3D fetal heart standard section identification method and system, electronic device and storage medium

Abstract

The application provides a 3D fetal heart standard section identification method and system, electronic equipment and a storage medium. The method comprises the following steps: generating a Gaussian noise image based on Gaussian distribution, inputting the Gaussian noise image and a 3D fetal heart image into an image conditional latent U-Net network model for feature fusion; inputting the fused features into a variational autoencoder (VAE) to generate a 2D standard section of the 3D fetal heart image; and inputting the 2D standard section into a regression model to obtain a 2D standard section of the 3D fetal heart image. The application generates a Gaussian noise image based on Gaussian distribution, performs noise processing on the photographed 3D fetal heart image, and generates a 2D standard section. The generated 2D standard section is generated based on the 3D fetal heart image and has a certain correlation with the 3D fetal heart image. The 2D standard section generated in this way is used as a reference section, and the finally obtained 2D fetal heart section is more standard.

Description

Technical Field

[0001] This invention relates to the field of image recognition, and more specifically, to a method, system, electronic device, and storage medium for recognizing 3D fetal heart standard cross-sections. Background Technology

[0002] Currently, the main steps of the 3D standard section recognition method are: randomly generate a 2D class standard section, and then regress the 2D class standard section into a 2D standard section through a regression model.

[0003] Since the current 2D standard cross section is randomly generated and has no relation to the scanned 3D fetal heart rate data, using the randomly generated 2D standard cross section as a reference cross section results in a low matching degree between the 2D standard cross section generated by the regression model and the 3D fetal heart rate data. Summary of the Invention

[0004] This invention addresses the technical problems existing in the prior art by providing a 3D fetal heart standard section recognition method, system, electronic device, and storage medium, which can overcome the problem of inaccurate 2D standard section generation in the prior art.

[0005] According to a first aspect of the present invention, a method for identifying a standard 3D fetal heart rate profile is provided, comprising:

[0006] Gaussian noise images are generated based on Gaussian distribution, and 3D fetal heart images are obtained.

[0007] The Gaussian noise image and the 3D fetal heart image are input into the Image Conditional Latent U-Net network model, and the Gaussian noise image and the 3D fetal heart image are fused in terms of features.

[0008] The fused features are input into the variational autoencoder (VAE) to generate a 2D-like standard section of the 3D fetal heart image;

[0009] The 2D standard cross-section is input into the regression model to obtain the 2D standard cross-section of the 3D fetal heart image.

[0010] Based on the above technical solution, the present invention can also be improved as follows.

[0011] Optionally, the image conditional latent U-Net network model includes a first encoder and a first decoder. The Gaussian noise image and the 3D fetal heart image are input into the image conditional latent U-Net network model, and feature fusion is performed on the Gaussian noise image and the 3D fetal heart image, including:

[0012] Based on the first encoder, the Gaussian noise image and the 3D fetal heart image are downsampled respectively to extract features at different levels;

[0013] Based on the first decoder, features at different levels of the Gaussian noise image and the 3D fetal heart image extracted by the first encoder are upsampled and fused to generate fused features.

[0014] Optionally, the step of downsampling the Gaussian noise image and the 3D fetal heart image based on the first encoder to extract features at different levels includes:

[0015] The Gaussian noise image and the 3D fetal heart image are downsampled by the first encoder, and the local and global features of the Gaussian noise image and the 3D fetal heart image are extracted respectively.

[0016] The first encoder downsamples the Gaussian noise image and the 3D fetal heart image according to preset image condition information, so that the extracted local and global features of the Gaussian noise image and the 3D fetal heart image meet the condition requirements. The image condition information includes text prompts, which guide the first encoder to generate a specific style of features and / or the target object of interest.

[0017] Optionally, the step of upsampling and fusing features at different levels of the Gaussian noise image and the 3D fetal heart image extracted by the first encoder based on the first decoder to generate fused features includes:

[0018] Based on the local and global features of the Gaussian noise image and the 3D fetal heart image that are fused with the image condition information transmitted by the first encoder, the local and global features are upsampled and fused to generate a fused feature map that meets the condition requirements.

