Three-dimensional modeling method and system for medical image data

Abstract

The invention belongs to the technical field of medical image processing, and discloses a three-dimensional modeling method and system for medical image data. The method comprises the following steps: constructing an image segmentation model and a three-dimensional reconstruction model by using a deep learning algorithm according to a plurality of historical medical image data and corresponding historical patient diagnosis data; acquiring real-time medical image data and corresponding real-time patient diagnosis data, and performing image segmentation on the real-time medical image data by using the image segmentation model to obtain a real-time segmented image; and performing three-dimensional reconstruction on the real-time segmented image and the real-time patient diagnosis data by using a three-dimensional reconstruction model to obtain a real-time three-dimensional model of the real-time medical image data. According to the method, the problems of insufficient segmentation precision, three-dimensional reconstruction distortion and low calculation efficiency in the prior art are solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical image processing, and particularly relates to a three-dimensional modeling method and system for medical image data. Background Art

[0002] Medical image data refers to the image materials obtained through imaging techniques during medical diagnosis and treatment. These data are usually generated by various medical imaging devices, such as X-ray films, computed tomography, magnetic resonance imaging, ultrasonic examinations, positron emission tomography, etc. Three-dimensional modeling of medical image data is an important part of medical image analysis, which is of great significance for disease diagnosis, treatment, and prognosis evaluation.

[0003] The existing three-dimensional modeling of medical image data has the following defects: 1) Insufficient segmentation accuracy: The image segmentation algorithms in the existing technology cannot accurately segment complex medical image data, especially when dealing with images with more noise and low contrast, the segmentation accuracy will be affected; 2) Distortion in three-dimensional reconstruction: The existing three-dimensional reconstruction technology introduces distortion during the reconstruction process. Especially when dealing with non-uniform or irregular structures, the reconstructed three-dimensional model may have a large difference from the actual anatomical structure; 3) Low computational efficiency: The three-dimensional modeling methods of the existing technology require a large amount of computing resources, resulting in slow processing speed, which is not conducive to rapid clinical diagnosis and treatment planning. Summary of the Invention

[0004] In order to solve the problems of insufficient segmentation accuracy, distortion in three-dimensional reconstruction, and low computational efficiency existing in the prior art, the purpose of the present invention is to provide a three-dimensional modeling method and system for medical image data.

[0005] The technical solution adopted by the present invention is as follows: A three-dimensional modeling method for medical image data, comprising the following steps: According to a number of historical medical image data and corresponding historical patient diagnosis data, use a deep learning algorithm to construct an image segmentation model and a three-dimensional reconstruction model; Collect real-time medical image data and corresponding real-time patient diagnosis data, and use the image segmentation model to perform image segmentation on the real-time medical image data to obtain a real-time segmented image; Use the three-dimensional reconstruction model to perform three-dimensional reconstruction on the real-time segmented image and the real-time patient diagnosis data to obtain a real-time three-dimensional model of the real-time medical image data.

[0006] Further, according to a number of historical medical image data and corresponding historical patient diagnosis data, using a deep learning algorithm to construct an image segmentation model and a three-dimensional reconstruction model, comprising the following steps: Collect a number of historical medical image data and corresponding historical patient diagnosis data, and preprocess them to obtain a number of preprocessed historical medical image data and a number of preprocessed historical patient diagnosis data; According to a number of preprocessed historical medical image data, use deep learning algorithms to construct an image segmentation model and generate a number of historical segmentation images; According to a number of historical segmentation images and corresponding preprocessed historical patient diagnosis data, use deep learning algorithms to construct a three-dimensional reconstruction model.

[0007] Furthermore, the preprocessing of historical medical image data includes data cleaning, Gaussian denoising, size normalization, image enhancement, and label annotation performed in sequence; The preprocessing of historical patient diagnosis data includes data cleaning, Gaussian denoising, and magnitude normalization performed in sequence.

[0008] Furthermore, the image segmentation model is constructed based on the CNN-FPN-SAM-Med2D algorithm, and the image segmentation model includes a graph feature extraction module based on the CNN algorithm, a feature fusion module based on the FPN algorithm, and an image segmentation module constructed based on the SAM-Med2D algorithm connected in sequence.

