Artery localization and detection method and apparatus, and electronic device and storage medium

Abstract

Provided in the embodiments of the present disclosure are an artery localization and detection method and apparatus, an electronic device, and a storage medium. The method comprises, during the process of continuously moving an image collection apparatus at a target part, iteratively executing at least one round of the following steps until the degree of integrity meets a preset condition: obtaining an image sequence collected by means of the image collection apparatus in the current round, stacking the image sequence into a pair of stacked images, and inputting the stacked images into a localization network, so as to obtain an image state output by means of the localization network; and using, as a target image, the last image frame in the image sequence obtained when the degree of integrity meets the preset condition, detecting an arterial blood vessel in the target image, and determining a detection result indicative of a health risk of the arterial blood vessel. In this way, when a user moves a image collection apparatus, an image can be recognized to assist the user in locating an arterial blood vessel, and the arterial blood vessel is detected when the degree of integrity meets a preset condition.

Description

Artery localization detection methods, devices, electronic equipment and storage media Technical Field

[0001] This disclosure relates to the field of medical image analysis technology, and more specifically, to an artery localization detection method, apparatus, electronic device, and storage medium. Background Technology

[0002] With an aging population and increased health awareness, deaths and serious cases of cardiovascular and cerebrovascular diseases are rising rapidly worldwide, leading to a growing demand for portable medical devices and medical imaging technologies.

[0003] Currently, medical images are typically acquired by combining imaging equipment with artificial intelligence algorithms, and then analyzed and processed to provide clinicians with more accurate and comprehensive diagnostic information, which helps in the early detection and treatment monitoring of diseases.

[0004] Imaging equipment has a high operational threshold, and the application of artificial intelligence algorithms is mainly focused on diagnostic functions. It relies heavily on the operator's medical expertise and skills to ensure the quality and accuracy of image acquisition and diagnostic interpretation. Therefore, these technologies are primarily used in professional medical environments and are not suitable for users without a medical background. Summary of the Invention

[0005] This disclosure provides an artery localization detection method, apparatus, electronic device, and storage medium to address the technical problem in the prior art where users without a medical background find it difficult to use imaging equipment.

[0006] According to one aspect of the present disclosure, an artery localization detection method is provided, comprising:

[0007] During the continuous movement of the image acquisition device at the target site, the following steps are iteratively executed at least once until the integrity meets the preset conditions: obtain the image sequence acquired by the image acquisition device in this round, stack the image sequence into a stacked image, input the stacked image into the positioning network, and obtain the image state output by the positioning network, wherein the image state represents the degree of integrity of the arterial vessel wall identified based on the corresponding stacked image;

[0008] The last frame of the image sequence obtained when the integrity meets the preset conditions is taken as the target image. The arteries in the target image are detected, and the detection results are determined. The detection results are used to indicate the health risks of the arteries.

[0009] According to another aspect of the present disclosure, an artery positioning detection device is provided, comprising:

[0010] The positioning and recognition module is used to iteratively execute at least one round of the following steps during the continuous movement of the image acquisition device at the target location until the integrity meets the preset conditions: obtaining the image sequence acquired by the image acquisition device in this round, stacking the image sequence into a stacked image, inputting the stacked image into the positioning network, and obtaining the image state output by the positioning network, wherein the image state represents the degree of integrity of the arterial vessel wall identified based on the corresponding stacked image;

[0011] The artery detection module is used to take the last frame of the image sequence obtained when the integrity meets the preset conditions as the target image, detect the arteries in the target image, determine the detection results, and use the detection results to indicate the health risks of the arteries.

[0012] According to another aspect of the present disclosure, an electronic device is provided, the electronic device including a memory, a processor, and a computer program stored in the memory, the processor executing the computer program to implement the steps of the method provided in any of the above embodiments.

[0013] According to another aspect of the present disclosure, a computer-readable storage medium is provided that stores a computer program thereon, which, when executed by a processor, implements the steps of the method provided in any of the above embodiments.

[0014] According to one aspect of the present disclosure, a computer program product is provided, including a computer program that is executed by a processor to perform the steps of the method provided in any of the above embodiments.

[0015] The beneficial effects of the technical solutions provided in this disclosure are:

[0016] By pre-training a localization network, the system analyzes and identifies the image status of images acquired by the user-operated image acquisition device in real time. The image status guides the user to adjust the position of the image acquisition device. When the completeness of the image status indication meets the preset conditions, the last frame of the acquired image sequence is used as the target image for arterial blood vessel detection. This solves the problem that users without a medical background cannot accurately locate arterial blood vessels, lowers the barrier to entry for using the image acquisition device, and allows for the detection of arterial blood vessels in the acquired target image after localization. The detection results can then be used to indicate the health risks of the arterial blood vessels, effectively reducing the complexity of user operation. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this disclosure, the accompanying drawings used in the description of the embodiments of this disclosure will be briefly introduced below.

[0018] Figure 1 is a schematic flowchart of an artery localization detection method provided in an embodiment of this disclosure.

[0019] Figure 2 is a schematic diagram of the steps of an artery localization detection method provided in an embodiment of this disclosure.

[0020] Figure 3 is a schematic diagram of a carotid artery ultrasound image provided in an embodiment of this disclosure.

[0021] Figure 4 is a schematic diagram of an artery positioning and detection device provided in an embodiment of this disclosure.

[0022] Figure 5 is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0023] The embodiments of this disclosure are described below with reference to the accompanying drawings. It should be understood that the embodiments described below with reference to the accompanying drawings are exemplary descriptions for explaining the technical solutions of the embodiments of this disclosure, and do not constitute a limitation on the technical solutions of the embodiments of this disclosure.

[0024] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the terms “comprising” and “including” as used in embodiments of this disclosure mean that the corresponding feature can be implemented as the presented feature, information, data, step, operation, element, and / or component, but do not exclude implementation as other features, information, data, step, operation, element, component, and / or combinations thereof supported by the art. It should be understood that when we say that an element is “connected” or “coupled” to another element, the one element can be directly connected or coupled to the other element, or it can mean that the one element and the other element are connected through an intermediate element. Furthermore, “connected” or “coupled” as used herein can include wireless connection or wireless coupling. The term “and / or” as used herein indicates at least one of the items defined by the term; for example, “A and / or B” can be implemented as “A,” or as “B,” or as “A and B.”

[0025] To make the objectives, technical solutions, and advantages of this disclosure clearer, the embodiments of this disclosure will be described in further detail below with reference to the accompanying drawings.

[0026] The following description of several exemplary embodiments illustrates the technical solutions of this disclosure and the technical effects produced by these solutions. It should be noted that the following embodiments can be referenced, learned from, or combined with each other. Identical terms, similar features, and similar implementation steps in different embodiments will not be repeated.

[0027] It is understood that in the artery localization detection method provided in the embodiments of this disclosure, any step of the method can be executed by a terminal and / or a server, and all steps in the method can be executed independently by the terminal or the server, or jointly by the terminal and the server.