[0019] Optionally, the Gaussian noise image and the 3D fetal heart image are input into the Image Conditional Latent U-Net network model, and feature fusion is performed on the Gaussian noise image and the 3D fetal heart image, including:

[0020] The generated fused feature map and the Gaussian noise image are then input back into the Image Conditional Latent U-Net network model to update the generated fused feature map;

[0021] The process is repeated multiple times to generate the final fused feature map.

[0022] Optionally, the variational autoencoder (VAE) includes a second encoder and a second decoder, inputting fused features into the variational autoencoder (VAE) to generate a 2D-like standard section of the 3D fetal heart image, including:

[0023] The second encoder maps the input fused feature map to a probability distribution in the latent space, the probability distribution being Gaussian, and samples a latent variable z from the probability distribution, the latent variable z being a low-dimensional representation of the fused feature map, the latent variable z containing key information of the fused feature map;

[0024] The second decoder decodes the latent variable z back to the original data space and generates the reconstructed 2D class standard section.

[0025] Optionally, the training process of the variational autoencoder (VAE) is as follows:

[0026] Obtain a training sample set, which includes multiple samples, each of which includes a fused feature map and a corresponding 2D class standard section;

[0027] The variational autoencoder (VAE) is trained based on the training sample set to obtain the variational autoencoder (VAE).

[0028] During the training of the variational autoencoder (VAE), the loss of the variational autoencoder (VAE) includes the KL divergence loss of the second encoder and the reconstruction loss of the second decoder. The KL divergence loss represents the difference between the probability distribution generated by the second encoder and the standard normal distribution, and the reconstruction loss represents the difference between the fused feature map output by the second decoder and the actual fused feature map.

[0029] The parameters of the variational autoencoder (VAE) are adjusted based on the loss to obtain the trained variational autoencoder (VAE).

[0030] According to a second aspect of the present invention, a 3D fetal heart rate standard section recognition system is provided, comprising:

[0031] The first generation module is used to generate Gaussian noise images based on Gaussian distribution and to obtain 3D fetal heart images;

[0032] The feature fusion module is used to input the Gaussian noise image and the 3D fetal heart image into the image conditional latent U-Net network model, and to perform feature fusion on the Gaussian noise image and the 3D fetal heart image;

[0033] The second generation module is used to input the fused features into the variational autoencoder (VAE) to generate a 2D-like standard section of the 3D fetal heart image;

[0034] The regression module is used to input the 2D standard cross-section into the regression model and obtain the 2D standard cross-section of the 3D fetal heart image through regression.

[0035] According to a third aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein the processor is configured to implement the steps of a 3D fetal heart standard section recognition method when executing a computer management program stored in the memory.

[0036] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which a computer management class program is stored, wherein the computer management class program, when executed by a processor, implements the steps of a 3D fetal heart standard section recognition method.

[0037] This invention provides a 3D fetal heart rate standard section recognition method, system, electronic device, and storage medium. It generates a Gaussian noise image based on a Gaussian distribution, adds noise to the captured 3D fetal heart rate image, and generates a 2D-type standard section. The generated 2D-type standard section is generated based on the 3D fetal heart rate image and has a certain correlation with the 3D fetal heart rate image. Using the 2D-type standard section generated in this way as a reference section, the final obtained 2D fetal heart rate section is more standard. Attached Figure Description

[0038] Figure 1 A flowchart of a 3D fetal heart standard section recognition method provided by the present invention;

[0039] Figure 2 This is a diagram illustrating the overall framework for 2D standard cross-section recognition of 3D fetal heart data according to an embodiment of the present invention.

[0040] Figure 3 This is a schematic diagram illustrating the working principle of the first encoder according to an embodiment of the present invention;

[0041] Figure 4 This is a structural diagram of a 3D fetal heart rate standard section recognition system provided in an embodiment of the present invention;

[0042] Figure 5 A schematic diagram of the hardware structure of a possible electronic device provided by the present invention;

[0043] Figure 6 This is a schematic diagram of the hardware structure of a possible computer-readable storage medium provided by the present invention. Detailed Implementation

[0044] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined with each other to form feasible technical solutions. Such combinations are not constrained by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0045] To address the shortcomings of existing technologies, this invention proposes a novel 3D fetal heart rate standard section recognition method. It is mainly based on the stable diffusion framework. First, an initial 2D standard section is generated using 3D fetal heart rate data. Then, the real 2D standard section is regressed from the 2D standard section. The aim is to improve the accuracy and realism of the generated section, and may also include improving the efficiency of the generation process, so as to better meet the needs of related fields for 2D standard sections.