[0009] Furthermore, according to a number of preprocessed historical medical image data, use deep learning algorithms to construct an image segmentation model and generate a number of historical segmentation images, including the following steps: Use the CNN algorithm to construct the basic network structure of the image segmentation model to obtain an initial graph feature extraction module; the basic network structure includes a number of convolutional layers and a number of pooling layers connected alternately in sequence; Add skip connections, extract the feature maps of all convolutional layers in the basic network structure, and use upsampling and horizontal connections according to a number of feature maps to construct a feature pyramid to obtain an initial feature fusion module; Use the SAM-Med2D algorithm to construct an initial image segmentation module, and connect the input end of the initial image segmentation module to the output end of the initial feature fusion module; Integrate the initial graph feature extraction module, the initial feature fusion module, and the initial image segmentation module to obtain an initial image segmentation model; Divide a number of preprocessed historical medical image data into a model training set and a model test set according to a ratio of 7:3; Input the model training set, optimize and train the initial image segmentation model to obtain an optimized image segmentation model and generate a number of historical segmentation images; Input the model test set, conduct model testing on the optimized image segmentation model, obtain the model test accuracy. If several model test accuracies are greater than the accuracy threshold, then output the final image segmentation model.

[0010] Furthermore, the 3D reconstruction model is constructed based on the cGAN-MLP algorithm, and the 3D reconstruction model includes a conditional information embedder and a conditional information processor both constructed based on the MLP algorithm, as well as a generator and a discriminator both constructed based on the RNN algorithm. The generator is respectively connected to the conditional information embedder and the discriminator, and the discriminator is connected to the conditional information processor.

[0011] Furthermore, according to several historical segmented images and the corresponding preprocessed historical patient diagnosis data, use the deep learning algorithm to construct a 3D reconstruction model, including the following steps: Set corresponding real 3D data for each historical segmented image and the corresponding preprocessed historical patient diagnosis data; Use the cGAN algorithm to construct an initial generator and an initial discriminator, and use the MLP algorithm to set a conditional information embedder for the initial generator and a conditional information processor for the initial discriminator; Integrate the initial generator, the initial discriminator, the conditional information embedder, and the conditional information processor to obtain an initial 3D reconstruction model; Combine the first loss function of the initial generator and the second loss function of the initial discriminator to obtain a comprehensive loss function; Extract the historical image features of the historical segmented images and the historical data features of the preprocessed historical patient diagnosis data; Use the conditional information embedder to perform conditional embedding on the historical image features and the historical data features to obtain historical conditional information embedded features; Use the initial generator to generate 3D data according to the historical conditional information embedded features to obtain generated 3D data; Use the conditional information processor to perform conditional information processing according to the real 3D data and the corresponding generated 3D data to obtain historical conditional information; Use the initial discriminator to perform data discrimination according to the real 3D data, the corresponding generated 3D data, and the historical conditional information to obtain historical data discrimination results; Traverse all historical segmented images and the corresponding preprocessed historical patient diagnosis data, repeat the above steps, and perform optimization training on the initial 3D reconstruction model; Use the comprehensive loss function to generate the historical loss values during the optimization training process. If the historical loss values are lower than the loss value threshold, then output the final 3D reconstruction model, otherwise, continue with the optimization training.

[0012] Further, collect real-time medical image data and corresponding real-time patient diagnosis data, and use an image segmentation model to perform image segmentation on the real-time medical image data to obtain a real-time segmented image, including the following steps: Collect real-time medical image data and corresponding real-time patient diagnosis data, and input the real-time medical image data into the image segmentation model; Use the graph feature extraction module of the image segmentation model to extract several real-time feature maps of the real-time medical image data; Use the feature fusion module of the image segmentation model to perform feature fusion on the several real-time feature maps to obtain real-time fusion features; Use the image segmentation module of the image segmentation model to perform image segmentation on the real-time fusion features to obtain a real-time segmented image.

[0013] Further, use a three-dimensional reconstruction model to perform three-dimensional reconstruction on the real-time segmented image and the real-time patient diagnosis data to obtain a real-time three-dimensional model of the real-time medical image data, including the following steps: Input the real-time segmented image and the real-time patient diagnosis data into the three-dimensional reconstruction model, and extract the real-time image features of the real-time segmented image and the real-time data features of the real-time patient diagnosis data; Use the conditional information embedder of the three-dimensional reconstruction model to perform conditional embedding on the real-time image features and the real-time data features to obtain real-time conditional information embedded features; Use the generator of the three-dimensional reconstruction model to perform three-dimensional data generation according to the real-time conditional information embedded features to obtain real-time three-dimensional data; Perform three-dimensional reconstruction according to the real-time three-dimensional data to obtain a real-time three-dimensional model of the real-time medical image data.

