware acceleration process is also included, including:

[0032] The palm ultrasound intima-media thickness measurement system constructs a recognition task for identifying the boundary between the intima and the media in the carotid artery image when acquiring the carotid artery image, and then calls a pre-configured hardware acceleration processor, assigns the established recognition task to the hardware acceleration processor for execution, and identifies the boundary in the carotid artery image.

[0033] Preferably, before step S2, a model building process is also included, in which the palm ultrasound intima-media thickness measurement system inputs the carotid artery image into a constructed boundary recognition model to identify the boundary between the intima and the media in the carotid artery image;

[0034] The model building process includes:

[0035] Step B1, the palm ultrasound intima-media thickness measurement system adds a multi-scale feature extraction module between the encoder and the decoder of the pre-built standard U-NET network model to extract features of different scales from the carotid artery image to obtain multi-scale features;

[0036] Step B2, the palm ultrasound intima-media thickness measurement system enhances the expression of multi-scale features by using a skip layer connection attention module of the standard U-NET network model after adding a multi-scale feature extraction module;

[0037] Step B3, the palm ultrasound intima-media thickness measurement system uses contrastive learning method to perform self-supervised pre-training on the standard U-NET network model after adding the multi-scale feature extraction module and the attention module, and performs formal training after the pre-training to obtain a boundary recognition model.

[0038] Preferably, step S3 comprises:

[0039] Step S31, the palmar ultrasound intima-media thickness measurement system establishes a reference coordinate system in the carotid artery image, and then obtains the coordinates of each pixel point on the boundary;

[0040] Step S32, the palm ultrasound intima-media thickness measurement system traverses each pixel point on the boundary, and for each pixel point, calculates the difference in the longitudinal coordinates of the intima and middle layer corresponding to the pixel point, then calculates the sum of all the differences and divides it by the total number of pixel points on the boundary to obtain the intima-media thickness measurement result.

[0041] The above technical solution has the following advantages or beneficial effects: the system of the present invention can realize automatic identification of the boundary between the intima and the middle layer of the carotid artery, and after identifying the boundary, the IMT measurement result is obtained by calculating the thickness of the boundary. Compared with the high-end ultrasound equipment in the prior art, the cost is lower, and compared with the method of thickness detection by experienced ultrasound doctors, automated detection can be realized to save labor costs and improve efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 FIG. 1 is a schematic structural diagram of a palmar ultrasound intima-media thickness measurement system in a preferred embodiment of the present invention;

[0043] Figure 2 The figure is a flow chart of a method for measuring the intima-media thickness of palmar ultrasound in a preferred embodiment of the present invention;

[0044] Figure 3 A schematic diagram of a sub-process of a model building process in a preferred embodiment of the present invention;

[0045] Figure 4 This is a schematic diagram of a sub-process of step S3 in a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0046] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. The present invention is not limited to this embodiment, and other embodiments may also fall within the scope of the present invention as long as they conform to the gist of the present invention.

[0047] In a preferred embodiment of the present invention, based on the above problems existing in the prior art, a palmar ultrasound intima-media thickness measurement system is provided. Figure 1 As shown, including:

[0048] An image acquisition module 1 is used to acquire a carotid artery image of a patient acquired by a handheld ultrasound device;

[0049] The boundary recognition module 2 is connected to the carotid artery image acquisition module 1 and is used to recognize the boundary between the intima and the middle layer in the carotid artery image;

[0050] The thickness measurement module 3 is connected to the boundary recognition module 2 and is used to calculate the thickness of the boundary to obtain the intima-media thickness measurement result.

[0051] Specifically, in order to solve the problem that the IMT (intima-media thickness) measurement in the prior art usually relies on high-end ultrasound equipment and experienced ultrasound doctors, the palm ultrasound IMT measurement system is used to automatically acquire the carotid artery image of the palm ultrasound device, and then identify the boundary between the intima and the media in the carotid artery image;

[0052] The recognition model in this embodiment can be recognized by the image recognition model in the prior art. The training data of the image recognition model uses an image set containing multiple standard carotid artery ultrasound images with associated labels. The image recognition model that can identify the boundary between the intima and the middle layer is obtained by training according to the conventional model training process, for example: 1. Support vector machine (SVM): The boundary between the intima and the middle layer in the ultrasound image is identified by training the SVM model. SVM can find the best hyperplane to segment data of different categories. 2. Convolutional neural network (CNN): CNN has a powerful feature extraction capability in the field of image processing, and can automatically learn the features of the intima and the middle layer from ultrasound images, thereby realizing automatic recognition of boundaries. 3. Gaussian mixture model (GMM) clustering: GMM is used to statistically model the grayscale features of the ultrasound imaging of the three-layer membrane of the carotid artery, and the probability density function distribution of the carotid artery lumen, the three-layer membrane of the wall and the surrounding tissue is obtained, and a probability model is established based on this. Then, the EM algorithm is used to estimate the grayscale mean and variance parameters of each category, and the characteristic parameter values ​​are mapped to the image space. Finally, the grayscale threshold method is used to detect the boundary between the intima and the media of the vascular wall to estimate the thickness.