[0046] Figure 1 A flowchart of a 3D fetal heart rate standard section recognition method provided by the present invention is shown below. Figure 1 As shown, the method includes:

[0047] Step 1: Generate a Gaussian noise image based on a Gaussian distribution and obtain a 3D fetal heart image.

[0048] See Figure 2 This is a framework diagram for recognizing 2D standard cross-sections of 3D fetal heart rate data. In this embodiment of the invention, noise is added to the acquired 3D fetal heart rate image; therefore, a noisy image needs to be generated first. In one embodiment of the invention, a Gaussian noise image is generated based on a Gaussian distribution. The size of the generated Gaussian noise image is 64*64.

[0049] Obtain the scanned 3D fetal heart image, and then cut the 3D fetal heart image into a 77*768 pixel size image.

[0050] Step 2: Input the Gaussian noise image and the 3D fetal heart image into the Image Conditional Latent U-Net network model, and perform feature fusion on the Gaussian noise image and the 3D fetal heart image.

[0051] Understandably, the generated Gaussian noise image and the 3D fetal heart image are input together into the image conditional latent U-Net network model for feature fusion.

[0052] In one possible embodiment of the present invention, the image conditional latent U-Net network model includes a first encoder and a first decoder. The Gaussian noise image and the 3D fetal heart image are input into the image conditional latent U-Net network model, and feature fusion is performed on the Gaussian noise image and the 3D fetal heart image, including:

[0053] Step 21: Based on the first encoder, downsample the Gaussian noise image and the 3D fetal heart image respectively to extract features at different levels.

[0054] Understandably, the image conditional latent U-Net network model mainly consists of a first encoder and a first decoder. The first encoder primarily extracts features from the Gaussian noise image and the 3D fetal heart image. During feature extraction, the first encoder mainly downsamples the Gaussian noise image and the 3D fetal heart image to extract features at different levels from the Gaussian noise image and the 3D fetal heart image, respectively.

[0055] In one possible embodiment of the present invention, the step of downsampling the Gaussian noise image and the 3D fetal heart image based on the first encoder to extract features at different levels includes:

[0056] The Gaussian noise image and the 3D fetal heart image are downsampled by the first encoder, and the local and global features of the Gaussian noise image and the 3D fetal heart image are extracted respectively.

[0057] The first encoder downsamples the Gaussian noise image and the 3D fetal heart image according to preset image condition information, so that the extracted local and global features of the Gaussian noise image and the 3D fetal heart image meet the condition requirements. The image condition information includes text prompts, which guide the encoder to generate a specific style of features and / or the target object of interest.

[0058] Understandably, when using the first encoder to extract features from the Gaussian noise image and the 3D fetal heart image respectively, the first encoder is provided with preset image condition information, such as prompting the first encoder to focus on extracting features from which region or which target object, and specifying the style of the feature map extracted by the first encoder. In this way, the feature map extracted by the first encoder will meet certain conditions for subsequent fusion.

[0059] Step 22: Based on the features of different levels of the Gaussian noise image and the 3D fetal heart image extracted by the encoder extracted by the first decoder, upsample and fuse them to generate fused features.

[0060] In one possible embodiment of the present invention, the step of upsampling and fusing features at different levels of the Gaussian noise image and the 3D fetal heart image extracted by the first encoder based on the first decoder to generate fused features includes:

[0061] Based on the local and global features of the Gaussian noise image and the 3D fetal heart image that are fused with the image condition information transmitted by the first encoder, the local and global features are upsampled and fused to generate a fused feature map that meets the condition requirements.

[0062] Understandably, in step 21, the first encoder extracts the local and global features of the Gaussian noise image and the 3D fetal heart image from the image condition information, and the first decoder upsamples and fuses the local and global features of the Gaussian noise image and the 3D fetal heart image to generate fused features.