[0014] A three-dimensional modeling system for medical image data, used to implement the three-dimensional modeling method. The system includes a model construction unit, an image segmentation unit, and a three-dimensional reconstruction unit connected in sequence.

[0015] The beneficial effects of the present invention are as follows: The present invention discloses a three-dimensional modeling method and system for medical image data. The constructed image segmentation model can more accurately segment complex medical image data, reduce the modeling error caused by inaccurate segmentation, and improve the segmentation accuracy; the constructed three-dimensional reconstruction model combines the real-time segmented image and the real-time patient diagnosis data for three-dimensional reconstruction, can more realistically restore the anatomical structure, and reduce the distortion problem in the reconstruction process; by segmenting the two-dimensional image and restoring the three-dimensional structure from the segmented image, the processing speed and calculation efficiency are improved, enabling the three-dimensional modeling to be completed quickly, especially suitable for clinical scenarios that require quick response; perform automated image segmentation and three-dimensional reconstruction, with a high degree of automation, reduce manual intervention, and reduce the error rate.

[0016] Other beneficial effects of the present invention will be further described in the specific implementation manners. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a flowchart of a three-dimensional modeling method for medical image data of the present invention.

[0018] Figure 2 is a structural block diagram of a three-dimensional modeling system for medical image data of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0019] The present invention will be further explained below with reference to the accompanying drawings and specific embodiments.

