Three-dimensional CT image fusion prediction method and system, medium and product
By adopting the fusion prediction method of residual neural network and attention mechanism in three-dimensional CT image processing, the problem of insufficient accuracy and generalization ability of three-dimensional CT image data analysis in the prior art is solved, and high-accurate image prediction is achieved, reducing the need for manual labeling and multiple detections, and improving diagnostic efficiency.
Patent Information
- Application Number
- CN202411737605.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-29
- Publication Date
- 2025-03-25
AI Technical Summary
The prior art processes three-dimensional CT image data with insufficient accuracy and generalization capabilities, and relies on manual annotation and multiple tests, which increases the burden on doctors and the radiation risk of patients.
A three-dimensional CT image fusion prediction method based on residual neural network and attention mechanism is adopted. By extracting the original features of two three-dimensional CT image data, feature fusion and prediction are performed, and predicting images of the third three-dimensional CT are generated, reducing manual annotation and multiple detections.
It improves the accuracy and generalization ability of medical imaging processing, reduces the need for manual annotation and multiple tests by patients, reduces the burden on doctors and patients, and improves diagnostic efficiency.
Smart Images
Figure CN119693246A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a three-dimensional CT image fusion prediction method, system, medium and product, and belongs to the technical field of machine learning image processing. Background Art
[0002] In recent years, with the development of medical imaging technology, computed tomography (CT) has become an important tool for diagnosing and treating cancer. Taking lung cancer as an example, doctors can clearly observe the structure and lesions of the lungs through CT images, so as to more accurately diagnose and evaluate the condition of lung cancer patients. However, due to the huge and complex amount of CT image data, most traditional image processing methods first slice the image into two dimensions and manually annotate it by doctors. However, this processing method is subject to the dual challenges of insufficient spatial information of the source image and difficulty in ensuring the accuracy of manual processing. In recent years, deep learning technology has attracted widespread attention in academia and the medical community, and has achieved remarkable results in the field of image processing. By constructing a multi-layer neural network, the deep learning method can automatically extract and learn high-level features in the image, greatly improving the accuracy and robustness of image analysis and processing.
[0003] In the field of 3D medical image processing, a variety of feature extraction and training methods have emerged, including visual feature extractors using 3D convolution and Transformer-based self-supervised segmentation pre-training models. Among them, some studies have proposed methods that combine deep learning features with radiological features to predict the risk of specific diseases. In addition, some studies have used Vision Transformer to build a self-supervised framework Swin UNETR for extracting features from 3D medical images. However, it is difficult to fully capture and utilize the complex structural information in 3D medical images by relying solely on self-supervised segmentation tasks. To overcome this limitation, some studies have used pre-trained segmentation large models to perform domain fine-tuning in multi-scale targets in 3D medical images, thereby solving the problems of poor performance and unstable predictions. At the same time, these studies also used data augmentation and regularization methods to further improve the accuracy and generalization performance of the model.
[0004] The above method uses convolutional neural networks to process three-dimensional medical image data, solves the problems of accuracy and generalization ability of volume data analysis in the field of medical image processing, and explores the application of neural network models in three-dimensional medical image processing. However, there are still limitations and shortcomings. First of all, the problem is still concentrated on the same classification and segmentation tasks as two-dimensional images, which require manual labeling and place higher requirements on the doctor's ability to read images. In addition, the detection of CT images not only increases the time patients are exposed to radiation environments, but also brings psychological pressure to patients. Only processing and analyzing existing images cannot solve the problem of the burden of multiple tests that patients currently face during diagnosis and treatment.
[0005] The information disclosed in this background technology section is only intended to enhance the understanding of the overall background of the invention and should not be regarded as an acknowledgement or any form of suggestion that the information constitutes the prior art already known to ordinary technicians in this field. Summary of the invention
[0006] The technical problem to be solved by the present invention is: how to optimize the neural network model to process three-dimensional CT image data, improve the accuracy and generalization ability of volume data analysis, and explore the problem of reducing the dependence on manual labeling and the burden of multiple tests on patients.
[0007] In order to solve the above technical problems, the present invention is implemented by adopting the following technical solutions.