[0028] The server can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server providing cloud computing services. The terminal can be a smartphone, tablet, laptop, desktop computer, smart voice interaction device (such as a smart speaker), wearable electronic device (such as a smartwatch), in-vehicle terminal, smart home appliance (such as a smart TV), AR / VR device, etc., but is not limited to these.

[0029] The embodiments of this disclosure will be described below with the server as the execution subject; however, this does not constitute a limitation on the embodiments of this disclosure.

[0030] Figure 1 is a schematic flowchart of an artery localization detection method provided in an embodiment of this disclosure. As shown in Figure 1, the technical solution provided in this embodiment includes the following steps:

[0031] Step S101: During the continuous movement of the image acquisition device at the target area, the following steps are iteratively executed at least once until the integrity meets the preset conditions:

[0032] The image sequence acquired by the image acquisition device in this round is obtained, the image sequence is stacked into a stacked image, the stacked image is input into the positioning network, and the image state output by the positioning network is obtained.

[0033] The image state indicates the degree of integrity of the arterial vessel wall as identified from the corresponding stacked images.

[0034] Specifically, the artery localization and detection method provided in this disclosure is used to help users without a medical background quickly locate arteries and determine the detection results. Specifically, the method can be divided into two parts: localization and recognition, and artery detection.

[0035] Among them, positioning and recognition refers to the user continuously acquiring images at the target site through an image acquisition device, performing positioning and recognition of the artery based on the acquired images, determining whether the identified images meet the detection requirements, and using the images that meet the detection requirements as the correctly positioned images.

[0036] It is understandable that the target location refers to the relevant part of the artery being detected, and the target location can be determined according to the actual needs of the artery being detected. The image refers to a medical image that can clearly display blood vessels, such as ultrasound images. The image acquisition device is the equipment used to acquire the corresponding type of image (e.g., an ultrasound probe), and its specific type can be determined according to actual needs.

[0037] Taking the carotid artery as the target artery and ultrasound image as an example, the artery localization detection method provided in this disclosure embodiment will be described in detail. Since the target artery is the carotid artery, the target location can be determined to be the neck region. Because users may not have a professional medical background and may find it difficult to accurately locate the carotid artery directly, the image acquisition device can be moved. Based on the image status output by the localization network, the localization accuracy of the current image can be fed back in real time to determine whether the accurate localization of the carotid artery has been completed. If not, the image acquisition device continues to move to help the user quickly find the location of the carotid artery. It is understood that if the user can accurately locate the carotid artery, only one round of the following steps is required.

[0038] In step S101, during the continuous movement of the image acquisition device at the target location, at least one round of the following steps is executed until the image status output by the positioning network can be used to determine whether the integrity meets the preset conditions. The preset conditions mean that the acquired image can meet the relevant requirements for arterial blood vessel detection and the arterial blood vessel wall can be clearly and completely displayed in the image.

[0039] The iterative execution steps include: obtaining the image sequence acquired by the image acquisition device in this round; stacking each image frame in the image sequence to obtain a stacked image. Due to the continuity of acquisition, the stacked image contains dynamic information and can reflect the arterial pulsation, assisting in arterial localization and improving the accuracy of localization. The image sequence includes multiple consecutive frames (e.g., three consecutive frames), the specific number of which can be determined according to actual needs.

[0040] The stacked images are input into a pre-trained localization network to obtain the image state output by the localization network. It can be understood that the localization network is a kind of image recognition and classification network. It can identify the image state that matches the current image from multiple image states and output it. The image state represents the degree of integrity of the blood vessel wall of the artery identified according to the corresponding stacked image. It can be used to reflect whether the artery is accurately located in the acquired image.

[0041] The localization network is trained using stacked sample images as training samples and the sample image states corresponding to the stacked sample images as labels. The specific network model structure and training method can be determined according to actual needs.

[0042] For example, since the carotid artery is on the side of the neck, the user can use the Adam's apple as the initial position of the image acquisition device and move the image acquisition device towards the side of the neck. During the movement, the image acquisition device will continuously acquire images and output the image status. When the completeness of the image status indication does not meet the preset conditions, the user will be prompted to continue moving the image acquisition device until the completeness of the image status indication meets the preset conditions, and then the user will be instructed to stop moving.

[0043] It is understood that the specific method of the mobile acquisition device can be continuous movement or intermittent movement, and the method of the image acquisition device to acquire images can be continuous acquisition or acquisition of multiple frames of images at intervals, etc. The embodiments disclosed herein do not limit this.

[0044] For example, if the user moves the image acquisition device while continuously moving it against the skin, the image acquisition device can continuously acquire multiple frames of images at a preset time interval. If the user moves the image acquisition device by touching the skin and then leaving the skin, and then moving to the next position and touching the skin again, the image acquisition device can only continuously acquire multiple frames of images at a preset time interval after touching the skin, and will standby and not acquire images when leaving the skin.

[0045] Step S102: Take the last frame of the image sequence obtained when the integrity meets the preset conditions as the target image, detect the arteries in the target image, and determine the detection result.

[0046] The test results are used to indicate health risks to arteries.

[0047] Specifically, in step S102, the last frame of the image sequence obtained when the integrity level determined in step S101 meets the preset conditions is taken as the target image. Arteries in the target image are detected, and the detection results are determined. These results are used to indicate the health risks of the arteries. After obtaining the detection results, feedback can be provided to the user, indicating the health risks.

[0048] Understandably, the testing of arterial vessels can include measuring the thickness of the arterial vessel walls, detecting the presence of plaques and calcifications, and the corresponding testing methods can be determined according to actual needs.

[0049] The technical solution provided in this disclosure uses a pre-trained positioning network to analyze and identify the image status of images acquired by the user-operated image acquisition device in real time. The image status guides the user to adjust the position of the image acquisition device. When the completeness of the image status indication meets the preset conditions, the last frame of the obtained image sequence is used as the target image for detecting arteries. This solves the problem that users without a medical background cannot accurately locate arteries, lowers the barrier to entry for using the image acquisition device, and allows for the detection of arteries in the target image after positioning. The detection results can be used to indicate the health risks of arteries, effectively reducing the complexity of user operation.

[0050] The method provided in this disclosure can effectively assist users in operation and automatically diagnose arterial blood vessels. It can reduce the complexity of user operation and can be widely used by users with and without medical backgrounds, effectively reducing operation time and improving diagnostic efficiency. At the same time, it has low hardware requirements, fast response speed, and high accuracy. It can be deployed in the cloud, on mobile devices and computers, and can be applied to the field of image-assisted diagnosis of blood vessels in the carotid artery, heart, abdomen, and other parts of the body.

[0051] In one possible implementation, the image state includes correct positioning and incorrect positioning; the type of incorrect positioning includes either near-correct positioning or incorrect positioning.

[0052] Correct localization means that the complete vessel wall of an artery is identified based on the corresponding stacked image; near-correct localization means that part of the vessel wall of an artery is identified based on the corresponding stacked image; incorrect localization means that the vessel wall of an artery cannot be identified based on the corresponding stacked image.