[0020] Embodiment 1: As Figure 1 shown, this embodiment provides a three-dimensional modeling method for medical image data, including the following steps: S1: According to a number of historical medical image data and corresponding historical patient diagnosis data, use a deep learning algorithm to construct an image segmentation model and a three-dimensional reconstruction model, including the following steps: S1-1: Collect a number of historical medical image data and corresponding historical patient diagnosis data, and preprocess them to obtain a number of preprocessed historical medical image data and a number of preprocessed historical patient diagnosis data; The preprocessing of historical medical image data includes data cleaning, Gaussian denoising, size normalization, image enhancement, and label annotation in sequence; Data cleaning is used to delete incomplete and low-quality image data; identify and correct errors in the data, such as incorrect labels or metadata; remove duplicate image data; Gaussian denoising applies a Gaussian filter to smooth the image data to reduce random noise and improve image quality; helps the model better learn the true features of the image; Size normalization adjusts all image data to a unified size for easy model processing; helps reduce the dependence of the model on the image size; Image enhancement applies a series of image processing techniques (such as rotation, scaling, flipping, contrast adjustment, etc.) to expand the dataset, improve the generalization ability of the model, simulate various imaging conditions that may be encountered in reality, and make the model more robust; Label annotation annotates the image data to mark the regions of interest (such as diseased tissues); usually requires the participation of professionals to ensure the accuracy and consistency of the annotation; The preprocessing of historical patient diagnosis data includes data cleaning, Gaussian denoising, and magnitude normalization in sequence; Data cleaning, removing invalid, incorrect, or missing data; ensuring data integrity and accuracy; Gaussian denoising, for numerical data, there may be measurement errors or outliers, and Gaussian denoising can reduce the impact of these noises; in diagnostic data, Gaussian denoising may involve handling statistical outliers; magnitude normalization, scaling feature data of different magnitudes into a small, unified interval, such as [0, 1] or [-1, 1]; helps prevent certain features from dominating in model training due to large numerical ranges. S1-2: Based on a number of preprocessed historical medical image data, use deep learning algorithms to construct an image segmentation model and generate a number of historical segmentation images. The image segmentation model is constructed based on the Convolutional Neural Networks (CNN)-Feature Pyramid Networks (FPN)-Segment Anything Model - Medicine 2Dimensionality (SAM-Med2D) algorithm, and the image segmentation model includes a graph feature extraction module based on the CNN algorithm, a feature fusion module based on the FPN algorithm, and an image segmentation module constructed based on the SAM-Med2D algorithm connected in sequence. The graph feature extraction module extracts hierarchical features in the image through a series of convolutional layers, activation functions, and pooling layers. The feature fusion module uses skip connections to fuse feature maps of different levels extracted by the CNN, constructs a feature pyramid through upsampling and lateral connections, and transfers high-level semantic information to the low level to enhance the semantic expression ability of low-level features. This can effectively combine low-level detail features and high-level semantic features, enabling the network to maintain a high accuracy when detecting tissue or lesion areas of different sizes in medical image data. SAM-Med2D uses an attention mechanism and a conditional generation model to predict segmentation masks in the image, and these masks are used for image segmentation generation. Based on a number of preprocessed historical medical image data, using deep learning algorithms to construct an image segmentation model and generate a number of historical segmentation images, including the following steps: S1-2-1: Use the CNN algorithm to construct the basic network structure of the image segmentation model to obtain the initial graph feature extraction module; the basic network structure includes a number of convolutional layers and a number of pooling layers connected alternately in sequence. The convolutional layer is used to extract local features, and the pooling layer is used to reduce the spatial dimension of the features. This structure can effectively extract key features in the image and provide rich information for subsequent segmentation tasks. S1-2-2: Add skip connections, extract the feature maps of all convolutional layers in the basic network structure, and use upsampling and lateral connections based on several feature maps to construct a feature pyramid to obtain an initial feature fusion module; Skip connections (such as residual connections) allow information in the network to propagate directly, reducing the vanishing gradient problem. The feature pyramid fuses feature maps of different scales through upsampling and lateral connections to retain detailed information; the feature pyramid network can improve the network's utilization of multi-scale features, helping the segmentation model to more accurately identify objects of different sizes; S1-2-3: Use the SAM-Med2D algorithm to construct an initial image segmentation module, and connect the input end of the initial image segmentation module to the output end of the initial feature fusion module; The SAM-Med2D algorithm can improve the accuracy and robustness of segmentation, especially when dealing with complex medical images; S1-2-4: Integrate the initial graph feature extraction module, the initial feature fusion module, and the initial image segmentation module to obtain an initial image segmentation model; S1-2-5: Divide several preprocessed historical medical image data into a model training set and a model test set according to a ratio of 7:3; S1-2-6: Input the model training set, optimize and train the initial image segmentation model to obtain an optimized image segmentation model, and generate several historical segmentation images; S1-2-7: Input the model test set, test the optimized image segmentation model to obtain the model test accuracy. If several model test accuracies are greater than the accuracy threshold, output the final image segmentation model; S1-3: Use a deep learning algorithm to construct a three-dimensional reconstruction model based on several historical segmentation images and the corresponding preprocessed historical patient diagnosis data; The three-dimensional reconstruction model is constructed based on the Conditional Generative Adversarial Network (cGAN)-Multilayer Perceptron (MLP) algorithm. The three-dimensional reconstruction model includes a conditional information embedder and a conditional information processor both based on the MLP algorithm, and a generator and a discriminator both based on the RNN algorithm. The generator is respectively connected to the conditional information embedder and the