[0053] The above system can realize automatic identification of the boundary between the intima and the middle layer of the carotid artery. After the boundary is identified, the IMT measurement result is obtained by calculating the thickness of the boundary. Compared with the high-end ultrasound equipment in the existing technology, the cost is lower. Compared with the thickness detection method performed by experienced ultrasound doctors, automated detection can be realized to save labor costs and improve efficiency.

[0054] In a preferred embodiment of the present invention, an image preprocessing module 4 is further included, which is connected to the carotid artery image acquisition module 1 and the boundary recognition module 2. Figure 1 As shown, including:

[0055] A contrast enhancement unit 41, configured to perform contrast adjustment on the carotid artery image using an adaptive contrast network model, and then update the carotid artery image to a contrast-adjusted carotid artery image;

[0056] A resolution reconstruction unit 42, connected to the contrast enhancement unit 41, is used to perform super-resolution reconstruction on the carotid artery image by using an adversarial network model, and then update the carotid artery image to a carotid artery image after resolution reconstruction;

[0057] An image denoising module 43, connected to the resolution reconstruction unit 42, for denoising the carotid artery image using a convolutional neural network, and then updating the carotid artery image to a denoised carotid artery image;

[0058] The edge enhancement module 44 is connected to the image denoising unit 43 and is used to perform edge enhancement on the carotid artery image using an edge enhancement network model, and then update the carotid artery image to an edge-enhanced carotid artery image.

[0059] Specifically, in this embodiment, the carotid artery image is further subjected to image preprocessing, and the preprocessing includes:

[0060] Contrast enhancement, using Adaptive Contrast Enhancement Network (ACE-Net): Design an adaptive contrast enhancement network to adaptively adjust the contrast according to the characteristics of the input image to enhance image details and edge information.

[0061] Super-resolution reconstruction uses a generative adversarial network (GAN) to perform super-resolution reconstruction to improve the resolution and clarity of the image.

[0062] Denoising, using denoising convolutional neural network (DnCNN): Design a convolutional neural network for denoising, retaining image details while suppressing noise.

[0063] Edge enhancement, using EdgeEnhanceNet: a deep learning network is designed to enhance the edge information of images and improve the clarity of boundaries.

[0064] Based on the above, a comprehensive image enhancement network (ComprehensiveEnhanceNet) is used: combining adaptive contrast enhancement, super-resolution reconstruction, denoising and edge enhancement technology to obtain a comprehensive image enhancement network to significantly improve the quality of carotid artery ultrasound images.

[0065] In a preferred embodiment of the present invention, Figure 1 As shown, it also includes a hardware acceleration module 5, which is connected to the boundary recognition module 2 and is used to construct a recognition task for recognizing the boundary between the inner membrane and the middle layer in the carotid artery image when acquiring the carotid artery image, and then call a pre-configured hardware acceleration processor to assign the established recognition task to the hardware acceleration processor to run, so as to recognize the boundary in the carotid artery image.

[0066] Specifically, in this embodiment, the hardware acceleration processor uses an integrated GPU (graphics processing unit (English: graphics processing unit, abbreviation: GPU)) or NPU (Neural network Processing Unit) to improve the computing performance of image processing and AI reasoning, and is used to accelerate the automatic boundary recognition and IMT measurement process.

[0067] In a preferred embodiment of the present invention, Figure 1As shown, it also includes a model building module 6, connected to the boundary recognition module 2, the boundary recognition module 2 inputs the carotid artery image into the boundary recognition model constructed by the model building module 6, and recognizes the boundary between the intima and the middle layer in the carotid artery image;

[0068] Model building module 6 includes:

[0069] A multi-scale feature extraction unit 61 is used to add a multi-scale feature extraction module between the encoder and the decoder of the pre-built standard U-NET network model to extract features of different scales on the carotid artery image to obtain multi-scale features;

[0070] The attention mechanism adding unit 62 is connected to the multi-scale feature extraction unit 61, and is used to connect the attention module to the skip layer of the standard U-NET network model after adding the multi-scale feature extraction module to enhance the expression of the multi-scale feature;

[0071] The pre-training unit 63 is connected to the attention mechanism adding unit 62, and is used to perform self-supervised pre-training using a contrastive learning method on the standard U-NET network model after adding a multi-scale feature extraction module and an attention module, and perform formal training after the pre-training to obtain a boundary recognition model.

