Method for determining standard section of ultrasonic image, program product, electronic equipment and storage medium

Abstract

The invention provides an ultrasonic image standard section determination method, a program product, electronic equipment and a storage medium, which are applied to the technical field of image processing, and the method comprises the following steps: obtaining a to-be-identified ultrasonic image; the to-be-recognized ultrasonic image is input into the anatomy detection model, an anatomy detection result output by the anatomy detection model is obtained, and the anatomy detection result comprises the position, the category and the confidence coefficient of the tissue structure in the to-be-recognized ultrasonic image; and judging whether the ultrasonic image to be identified is the ultrasonic image standard section or not according to the anatomical detection result. The ultrasonic image to be recognized is input into the anatomical detection model to obtain the corresponding anatomical detection result, so that whether the ultrasonic image to be recognized meets the standard section requirement or not is automatically judged. Compared with a mode in the prior art, the scheme provided by the embodiment of the invention can improve the accuracy of determining the standard section of the ultrasonic image and the recognition efficiency, and realizes automatic analysis and quality control of the to-be-recognized ultrasonic image.

Description

Technical Field

[0001] The present application relates to the technical field of image processing. Specifically, it relates to a method for determining a standard section of an ultrasound image, a program product, an electronic device, and a storage medium. Background Art

[0002] In recent years, the application of ultrasound imaging in clinical diagnosis has become more and more extensive, especially in the fields of obstetrics and gynecology, cardiology, abdominal ultrasound, etc. Ultrasound imaging has become an important tool for disease screening, diagnosis, and follow-up due to its non-invasive, real-time, and low-cost advantages. However, the quality of ultrasound images is easily affected by various factors, including the doctor's operation technique, probe angle, patient position, and equipment performance. These factors make it a major problem to obtain standard sections during the diagnosis process. Traditional ultrasound image quality control mainly relies on the doctor's experience judgment, lacking objective and unified standards, which not only increases the doctor's workload but also may lead to subjectivity and inconsistency in the diagnosis results. Summary of the Invention

[0003] The purpose of the embodiments of the present application is to provide a method for determining a standard section of an ultrasound image, so as to solve the technical problem of low accuracy in determining the standard section of an ultrasound image in the prior art.

[0004] In a first aspect, the embodiments of the present application provide a method for determining a standard section of an ultrasound image, including: obtaining an ultrasound image to be recognized; inputting the ultrasound image to be recognized into an anatomical detection model to obtain an anatomical detection result output by the anatomical detection model, where the anatomical detection result includes the position, category, and confidence of the tissue structure in the ultrasound image to be recognized; and judging whether the ultrasound image to be recognized is a standard section of the ultrasound image according to the anatomical detection result.

[0005] In the above solution, by inputting the ultrasound image to be recognized into the anatomical detection model to obtain the corresponding anatomical detection result, it is thus possible to automatically judge whether the ultrasound image to be recognized meets the requirements of the standard section. Compared with the method of relying on doctor judgment to determine the standard section of an ultrasound image in the prior art, the solution provided by the embodiments of the present application can improve the accuracy and recognition efficiency of determining the standard section of an ultrasound image, and realizes the automatic analysis and quality control of the ultrasound image to be recognized.

[0006] In an alternative embodiment, the anatomical detection model includes a high-resolution detection head and / or a receptive field module, wherein the high-resolution detection head and the receptive field module are used to detect small targets in the ultrasound image to be recognized. In the above solution, since small targets in ultrasound images usually have low resolution and blurred boundaries, therefore, by introducing a high-resolution detection head, small target details can be captured using high-resolution features, or, by introducing a receptive field module, the receptive field can be extended to cover the context information of small targets, thereby enhancing the detailed features of small targets and improving the detection accuracy.

[0007] In an alternative embodiment, the anatomical detection model is a YOLO model, the high-resolution detection head and the receptive field module are arranged in the P3 layer of the YOLO model, and the receptive field module is arranged after the high-resolution detection head. In the above solution, the YOLO model can be used to detect key anatomical tissue structures in the ultrasound image to be recognized. Among them, the receptive field module is arranged after the high-resolution detection head so that the receptive field module acts on the high-resolution features that have been enhanced by the high-resolution detection head, so as to ensure that on the basis of retaining details, the global context information is further integrated, thereby improving the detection accuracy.

[0008] In an alternative embodiment, before inputting the ultrasound image to be recognized into the anatomical detection model and obtaining the anatomical detection result output by the anatomical detection model, the method further includes: training a neural network model using the following steps to obtain the anatomical detection model: obtaining a sample data set; inputting the sample data set into the neural network model for training to obtain the loss value corresponding to the neural network model, where the loss value includes at least one of the following: a topological loss value and a structural connectivity loss value, the topological loss value represents the distance between two tissue structures in the predicted image, and the structural connectivity loss value represents the distance between the tissue structure in the predicted image and the tissue structure in the real image; optimizing the neural network model according to the loss value to obtain the anatomical detection model. In the above solution, the topological loss value is calculated based on a loss function constrained by anatomical relationships, and the detection rationality is improved through geometric constraints to ensure that the relative positions of the targets conform to anatomical prior knowledge; the structural connectivity loss value is calculated based on a loss function that constrains the rationality of anatomical structures, ensuring that the anatomical structures conform to reasonable anatomical laws and improving the detection accuracy.

[0009] In an alternative embodiment, the obtaining of the sample data set includes: obtaining an original ultrasound image; performing data augmentation processing on the original ultrasound image to obtain the sample images in the sample data set, where the data augmentation processing includes at least one of the following: speckle noise simulation, artifact generation, anatomical structure deformation enhancement, and spatial constraint enhancement. In the above solution, by performing data augmentation on the original ultrasound image, the diversity of the sample data is increased, the generalization ability and robustness of the trained model are improved, and thus the detection accuracy is improved.

[0010] In an alternative embodiment, the determining whether the ultrasound image to be recognized is a standard ultrasound image section according to the anatomical detection result includes: filtering the anatomical detection result based on anatomical rules to obtain a filtered detection result, where the filtered detection result does not include unreasonable detection results; determining whether the ultrasound image to be recognized is the standard ultrasound image section through a predefined determination rule and the filtered detection result. In the above solution, by filtering the anatomical detection result, unreasonable detection results in the anatomical detection result can be filtered out, and at the same time, the standard ultrasound image section is determined through a predefined determination rule, so that the accuracy of determining the standard ultrasound image section can be improved.

[0011] In an alternative embodiment, after determining whether the ultrasound image to be recognized is a standard ultrasound image section according to the anatomical detection result, the method further includes: calculating an image score corresponding to the ultrasound image to be recognized according to the anatomical detection result; if the ultrasound image to be recognized is the standard ultrasound image section and the image score is greater than a score threshold, then retain the ultrasound image to be recognized, otherwise delete the ultrasound image to be recognized. In the above solution, after determining whether the ultrasound image to be recognized is a standard ultrasound image section, the ultrasound image to be recognized can be further scored. The higher the score, the better the image quality and the more it meets the requirements of the standard section. Therefore, the ultrasound image to be recognized can be retained when the image score is greater than the score threshold.