discriminator, and the discriminator is connected to the conditional information processor; A conditional information embedder for processing conditional information including image features and data features to obtain conditional information embedding features in a sequence format, thereby integrating multi-modal conditional information into the generation process; a generator for generating three-dimensional data based on the conditional information embedding features output by processing the conditional information embedder and random noise; a conditional information processor for processing additional conditional information of the three-dimensional data to help the discriminator more accurately judge the authenticity of the three-dimensional data; a discriminator for analyzing whether the generated three-dimensional data is real and conforms to the given conditional information; the generator and the discriminator compete with each other through an adversarial training process, the generator attempts to generate three-dimensional data that can deceive the discriminator, while the discriminator attempts to better identify real and fake three-dimensional data, achieving accurate and efficient generation of three-dimensional data; According to a number of historical segmentation images and corresponding preprocessed historical patient diagnosis data, using a deep learning algorithm, a three-dimensional reconstruction model is constructed, including the following steps: S1-3-1: Set corresponding real three-dimensional data for each historical segmentation image and corresponding preprocessed historical patient diagnosis data; Set real three-dimensional data for each historical segmentation image and corresponding preprocessed patient diagnosis data as a reference standard; S1-3-2: Use the cGAN algorithm to construct an initial generator and an initial discriminator, and use the MLP algorithm to set a conditional information embedder for the initial generator and a conditional information processor for the initial discriminator; S1-3-3: Integrate the initial generator, the initial discriminator, the conditional information embedder, and the conditional information processor to obtain an initial three-dimensional reconstruction model; S1-3-4: Combine the first loss function of the initial generator and the second loss function of the initial discriminator to obtain a comprehensive loss function; The model can be more comprehensively optimized through the comprehensive loss function, and the training of the generator and the discriminator can be balanced; S1-3-5: Extract the historical image features of the historical segmentation images and the historical data features of the preprocessed historical patient diagnosis data; S1-3-6: Use the conditional information embedder to perform conditional embedding on the historical image features and historical data features to obtain historical conditional information embedding features; S1-3-7: Use the initial generator to generate three-dimensional data according to the historical conditional information embedding features to obtain generated three-dimensional data; S1-3-8: Use the conditional information processor to perform conditional information processing according to the real three-dimensional data and the corresponding generated three-dimensional data to obtain historical conditional information; S1-3-9: Use the initial discriminator to perform data discrimination based on real 3D data, corresponding generated 3D data, and historical conditional information to obtain historical data discrimination results; S1-3-10: Traverse all historical segmentation images and corresponding preprocessed historical patient diagnosis data, repeat the above steps to optimize and train the initial 3D reconstruction model; S1-3-11: Use the comprehensive loss function to generate historical loss values during the optimization training process. If the historical loss value is lower than the loss value threshold, output the final 3D reconstruction model; otherwise, continue with the optimization training; The 3D reconstruction model can utilize the patient's diagnosis data and segmentation images to generate realistic 3D data, has broad application potential in the fields of medical imaging, surgical planning, disease diagnosis, etc., and can improve the quality and efficiency of medical services; S2: Collect real-time medical image data and corresponding real-time patient diagnosis data, and use an image segmentation model to perform image segmentation on the real-time medical image data to obtain real-time segmentation images, including the following steps: S2-1: Collect real-time medical image data and corresponding real-time patient diagnosis data, and input the real-time medical image data into the image segmentation model; S2-2: Use the graph feature extraction module of the image segmentation model to extract several real-time feature maps of the real-time medical image data; S2-3: Use the feature fusion module of the image segmentation model to perform feature fusion on several real-time feature maps to obtain real-time fusion features; S2-4: Use the image segmentation module of the image segmentation model to perform image segmentation on the real-time fusion features to obtain real-time segmentation images; The image segmentation module uses the fused features to perform the segmentation task and outputs segmentation masks that mark the regions of interest in the image; S3: Use the 3D reconstruction model to perform 3D reconstruction on the real-time segmentation images and real-time patient diagnosis data to obtain the real-time 3D model of the real-time medical image data, including the following steps: S3-1: Input the real-time segmentation images and real-time patient diagnosis data into the 3D reconstruction model, and extract the real-time image features of the real-time segmentation images and the real-time data features of the real-time patient diagnosis data; Feature extraction is the basis for understanding the input data, provides the necessary information for 3D reconstruction, and ensures that the reconstruction process can take into account the details of the image and the specific diagnosis information of the patient; S3-2: Use the conditional information embedder of the 3D reconstruction model to perform conditional embedding on the real-time image features and real-time data features to obtain real-time conditional information embedded features; By embedding conditional information, the generator can be guided to more accurately generate three-dimensional data corresponding to specific diagnoses and image features, improving the pertinence and accuracy of reconstruction; S3-3: Use the generator of the three-dimensional reconstruction model to embed features according to real-time conditional information and generate three-dimensional data to obtain real-time three-dimensional data; Generating three-dimensional data is the core of the reconstruction process. Restoring the three-dimensional structure from two-dimensional images is crucial for understanding complex anatomical structures and pathological conditions; S3-4: Perform three-dimensional reconstruction based on the real-time three-dimensional data to obtain a real-time three-dimensional model of the real-time medical image data.