[0072] Specifically, in this embodiment, the boundary recognition model adopts a network model of the U-NET architecture, and improves on this basis;

[0073] First, the improved U-Net architecture introduces a multi-scale feature extraction module. A multi-scale feature extraction module (MSFE) is added between the encoder and decoder of U-Net to capture features of different scales and improve the model's ability to recognize patches and boundaries.

[0074] Secondly, add an attention module: add an attention mechanism to the jump connection of U-Net to enhance the expression ability of important features and suppress irrelevant information.

[0075] In addition, self-supervised pre-training is adopted: before formal training, contrastive learning is used for self-supervised pre-training to enhance the feature representation ability of the model.

[0076] Furthermore, multi-task learning is set up, and an adaptive multi-task loss function is used: the inner membrane and middle layer boundary are detected simultaneously, and the losses of different tasks are adjusted through adaptive weights to ensure the balanced development of the model on multiple tasks.

[0077] In addition, Bayesian deep learning is used to estimate the uncertainty of the model: Bayesian neural network is used to estimate the uncertainty of the model, mark high uncertainty areas, and prompt the need for manual review to improve the reliability of the system.

[0078] Moreover, the boundary recognition model adopts a lightweight design, a lightweight model based on NAS (Neural Architecture Search): a lightweight and efficient model structure is automatically designed through neural architecture search (NAS), which is suitable for real-time reasoning on edge devices.

[0079] The above improvements are integrated into U-Net to form an innovative U-Net architecture. With these improvements, we can significantly improve the performance of U-Net in carotid IMT measurement and identification, ensure the efficiency and accuracy of the system, and simplify the operation process and reduce the dependence on operator experience.

[0080] In a preferred embodiment of the present invention, Figure 1 As shown, the thickness measurement module 3 includes:

[0081] A coordinate establishing unit 31 is used to establish a reference coordinate system in the carotid artery image and then obtain the coordinates of each pixel point on the boundary;

[0082] The thickness calculation unit 32 is connected to the coordinate establishment unit 31, and is used to traverse each pixel point on the boundary. For each pixel point, the difference between the longitudinal coordinates of the inner membrane and the middle layer corresponding to the pixel point is calculated, and then the sum of all the differences is calculated and divided by the total number of pixel points on the boundary to obtain the IMT measurement result.

[0083] Specifically, in this embodiment, the intima-media thickness (IMT) is calculated based on the identified boundary position. The specific calculation formula is:

[0084]

[0085] Where yintima(i) and ymedia(i) represent the ordinates of the intima and media at the i-th pixel on the boundary, respectively, and N is the total number of pixels on the boundary.

[0086] The present invention also provides a palm ultrasound intima-media thickness measurement method, using the palm ultrasound intima-media thickness measurement system as above, such as Figure 2 As shown, including:

[0087] Step S1, a palm ultrasound intima-media thickness measurement system obtains a carotid artery image of a patient acquired by a palm ultrasound device;

[0088] Step S2, the palmar ultrasound intima-media thickness measurement system identifies the boundary between the intima and the media in the carotid artery image;

[0089] Step S3: The palmar ultrasound intima-media thickness measurement system calculates the thickness of the boundary to obtain an intima-media thickness measurement result.

[0090] In a preferred embodiment of the present invention, after step S1, an image preprocessing process is also included, including:

[0091] Step A1, the palmar ultrasound intima-media thickness measurement system uses an adaptive contrast network model to adjust the contrast of the carotid artery image, and then updates the carotid artery image to a contrast-adjusted carotid artery image;

[0092] Step A2, the palmar ultrasound intima-media thickness measurement system uses an adversarial network model to perform super-resolution reconstruction on the carotid artery image, and then updates the carotid artery image to a carotid artery image after resolution reconstruction;

[0093] Step A3, the palmar ultrasound intima-media thickness measurement system uses a convolutional neural network to perform denoising on the carotid artery image, and then updates the carotid artery image to the denoised carotid artery image;

[0094] Step A4: the palmar ultrasound intima-media thickness measurement system uses an edge enhancement network model to perform edge enhancement on the carotid artery image, and then updates the carotid artery image to an edge-enhanced carotid artery image.