[0053] The location network is generated in the following way:

[0054] Obtain a set of stacked sample images, which includes at least one stacked sample image;

[0055] Based on the sample stacked image set, multiple batches of sample subsets are constructed. Each sample subset includes multiple sample stacked images and the corresponding sample image state for each sample stacked image.

[0056] Based on the sample subsets of each batch, the initial localization network is trained in batches with the goal of minimizing the loss function, until the preset training stopping condition is met, and the localization network is obtained.

[0057] The loss function is a multivariate cross-entropy loss function that describes the difference between the predicted image state of the localization network corresponding to the stacked image samples and the sample image state.

[0058] Specifically, in order to better assist users in locating arteries, the image state in this embodiment includes correct positioning and incorrect positioning, wherein the type of incorrect positioning includes either correct positioning or incorrect positioning, and the incorrect positioning is further divided into finer-grained categories.

[0059] Accordingly, when pre-training the localization network, the labels of the training samples include three types of image states: correct localization, near-correct localization, and incorrect localization.

[0060] In this context, "correct localization" means that the complete vessel wall of an artery is identified from the corresponding stacked image; that is, the proximal and distal ends of the vessel wall are clearly visible in the image. "Near-correct localization" means that a portion of the vessel wall is identified from the corresponding stacked image; that is, the proximal or distal ends of a portion of the vessel wall are not visible or not clearly presented in the image. "Incorrect localization" means that the vessel wall cannot be identified from the corresponding stacked image; that is, the artery is not included in the image.

[0061] When training the localization network, it is necessary to construct a sample stacked image set. The sample stacked image set includes at least one sample stacked image. It can be understood that the sample stacked image is obtained by stacking each sample image in the sample image sequence. Each sample image refers to a continuous image obtained by acquiring videos of target parts of people of different ages and body mass indices through an image acquisition device.

[0062] In this embodiment, the localization network employs a batch training method. Therefore, based on the stacked sample images, multiple batches of sample subsets are constructed. Each sample subset includes multiple stacked sample images and the corresponding sample image state for each stacked sample image. For each stacked sample image, its labeled tags (sample image states) are manually annotated by the observer based on the sample images.

[0063] After acquiring multiple sample subsets, the initial localization network is trained in batches according to the sample subsets of each batch. In each training, the multivariate cross-entropy loss function, which describes the difference between the predicted image state of the localization network corresponding to the stacked sample images and the sample image state, is used as the loss function during training. The parameters of the localization network are updated with the goal of minimizing the loss function until the preset training stopping condition is met, and the localization network is obtained.

[0064] Understandably, the training stopping condition could be reaching a preset number of training iterations or determining network convergence, which can be determined according to actual needs.

[0065] Figure 2 is a schematic diagram of the steps of an artery localization detection method provided in an embodiment of this disclosure. As shown in Figure 2, the image acquisition device is an ultrasonic probe. When the trained localization network is applied, the probe acquires images when the user moves the probe, and obtains stacked images based on the acquired images and inputs them into the localization network. The localization network provides feedback to the user on whether the probe is in the correct position based on the output image status (correct positioning, near-correct positioning, or incorrect positioning).

[0066] If the probe is not in the correct position, the system will provide feedback to the user on whether the probe is close to the correct position. Based on whether it is close to the correct position, the system will indicate to the user whether a slight movement is needed until the probe is confirmed to be in the correct position (the image status is "correctly positioned"). The last frame of the image sequence acquired at this time will be used as the target image. The detection result will be determined based on the detection of the target image.

[0067] The technical solution provided in this disclosure defines image states including correct positioning, near-correct positioning, and misaligned positioning. Based on the real-time acquired images, it can provide feedback to the user on whether the position of the image acquisition device is correct and whether it is close to the artery. Compared with solutions that directly provide yes or no feedback, setting near-correct positioning allows the user to move the image acquisition device more slowly to avoid the problem of incorrect positioning caused by moving too fast or moving too large a range. This can significantly reduce the user's operation time and help users without a medical background to accurately locate the artery.

[0068] In one possible implementation, multiple batches of sample subsets are constructed based on the sample stacked image set, including:

[0069] For multiple observers, obtain the labeled sample set corresponding to each observer. Each labeled sample set includes each sample stacked image in the sample stacked image set, as well as the sample image state labeled by the observer for each sample stacked image. The sample image state is determined by the observer based on the last frame image in the corresponding sample image sequence of the sample stacked image.

[0070] Based on the similarity between the labeled sample sets corresponding to each observer, multiple target observers are selected from each observer, and the similarity between the labeled sample sets corresponding to each target observer is greater than a preset threshold.

[0071] Based on the labeled sample set corresponding to each target observer, multiple batches of sample subsets are constructed, and the number of sample stacked images from each labeled sample set is consistent in each sample subset.

[0072] Specifically, in order to improve the quality of the sample subset during training, this embodiment of the disclosure introduces diversity by using annotations from multiple observers. For multiple observers, the sample image state of the last frame image in the corresponding sample image sequence for each observer is obtained independently, and the sample image state is used as the image state of the corresponding sample stack image.

[0073] Understandably, in practical applications, one can obtain the image state labeled by the observer for each image frame in the video, obtain multiple image sequences by segmenting the video frames (including direct segmentation and overlapping segmentation), and then stack the images in the image sequences to obtain stacked images, thus constructing a sample stacked image set.

[0074] Obtain the image state of each stacked image in the sample stacked image set labeled by each observer, and construct the labeled sample set corresponding to each observer. Each labeled sample set includes each stacked image in the sample stacked image set, as well as the sample image state labeled by the sample stacked image.

[0075] Calculate the similarity between the labeled sample sets corresponding to each observer, and select multiple target observers from each observer based on the similarity.

[0076] Understandably, the purpose of screening observers is to remove observers who are too different from other observers, so as to reduce the possibility of sample labeling errors. Therefore, it is necessary to determine that the similarity between the labeled sample sets corresponding to each target observer is greater than a preset threshold.

[0077] It should be noted that the method for determining that the similarity is greater than the preset threshold can be either to determine the specific value of the preset threshold based on actual needs and then filter, or to combine clustering algorithms to remove observers with low similarity to most observers based on the similarity between the labeled sample sets corresponding to each observer.

[0078] Since the training of the localization network is based on batch training (such as mini-batch learning algorithm), after the target observer is determined, multiple batches of sample subsets are constructed according to the labeled sample set corresponding to each target observer. Each sample subset needs to contain labeled data from multiple target observers evenly, that is, the number of stacked images of samples from each labeled sample set is consistent. It can be understood that consistent number is not completely equivalent to the same number, and a small amount of difference is allowed.

[0079] The technical solution provided in this disclosure uses labeled data from multiple observers to construct training samples for the localization network. This can reduce the model's bias towards a single label, ensure that the number of stacked images from each labeled sample set in each sample subset is consistent, help maintain the balance of the dataset, avoid the model being biased towards certain specific observer labels, and improve the model's performance and reliability.