[0012] In an alternative embodiment, after obtaining the ultrasound image to be recognized, the method further includes: inputting the ultrasound image to be recognized into an organ classification model to obtain an organ classification result output by the organ classification model; determining an anatomical detection model corresponding to the ultrasound image to be recognized according to the organ classification result. In the above solution, according to the organ classification result, the corresponding anatomical detection model can be called to ensure the accurate recognition of key anatomical organizational structures, thereby improving the detection accuracy.

[0013] In an alternative embodiment, before inputting the ultrasound image to be recognized into the anatomical detection model and obtaining the anatomical detection result output by the anatomical detection model, the method further includes: for an organ, training a neural network model using the following steps to obtain an anatomical detection model corresponding to the organ: obtaining a sample data set corresponding to the organ; inputting the sample data into the neural network model for training to obtain an anatomical detection model corresponding to the organ, where different confidence thresholds are used during the training of different organs. In the above solution, the anatomical detection model corresponding to each organ can be trained separately. At the same time, since the detection difficulty of anatomical structures based on different organs is different, different confidence thresholds can be configured for different organs to ensure the accurate recognition of key anatomical tissue structures, thereby improving the detection accuracy.

[0014] In a second aspect, an embodiment of the present application provides an apparatus for determining a standard section of an ultrasound image, including: an acquisition module for acquiring an ultrasound image to be recognized; a first input module for inputting the ultrasound image to be recognized into an anatomical detection model to obtain an anatomical detection result output by the anatomical detection model, where the anatomical detection result includes the position, category, and confidence of the tissue structure in the ultrasound image to be recognized; a judgment module for judging whether the ultrasound image to be recognized is a standard section of the ultrasound image according to the anatomical detection result.

[0015] In the above solution, by inputting the ultrasound image to be recognized into the anatomical detection model to obtain the corresponding anatomical detection result, it is thus possible to automatically judge whether the ultrasound image to be recognized meets the requirements of the standard section. Compared with the method in the prior art that relies on doctors' judgment to determine the standard section of the ultrasound image, the solution provided by the embodiment of the present application can improve the accuracy and recognition efficiency of determining the standard section of the ultrasound image, and realizes the automated analysis and quality control of the ultrasound image to be recognized.

[0016] In an alternative embodiment, the anatomical detection model includes a high-resolution detection head and / or a receptive field module, where the high-resolution detection head and the receptive field module are used to detect small targets in the ultrasound image to be recognized. In the above solution, since small targets in ultrasound images usually have low resolution and blurred boundaries, therefore, by introducing a high-resolution detection head, the details of small targets can be captured using high-resolution features, or by introducing a receptive field module, the receptive field can be extended to cover the context information of small targets, thereby enhancing the detailed features of small targets and further improving the detection accuracy.

[0017] In an alternative embodiment, the anatomical detection model is a YOLO model. The high-resolution detection head and the receptive field module are arranged in the P3 layer of the YOLO model, and the receptive field module is arranged after the high-resolution detection head. In the above solution, the YOLO model can be used to detect key anatomical tissue structures in the ultrasound image to be recognized. Among them, the receptive field module is arranged after the high-resolution detection head, so that the receptive field module acts on the high-resolution features that have been enhanced by the high-resolution detection head, so as to further integrate global context information on the basis of retaining details, thereby improving the accuracy of detection.

[0018] In an alternative embodiment, the device for determining the standard ultrasound image section further includes: a first training module, configured to train a neural network model by the following steps to obtain the anatomical detection model: obtaining a sample data set; inputting the sample data set into the neural network model for training to obtain a loss value corresponding to the neural network model, where the loss value includes at least one of the following: a topological loss value and a structural connectivity loss value, the topological loss value represents the distance between two tissue structures in the predicted image, and the structural connectivity loss value represents the distance between the tissue structure in the predicted image and the tissue structure in the real image; optimizing the neural network model according to the loss value to obtain the anatomical detection model. In the above solution, the topological loss value is calculated based on a loss function constrained by anatomical relationships, and the detection rationality is improved through geometric constraints to ensure that the relative positions of the targets conform to anatomical prior knowledge; the structural connectivity loss value is calculated based on a loss function that constrains the rationality of anatomical structures, ensuring that the anatomical structures conform to reasonable anatomical laws and improving the accuracy of detection.

[0019] In an alternative embodiment, the first training module is specifically configured to: obtain an original ultrasound image; perform data augmentation processing on the original ultrasound image to obtain a sample image in the sample data set, where the data augmentation processing includes at least one of the following: speckle noise simulation, artifact generation, anatomical structure deformation enhancement, and spatial constraint enhancement. In the above solution, by performing data augmentation on the original ultrasound image, the diversity of sample data is increased, the generalization ability and robustness of the trained model are improved, and the accuracy of detection is further improved.

[0020] In an alternative embodiment, the determination module is specifically configured to: filter the anatomical detection result based on anatomical rules to obtain a filtered detection result, where the filtered detection result does not include unreasonable detection results; and determine whether the ultrasound image to be recognized is the standard section of the ultrasound image through predefined determination rules and the filtered detection result. In the above solution, by filtering the anatomical detection result, unreasonable detection results in the anatomical detection result can be filtered out, and at the same time, the standard section of the ultrasound image is determined through predefined determination rules, thereby improving the accuracy of determining the standard section of the ultrasound image.

[0021] In an alternative embodiment, the apparatus for determining the standard section of the ultrasound image further includes: a calculation module, configured to calculate an image score corresponding to the ultrasound image to be recognized according to the anatomical detection result; and a retention module, configured to retain the ultrasound image to be recognized if the ultrasound image to be recognized is the standard section of the ultrasound image and the image score is greater than a score threshold, otherwise delete the ultrasound image to be recognized. In the above solution, after determining whether the ultrasound image to be recognized is the standard section of the ultrasound image, the image to be recognized can be further scored. The higher the score, the better the image quality and the more it meets the requirements of the standard section. Therefore, the ultrasound image to be recognized can be retained when the image score is greater than the score threshold.

[0022] In an alternative embodiment, the apparatus for determining the standard section of the ultrasound image further includes: a second input module, configured to input the ultrasound image to be recognized into an organ classification model to obtain an organ classification result output by the organ classification model; and a determination module, configured to determine an anatomical detection model corresponding to the ultrasound image to be recognized according to the organ classification result. In the above solution, according to the organ classification result, the corresponding anatomical detection model can be called to ensure the accurate recognition of key anatomical organizational structures, thereby improving the accuracy of detection.

[0023] In an alternative embodiment, the apparatus for determining the standard section of the ultrasound image further includes: a second training module, configured to train a neural network model for an organ by using the following steps to obtain an anatomical detection model corresponding to the organ: obtain a sample data set corresponding to the organ; input the sample data into the neural network model for training to obtain an anatomical detection model corresponding to the organ, where different confidence thresholds are used during the training of different organs. In the above solution, the anatomical detection model corresponding to each organ can be trained separately. At the same time, since the detection difficulty of the anatomical structures of different organs is different, different confidence thresholds can be configured for different organs to ensure the accurate recognition of key anatomical organizational structures, thereby improving the accuracy of detection.