[0095] In a preferred embodiment of the present invention, a hardware acceleration process is also included, including:

[0096] When acquiring a carotid artery image, the palm ultrasound intima-media thickness measurement system constructs a recognition task for identifying the boundary between the intima and the media in the carotid artery image, then calls a pre-configured hardware acceleration processor, assigns the established recognition task to the hardware acceleration processor for operation, and identifies the boundary in the carotid artery image.

[0097] In a preferred embodiment of the present invention, Figure 3 As shown, before step S2, a model building process is also included. In step S2, the palmar ultrasound intima-media thickness measurement system inputs the carotid artery image into the constructed boundary recognition model to identify the boundary between the intima and the media in the carotid artery image;

[0098] The model building process includes:

[0099] Step B1, the palmar ultrasound intima-media thickness measurement system adds a multi-scale feature extraction module between the encoder and the decoder of the pre-built standard U-NET network model to extract features of different scales from the carotid artery image to obtain multi-scale features;

[0100] Step B2, the palm ultrasound intima-media thickness measurement system uses a skip layer connection attention module of the standard U-NET network model after adding a multi-scale feature extraction module to enhance the expression of multi-scale features;

[0101] Step B3, the palm ultrasound intima-media thickness measurement system uses contrastive learning method to perform self-supervised pre-training on the standard U-NET network model after adding the multi-scale feature extraction module and the attention module, and performs formal training after the pre-training to obtain the boundary recognition model.

[0102] In a preferred embodiment of the present invention, Figure 4 As shown, step S3 includes:

[0103] Step S31, the palmar ultrasound intima-media thickness measurement system establishes a reference coordinate system in the carotid artery image, and then obtains the coordinates of each pixel point on the boundary;

[0104] Step S32, the palm ultrasound intima-media thickness measurement system traverses each pixel point on the boundary, and for each pixel point, calculates the difference in the longitudinal coordinates of the intima and middle layer corresponding to the pixel point, then calculates the sum of all the differences and divides it by the total number of pixels on the boundary to obtain the intima-media thickness measurement result.

[0105] The above are only preferred embodiments of the present invention, and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the contents of this specification and illustrations should be included in the protection scope of the present invention.

Claims

1. A palmar ultrasound intima-media thickness measurement system, characterized in that: include: An image acquisition module, used to acquire a carotid artery image of a patient acquired by a handheld ultrasound device; a boundary recognition module, connected to the carotid artery image acquisition module, and used to recognize the boundary between the intima and the middle layer in the carotid artery image; The thickness measurement module is connected to the boundary recognition module and is used to calculate the thickness of the boundary to obtain the intima-media thickness measurement result.

2. The palmar ultrasound intima-media thickness measurement system according to claim 1, characterized in that: It also includes an image preprocessing module, which is connected to the carotid artery image acquisition module and the boundary recognition module, and includes: a contrast enhancement unit, configured to perform contrast adjustment on the carotid artery image using an adaptive contrast network model, and then update the carotid artery image to the carotid artery image after contrast adjustment; a resolution reconstruction unit, connected to the contrast enhancement unit, configured to perform super-resolution reconstruction on the carotid artery image by using an adversarial network model, and then update the carotid artery image to the carotid artery image after resolution reconstruction; an image denoising unit, connected to the resolution reconstruction unit, and configured to perform denoising processing on the carotid artery image using a convolutional neural network, and then update the carotid artery image to the denoised carotid artery image; An edge enhancement unit is connected to the image denoising unit and is used to perform edge enhancement on the carotid artery image using an edge enhancement network model, and then update the carotid artery image to the edge-enhanced carotid artery image.

3. The palmar ultrasound intima-media thickness measurement system according to claim 1, characterized in that: The system also includes a hardware acceleration module, which is connected to the boundary recognition module and is used to construct a recognition task for recognizing the boundary between the intima and the middle layer in the carotid artery image when acquiring the carotid artery image, and then call a pre-configured hardware acceleration processor to assign the established recognition task to the hardware acceleration processor for execution, so as to recognize and obtain the boundary in the carotid artery image.

4. The palmar ultrasound intima-media thickness measurement system according to claim 1, characterized in that: It also includes a model building module connected to the boundary recognition module, wherein the boundary recognition module inputs the carotid artery image into the boundary recognition model constructed by the model building module to recognize the boundary between the intima and the middle layer in the carotid artery image; The model building module includes: A multi-scale feature extraction unit is used to add a multi-scale feature extraction module between the encoder and the decoder of the pre-built standard U-NET network model to extract features of different scales on the carotid artery image to obtain multi-scale features; An attention mechanism adding unit, connected to the multi-scale feature extraction unit, for enhancing the expression of multi-scale features by connecting the attention module to the skip layer of the standard U-NET network model after adding the multi-scale feature extraction module; A pre-training unit is connected to the attention mechanism adding unit and is used to perform self-supervised pre-training on the standard U-NET network model after adding the multi-scale feature extraction module and the attention module by using a contrastive learning method, and perform formal training after the pre-training to obtain the boundary recognition model.