[0080] In one possible implementation, stacked images are input into a localization network to obtain the image state output by the localization network, including:

[0081] Feature extraction is performed on the stacked images, and the first feature map is output.

[0082] Convert the first feature map into a one-dimensional vector and output the first feature vector;

[0083] The first feature vector is converted into the predicted probability corresponding to each image state, and the image state with the highest probability is obtained and output.

[0084] Specifically, the stacked images are input into the localization network, which processes the stacked images through three steps: feature extraction, feature integration, and vector prediction.

[0085] Feature extraction refers to the automatic extraction of features from the stacked images by the localization network through a series of convolutional and pooling layers, outputting a first feature map, which contains abstract feature information of the stacked images.

[0086] Feature integration refers to converting the first feature map from a multidimensional array into a one-dimensional array, converting each pixel value in the first feature map into an element in a one-dimensional vector, and outputting the first feature vector, which contains all the feature information extracted from the stacked images.

[0087] Vector prediction refers to inputting the first feature vector into one or more fully connected (dense layers), processing the first feature vector by the fully connected layers, and connecting the output of the fully connected layers to the classification layer. The classification layer outputs the predicted probability corresponding to each image state, and the image state with the highest probability is selected as the image state predicted by the localization network and output.

[0088] It is understood that each step implemented by the above-mentioned positioning network is implemented by the specific modules contained in the positioning network. This disclosure only limits the method steps required to be implemented by the positioning network. Network models with the same function can have many different implementation methods. The specific architecture of the positioning network and the type of each module can be determined according to actual needs. This disclosure does not limit this.

[0089] For example, consider the huge market demand and potential in the medical device field. With an aging population and increasing health awareness, the demand for medical imaging technology is constantly growing. Portable ultrasound machines, as low-cost, radiation-free imaging devices, have a large market potential. Furthermore, AI-assisted imaging devices can effectively lower the operational threshold, and have significant market potential in homes, communities, and elderly care facilities.

[0090] The number of deaths and serious cases of cardiovascular and cerebrovascular diseases is increasing rapidly worldwide, leading to a growing demand for portable medical devices that can be used for early stroke risk screening, and have promising applications and opportunities in the global market.

[0091] This disclosure discloses a network structure for a positioning network designed for portable devices. A positioning network based on a lightweight deep convolutional neural network is designed. The positioning network consists of an image input layer, convolutional layers, convolutional modules, pooling layers, and fully connected layers. Each convolutional module contains a 3×3 depthwise separable convolutional layer and a 1×1 convolutional layer. After each of the two convolutional layers, there is a normalization layer and an activation layer. The normalization layer uses batch normalization, the activation layer uses a linear rectified function, and finally, the predicted image state value is output using softmax (normalized exponential function).

[0092] Using the cross-entropy loss function as the objective optimization function, the localization network is trained using stochastic gradient descent (Adam) with adaptive first-order and second-order moment estimation. It is possible to reduce the number of convolutional kernels in the convolutional layers, thereby reducing the weight size of the network model without shrinking the receptive field, thus lowering the computational requirements and improving the computational speed.

[0093] It is understood that the above-described positioning network architecture and module design are merely specific examples to illustrate the methods provided in this disclosure. In addition, other architectures and module designs may be adopted according to actual needs, and this disclosure does not limit them.

[0094] The technical solution provided in this disclosure can extract and integrate features from the input stacked images and ultimately predict the image state by constructing and pre-training a localization network. Compared with existing related solutions, this technical solution does not need to obtain the segmentation information of the intima-media during the localization process. It only needs to use the original image information of the image acquisition device to assist users in accurately locating arterial blood vessels, which is more real-time.

[0095] In one possible implementation, arteries in the target image are detected, and the detection results are determined, including:

[0096] The target image is input into a pre-trained segmentation network to obtain a binary image output by the segmentation network. The binary image is used to segment the intima-media region and non-intima-media region of the arterial wall. The segmentation network is trained using sample images as training samples and the corresponding sample binary images are used as labels.

[0097] The coordinates of multiple inner membrane edge points and multiple outer membrane edge points are extracted from the inner and middle membrane regions of the binary image;

[0098] Based on the coordinates of multiple inner membrane edge points and multiple outer membrane edge points, at least one matching point pair is determined. The matching point pair includes one inner membrane matching point and one outer membrane matching point, and satisfies the following conditions: the inner membrane matching point is the inner membrane edge point with the closest pixel distance to the outer membrane matching point, and the outer membrane matching point is the outer membrane edge point with the closest pixel distance to the inner membrane matching point.

[0099] Convert the pixel distance between the inner membrane matching point and the outer membrane matching point in at least one matching point pair into the actual distance, and use it as the inner and middle membrane thickness value;

[0100] The test results are determined based on the inner membrane thickness value and the preset inner membrane thickness threshold.

[0101] Specifically, Figure 3 is a schematic diagram of a carotid artery ultrasound image provided in an embodiment of this disclosure. As shown in Figure 3, the vessel wall of an artery can be divided into the intima, media, and adventitia. The proximal and distal ends of the vessel wall refer to their positional relationship relative to the image acquisition device. The intima is adjacent to the vessel lumen. The intima-media thickness refers to the thickness of the intima and media. The intima-media thickness value of the vessel wall of an artery is an important indicator for measuring the health risk of an artery.

[0102] Understandably, when the image type is ultrasound, the acoustic properties of the vessel adventitia cause strong reflection of ultrasound, making it prone to artifacts on its posterior side. This affects the representation of the proximal intima-media layer, leading to measurement errors. Therefore, measurements need to be taken on the distal vessel wall to ensure accuracy. In practical applications, the intima-media region of the vessel wall labeled in the binary image and the sample binary image refers to the intima-media region of the distal vessel wall. If the image type is other, the labeling location can be determined according to actual needs.

[0103] The target image determined by the localization network is input into a pre-trained segmentation network to obtain a binary image output by the segmentation network. The binary image is used to segment the intima-media region and non-intima-media region of the arterial wall.

[0104] The segmentation network can convert the input image into a binary image that can indicate the inner and outer membrane regions. The segmentation network is trained using sample images as training samples and the corresponding binary images as labels. The specific network model structure and training method can be determined according to actual needs.

[0105] The sample binary image is a binary image relative to the sample image with labeled inner and outer membrane edge points. In the labeling process, several edge points can be manually labeled on the inner and outer membrane edges of the sample image. Interpolation is then used to generate several corresponding edge points between adjacent inner and outer membrane edge points. The image with the labeled edge points is then converted into a binary image to obtain the sample binary image.

[0106] For example, during the training of a segmentation network model, the Dice similarity coefficient between the segmentation results of the inner and outer membranes in the binary image output by the segmentation network and the segmentation results of the inner and outer membranes in the sample binary image can be used to determine whether the segmentation network has converged. If it is determined that it has not converged, training continues until it is determined that the segmentation network has converged, and the trained segmentation network is obtained.