[0024] In a third aspect, an embodiment of the present application provides a computer program product, including computer program instructions, which, when read and run by a processor, execute the method for determining a standard cross-section of an ultrasonic image as described in the first aspect.

[0025] In a fourth aspect, an embodiment of the present application provides an electronic device, including: a processor, a memory, and a bus; the processor and the memory communicate with each other through the bus; the memory stores computer program instructions executable by the processor, and the processor can execute the method for determining a standard cross-section of an ultrasonic image as described in the first aspect by invoking the computer program instructions.

[0026] In a fifth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores computer program instructions, and when the computer program instructions are run by a computer, the computer is enabled to execute the method for determining a standard cross-section of an ultrasonic image as described in the first aspect.

[0027] To make the above objects, features, and advantages of the present application more obvious and understandable, specific embodiments of the present application are hereinafter given, and in conjunction with the accompanying drawings, the detailed description is as follows. Description of the Drawings

[0028] To more clearly illustrate the technical solutions in the embodiments of the present application, the accompanying drawings required to be used in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present application, and thus should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can also be obtained based on these drawings without creative efforts.

[0029] Figure 1 It is a flowchart of a method for determining a standard cross-section of an ultrasonic image provided by an embodiment of the present application;

[0030] Figure 2 It is a schematic structural diagram of an anatomical detection model provided by an embodiment of the present application;

[0031] Figure 3 It is a structural block diagram of a device for determining a standard cross-section of an ultrasonic image provided by an embodiment of the present application;

[0032] Figure 4 It is a structural block diagram of an electronic device provided by an embodiment of the present application. Detailed Embodiments

[0033] Next, the technical solutions in the embodiments of the present application will be described in conjunction with the accompanying drawings in the embodiments of the present application.

[0034] Please refer to Figure 1 , Figure 1The flowchart of a method for determining a standard section of an ultrasonic image provided by an embodiment of the present application. This method can be, but is not limited to, executed by an electronic device. Figure 4 shows the possible structure of the electronic device, which can be specifically referred to the description about Figure 4 below. Among them, the method for determining the standard section of the ultrasonic image specifically may include the following steps:

[0035] Step S101: Obtain the ultrasonic image to be recognized.

[0036] Step S102: Input the ultrasonic image to be recognized into the anatomical detection model, and obtain the anatomical detection result output by the anatomical detection model. Among them, the anatomical detection result includes the position, category, and confidence of the organizational structure in the ultrasonic image to be recognized.

[0037] Step S103: Determine whether the ultrasonic image to be recognized is a standard section of the ultrasonic image according to the anatomical detection result.

[0038] Specifically, in the above step S101, the ultrasonic image to be recognized is an image obtained by using ultrasonic technology to image the internal tissues and organs of the human body. It can be understood that for different internal tissues and organs of the human body, the ultrasonic image to be recognized can be divided into ultrasonic images corresponding to different organs, such as: the ultrasonic image corresponding to the liver, the ultrasonic image corresponding to the kidney, the ultrasonic image corresponding to the heart, etc.

[0039] It should be noted that the embodiments of the present application do not specifically limit the specific implementation manner of obtaining the ultrasonic image to be recognized, and those skilled in the art can make appropriate adjustments according to the actual situation. For example, the ultrasonic image transmitted in real time by the ultrasonic device can be received; or, the historical image stored in advance can be read from the cloud or locally, etc.

[0040] In the above step S102, the anatomical detection model is used to identify the key anatomical organizational structures in the ultrasonic image to be recognized. Among them, the key anatomical organizational structures refer to the human tissues and structures that can be clearly displayed during ultrasonic examination and are of great significance for disease diagnosis, anatomical positioning, etc., such as: the right branch of the portal vein, inferior vena cava, gallbladder, hepatic vein, etc. corresponding to the liver, the renal cortex, renal medulla, renal pelvis, etc. corresponding to the kidney, the left ventricle, right ventricle, aorta, etc. corresponding to the heart. It can be understood that the input of the anatomical detection model is the ultrasonic image to be recognized, and the output is the position, category, and confidence of the organizational structure in the ultrasonic image to be recognized.

[0041] It should be noted that the specific implementation manner of the above anatomical detection model in the embodiments of the present application is not specifically limited, and those skilled in the art can make appropriate adjustments according to the actual situation. For example: YOLO model, U-Net model, Faster R–CNN model, etc.

[0042] In the above step S103, the standard ultrasound image section refers to the ultrasound scanning plane with a standardized operation method and a specific observation angle adopted during the ultrasound examination process to clearly and accurately display the structure, morphology, position, and mutual relationship of internal organs or tissues in the human body. In the embodiments of the present application, according to the anatomical detection result obtained in the above step S102, it can be further determined whether the to-be-identified ultrasound image is a standard ultrasound image section.

[0043] It should be noted that the specific implementation manner of determining whether the to-be-identified ultrasound image is a standard ultrasound image section in the embodiments of the present application is not specifically limited, and those skilled in the art can make appropriate adjustments according to the actual situation. For example, it can be determined whether the to-be-identified ultrasound image is a standard ultrasound image section according to the number of key anatomical tissue structures; or it can be determined whether the to-be-identified ultrasound image is a standard ultrasound image section according to the positional relationship of key anatomical tissue structures; or it can be determined whether the to-be-identified ultrasound image is a standard ultrasound image section according to the morphology of key anatomical tissue structures, etc.

[0044] In the above solution, by inputting the to-be-identified ultrasound image into the anatomical detection model to obtain the corresponding anatomical detection result, it is thereby realized to automatically determine whether the to-be-identified ultrasound image meets the requirements of the standard section. Compared with the prior art method that relies on doctors to judge and determine the standard ultrasound image section, the solution provided by the embodiments of the present application can improve the accuracy and recognition efficiency of determining the standard ultrasound image section, and realizes the automatic analysis and quality control of the to-be-identified ultrasound image.

[0045] Furthermore, on the basis of the above embodiments, the above anatomical detection model may include a high-resolution detection head and / or a receptive field module (Receptive Field Block, RFB), wherein the high-resolution detection head and the receptive field module are used to detect small targets in the to-be-identified ultrasound image.

[0046] Specifically, the small targets in the to-be-identified ultrasound image refer to the interesting structures or lesions with relatively small sizes in the ultrasound image, such as bile ducts, small blood vessels, micro-nodules, etc., which usually have low resolution and blurred boundaries in the ultrasound image. Therefore, a high-resolution detection head or a receptive field module can be added to the anatomical detection model to detect small targets in the to-be-identified ultrasound image.

[0047] As an implementation, the high-resolution detection head can further extract and enhance the detailed features of small targets through convolutional layers and upsampling operations, and optimize the performance of the high-resolution detection head through a multi-task loss function (e.g., classification loss, bounding box regression loss, etc.).