[0008] In a first aspect, the present invention provides a three-dimensional CT image fusion prediction method, comprising: Obtaining two three-dimensional CT image data of the patient's part to be tested at intervals of a preset inspection period; According to the two three-dimensional CT image data, based on the pre-trained three-dimensional CT image fusion prediction model, prediction is performed to obtain a predicted image of the third three-dimensional CT; The data processing process of the three-dimensional CT image fusion prediction model includes: Based on the residual neural network, all original features of the two three-dimensional CT image data are extracted; Using a fusion network based on an attention mechanism to fuse all the original features, a 3D CT image fusion feature is obtained; Using a prediction head including an attention module to predict the fusion features of the three-dimensional CT image, to obtain a third three-dimensional CT image feature map; Through transfer learning, upsampling is used to restore the third three-dimensional CT image feature map to obtain the third three-dimensional CT predicted image.
[0009] Furthermore, after acquiring two three-dimensional CT image data of the patient's to-be-detected part separated by a preset inspection period, the method further includes: Normalizing the two three-dimensional CT image data using a standard deviation and a mean value to obtain the two normalized three-dimensional CT image data; The Gaussian filtering method is used to remove noise from each transverse frame data in the normalized two-dimensional CT image data.
[0010] Furthermore, the residual neural network is used to extract all original features of the two 3D CT image data, including: Using a three-dimensional convolution kernel in each layer of the residual neural network, the spatial structure features and hierarchical features of the two three-dimensional CT image data for processing are extracted to obtain multi-scale information of the two three-dimensional CT image data; Using the upsampling process of the UNet structure, the multi-scale information of the two 3D CT image data is fused to obtain all the original features of the two 3D CT image data; All original features of the two 3D CT image data are output from the output layer of the residual neural network.
[0011] Furthermore, all the original features are fused using a fusion network based on an attention mechanism to obtain 3D CT image fusion features, including: An attention mechanism is introduced to perform weighted processing on all original features of the two 3D CT image data to obtain a weighted fused feature map; Use the Encoder-Decoder module to process the weighted fused feature map to obtain the structural and semantic associations between adjacent features; The attention mechanism is introduced again to make fusion decisions on the structural and semantic associations between the adjacent features to obtain the fusion features of the three-dimensional CT image.
[0012] Furthermore, the Encoder-Decoder module includes: The Encoder module is used to gradually extract the high-level feature representation in the weighted fused feature map to obtain the encoded feature map; The Decoder module is used to decode the encoded feature map back to the original feature space to obtain the structure and semantic associations between adjacent features.
[0013] Furthermore, the three-dimensional CT image fusion features are predicted using a prediction head including an attention module to obtain a third three-dimensional CT image feature map, including: One or more attention modules are embedded in the network architecture of the prediction head; Inputting the 3D CT image fusion features into a prediction head including an attention module; During the prediction process, the attention module dynamically adjusts the weights of different features or positions according to the content of the 3D CT image fusion features to obtain features processed by the attention module; The prediction head is used to process and predict the features processed by the attention module to obtain the third 3D CT image feature map.
[0014] Furthermore, by transfer learning, upsampling is used to restore the third three-dimensional CT image feature map to obtain a predicted image of the third three-dimensional CT, including: Based on the transfer learning strategy, the prior knowledge in the field of 2D image processing is transferred to 3D image processing; Using the three-dimensional CT image fusion features, training an image restoration network to learn to upsample the low-resolution three-dimensional CT image fusion features to high resolution while maintaining details and clarity, thereby obtaining a trained image restoration network, wherein the image restoration network includes an upsampling module and prior knowledge in the field of two-dimensional image processing; The third three-dimensional CT image feature map is input into a trained image restoration network to obtain a predicted image of the third three-dimensional CT.