[0107] The coordinates of multiple inner membrane edge points and multiple outer membrane edge points are extracted from the inner and middle membrane regions of the binary image output by the localization network.

[0108] For example, edge detection and contour extraction can be performed on the inner and middle membrane regions in a binary image to determine all pixels on the inner and outer membrane edges, which are then used as inner and outer membrane edge points, and their corresponding coordinates can be determined.

[0109] The embodiments of this disclosure calculate the thickness of the inner and middle membranes based on the symmetric point correspondence algorithm. For the multiple inner membrane edge points and multiple outer membrane edge points obtained, the distance between each edge point can be calculated by the coordinates of each edge point, and at least one matching point pair is determined based on the nearest neighbor principle of pixel distance.

[0110] The matching point pair includes an inner membrane matching point and an outer membrane matching point, and satisfies the following conditions: the inner membrane matching point is the inner membrane edge point that is closest to the outer membrane matching point in terms of pixel distance, and the outer membrane matching point is the outer membrane edge point that is closest to the inner membrane matching point.

[0111] It is understandable that for inner membrane edge points A1-An and outer membrane edge points B1-Bn, there may be a situation where, for a given inner membrane edge point A1, the outer membrane edge point closest to A1 pixel distance is B2, but for a given outer membrane edge point B2, the inner membrane edge point closest to B2 pixel distance is A2. Therefore, it is necessary to consider both inner and outer membrane edge points and determine the matching point by finding the closest pixel distance between them.

[0112] For example, the coordinates of the edge points of the inner and outer membranes can be extracted from left to right on a two-dimensional segmented image. For each inner membrane edge point Ai, the nearest outer membrane edge point Bj is found. Then, the nearest inner membrane edge point to outer membrane edge point Bj is found. If the inner membrane edge point equals Ai, the match is successful. The first inner and outer membrane edge point obtained by the above algorithm is used as the initial point. By traversing all inner and outer membrane edge points according to the above algorithm, a series of successfully matched point pairs can be obtained.

[0113] After determining at least one matching point pair, the pixel distance between the intima matching point and the adventitia matching point in the at least one matching point pair is converted into the actual distance (that is, the pixel distance is converted into the actual thickness value of the intima-media of the blood vessel wall), and used as the intima-media thickness value.

[0114] Understandably, when there are multiple matching point pairs, the maximum thickness value, minimum thickness value, or average thickness value can be obtained based on the inner membrane thickness values ​​calculated from different matching point pairs.

[0115] It should be noted that in clinical medicine, intima-media thickness can be used to assess cardiovascular disease risk. For example, clinically, the normal value for carotid intima-media thickness is less than 0.9 mm. If the thickness is between 1 mm and 1.5 mm, carotid artery sclerosis or carotid intima-media thickening may be considered. If the thickness is greater than 1.5 mm, plaque may be present.

[0116] In this embodiment of the disclosure, different intima-media thickness thresholds can be preset for arteries at different locations, and the specific values ​​can be adjusted according to actual needs when setting the thresholds to meet the user's needs for health risk monitoring.

[0117] After determining the inner membrane thickness value, compare the inner membrane thickness value with the preset inner membrane thickness threshold. Determine the test result based on the relationship between the two values. The specific way to display the test result can be determined according to actual needs.

[0118] For example, if the intima-media thickness thresholds for the carotid artery are set to 0.9mm, 1.2mm, and 1.5mm, and the intima-media thickness value is determined to be between 0.9mm and 1.2mm based on the size relationship, the user is alerted to a mild health risk; if the intima-media thickness value is determined to be between 1.2mm and 1.5mm, the user is alerted to a moderate health risk; and if the intima-media thickness value is determined to be greater than 1.5mm, the user is alerted to a serious health risk and is advised to seek medical attention for a physical examination.

[0119] The technical solution provided in this disclosure can automatically segment the acquired target image to obtain a binary image that indicates the segmentation of the intima-media in arteries by constructing and pre-training a segmentation network, and calculate the intima-media thickness value. This can reduce the error in thickness calculation caused by subjective judgment of the segmentation of the intima-media, and ensure the standardization of the intima-media segmentation and thickness calculation methods. Regardless of whether the user has a medical background, the thickness calculation can be accurately achieved.

[0120] Compared to existing related solutions, this technical solution does not require binary images of blood vessels. It can directly generate a binary image indicating intima-media segmentation based on the original target image, which can produce continuous and dense intima-media segmentation results. Without the need for interpolation, the calculation of intima-media thickness based on the acquired binary image is more accurate.

[0121] In one possible implementation, the pixel distance between the inner membrane matching point and the outer membrane matching point in at least one matching point pair is converted into an actual distance, before which the following is also included:

[0122] For two adjacent matching point pairs, determine the inner membrane edge curve segment and the outer membrane edge curve segment between the two matching point pairs;

[0123] The target intima edge point is determined from each intima edge point on the intima edge curve segment, and the target epimembrane edge point is determined from each epimembrane edge point on the epimembrane edge curve segment. The distance ratio between the target intima edge point and the target epimembrane edge point on their respective edge curve segments is consistent.

[0124] The target inner membrane edge point and the target outer membrane edge point are used as new matching point pairs.

[0125] Specifically, in order to further improve the accuracy of the inner membrane thickness value, this embodiment of the disclosure considers that after the first extraction of matching point pairs, the distribution of the obtained matching point pairs may be uneven, and there may be a large distance between the matching point pairs, which cannot accurately reflect the inner membrane thickness.

[0126] Therefore, based on the first obtained matching point pairs, the embodiments of this disclosure generate new matching point pairs by interpolating between adjacent matching point pairs.

[0127] For example, for any two adjacent matching points, the inner membrane edge between the two matching points is called the inner membrane edge curve segment, and the outer membrane edge between the two matching points is called the outer membrane edge curve segment. There may be multiple inner membrane edge points on the inner membrane edge curve segment and multiple outer membrane edge points on the outer membrane edge curve segment.

[0128] Based on the consistent distance ratios between the target inner membrane edge point and the target outer membrane edge point on their respective edge curves, the target inner membrane edge point and the target outer membrane edge point are determined.

[0129] It is understandable that the endometrial edge curve segment and the outer membrane edge curve segment are actually two curve segments. The target endometrial edge point and the target outer membrane edge point divide their respective edge curve segments into two parts. The consistent distance ratio means that the ratio of the arc lengths of the corresponding curves in the two parts obtained by dividing each edge curve segment is consistent.

[0130] For example, taking the fixed outer membrane edge point and interpolation to determine the corresponding inner membrane edge point as an example, the method of determining the target outer membrane edge point and the target inner membrane edge point will be explained in detail:

[0131] For any two adjacent matching point pairs, calculate the curve arc length of the inner membrane edge curve segment between the two inner membrane matching points, and the curve arc length of the outer membrane edge curve segment between the two outer membrane matching points.