[0048] As another implementation, the receptive field module simulates the receptive field mechanism in the human visual system and uses a multi-branch convolutional structure to fuse features of different scales. The receptive field module contains multiple parallel convolutional branches, and each branch uses a different convolutional kernel size (e.g., 1×1, 3×3, 5×5) to capture features of different scales, and splices (Concatenation) or weighted fuses the feature maps of multiple branches to generate a feature map with rich context information.

[0049] In the above solution, since small targets in ultrasonic images usually have low resolution and blurred boundaries, therefore, by introducing a high-resolution detection head, the detailed features of small targets can be captured using high-resolution features, or by introducing a receptive field module, the receptive field can be extended to cover the context information of small targets, thereby enhancing the detailed features of small targets and further improving the detection accuracy.

[0050] Furthermore, based on the above embodiments, the above anatomical detection model can be a YOLO model, and the high-resolution detection head and the receptive field module are set in the P3 layer of the YOLO model, and the receptive field module is set after the high-resolution detection head.

[0051] Specifically, small targets in ultrasonic images usually have low resolution and blurred boundaries, so an additional high-resolution detection head can be added to the shallow feature map (e.g., P3 layer) of the YOLO model to make full use of high-resolution features to capture the details of small targets. As described in the above embodiments, this high-resolution detection head further extracts and enhances the detailed features of small targets through convolutional layers and upsampling operations.

[0052] The receptive field module is a module for expanding the receptive field, which can capture the context information of the target and is particularly suitable for small target detection. As described in the above embodiments, the receptive field module contains multiple parallel convolutional branches, and each branch uses a different convolutional kernel size to capture features of different scales, and splices (Concatenation) or weighted fuses the feature maps of multiple branches to generate a feature map with rich context information.

[0053] Among them, the receptive field module expands the receptive field through multi-branch convolutions (1x1, 3x3 dilated convolution, 5x5 dilated convolution) and fuses multi-scale context information. This is particularly important for small object detection because small objects need to combine surrounding environmental information (such as the relative position of blood vessels and surrounding tissues) for accurate identification. As an implementation, the receptive field module can act on the enhanced high-resolution features to further integrate global context information while preserving details.

[0054] Please refer to Figure 2 , Figure 2 which is a schematic diagram of the architecture of an anatomical detection model provided by an embodiment of the present application. The following briefly introduces Figure 2 the shown anatomical detection model.

[0055] Silence: As the starting point of model data input, it receives the image data to be processed; Conv: Represents the convolutional layer, which is used to extract image features; RepNCSPELAN4: Is a specific structure in the network, which may be used to construct a feature pyramid for multi-scale feature extraction; CBLiner, CBFuse: Modules used for linear transformation and fusion of features; Upsample: Upsampling operation, which is used to increase the size of the feature map; Concat: Concatenation operation, which concatenates different feature maps in the channel dimension or other appropriate dimensions; HATHead: High-resolution detection head; RFB: Receptive field module, which expands the receptive field of the model through convolutional kernels of different sizes to capture information at different scales.

[0056] Among them, the high-resolution detection head is deployed after Figure 2 the 15th module and before the 16th module in Figure 2 and the receptive field module is deployed after

[0057] the 16th module and before the 17th module in

[0058] In the above solution, the YOLO model can be used to detect key anatomical tissue structures in the ultrasound image to be recognized. Among them, the receptive field module is set after the high-resolution detection head, so that the receptive field module acts on the high-resolution features that have been enhanced by the high-resolution detection head to further integrate global context information while preserving details, thereby improving the accuracy of detection.

[0059] Step 1), obtain a sample data set.

[0060] Step 2), input the sample data set into the neural network model for training to obtain the loss value corresponding to the neural network model.

[0061] Step 3), optimize the neural network model according to the loss value to obtain the anatomical detection model.

[0062] Specifically, in the above step 1), the sample data set includes multiple sample images and the labels corresponding to the sample images. Among them, the above labels include the position and category of the tissue structure in the sample image. The specific implementation manner of obtaining the sample data set in the embodiments of the present application is not specifically limited, and those skilled in the art can make appropriate adjustments according to the actual situation. For example, the sample images transmitted in real time by the ultrasonic device can be received; or, the sample images stored in advance can be read from the cloud or locally, etc.

[0063] In the above steps 2)-3), the anatomical detection model can be obtained by inputting the sample data set into the neural network model for training. Among them, the loss value in the training process can include at least one of the following: localization loss value L bbox , classification loss value L cls , confidence loss value L conf , topological loss value L Topo (Topological Loss) and structural connectivity loss value L hausdorff .

[0064] Among them, the localization loss value L bbox represents the error between the predicted box in the predicted image and the ground truth box in the real image; the classification loss value L cls represents the loss of incorrect prediction of the target category; the confidence loss value L conf represents the confidence error of calculating whether the predicted box contains the target.

[0065] The topological loss value L Topo represents the distance between two tissue structures in the predicted image. There is a fixed topological relationship between anatomical structures (for example, the right branch of the portal vein must be adjacent to the gallbladder), and the detection rationality can be improved through geometric constraints to ensure that the relative position of the target conforms to the anatomical prior. As an implementation manner, the topological loss value L Topo can be calculated using the following formula:

[0066] L Topo =∑ i ∑ j max(0, d(S i , S j ) - d max );

[0067] Among them, Si and S j are two detected anatomical structures, d(S i , S j ) is the Euclidean distance between the two, and d max is the maximum distance that is anatomically reasonable (medical setting prior knowledge).

[0068] The structural connectivity loss value L Hausdorff characterizes the distance between the organizational structure in the predicted image and the organizational structure in the real image. In medical imaging (especially ultrasound images), anatomical structures usually have fixed topological relationships. For example, the right branch of the portal vein and the gallbladder should be together, the hepatic vein and the inferior vena cava should be in the same area, etc. This structural connectivity is basic knowledge in the medical field, and in object detection, it is very important to maintain the rationality of anatomical structures.

[0069] In traditional object detection tasks, the model focuses on detecting the position and size of a single object (such as a tumor, an organ), while the structural connectivity loss further considers the relative positions and connection relationships between multiple objects. The purpose of designing this loss is to ensure that the detected anatomical structures do not violate these anatomical relationships.

[0070] The Hausdorff distance is a measure of the similarity between two sets (or point sets). It calculates the distance of the farthest point pair between two point sets, that is, the maximum distance from a point in one point set to the nearest point in the other point set. For the problem of detecting anatomical structures, the Hausdorff distance can evaluate the maximum deviation between the predicted anatomical structure and the real structure. The larger this deviation is, the worse the connectivity between the position of the anatomical structure detected by the model and the real structure. If the Hausdorff distance between the predicted anatomical structures exceeds a certain threshold, it means that the connectivity between the anatomical structures violates the medical prior knowledge. At this time, we need to penalize this unreasonable prediction result.

[0071] As an implementation, the structural connectivity loss value L Hausdorff can be calculated using the following formula:

[0072] L Hausdorff = max(max x∈P min y∈Q ||x - y||, max y∈Q min x∈P ||x - y||);

[0073] where P is the set of points of the predicted anatomical structure, and Q is the set of points of the real anatomical structure.