[0021] Embodiment 2: As Figure 2 shown, this embodiment provides a three-dimensional modeling system for medical image data to implement the three-dimensional modeling method. The system includes a model construction unit, an image segmentation unit, and a three-dimensional reconstruction unit connected in sequence; The model construction unit is used to construct an image segmentation model and a three-dimensional reconstruction model using a deep learning algorithm based on a number of historical medical image data and corresponding historical patient diagnosis data; The image segmentation unit is used to collect real-time medical image data and corresponding real-time patient diagnosis data, and use the image segmentation model to perform image segmentation on the real-time medical image data to obtain real-time segmented images; The three-dimensional reconstruction unit is used to perform three-dimensional reconstruction on the real-time segmented images and real-time patient diagnosis data using the three-dimensional reconstruction model to obtain a real-time three-dimensional model of the real-time medical image data.

[0022] The present invention discloses a three-dimensional modeling method and system for medical image data. The constructed image segmentation model can more accurately segment complex medical image data, reduce the modeling error caused by inaccurate segmentation, and improve the segmentation accuracy; the constructed three-dimensional reconstruction model combines real-time segmented images and real-time patient diagnosis data for three-dimensional reconstruction, can more realistically restore the anatomical structure, and reduces the distortion problem in the reconstruction process; by segmenting two-dimensional images and restoring the three-dimensional structure from the segmented images, the processing speed and computational efficiency are improved, enabling three-dimensional modeling to be completed quickly, especially suitable for clinical scenarios that require rapid response; automated image segmentation and three-dimensional reconstruction are performed, with a high degree of automation, reducing manual intervention and lowering the error rate.

[0023] The specific embodiments described above further elaborate on the objective, technical solution and beneficial effects of the present invention. It should be understood that the above description is only the specific embodiments of the present invention and is not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A three-dimensional modeling method for medical image data, characterized in that: The steps include: Based on several historical medical imaging data and corresponding historical patient diagnosis data, a deep learning algorithm is used to construct an image segmentation model and a 3D reconstruction model; Collect real-time medical image data and corresponding real-time patient diagnosis data, and use an image segmentation model to perform image segmentation on the real-time medical image data to obtain a real-time segmented image; The real-time segmented image and the real-time patient diagnosis data are three-dimensionally reconstructed using the three-dimensional reconstruction model to obtain a real-time three-dimensional model of the real-time medical imaging data.

2. The three-dimensional modeling method of medical image data according to claim 1, characterized in that: Based on several historical medical imaging data and corresponding historical patient diagnosis data, a deep learning algorithm is used to construct an image segmentation model and a 3D reconstruction model, including the following steps: Collecting a number of historical medical image data and corresponding historical patient diagnosis data, and preprocessing them to obtain a number of preprocessed historical medical image data and a number of preprocessed historical patient diagnosis data; Based on some pre-processed historical medical imaging data, a deep learning algorithm is used to build an image segmentation model and generate some historical segmentation images; Based on several historical segmentation images and the corresponding pre-processed historical patient diagnosis data, a three-dimensional reconstruction model is constructed using a deep learning algorithm.

3. The three-dimensional modeling method of medical image data according to claim 2, characterized in that: The preprocessing of the historical medical image data includes data cleaning, Gaussian denoising, size normalization, image enhancement and labeling in sequence; The preprocessing of the historical patient diagnosis data includes data cleaning, Gaussian denoising and magnitude normalization performed in sequence.

4. The three-dimensional modeling method of medical image data according to claim 2, characterized in that: The image segmentation model is constructed based on the CNN-FPN-SAM-Med2D algorithm, and the image segmentation model includes a graph feature extraction module based on the CNN algorithm, a feature fusion module based on the FPN algorithm, and an image segmentation module constructed based on the SAM-Med2D algorithm, which are connected in sequence.

5. The three-dimensional modeling method of medical image data according to claim 4, characterized in that: Based on some pre-processed historical medical image data, a deep learning algorithm is used to build an image segmentation model and generate some historical segmentation images, including the following steps: Using the CNN algorithm, a basic network structure of an image segmentation model is constructed to obtain an initial graph feature extraction module; the basic network structure includes a plurality of convolutional layers and a plurality of pooling layers that are alternately connected in sequence; Add skip connections to extract feature maps of all convolutional layers in the basic network structure, and build a feature pyramid based on several feature maps using upsampling and lateral connections to obtain the initial feature fusion module; The initial image segmentation module is constructed using the SAM-Med2D algorithm, and the input end of the initial image segmentation module is connected to the output end of the initial feature fusion module; Integrate the initial graph feature extraction module, the initial feature fusion module and the initial image segmentation module to obtain an initial image segmentation model; Divide a number of pre-processed historical medical image data into a model training set and a model test set in a ratio of 7:3; Input the model training set, optimize the initial image segmentation model, obtain the optimized image segmentation model, and generate several historical segmentation images; The model test set is input to perform model testing on the optimized image segmentation model to obtain the model test accuracy. If the model test accuracy is greater than the accuracy threshold, the final image segmentation model is output.