[0132] From the outer membrane edge curve segment, select a target outer membrane edge point (not a matching point), and calculate the ratio (i.e., distance ratio) of the curve arc length from this outer membrane edge point to the adjacent left-side matching point to the curve arc length of the outer membrane edge curve segment. It can be understood that the distance ratio can also be the ratio calculated using the above method with the right side as a reference, or the ratio of the curve arc length formed by the target outer membrane edge point and the left-side matching point to the curve arc length formed by the target outer membrane edge point and the right-side matching point, etc.

[0133] Multiply the calculated ratio by the curve length of the intima edge curve segment to estimate the curve length between the target intima edge point corresponding to the current target outer membrane edge point and the matching point of the left intima.

[0134] The target endometrial edge point is interpolated in the endometrial edge curve segment based on the arc length of the curve between the calculated target endometrial edge point and the matching point of the left endometrium. It can be understood that the target endometrial point can be a new endometrial edge point generated on the endometrial edge curve segment, or the endometrial edge point with the smallest error selected from the original endometrial edge points.

[0135] The acquired target inner membrane edge point and its corresponding outer membrane edge point are used as new matching point pairs.

[0136] It is understandable that when calculating the thickness value of the inner and middle membranes, for the initial matching point pair (i.e. the first matching point pair obtained) and the newly added matching point pair, the pixel distance between the corresponding inner membrane edge point and the outer membrane edge point of each matching point pair can be used to determine an inner and middle membrane thickness value.

[0137] Using the above method, multiple new matching point pairs can be determined by interpolation between adjacent matching point pairs based on the first-time obtained matching point pairs. This can reduce the error in calculating the intima-media thickness caused by the uneven distribution of the first-time extracted matching point pairs when further calculating the intima-media thickness based on the matching point pairs. It can also maintain high accuracy in areas with large changes in vessel wall curvature, effectively improving the accuracy of intima-media thickness calculation.

[0138] In one possible implementation, the target image is input into a pre-trained segmentation network to obtain a binary image output by the segmentation network, including:

[0139] Feature extraction is performed on the current image of the target at multiple different scales, and second feature maps at multiple different scales are output.

[0140] For each second feature map, adjust the second feature map according to the attention weights in the channel and space, and output the corresponding adjusted second feature map for each second feature map;

[0141] All adjusted second feature maps are fused to obtain a fused feature map. The fused feature map is then mapped to a binary image and output. The binary image has the same size as the target current image.

[0142] Specifically, the target image is input into the segmentation network, which processes the stacked images through three steps: feature encoding, attention weighting, and feature decoding.

[0143] Feature encoding refers to scaling the target image at different scales to obtain multiple images at different scales, and extracting features from each image at each scale to determine a corresponding feature map and outputting it, that is, obtaining multiple second feature maps at different scales.

[0144] The localization network application uses channel attention and spatial attention mechanisms to adjust each second feature map. The channel attention mechanism emphasizes the channels that contribute most to the task and suppresses irrelevant or redundant channels by assigning different weights to each channel. The spatial attention mechanism highlights the regions that contribute most to the task and suppresses irrelevant or redundant regions by weighting specific spatial locations in the feature map.

[0145] The attention weights in the channels and space are determined by training the localization network. The attention weights include channel attention weights (weight vectors) and spatial attention weights (spatial heatmaps). Attention weighting means that for each second feature map, the second feature map is weighted according to the channel attention weights, with the channel weights corresponding to each channel, to obtain the channel feature map after channel attention adjustment. According to the spatial attention weights, each pixel in the channel feature map is weighted to obtain the spatial feature map after spatial attention adjustment. The spatial feature map is used as the adjusted second feature map, and the corresponding adjusted second feature map for each second feature map is output.

[0146] Feature decoding refers to finding a second feature map of the corresponding size for each adjusted second feature map through deconvolution and upsampling operations, stacking them, restoring the resolution of the adjusted second feature map to be the same as the target image, obtaining each restored feature map, fusing each restored feature map to obtain a fused feature map, mapping the fused feature map to a binary image and outputting it, the binary image being the same size as the current target image.

[0147] It is understood that each step implemented by the segmentation network is implemented by the specific modules contained in the segmentation network. This disclosure only limits the method steps required to be implemented by the segmentation network. Network models with the same function can have many different implementations. The specific architecture of the segmentation network and the type of each module can be determined according to actual needs. This disclosure does not limit this.

[0148] For example, starting from the same network structure as the localization network, a lightweight U-shaped (U-Net network structure) attention convolutional network segmentation network can be designed. The segmentation network consists of an encoder, a decoder, and an attention module.

[0149] Based on the U-Net network structure, a segmentation network including an encoder, attention module, and decoder is constructed to handle the segmentation task of the inner and middle membranes in images. It automatically identifies and accurately marks the boundaries of the inner and middle membranes, reducing the errors that may occur when performing image segmentation manually and improving the accuracy of calculating the thickness value of the inner and middle membranes.

[0150] The encoder consists of multiple convolutional modules and pooling layers. Each convolutional module contains a 3×3 depthwise separable convolution. Compared with conventional convolution, separable convolution can significantly reduce the weight size of the model. With batch normalization layers and linear rectified activation functions, the encoder can generate multi-scale image features.

[0151] The attention mechanism, situated between the encoder and decoder, consists of a spatial attention module and a channel attention module. It can generate spatial heatmaps and channel weights, and extract effective spatial and channel information.

[0152] The decoder contains multiple convolutional modules and upsampling operations identical to those in the encoder, fuses multi-scale features, and maps the features to a binary image of the same size as the input image using a sigmoid function.

[0153] When training a segmentation network, the goal can be to minimize the Dice similarity coefficient, and the segmentation network can be trained using stochastic gradient descent (Adam) with adaptive first and second moment estimation.

[0154] By designing localization and segmentation networks with the aim of serving portable devices, the hardware requirements of the network model can be reduced. A portable ultrasound device based on artificial intelligence assistance is designed. As a low-cost, radiation-free, and easy-to-operate imaging device, the portable ultrasound device has the advantages of easy operation, short response time, high accuracy, and low hardware requirements, and has broad application prospects.

[0155] It can provide clinicians with more accurate and comprehensive diagnostic information in the field of medical imaging, which helps in early disease detection and treatment monitoring. It can also be applied to users without a medical background to assess health risks.

[0156] It is understood that the above architecture and module design of the segmentation network are only used as a specific example to illustrate the method provided in this disclosure. In addition, other architectures and module designs can be adopted according to actual needs, and this disclosure does not limit them.

[0157] It is understood that the embodiments described herein are merely specific examples to illustrate the present disclosure in detail, and the described embodiments are intended only to facilitate the understanding of the present disclosure and do not constitute any limitation thereof.

[0158] Figure 4 is a schematic diagram of an artery positioning detection device provided in an embodiment of this disclosure. As shown in Figure 4, the artery positioning detection device 400 includes:

[0159] The positioning and recognition module 401 is used to iteratively execute at least one round of the following steps during the continuous movement of the image acquisition device at the target location until the integrity meets the preset conditions: obtaining the image sequence acquired by the image acquisition device in this round, stacking the image sequence into a stacked image, inputting the stacked image into the positioning network, and obtaining the image state output by the positioning network, wherein the image state represents the integrity of the arterial vessel wall identified according to the corresponding stacked image.