[0074] The final loss function L in the embodiments of the present application total can be:

[0075] L total = λ1(λ coord L bbox + L conf + L cls ) + λ2L Topo + λ3L Hausdorff ;

[0076] where λ1, λ2, and λ3 are the weights of each loss, and λ coord is the balance coefficient, enhancing the weight of the positioning loss.

[0077] In the above solution, the topological loss value is calculated based on the loss function constrained by anatomical relationships, and the detection rationality is improved through geometric constraints to ensure that the relative positions of the targets conform to anatomical prior knowledge; the structural connectivity loss value is calculated based on the loss function that constrains the rationality of anatomical structures, ensuring that the anatomical structures conform to reasonable anatomical laws and improving the accuracy of detection.

[0078] Further, based on the above embodiments, the steps of obtaining the sample data set can specifically include the following steps:

[0079] Step 1), obtain the original ultrasound image.

[0080] Step 2), perform data augmentation processing on the original ultrasound image to obtain the sample images in the sample data set.

[0081] Specifically, in the above step 1), the original ultrasound image refers to an ultrasound image that has not been processed or has only been partially processed. Among them, the embodiments of the present application do not specifically limit the specific implementation manner of obtaining the original ultrasound image, and those skilled in the art can make appropriate adjustments according to the actual situation. For example, the original ultrasound image transmitted in real time by the ultrasound device can be received; or, the historical images stored in advance can be read from the cloud or locally, etc.

[0082] In the above step 2), the data augmentation processing may include at least one of the following: speckle noise simulation, artifact generation, anatomical structure deformation enhancement, and spatial constraint enhancement.

[0083] For speckle noise simulation, the speckle noise in the ultrasound image can usually be simulated by adding Rayleigh Distribution noise because the Rayleigh distribution can better describe the random scattering phenomenon in the ultrasound signal. Specific implementation manner:

[0084] Step A: For each pixel value I(x, y) in the image, generate a random noise N(x, y) with a Rayleigh distribution, where the scale parameter σ of the Rayleigh distribution can be adjusted according to the actual noise level.

[0085] Step B: Superimpose the noise on the original image, I noisy (x, y) = I(x, y) + α·N(x, y), where α is the noise intensity coefficient used to control the strength of the noise.

[0086] Step C: Normalize the result to ensure that the pixel values are within a reasonable range.

[0087] Regarding the generation of artifacts, artifacts can be divided into motion artifacts and acoustic shadows. Among them, motion artifacts can be generated by simulating probe movement or patient breathing movement, while acoustic shadows can be generated by simulating the attenuation and reflection of sound waves in tissues. Specific implementation methods:

[0088] Step A: Motion artifacts: Use image translation, rotation, or blurring operations to simulate probe movement. For example, apply translation or rotation in random directions to the image, and then use Gaussian blur to simulate the motion blur effect.

[0089] Step B: Acoustic shadows: Randomly select regions in the image to simulate regions where sound waves cannot penetrate (such as bones or gas), and set the pixel values of these regions to low intensity or zero.

[0090] Step C: Combine motion artifacts and acoustic shadows to generate a more complex artifact effect.

[0091] Regarding the enhancement of anatomical structure deformation, the enhancement of anatomical structure deformation is based on elastic transformation (ElasticDeformation) to simulate tissue deformation and enhance the robustness of the model to morphological changes. Elastic transformation (ElasticDeformation) is a commonly used data augmentation method that can simulate the deformation of tissues under stress. This method performs a non-linear transformation on the image through a random displacement field. Specific implementation methods:

[0092] Step A: Generate a random displacement field Δ(x, y), where each displacement vector (Δ x , Δ y ) follows a Gaussian distribution.

[0093] Step B: For each pixel (x, y) in the image, perform interpolation calculation according to the displacement field I deformed (x, y) = I(x + Δ x , y + Δ y ).

[0094] Step C: Use bilinear interpolation or cubic spline interpolation to ensure the smoothness of the deformed image while controlling the deformation intensity to avoid anatomical structure distortion caused by excessive distortion.

[0095] Regarding spatial constraint enhancement, in medical images, the relative positions and topological relationships of anatomical structures must be kept reasonable. Therefore, spatial constraint enhancement can be introduced during data augmentation. Specific implementation methods:

[0096] Use the annotation information of anatomical structures (such as segmentation masks, etc.) as constraint conditions to ensure that the relative positions of key anatomical structures remain unchanged or change within a reasonable range during the deformation or enhancement process. For example, the deformation can be constrained by calculating the distance loss between key structures.

[0097] Furthermore, as an implementation method, before performing data augmentation on the original ultrasound image, the original ultrasound image can be preprocessed, including operations such as denoising, normalization, and contrast enhancement, to improve the detection accuracy of the subsequent model. As another implementation method, after performing data augmentation on the original ultrasound image, the image can be adjusted to a unified resolution and size for easy model processing. As yet another implementation method, the original ultrasound image can be preprocessed before performing data augmentation on the original ultrasound image, and the image can be adjusted to a unified resolution and size after performing data augmentation on the original ultrasound image.

[0098] In the above solution, by performing data augmentation on the original ultrasound image, the diversity of sample data is increased, the generalization ability and robustness of the trained model are improved, and thus the detection accuracy is improved.

[0099] Furthermore, based on the above embodiments, the above step S103 can specifically include the following steps:

[0100] Step 1): Filter the anatomical detection results based on anatomical rules to obtain the filtered detection results, where the filtered detection results do not include unreasonable detection results.

[0101] Step 2): Determine whether the ultrasound image to be recognized is a standard ultrasound image section based on the predefined judgment rules and the filtered detection results.

[0102] Specifically, in the above step 1), the unreasonable detection results in the anatomical detection results can be filtered based on anatomical rules to obtain the filtered detection results. As an implementation method, unreasonable detection results can be filtered based on medical prior knowledge. After non-maximum suppression (NMS), the detection results are logically verified through a rule engine.

[0103] It should be noted that the embodiments of the present application do not specifically limit the specific implementation manners of anatomical rules, and those skilled in the art can make appropriate adjustments according to actual situations. For example, if the right branch of the portal vein is detected, the gallbladder must be detected near it, otherwise it is regarded as a false detection; or, the left hepatic vein and the right hepatic vein cannot appear simultaneously in the same image.

[0104] In the above step 2), it is possible to determine whether the ultrasonic image to be recognized is a standard section of the ultrasonic image through predefined judgment rules and the filtered detection results. It should be noted that the embodiments of the present application do not specifically limit the specific implementation manners of the predefined judgment rules, and those skilled in the art can make appropriate adjustments according to actual situations. For example, the predefined rules may include the number, positional relationship, etc. of key anatomical organizational structures. For example: at the first porta hepatis of the liver, the right branch of the portal vein, the inferior vena cava, and the gallbladder need to be detected. Oblique section of the second porta hepatis of the liver: the right vein, the middle vein, the left vein, and the inlet of the inferior vena cava need to be detected.