6. The three-dimensional modeling method of medical image data according to claim 2, characterized in that: The three-dimensional reconstruction model is constructed based on the cGAN-MLP algorithm, and the three-dimensional reconstruction model includes a conditional information embedder and a conditional information processor both constructed based on the MLP algorithm, and a generator and a discriminator both constructed based on the RNN algorithm. The generator is connected to the conditional information embedder and the discriminator respectively, and the discriminator is connected to the conditional information processor.

7. The three-dimensional modeling method of medical image data according to claim 6, characterized in that: Based on several historical segmentation images and the corresponding pre-processed historical patient diagnosis data, a deep learning algorithm is used to construct a 3D reconstruction model, including the following steps: Setting corresponding real three-dimensional data for each historical segmented image and corresponding pre-processed historical patient diagnosis data; Use the cGAN algorithm to build the initial generator and the initial discriminator, and use the MLP algorithm to set the conditional information embedder for the initial generator and the conditional information processor for the initial discriminator; Integrate the initial generator, the initial discriminator, the conditional information embedder and the conditional information processor to obtain an initial 3D reconstruction model; Combine the first loss function of the initial generator and the second loss function of the initial discriminator to obtain a comprehensive loss function; Extracting historical image features of historical segmented images and historical data features of preprocessed historical patient diagnosis data; Use the conditional information embedder to conditionally embed historical image features and historical data features to obtain historical conditional information embedding features; Using the initial generator, embedding features according to historical condition information, generating three-dimensional data, and obtaining generated three-dimensional data; Using a condition information processor, condition information processing is performed according to the real three-dimensional data and the corresponding generated three-dimensional data to obtain historical condition information; Using the initial discriminator, data discrimination is performed based on the real three-dimensional data, the corresponding generated three-dimensional data, and the historical condition information to obtain the historical data discrimination result; Traverse all historical segmentation images and corresponding pre-processed historical patient diagnosis data, repeat the above steps, and optimize the initial 3D reconstruction model; Use the comprehensive loss function to generate the historical loss value of the optimization training process. If the historical loss value is lower than the loss value threshold, the final 3D reconstruction model is output; otherwise, the optimization training continues.

8. The three-dimensional modeling method of medical image data according to claim 5, characterized in that: Collecting real-time medical image data and corresponding real-time patient diagnosis data, and using an image segmentation model to perform image segmentation on the real-time medical image data to obtain a real-time segmented image, includes the following steps: Collecting real-time medical imaging data and corresponding real-time patient diagnostic data, and inputting the real-time medical imaging data into an image segmentation model; Using the graph feature extraction module of the image segmentation model, extracting several real-time feature graphs of real-time medical imaging data; Use the feature fusion module of the image segmentation model to fuse several real-time feature maps to obtain real-time fusion features; The image segmentation module of the image segmentation model is used to perform image segmentation on the real-time fusion features to obtain a real-time segmented image.

9. The three-dimensional modeling method of medical image data according to claim 7, characterized in that: Using the three-dimensional reconstruction model, the real-time segmented image and the real-time patient diagnosis data are three-dimensionally reconstructed to obtain a real-time three-dimensional model of the real-time medical imaging data, including the following steps: inputting the real-time segmentation image and the real-time patient diagnosis data into the three-dimensional reconstruction model, and extracting the real-time image features of the real-time segmentation image and the real-time data features of the real-time patient diagnosis data; Using the conditional information embedder of the 3D reconstruction model, conditionally embed the real-time image features and the real-time data features to obtain the real-time conditional information embedding features; Using a generator of a three-dimensional reconstruction model, embedding features according to real-time condition information, generating three-dimensional data, and obtaining real-time three-dimensional data; Based on the real-time three-dimensional data, three-dimensional reconstruction is performed to obtain a real-time three-dimensional model of the real-time medical imaging data.

10. A three-dimensional modeling system for medical image data, used to implement the three-dimensional modeling method according to any one of claims 1 to 9, characterized in that: The system comprises a model building unit, an image segmentation unit and a three-dimensional reconstruction unit which are connected in sequence.

Images