5. The palmar ultrasound intima-media thickness measurement system according to claim 1, characterized in that: The thickness measurement module comprises: A coordinate establishing unit, used to establish a reference coordinate system in the carotid artery image, and then obtain the coordinates of each pixel point on the boundary; The thickness calculation unit is connected to the coordinate establishment unit and is used to traverse each pixel point on the boundary. For each pixel point, the difference between the longitudinal coordinates of the inner membrane and the middle layer corresponding to the pixel point is calculated, and then the sum of all the differences is calculated and divided by the total number of pixel points on the boundary to obtain the measurement result of the inner membrane and the middle layer thickness.

6. A palmar ultrasound intima-media thickness measurement method, characterized in that: The palm ultrasound intima-media thickness measurement system according to any one of claims 1 to 5 comprises: Step S1, the palm ultrasound intima-media thickness measurement system obtains a carotid artery image of a patient collected by a palm ultrasound device; Step S2, the palmar ultrasound intima-media thickness measurement system identifies the boundary between the intima and the media in the carotid artery image; Step S3, the palmar ultrasound intima-media thickness measurement system calculates the thickness of the boundary to obtain an intima-media thickness measurement result.

7. The palmar ultrasound intima-media thickness measurement method according to claim 6, characterized in that: After step S1, an image preprocessing process is also included, including: Step A1, the palmar ultrasound intima-media thickness measurement system uses an adaptive contrast network model to perform contrast adjustment on the carotid artery image, and then updates the carotid artery image to the carotid artery image after contrast adjustment; Step A2, the palmar ultrasound intima-media thickness measurement system uses an adversarial network model to perform super-resolution reconstruction on the carotid artery image, and then updates the carotid artery image to the carotid artery image after super-resolution reconstruction; Step A3, the palmar ultrasound intima-media thickness measurement system uses a convolutional neural network to perform denoising on the carotid artery image, and then updates the carotid artery image to the denoised carotid artery image; Step A4: the palmar ultrasound intima-media thickness measurement system uses an edge enhancement network model to perform edge enhancement on the carotid artery image, and then updates the carotid artery image to the edge-enhanced carotid artery image.

8. The palmar ultrasound intima-media thickness measurement method according to claim 6, characterized in that: Also included are hardware acceleration processes, including: The palm ultrasound intima-media thickness measurement system constructs a recognition task for identifying the boundary between the intima and the media in the carotid artery image when acquiring the carotid artery image, and then calls a pre-configured hardware acceleration processor, assigns the established recognition task to the hardware acceleration processor for execution, and identifies the boundary in the carotid artery image.

9. The palmar ultrasound intima-media thickness measurement method according to claim 6, characterized in that: The method further includes a model building process before step S2, wherein the palmar ultrasound intima-media thickness measurement system inputs the carotid artery image into a constructed boundary recognition model to identify the boundary between the intima and the media in the carotid artery image; The model building process includes: Step B1, the palm ultrasound intima-media thickness measurement system adds a multi-scale feature extraction module between the encoder and the decoder of the pre-built standard U-NET network model to extract features of different scales from the carotid artery image to obtain multi-scale features; Step B2, the palm ultrasound intima-media thickness measurement system enhances the expression of multi-scale features by using a skip layer connection attention module of the standard U-NET network model after adding a multi-scale feature extraction module; Step B3, the palm ultrasound intima-media thickness measurement system uses contrastive learning method to perform self-supervised pre-training on the standard U-NET network model after adding the multi-scale feature extraction module and the attention module, and performs formal training after the pre-training to obtain a boundary recognition model.

10. The palmar ultrasound intima-media thickness measurement method according to claim 6, characterized in that: The step S3 comprises: Step S31, the palmar ultrasound intima-media thickness measurement system establishes a reference coordinate system in the carotid artery image, and then obtains the coordinates of each pixel point on the boundary; Step S32, the palm ultrasound intima-media thickness measurement system traverses each pixel point on the boundary, and for each pixel point, calculates the difference in the longitudinal coordinates of the intima and middle layer corresponding to the pixel point, then calculates the sum of all the differences and divides it by the total number of pixel points on the boundary to obtain the intima-media thickness measurement result.

Images