[0063] In one possible embodiment of the present invention, the Gaussian noise image and the 3D fetal heart image are input into an image conditional latent U-Net network model, and feature fusion is performed on the Gaussian noise image and the 3D fetal heart image, including:

[0064] The generated fused feature map and the Gaussian noise image are then input into the image conditional latent U-Net to update the generated fused feature map; this fusion update is repeated multiple times to generate the final fused feature map.

[0065] Understandably, after initially fusing the features of the Gaussian noise image and the 3D fetal heart image to generate the fused features, the initially generated fused features and the features of the Gaussian noise image are fused again. This process is repeated multiple times to generate the final fused feature map.

[0066] Step 3: Input the fused features into the variational autoencoder (VAE) to generate a 2D-class standard section of the 3D fetal heart image.

[0067] Understandably, after step 2 generates the fused feature map, this step inputs the fused feature map into the variational autoencoder (VAE) to generate a 2D class-standard cross-section of the 3D fetal heart image. The variational autoencoder (VAE) includes a second encoder and a second decoder. See [link to relevant documentation]. Figure 3The second encoder maps the input fused feature map to a probability distribution in the latent space, the probability distribution being Gaussian, and samples a latent variable z from the probability distribution, the latent variable z being a low-dimensional representation of the fused feature map, the latent variable z containing key information of the fused feature map; the second decoder decodes the latent variable z to restore it back to the original data space x'=d(z), generating a reconstructed 2D class standard section.

[0068] The variational autoencoder (VAE) is trained through self-supervised learning. The training process of the variational autoencoder (VAE) is as follows:

[0069] Obtain a training sample set, which includes multiple samples, each of which includes a fused feature map and a corresponding 2D class standard section;

[0070] The variational autoencoder (VAE) is trained based on the training sample set to obtain the variational autoencoder (VAE).

[0071] During the training of the variational autoencoder (VAE), the loss of the variational autoencoder (VAE) includes the KL divergence loss of the second encoder and the reconstruction loss of the second decoder. The KL divergence loss characterizes the difference between the probability distribution generated by the second encoder and the standard normal distribution, and the reconstruction loss characterizes the difference between the fused feature map output by the second decoder and the actual fused feature map.

[0072] The parameters of the variational autoencoder (VAE) are adjusted based on the loss to obtain the trained variational autoencoder (VAE).

[0073] Understandably, training a variational autoencoder (VAE) primarily involves training a process from fused features to a 2D class-standard cross-section. The fused feature map generated in step 2 is input into the trained VAE, which outputs a 2D class-standard cross-section of the 3D fetal heart image.

[0074] Step 4: Input the 2D standard cross-section into the regression model to obtain the 2D standard cross-section of the 3D fetal heart image.

[0075] Understandably, the regression model learns transformation parameters from a 2D class-standard section to a 2D standard section through a regression task. During the regression process, the 2D class-standard section is transformed using initial transformation parameters. The resulting 2D standard section is then compared to the actual 2D standard section, and the transformation parameters of the regression model are adjusted based on the differences. Finally, based on the regression model, the 2D class-standard section of the 3D fetal heart image generated in step 3 is transformed into a 2D standard section.

[0076] See Figure 4The diagram shows a 3D fetal heart rate standard section recognition system provided in an embodiment of the present invention. The recognition system includes:

[0077] The first generation module 401 is used to generate a Gaussian noise image based on a Gaussian distribution and to obtain a 3D fetal heart image;

[0078] The feature fusion module 402 is used to input the Gaussian noise image and the 3D fetal heart image into the image conditional latent U-Net network model, and to perform feature fusion on the Gaussian noise image and the 3D fetal heart image;

[0079] The second generation module 403 is used to input the fused features into the variational autoencoder (VAE) to generate a 2D-like standard section of the 3D fetal heart image.

[0080] The regression module 404 is used to input the 2D standard cross-section into the regression model and obtain the 2D standard cross-section of the 3D fetal heart image through regression.

[0081] It is understood that the 3D fetal heart rate standard section recognition system provided by the present invention corresponds to the 3D fetal heart rate standard section recognition method provided in the foregoing embodiments. The relevant technical features of the 3D fetal heart rate standard section recognition system can be referred to the relevant technical features of the 3D fetal heart rate standard section recognition method, and will not be repeated here.