[0160] The artery detection module 402 is used to take the last frame of the image sequence obtained when the integrity meets the preset conditions as the target image, detect the arteries in the target image, determine the detection results, and the detection results are used to indicate the health risks of the arteries.

[0161] In one possible implementation, the image state includes correct positioning and incorrect positioning; the type of incorrect positioning includes either near-correct positioning or incorrect positioning.

[0162] Correct localization means that the complete vessel wall of an artery is identified based on the corresponding stacked image; near-correct localization means that part of the vessel wall of an artery is identified based on the corresponding stacked image; incorrect localization means that the vessel wall of an artery cannot be identified based on the corresponding stacked image.

[0163] The location network is generated in the following way:

[0164] Obtain a set of stacked sample images, which includes at least one stacked sample image;

[0165] Based on the sample stacked image set, multiple batches of sample subsets are constructed. Each sample subset includes multiple sample stacked images and the corresponding sample image state for each sample stacked image.

[0166] Based on the sample subsets of each batch, the initial localization network is trained in batches with the goal of minimizing the loss function, until the preset training stopping condition is met, and the localization network is obtained.

[0167] The loss function is a multivariate cross-entropy loss function that describes the difference between the predicted image state of the localization network corresponding to the stacked image samples and the sample image state.

[0168] In one possible implementation, the artery localization detection device also includes a sample construction module;

[0169] The sample building module is used for:

[0170] For multiple observers, obtain the labeled sample set corresponding to each observer. Each labeled sample set includes each sample stacked image in the sample stacked image set, as well as the sample image state labeled by the observer for each sample stacked image. The sample image state is determined by the observer based on the last frame image in the corresponding sample image sequence of the sample stacked image.

[0171] Based on the similarity between the labeled sample sets corresponding to each observer, multiple target observers are selected from each observer, and the similarity between the labeled sample sets corresponding to each target observer is greater than a preset threshold.

[0172] Based on the labeled sample set corresponding to each target observer, multiple batches of sample subsets are constructed, and the number of sample stacked images from each labeled sample set is consistent in each sample subset.

[0173] In one possible implementation, the location identification module is used for:

[0174] Feature extraction is performed on the stacked images, and the first feature map is output.

[0175] Convert the first feature map into a one-dimensional vector and output the first feature vector;

[0176] The first feature vector is converted into the predicted probability corresponding to each image state, and the image state with the highest probability is obtained and output.

[0177] In one possible implementation, the artery detection module is used for:

[0178] The target image is input into a pre-trained segmentation network to obtain a binary image output by the segmentation network. The binary image is used to segment the intima-media region and non-intima-media region of the arterial wall. The segmentation network is trained using sample images as training samples and the corresponding sample binary images are used as labels.

[0179] The coordinates of multiple inner membrane edge points and multiple outer membrane edge points are extracted from the inner and middle membrane regions of the binary image;

[0180] Based on the coordinates of multiple inner membrane edge points and multiple outer membrane edge points, at least one initial matching point pair is determined. The matching point pair includes one inner membrane matching point and one outer membrane matching point, and satisfies the following conditions: the inner membrane matching point is the inner membrane edge point with the closest pixel distance to the outer membrane matching point, and the outer membrane matching point is the outer membrane edge point with the closest pixel distance to the inner membrane matching point.

[0181] Convert the pixel distance between the inner membrane matching point and the outer membrane matching point in at least one matching point pair into the actual distance, and use it as the inner and middle membrane thickness value;

[0182] The test results are determined based on the inner membrane thickness value and the preset inner membrane thickness threshold.

[0183] In one possible implementation, the artery detection module is used for:

[0184] For two adjacent matching point pairs, determine the inner membrane edge curve segment and the outer membrane edge curve segment between the two matching point pairs;

[0185] The target intima edge point is determined from each intima edge point on the intima edge curve segment, and the target epimembrane edge point is determined from each epimembrane edge point on the epimembrane edge curve segment. The distance ratio between the target intima edge point and the target epimembrane edge point on their respective edge curve segments is consistent.

[0186] The target inner membrane edge point and the target outer membrane edge point are used as new matching point pairs.

[0187] In one possible implementation, the artery detection module is used for:

[0188] Feature extraction is performed on the current image of the target at multiple different scales, and second feature maps at multiple different scales are output.

[0189] For each second feature map, adjust the second feature map according to the attention weights in the channel and space, and output the corresponding adjusted second feature map for each second feature map;

[0190] All adjusted second feature maps are fused to obtain a fused feature map. The fused feature map is then mapped to a binary image and output. The binary image has the same size as the target current image.

[0191] The apparatus of this disclosure embodiment can execute the method provided in this disclosure embodiment, and the implementation principle is similar. The actions performed by each module in the apparatus of each disclosure embodiment correspond to the steps in the method of each disclosure embodiment. For detailed functional descriptions of each module of the apparatus, please refer to the descriptions in the corresponding methods shown above, which will not be repeated here.

[0192] Furthermore, in this disclosure, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Moreover, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0193] This disclosure provides an electronic device (computer device / equipment / system) including a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the steps of the method provided in any optional embodiment of this disclosure and achieve the corresponding technical effects.

[0194] In one optional embodiment, an electronic device is provided. Figure 5 is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. As shown in Figure 5, the electronic device 500 includes a processor 501 and a memory 503. The processor 501 and the memory 503 are connected, for example, via a bus 502. Optionally, the electronic device 500 may further include a transceiver 504, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 504 is not limited to one type, and the structure of the electronic device 500 does not constitute a limitation on the embodiments of this disclosure.

[0195] Processor 501 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with this disclosure. Processor 501 may also be a combination that implements computational functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0196] Bus 502 may include a pathway for transmitting information between the aforementioned components. Bus 502 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 502 may be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in Figure 5, but this does not indicate that there is only one bus or one type of bus.

[0197] The memory 503 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media, other magnetic storage devices, or any other medium capable of carrying or storing computer programs and capable of being read by a computer, without limitation.

[0198] The memory 503 is used to store computer programs that execute embodiments of the present disclosure, and the execution is controlled by the processor 501. The processor 501 is used to execute the computer programs stored in the memory 503 to implement the steps shown in the foregoing method embodiments.

[0199] The electronic devices in this disclosure may include, but are not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), wearable devices, and fixed terminals such as digital TVs and desktop computers.

[0200] This disclosure provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can implement the steps and corresponding content of the aforementioned method embodiments.

[0201] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, can implement the steps and corresponding content of the aforementioned method embodiments.

[0202] It should be noted that the computer-readable storage medium described above in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof.