[0015] In a second aspect, the present invention provides a three-dimensional CT image fusion prediction system, comprising the following modules: A data acquisition module, used to acquire two three-dimensional CT image data of the patient's part to be detected at intervals of a preset inspection cycle; An impact prediction module, used for making predictions based on the two three-dimensional CT image data and a pre-trained three-dimensional CT image fusion prediction model; An image output module is used to obtain the predicted image of the third three-dimensional CT; The data processing process of the three-dimensional CT image fusion prediction model includes: Based on the residual neural network, all original features of the two three-dimensional CT image data are extracted; Using a fusion network based on an attention mechanism to fuse all the original features, a 3D CT image fusion feature is obtained; Using a prediction head including an attention module to predict the fusion features of the three-dimensional CT image, to obtain a third three-dimensional CT image feature map; Through transfer learning, upsampling is used to restore the third three-dimensional CT image feature map to obtain the third three-dimensional CT predicted image.
[0016] In a third aspect, the present invention provides a computer-readable storage medium having a computer program / instruction stored thereon, wherein the computer program / instruction, when executed by a processor, implements the steps of the method described in the first aspect.
[0017] In a fourth aspect, the present invention provides a computer program product, comprising a computer program / instruction, characterized in that when the computer program / instruction is executed by a processor, the steps of the method described in the first aspect are implemented.
[0018] Compared with the prior art, the present invention has the following beneficial effects: The present invention uses two three-dimensional CT image data of the patient's part to be detected extracted by the residual neural network at intervals of a preset inspection cycle and feature fusion performed by a fusion network using an attention mechanism to ensure the high accuracy of the predicted image, improve the accuracy and generalization ability of medical image processing, and help doctors obtain predictive image data without requiring patients to undergo multiple CT tests, thereby greatly reducing the burden of doctors in processing a large number of images of suspected patients and improving diagnostic efficiency. Doctors can focus more on the analysis of the predicted CT images, providing a more accurate visualization basis for preventive treatment, while also reducing the need for patients to undergo multiple CT image irradiations, thereby reducing the risk and pressure on patients.
[0019] The present invention obtains two 3D CT image data of the patient's part to be tested at intervals of a preset inspection cycle, and uses a pre-trained 3D CT image fusion prediction model for prediction to obtain the predicted image of the third 3D CT, thereby optimizing the neural network model to process 3D CT image data, improving the accuracy and generalization ability of volume data analysis, and reducing the dependence on manual annotation and the burden of multiple tests on patients. It provides a new method for the medical field to conduct prospective and comprehensive research and analysis on patients based on a small number of test results. It not only helps to improve the quality and efficiency of medical services, but also provides new ideas and technical support for future medical research. BRIEF DESCRIPTION OF THE DRAWINGS
[0020] Figure 1 FIG. 4 is a flow chart of a three-dimensional CT image fusion prediction method provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0021] The technical solution of the present invention is described in detail below through the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations on the technical solution of the present invention. The embodiments of the present invention and the technical features in the embodiments may be combined with each other unless there is a conflict.
[0022] The term "and / or" is only a description of the association relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " generally indicates that the related objects are in an "or" relationship. Example
[0023] like Figure 1 As shown, this embodiment introduces a three-dimensional CT image fusion prediction method, including: Step 1: Obtain two three-dimensional CT image data of the patient's part to be tested at intervals of a preset inspection cycle.
[0024] By comparing two 3D CT image data with a preset examination period, we can preliminarily observe the changes in the patient's part to be tested, provide doctors with a preliminary reference for the patient's condition, and assist in diagnosis. At the same time, it can also ensure the continuity and consistency of the data, which helps the 3D CT image fusion prediction model learn the changing patterns within the time interval.
[0025] After obtaining two three-dimensional CT image data of the patient's part to be detected at intervals of a preset inspection period, the method further includes: Normalizing the two three-dimensional CT image data using a standard deviation and a mean value to obtain the two normalized three-dimensional CT image data; The Gaussian filtering method is used to remove noise from each transverse frame data in the normalized two-dimensional CT image data.
[0026] Step 2: According to the two three-dimensional CT image data, based on the pre-trained three-dimensional CT image fusion prediction model, prediction is performed to obtain a predicted image of the third three-dimensional CT.