[0082] Please see Figure 5 , Figure 5 This is a schematic diagram illustrating an embodiment of the electronic device provided in this invention. For example... Figure 5 As shown, this embodiment of the invention provides an electronic device 500, including a memory 510, a processor 520, and a computer program 511 stored in the memory 510 and executable on the processor 520. When the processor 520 executes the computer program 511, it performs the following steps: generating a Gaussian noise image based on a Gaussian distribution and obtaining a 3D fetal heart image; inputting the Gaussian noise image and the 3D fetal heart image into an image conditional latent U-Net to generate a 2D class standard section of the 3D fetal heart image; inputting the 2D class standard section into a variational autoencoder (VAE) to generate a 2D standard section of the 3D fetal heart image.

[0083] Please see Figure 6 , Figure 6 This is a schematic diagram illustrating an embodiment of a computer-readable storage medium provided by the present invention. (See diagram below.) Figure 6As shown, this embodiment provides a computer-readable storage medium 600, on which a computer program 611 is stored. When the computer program 611 is executed by a processor, it performs the following steps: generating a Gaussian noise image based on a Gaussian distribution and obtaining a 3D fetal heart image; inputting the Gaussian noise image and the 3D fetal heart image into an image conditional latent U-Net to generate a 2D class standard section of the 3D fetal heart image; inputting the 2D class standard section into a variational autoencoder (VAE) to generate a 2D standard section of the 3D fetal heart image.

[0084] This invention provides a method, system, electronic device, and storage medium for recognizing standard 3D fetal heart rate cross-sections. It generates Gaussian noise images based on a Gaussian distribution, adds noise to the captured 3D fetal heart rate images, and generates 2D-type standard cross-sections. These 2D-type standard cross-sections are generated based on the 3D fetal heart rate images and have a certain correlation with them. Using these 2D-type standard cross-sections as reference cross-sections results in more standardized 2D fetal heart rate cross-sections.

[0085] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0086] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0087] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0088] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0089] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0090] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0091] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for identifying standard cross-sections of a 3D fetal heartbeat, characterized in that, include: Gaussian noise images are generated based on Gaussian distribution, and 3D fetal heart images are obtained. The Gaussian noise image and the 3D fetal heart image are input into the Image Conditional Latent U-Net network model, and the Gaussian noise image and the 3D fetal heart image are fused in terms of features. The fused features are input into the variational autoencoder (VAE) to generate a 2D-like standard section of the 3D fetal heart image; The 2D standard cross-section is input into the regression model to obtain the 2D standard cross-section of the 3D fetal heart image; The variational autoencoder (VAE) includes a second encoder and a second decoder. The fused features are input into the VAE to generate a 2D-like standard section of the 3D fetal heart image, including: The second encoder maps the input fused feature map to a probability distribution in the latent space, the probability distribution being Gaussian, and samples a latent variable z from the probability distribution, the latent variable z being a low-dimensional representation of the fused feature map, the latent variable z containing key information of the fused feature map; The second decoder decodes the latent variable z back to the original data space, generating a reconstructed 2D class standard section; The training process of the variational autoencoder (VAE) is as follows: Obtain a training sample set, which includes multiple samples, each of which includes a fused feature map and a corresponding 2D class standard section; The variational autoencoder (VAE) is trained based on the training sample set to obtain the variational autoencoder (VAE). During the training of the variational autoencoder (VAE), the loss of the variational autoencoder (VAE) includes the KL divergence loss of the second encoder and the reconstruction loss of the second decoder. The KL divergence loss represents the difference between the probability distribution generated by the second encoder and the standard normal distribution, and the reconstruction loss represents the difference between the fused feature map output by the second decoder and the actual fused feature map. The parameters of the variational autoencoder (VAE) are adjusted based on the loss to obtain the trained variational autoencoder (VAE).