[0105] In the above solution, by filtering the anatomical detection results, unreasonable detection results in the anatomical detection results can be filtered out, and at the same time, the standard section of the ultrasonic image is judged through predefined judgment rules, thereby improving the accuracy of determining the standard section of the ultrasonic image.

[0106] Further, on the basis of the above embodiments, after the above step S103, the method for determining the standard section of the ultrasonic image provided by the embodiments of the present application may further include the following steps:

[0107] Step 1), calculate the image score corresponding to the ultrasonic image to be recognized according to the anatomical detection results.

[0108] Step 2), if the ultrasonic image to be recognized is a standard section of the ultrasonic image and the image score is greater than the score threshold, retain the ultrasonic image to be recognized, otherwise delete the ultrasonic image to be recognized.

[0109] Specifically, in the above step 1), the image score corresponding to the ultrasonic image to be recognized can be calculated according to the anatomical detection results such as the number, positional relationship, and morphology of the detected key anatomical organizational structures. For example, the scoring rules may include: quantity (whether the detected key anatomical organizational structures are complete), positional relationship (whether the relative positions between the key anatomical organizational structures meet the standards), morphology (whether the morphology of the key anatomical organizational structures is normal (such as whether the blood vessels are complete and the boundaries are clear)), etc.

[0110] As an implementation manner, the numerical range of the image score can be 0 - 100 points, and the higher the score, the better the image quality and the more in line with the requirements of the standard section.

[0111] In step 2) above, if the ultrasonic image to be recognized meets the following conditions: the detected key anatomical organizational structures meet the requirements of the standard cross-section of the ultrasonic image of the organ, and the image score is greater than the score threshold (for example: 90 points), then the ultrasonic image to be recognized is retained; otherwise, the ultrasonic image to be recognized is deleted.

[0112] As an implementation, after retaining the ultrasonic image to be recognized, the retained ultrasonic image to be recognized can also be automatically named according to the organ category and the type of standard cross-section.

[0113] In the above solution, after determining whether the ultrasonic image to be recognized is a standard cross-section of the ultrasonic image, the image to be recognized can be further scored. The higher the score, the better the image quality and the more it meets the requirements of the standard cross-section. Therefore, the ultrasonic image to be recognized can be retained when the image score is greater than the score threshold.

[0114] Furthermore, on the basis of the above embodiments, after step S101 above, the method for determining the standard cross-section of the ultrasonic image provided by the embodiments of the present application may further include the following steps:

[0115] Step 1), input the ultrasonic image to be recognized into the organ classification model to obtain the organ classification result output by the organ classification model.

[0116] Step 2), determine the anatomical detection model corresponding to the ultrasonic image to be recognized according to the organ classification result.

[0117] Specifically, in step 1) above, the organ classification model can be implemented by using a deep learning classification model, such as: ResNet, EfficientNet or MobileNet, etc. Among them, the input of the organ classification model is the ultrasonic image to be recognized, and the output of the organ classification model is the organ classification result, such as: liver, kidney, heart, etc.

[0118] In step 2) above, different organs can correspond to different anatomical detection models, and the anatomical detection models of each organ can be trained separately. Therefore, according to the above organ classification result, the organ category to which the image to be recognized belongs can be determined, and the corresponding anatomical detection model can be called.

[0119] In the above solution, according to the organ classification result, the corresponding anatomical detection model can be called to ensure the accurate recognition of the key anatomical organizational structures, thereby improving the accuracy of detection.

[0120] Further, based on the above embodiments, the specific implementation of training the anatomical detection model will be introduced below. Before the above step S102, for a method of determining a standard cross-section of an ultrasonic image provided by an embodiment of the present application, for an organ, the following steps can be used to train a neural network model to obtain an anatomical detection model corresponding to the organ:

[0121] Step 1), obtain a sample data set corresponding to the organ.

[0122] Step 2), input the sample data into the neural network model for training to obtain an anatomical detection model corresponding to the organ.

[0123] Specifically, in the above step 1), the sample data set includes multiple sample images and labels corresponding to the sample images. Among them, the above labels include the positions and categories of tissue structures in the sample images. The embodiments of the present application do not specifically limit the specific implementation of obtaining the sample data set, and those skilled in the art can make appropriate adjustments according to the actual situation. For example, sample images transmitted in real time by an ultrasonic device can be received; or, sample images stored in advance can be read from the cloud or locally, etc.

[0124] In the above step 2), since the detection difficulties of the anatomical structures of different organs are different (such as liver blood vessels and heart valves), a fixed confidence threshold may lead to missed detections or false detections. Therefore, according to the confidence distribution of the targets of each organ statistically calculated during the training stage, the threshold can be dynamically adjusted during inference (such as the liver detection threshold = 0.6, the heart detection threshold = 0.5); that is to say, the confidence threshold can be adaptively adjusted according to the organ type during training.

[0125] In the above solution, the anatomical detection model corresponding to each organ can be trained separately. At the same time, since the detection difficulties of the anatomical structures of different organs are different, different confidence thresholds can be configured for different organs to ensure the accurate recognition of key anatomical tissue structures, thereby improving the detection accuracy.

[0126] The embodiment of this application provides an Intelligent Retention Quality Control Network (IRQCNet). This technology combines an organ classification model and an organ-specific anatomical detection model. First, it classifies the ultrasound image to be recognized, and then uses a specific anatomical detection model for different organs to identify key anatomical organizational structures. Then, it determines whether the current image meets the standard section requirements of the organ based on the detected organizational structures, and scores the image to be recognized according to the quantity, positional relationship, shape, etc. of the organizational structures. Finally, it automatically retains and names the image. This automated scoring and image retention mechanism can significantly improve the quality of ultrasound images and reduce the workload of doctors.

[0127] The embodiment of this application also proposes a YOLO model for different organs and their specific anatomical structures. By accurately positioning and detecting the organizational structures of specific organs, it ensures the accuracy of detection. To enhance the robustness of the model, a high-resolution detection head and a receptive field module are introduced into the feature map of the P3 layer of YOLO to better capture the details of small targets. These designs of feature fusion and extended receptive field improve the detection ability of the model for small targets. In addition, by introducing L Topo 、L Hausdorff These loss functions not only improve the detection accuracy of the model but also ensure the medical rationality of the detection results, further enhancing the quality control ability of ultrasound images.

[0128] Please refer to Figure 3 , Figure 3 which is the structural block diagram of a device for determining the standard section of an ultrasound image provided by the embodiment of this application. The device 300 for determining the standard section of an ultrasound image includes: an acquisition module 301 for acquiring the ultrasound image to be recognized; a first input module 302 for inputting the ultrasound image to be recognized into the anatomical detection model to obtain the anatomical detection result output by the anatomical detection model, where the anatomical detection result includes the position, category, and confidence of the organizational structures in the ultrasound image to be recognized; and a judgment module 303 for judging whether the ultrasound image to be recognized is the standard section of the ultrasound image according to the anatomical detection result.