[0203] In this disclosure, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0204] Computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof, including but not limited to object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0205] The terms “first,” “second,” “third,” “fourth,” “1,” “2,” etc. (if present) in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in a sequence other than that shown in the figures or text.

[0206] It should be understood that although arrows indicate various operation steps in the flowcharts of the embodiments of this disclosure, the order in which these steps are implemented is not limited to the order indicated by the arrows. Unless explicitly stated herein, in some implementation scenarios of the embodiments of this disclosure, the implementation steps in each flowchart can be executed in other orders as required. Furthermore, some or all of the steps in each flowchart may include multiple sub-steps or multiple stages based on the actual implementation scenario. Some or all of these sub-steps or stages can be executed at the same time, and each sub-step or stage can also be executed at different times. In scenarios where execution times differ, the execution order of these sub-steps or stages can be flexibly configured as required, and the embodiments of this disclosure do not limit this.

[0207] The above description is only an optional implementation method for some implementation scenarios of this disclosure. It should be noted that for those skilled in the art, other similar implementation methods based on the technical concept of this disclosure without departing from the technical concept of this disclosure also fall within the protection scope of the embodiments of this disclosure.

Claims

1. A method for arterial localization detection, characterized in that, include: During the continuous movement of the image acquisition device at the target area, the following steps are iteratively executed at least once until the completeness meets the preset conditions: The image sequence acquired by the image acquisition device in this round is obtained, the image sequence is stacked into a stacked image, the stacked image is input into the positioning network, and the image state output by the positioning network is obtained, wherein the image state represents the degree of integrity of the blood vessel wall of the artery identified according to the corresponding stacked image; The last frame of the image sequence obtained when the integrity meets the preset conditions is taken as the target image. The arteries in the target image are detected, and the detection results are determined. The detection results are used to indicate the health risk of the arteries.

2. The artery localization detection method according to claim 1, characterized in that, The image state includes correct positioning and incorrect positioning; the type of incorrect positioning includes either near-correct positioning or incorrect positioning. The term "correct positioning" means that the complete vessel wall of an artery is identified based on the corresponding stacked image; "nearly correct positioning" means that a portion of the vessel wall of an artery is identified based on the corresponding stacked image; and "incorrect positioning" means that the vessel wall of an artery cannot be identified based on the corresponding stacked image. The positioning network is generated in the following way: Obtain a set of stacked sample images, wherein the set of stacked sample images includes at least one stacked sample image; Based on the sample stacked image set, multiple batches of sample subsets are constructed, each sample subset including multiple sample stacked images and the corresponding sample image state of each sample stacked image; Based on the sample subsets of each batch, with minimizing the loss function as the optimization objective, the initial localization network is trained in batches until the preset training stopping condition is met, thereby obtaining the localization network; The loss function is a multivariate cross-entropy loss function that describes the difference between the predicted image state of the localization network corresponding to the stacked sample images and the sample image state.

3. The artery localization detection method according to claim 2, characterized in that, The step of constructing multiple batches of sample subsets based on the stacked image set includes: For multiple observers, obtain the labeled sample set corresponding to each observer. Each labeled sample set includes each sample stacked image in the sample stacked image set, and the sample image state labeled by the observer for each sample stacked image. The sample image state is determined by the observer based on the last frame image in the corresponding sample image sequence of the sample stacked image. Based on the similarity between the labeled sample sets corresponding to each observer, multiple target observers are selected from each observer, and the similarity between the labeled sample sets corresponding to each target observer is greater than a preset threshold. Based on the labeled sample set corresponding to each target observer, multiple batches of sample subsets are constructed, and the number of sample stacked images from each labeled sample set is consistent in each sample subset.

4. The artery localization detection method according to claim 1, characterized in that, The step of inputting the stacked image into the localization network to obtain the image state output by the localization network includes: Feature extraction is performed on the stacked images to output a first feature map; The first feature map is converted into a one-dimensional vector, and the first feature vector is output. The first feature vector is converted into the predicted probability corresponding to each image state, and the image state with the highest probability is obtained and output.

5. The artery localization detection method according to any one of claims 1-4, characterized in that, The step of detecting arteries in the target image and determining the detection results includes: The target image is input into a pre-trained segmentation network to obtain a binary image output by the segmentation network. The binary image is used to segment the intima-media region and non-intima-media region of the arterial wall. The segmentation network is trained using sample images as training samples and the corresponding sample binary images are labeled. The coordinates of multiple inner membrane edge points and multiple outer membrane edge points are extracted from the inner and middle membrane regions of the binary image; Based on the coordinates of the plurality of inner membrane edge points and the plurality of outer membrane edge points, at least one matching point pair is determined. The matching point pair includes one inner membrane matching point and one outer membrane matching point, and satisfies that the inner membrane matching point is the inner membrane edge point with the closest pixel distance to the outer membrane matching point, and the outer membrane matching point is the outer membrane edge point with the closest pixel distance to the inner membrane matching point. The pixel distance between the inner membrane matching point and the outer membrane matching point in the at least one matching point pair is converted into an actual distance and used as the inner and middle membrane thickness value. The detection result is determined based on the inner membrane thickness value and the preset inner membrane thickness threshold.

6. The artery localization detection method according to claim 5, characterized in that, Before converting the pixel distance between the inner membrane matching point and the outer membrane matching point in the at least one matching point pair into an actual distance, the process further includes: For two adjacent matching point pairs, determine the inner membrane edge curve segment and the outer membrane edge curve segment between the two matching point pairs; The target inner membrane edge point is determined from each inner membrane edge point on the inner membrane edge curve segment, and the target outer membrane edge point is determined from each outer membrane edge point on the outer membrane edge curve segment. The distance ratio between the target inner membrane edge point and the target outer membrane edge point on their respective edge curve segments is consistent. The target inner membrane edge point and the target outer membrane edge point are used as newly added matching point pairs.

7. The artery localization detection method according to claim 5, characterized in that, The step of inputting the target image into a pre-trained segmentation network to obtain a binary image output by the segmentation network includes: Feature extraction is performed on the current image of the target at multiple different scales, and second feature maps at multiple different scales are output. For each second feature map, the second feature map is adjusted according to the attention weights in the channel and space, and the adjusted second feature map is output for each second feature map. All adjusted second feature maps are fused to obtain a fused feature map. The fused feature map is then mapped to a binary image and output. The binary image has the same size as the target current image.

8. An artery positioning and detection device, characterized in that, include: The positioning and recognition module is used to iteratively execute at least one round of the following steps during the continuous movement of the image acquisition device at the target location until the integrity meets the preset conditions: obtaining the image sequence acquired by the image acquisition device in this round, stacking the image sequence into a stacked image, inputting the stacked image into the positioning network, and obtaining the image state output by the positioning network, wherein the image state represents the integrity of the arterial vessel wall identified based on the corresponding stacked image; The artery detection module is used to take the last frame of the image sequence obtained when the integrity meets the preset conditions as the target image, detect the arteries in the target image, and determine the detection result. The detection result is used to indicate the health risk of the arteries.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-7.

11. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-7.

Images