2. The 3D fetal heart standard section identification method according to claim 1, characterized in that, The image conditional latent U-Net network model includes a first encoder and a first decoder. The Gaussian noise image and the 3D fetal heart image are input into the image conditional latent U-Net network model, and feature fusion is performed on the Gaussian noise image and the 3D fetal heart image, including: Based on the first encoder, the Gaussian noise image and the 3D fetal heart image are downsampled respectively to extract features at different levels; Based on the first decoder, features at different levels of the Gaussian noise image and the 3D fetal heart image extracted by the first encoder are upsampled and fused to generate fused features.

3. The 3D fetal heart standard section identification method according to claim 2, characterized in that, The step involves downsampling the Gaussian noise image and the 3D fetal heart image based on the first encoder to extract features at different levels, including: The Gaussian noise image and the 3D fetal heart image are downsampled by the first encoder, and the local and global features of the Gaussian noise image and the 3D fetal heart image are extracted respectively. The first encoder downsamples the Gaussian noise image and the 3D fetal heart image according to preset image condition information, so that the extracted local and global features of the Gaussian noise image and the 3D fetal heart image meet the condition requirements. The image condition information includes text prompts, which guide the first encoder to generate a specific style of features and / or the target object of interest.

4. The 3D fetal heart standard section identification method according to claim 3, characterized in that, The step of upsampling and fusing features at different levels extracted from the Gaussian noise image and the 3D fetal heart image by the first decoder and the first encoder to generate fused features includes: Based on the local and global features of the Gaussian noise image and the 3D fetal heart image that are fused with the image condition information transmitted by the first encoder, the local and global features are upsampled and fused to generate a fused feature map that meets the condition requirements.

5. The 3D fetal heart standard section identification method according to claim 4, characterized in that, The Gaussian noise image and the 3D fetal heart image are input into the Image Conditional Latent U-Net network model, and feature fusion is performed on the Gaussian noise image and the 3D fetal heart image, including: The generated fused feature map and the Gaussian noise image are then input back into the Image Conditional Latent U-Net network model to update the generated fused feature map; The process is repeated multiple times to generate the final fused feature map.

6. A 3D fetal heart rate standard section recognition system, characterized in that, include: The first generation module is used to generate Gaussian noise images based on Gaussian distribution and to obtain 3D fetal heart images; The feature fusion module is used to input the Gaussian noise image and the 3D fetal heart image into the image conditional latent U-Net network model, and to perform feature fusion on the Gaussian noise image and the 3D fetal heart image; The second generation module is used to input the fused features into the variational autoencoder (VAE) to generate a 2D-like standard section of the 3D fetal heart image; The regression module is used to input the 2D standard cross-section into the regression model and obtain the 2D standard cross-section of the 3D fetal heart image through regression. The variational autoencoder (VAE) includes a second encoder and a second decoder. The fused features are input into the VAE to generate a 2D-like standard section of the 3D fetal heart image, including: The second encoder maps the input fused feature map to a probability distribution in the latent space, the probability distribution being Gaussian, and samples a latent variable z from the probability distribution, the latent variable z being a low-dimensional representation of the fused feature map, the latent variable z containing key information of the fused feature map; The second decoder decodes the latent variable z back to the original data space, generating a reconstructed 2D class standard section; The training process of the variational autoencoder (VAE) is as follows: Obtain a training sample set, which includes multiple samples, each of which includes a fused feature map and a corresponding 2D class standard section; The variational autoencoder (VAE) is trained based on the training sample set to obtain the variational autoencoder (VAE). During the training of the variational autoencoder (VAE), the loss of the variational autoencoder (VAE) includes the KL divergence loss of the second encoder and the reconstruction loss of the second decoder. The KL divergence loss represents the difference between the probability distribution generated by the second encoder and the standard normal distribution, and the reconstruction loss represents the difference between the fused feature map output by the second decoder and the actual fused feature map. The parameters of the variational autoencoder (VAE) are adjusted based on the loss to obtain the trained variational autoencoder (VAE).

7. An electronic device, characterized in that, The device includes a memory and a processor, wherein the processor is used to execute computer management programs stored in the memory to implement the steps of the 3D fetal heart standard section recognition method according to any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, It stores a computer management program, which, when executed by a processor, implements the steps of the 3D fetal heart standard section recognition method according to any one of claims 1-5.

Images