[0129] In the above solution, by inputting the ultrasound image to be recognized into the anatomical detection model to obtain the corresponding anatomical detection result, it realizes the automatic judgment of whether the ultrasound image to be recognized meets the standard section requirements. Compared with the method of relying on doctors' judgment to determine the standard section of ultrasound images in the prior art, the solution provided by the embodiment of this application can improve the accuracy and recognition efficiency of determining the standard section of ultrasound images, and realizes the automated analysis and quality control of the ultrasound image to be recognized.

[0130] Further, based on the above embodiments, the anatomical detection model includes a high-resolution detection head and / or a receptive field module, wherein the high-resolution detection head and the receptive field module are used to detect small targets in the ultrasonic image to be recognized.

[0131] In the above solution, since small targets in ultrasonic images usually have low resolution and blurred boundaries, therefore, by introducing a high-resolution detection head, small target details can be captured using high-resolution features, or, by introducing a receptive field module, the receptive field can be extended to cover the context information of small targets, thereby enhancing the detailed features of small targets and further improving the detection accuracy.

[0132] Further, based on the above embodiments, the anatomical detection model is a YOLO model, the high-resolution detection head and the receptive field module are arranged in the P3 layer of the YOLO model, and the receptive field module is arranged after the high-resolution detection head.

[0133] In the above solution, the YOLO model can be used to detect key anatomical organizational structures in the ultrasonic image to be recognized. Among them, the receptive field module is arranged after the high-resolution detection head so that the receptive field module acts on the high-resolution features that have been enhanced by the high-resolution detection head, to ensure that on the basis of retaining details, the global context information is further integrated, thereby improving the detection accuracy.

[0134] Further, based on the above embodiments, the device 300 for determining the standard section of the ultrasonic image further includes: a first training module, which is used to train a neural network model through the following steps to obtain the anatomical detection model: obtain a sample data set; input the sample data set into the neural network model for training to obtain the loss value corresponding to the neural network model, where the loss value includes at least one of the following: a topological loss value and a structural connectivity loss value, the topological loss value represents the distance between two organizational structures in the predicted image, and the structural connectivity loss value represents the distance between the organizational structure in the predicted image and the organizational structure in the real image; optimize the neural network model according to the loss value to obtain the anatomical detection model.

[0135] In the above solution, the topological loss value is calculated based on a loss function constrained by anatomical relationships, and the detection rationality is improved through geometric constraints to ensure that the relative positions of the targets conform to anatomical prior knowledge; the structural connectivity loss value is calculated based on a loss function that constrains the rationality of anatomical structures to ensure that the anatomical structures conform to reasonable anatomical laws and improve the detection accuracy.

[0136] Further, based on the above embodiments, the first training module is specifically configured to: obtain an original ultrasound image; perform data augmentation processing on the original ultrasound image to obtain a sample image in the sample dataset, where the data augmentation processing includes at least one of the following: speckle noise simulation, artifact generation, anatomical structure deformation enhancement, and spatial constraint enhancement.

[0137] In the above solution, by performing data augmentation on the original ultrasound image, the diversity of sample data is increased, the generalization ability and robustness of the trained model are improved, and thus the detection accuracy is improved.

[0138] Further, based on the above embodiments, the determination module 303 is specifically configured to: filter the anatomical detection result based on anatomical rules to obtain a filtered detection result, where the filtered detection result does not include unreasonable detection results; determine whether the ultrasound image to be recognized is the standard section of the ultrasound image according to a predefined determination rule and the filtered detection result.

[0139] In the above solution, by filtering the anatomical detection result, unreasonable detection results in the anatomical detection result can be filtered out, and at the same time, the standard section of the ultrasound image is determined according to a predefined determination rule, so that the accuracy of determining the standard section of the ultrasound image can be improved.

[0140] Further, based on the above embodiments, the apparatus 300 for determining the standard section of the ultrasound image further includes: a calculation module, configured to calculate an image score corresponding to the ultrasound image to be recognized according to the anatomical detection result; a retention module, configured to retain the ultrasound image to be recognized if the ultrasound image to be recognized is the standard section of the ultrasound image and the image score is greater than a score threshold, otherwise delete the ultrasound image to be recognized.

[0141] In the above solution, after determining whether the ultrasound image to be recognized is the standard section of the ultrasound image, the ultrasound image to be recognized can be further scored. The higher the score, the better the image quality and the more in line with the requirements of the standard section. Therefore, the ultrasound image to be recognized can be retained when the image score is greater than the score threshold.

[0142] Further, based on the above embodiments, the apparatus 300 for determining the standard section of the ultrasound image further includes: a second input module, configured to input the ultrasound image to be recognized into an organ classification model to obtain an organ classification result output by the organ classification model; a determination module, configured to determine an anatomical detection model corresponding to the ultrasound image to be recognized according to the organ classification result.

[0143] In the above solution, according to the organ classification result, the corresponding anatomical detection model can be called to ensure the accurate identification of key anatomical organizational structures, thereby improving the accuracy of detection.

[0144] Further, on the basis of the above embodiment, the determining device 300 for the standard section of the ultrasonic image further includes: a second training module, configured to train a neural network model for an organ by using the following steps to obtain the anatomical detection model corresponding to the organ: obtaining a sample data set corresponding to the organ; inputting the sample data into the neural network model for training to obtain the anatomical detection model corresponding to the organ, wherein different confidence thresholds are used during the training of different organs.

[0145] In the above solution, the anatomical detection models corresponding to each organ can be trained separately. At the same time, since the detection difficulty of anatomical structures based on different organs is different, different confidence thresholds can be configured for different organs to ensure the accurate identification of key anatomical organizational structures, thereby improving the accuracy of detection.

[0146] Please refer to Figure 4 , Figure 4 which is a structural block diagram of an electronic device provided by an embodiment of the present application. The electronic device 400 includes: at least one processor 401, at least one communication interface 402, at least one memory 403, and at least one communication bus 404. Among them, the communication bus 404 is used to realize the direct connection and communication of these components, the communication interface 402 is used to communicate with other node devices for signaling or data, and the memory 403 stores machine-readable instructions executable by the processor 401. When the electronic device 400 runs, the processor 401 communicates with the memory 403 through the communication bus 404, and when the machine-readable instructions are called by the processor 401, the above method for determining the standard section of the ultrasonic image is executed.

[0147] For example, the processor 401 of the embodiment of the present application can read a computer program from the memory 403 through the communication bus 404 and execute the computer program to implement the following method: obtaining an ultrasonic image to be recognized; inputting the ultrasonic image to be recognized into the anatomical detection model to obtain the anatomical detection result output by the anatomical detection model, wherein the anatomical detection result includes the position, category, and confidence of the organizational structure in the ultrasonic image to be recognized; judging whether the ultrasonic image to be recognized is a standard section of the ultrasonic image according to the anatomical detection result.

[0148] Among them, the processor 401 includes one or more, which can be an integrated circuit chip with signal processing capabilities. The above-mentioned processor 401 can be a general-purpose processor, including a central processing unit (CPU), a microcontroller unit (MCU), a network processor (NP), or other conventional processors; it can also be a dedicated processor, including a neural-network processing unit (NPU), a graphics processing unit (GPU), a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. Moreover, when there are multiple processor 401s, a part of them can be general-purpose processors and another part can be dedicated processors.

[0149] The memory 403 includes one or more, which can be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), etc.

[0150] It can be understood that Figure 4 the structure shown is only schematic, and the electronic device 400 may also include more or fewer components than those Figure 4 shown, or have a different configuration from that Figure 4 shown. Figure 4The components shown in [the figure] can be implemented using hardware, software, or a combination thereof. In the embodiments of the present application, the electronic device 400 can be, but is not limited to, physical devices such as desktop computers, laptop computers, smartphones, smart wearable devices, vehicle-mounted devices, etc., and can also be virtual devices such as virtual machines. Additionally, the electronic device 400 does not necessarily have to be a single device and can also be a combination of multiple devices, such as a server cluster, and so on.

[0151] The embodiments of the present application also provide a computer program product, including a computer program stored on a computer-readable storage medium. The computer program includes computer program instructions. When the computer program instructions are executed by a computer, the computer can execute the steps of the method for determining the standard ultrasound image section in the above embodiments, for example, including: Step S101: Obtain the ultrasound image to be recognized. Step S102: Input the ultrasound image to be recognized into the anatomical detection model to obtain the anatomical detection result output by the anatomical detection model, where the anatomical detection result includes the position, category, and confidence level of the tissue structure in the ultrasound image to be recognized. Step S103: Determine whether the ultrasound image to be recognized is a standard ultrasound image section according to the anatomical detection result.

[0152] The embodiments of the present application also provide a computer-readable storage medium. The computer-readable storage medium stores computer program instructions. When the computer program instructions are run by a computer, the computer executes the method for determining the standard ultrasound image section described in the foregoing method embodiments.

[0153] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division, and there can be other division methods in actual implementation. Also, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some communication interfaces. The indirect coupling or communication connection of the devices or units can be in electrical, mechanical, or other forms.

[0154] In addition, the units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0155] Furthermore, in each embodiment of the present application, the various functional modules can be integrated together to form an independent part, or each module can exist separately, or two or more modules can be integrated to form an independent part.

[0156] It should be noted that when a function is implemented in the form of a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present application. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0157] In this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0158] The above are only the embodiments of the present application and are not used to limit the protection scope of the present application. For those skilled in the art, the present application can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for determining a standard cross-section of an ultrasonic image, characterized in that, Including: Obtain the ultrasound image to be recognized; Input the ultrasound image to be recognized into the anatomical detection model, and obtain the anatomical detection result output by the anatomical detection model. Among them, the anatomical detection result includes the position, category, and confidence of the tissue structure in the ultrasound image to be recognized; Judge whether the ultrasound image to be recognized is a standard section of the ultrasound image according to the anatomical detection result.

2. The method for determining the standard cross-section of an ultrasonic image according to claim 1, characterized in that, The anatomical detection model includes a high-resolution detection head and / or a receptive field module. Among them, the high-resolution detection head and the receptive field module are used to detect small targets in the ultrasound image to be recognized.

3. The method for determining the standard cross-section of an ultrasonic image according to claim 2, characterized in that The anatomical detection model is a YOLO model. The high-resolution detection head and the receptive field module are set in the P3 layer of the YOLO model, and the receptive field module is set after the high-resolution detection head.

4. The method for determining the standard cross-section of an ultrasonic image according to any one of claims 1 to 3, characterized in that, Before inputting the ultrasound image to be recognized into the anatomical detection model and obtaining the anatomical detection result output by the anatomical detection model, the method further includes: Train the neural network model using the following steps to obtain the anatomical detection model: Obtain a sample data set; Input the sample data set into the neural network model for training to obtain the loss value corresponding to the neural network model. Among them, the loss value includes at least one of the following: topological loss value and structural connectivity loss value. The topological loss value represents the distance between two tissue structures in the predicted image, and the structural connectivity loss value represents the distance between the tissue structure in the predicted image and the tissue structure in the real image; Optimize the neural network model according to the loss value to obtain the anatomical detection model.

5. The method for determining the standard cross-section of an ultrasonic image according to claim 4, characterized in that, The obtaining of the sample data set includes: Obtain the original ultrasound image; Perform data augmentation processing on the original ultrasound image to obtain the sample images in the sample data set. Among them, the data augmentation processing includes at least one of the following: speckle noise simulation, artifact generation, anatomical structure deformation enhancement, and spatial constraint enhancement.

6. The method for determining the standard cross-section of an ultrasonic image according to any one of claims 1 to 3, characterized in that, The judging whether the ultrasound image to be recognized is a standard section of the ultrasound image according to the anatomical detection result includes: Filter the anatomical detection result based on anatomical rules to obtain the filtered detection result. Among them, the filtered detection result does not include unreasonable detection results; Judge whether the ultrasound image to be recognized is the standard section of the ultrasound image through predefined judgment rules and the filtered detection result.

7. The method for determining the standard cross-section of an ultrasonic image according to claim 6, characterized in that, After judging whether the ultrasound image to be recognized is a standard section of the ultrasound image according to the anatomical detection result, the method further includes: Calculate the image score corresponding to the ultrasound image to be recognized according to the anatomical detection result; If the ultrasound image to be recognized is the standard section of the ultrasound image and the image score is greater than the score threshold, retain the ultrasound image to be recognized, otherwise delete the ultrasound image to be recognized.

8. The method for determining the standard section of an ultrasonic image according to any one of claims 1 to 3, characterized in that, After obtaining the ultrasound image to be recognized, the method further includes: Input the ultrasound image to be recognized into the organ classification model to obtain the organ classification result output by the organ classification model; Determine an anatomical detection model corresponding to the to-be-recognized ultrasound image according to the organ classification result.

9. The method for determining the standard cross-section of an ultrasonic image according to claim 8, wherein, Before inputting the to-be-recognized ultrasound image into the anatomical detection model and obtaining the anatomical detection result output by the anatomical detection model, the method further includes: For one organ, train a neural network model by using the following steps to obtain an anatomical detection model corresponding to this organ: Obtain a sample data set corresponding to this organ; Input the sample data into the neural network model for training to obtain an anatomical detection model corresponding to this organ, wherein different confidence thresholds are used during the training of different organs.

10. A computer program product, characterized in that, It includes computer program instructions, and when the computer program instructions are read and run by a processor, the method for determining the standard section of an ultrasound image according to any one of claims 1-9 is executed.

11. An electronic device, characterized in that, It includes: A processor, a memory, and a bus; The processor and the memory complete communication with each other through the bus; The memory stores computer program instructions executable by the processor, and the processor can execute the method for determining the standard section of an ultrasound image according to any one of claims 1-9 by invoking the computer program instructions.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions, and when the computer program instructions are run by a computer, the computer is caused to execute the method for determining the standard section of an ultrasound image according to any one of claims 1-9.

Images