[0027] Using the pre-trained model, the two 3D CT image data are deeply analyzed and processed, and the features of the two 3D CT image data are integrated to predict the third 3D CT image that the patient may obtain in the future. This improves the accuracy of the prediction, reduces the time doctors spend processing large amounts of image data, and improves work efficiency.
[0028] The data processing process of the three-dimensional CT image fusion prediction model includes: Step 2.1: Based on the residual neural network, all original features of the two 3D CT image data are extracted.
[0029] By utilizing the deep learning capability of the residual neural network, rich feature information, including anatomical structure, pathological changes, etc., is extracted from the two 3D CT image data. Based on the residual neural network, all the original features of the two 3D CT image data are extracted to improve the accuracy and efficiency of feature extraction.
[0030] In this embodiment, all original features of the two 3D CT image data are extracted based on the residual neural network, including: Using a three-dimensional convolution kernel in each layer of the residual neural network, the spatial structure features and hierarchical features of the two three-dimensional CT image data for processing are extracted to obtain multi-scale information of the two three-dimensional CT image data; Using the upsampling process of the UNet structure, the multi-scale information of the two 3D CT image data is fused to obtain all the original features of the two 3D CT image data; All original features of the two 3D CT image data are output from the output layer of the residual neural network.
[0031] Step 2.2: Use a fusion network based on an attention mechanism to fuse all the original features to obtain three-dimensional CT image fusion features.
[0032] Through the attention mechanism, all the original features of the two 3D CT image data are weighted and fused, which can highlight the important feature information and suppress the unimportant information, thereby improving the accuracy of the prediction, and realizing the deep fusion and optimization of the features, thereby improving the accuracy and reliability of the predicted images.
[0033] In this embodiment, all the original features are fused using a fusion network based on an attention mechanism to obtain 3D CT image fusion features, including: An attention mechanism is introduced to perform weighted processing on all original features of the two 3D CT image data to obtain a weighted fused feature map; Using an Encoder-Decoder module to process the weighted fused feature map to obtain the structural and semantic associations between adjacent features, wherein the Encoder-Decoder module includes an Encoder module and a Decoder module; The Encoder module is used to gradually extract high-level feature representations from the weighted fused feature map to obtain a coded feature map; The Decoder module is used to decode the encoded feature map back to the original feature space to obtain the structure and semantic association between adjacent features.
[0034] The attention mechanism is introduced again to make fusion decisions on the structural and semantic associations between the adjacent features to obtain the fusion features of the three-dimensional CT image.
[0035] Step 2.3: Use the prediction head including the attention module to predict the fusion features of the three-dimensional CT image to obtain a third three-dimensional CT image feature map.
[0036] Using the prediction capability of the prediction head, the fused features are converted into feature maps of the third 3D CT image.
[0037] The feature map of the third 3D CT image contains key information of the predicted image, such as anatomical structure and pathological changes, etc. The conversion from features to predicted images is realized, providing a basis for subsequent upsampling and restoration.
[0038] In this embodiment, the prediction head including the attention module is used to predict the 3D CT image fusion features to obtain a third 3D CT image feature map, including: One or more attention modules are embedded in the network architecture of the prediction head; Inputting the 3D CT image fusion features into a prediction head including an attention module; During the prediction process, the attention module dynamically adjusts the weights of different features or positions according to the content of the 3D CT image fusion features to obtain features processed by the attention module; The prediction head is used to process and predict the features processed by the attention module to obtain the third 3D CT image feature map.
[0039] Step 2.3: Through transfer learning, upsampling is used to restore the third three-dimensional CT image feature map to obtain the predicted image of the third three-dimensional CT.
[0040] By using the ability of transfer learning, the third 3D CT image feature map is restored to a high-quality third 3D CT predicted image. The predicted image of the third 3D CT has a high degree of similarity with the real image and can be used to assist doctors in diagnosis and treatment, providing doctors with an intuitive and accurate basis for disease assessment and treatment.
[0041] In this embodiment, the third three-dimensional CT image feature map is restored by upsampling through transfer learning to obtain a predicted image of the third three-dimensional CT, including: Based on the transfer learning strategy, the prior knowledge in the field of 2D image processing is transferred to 3D image processing; Using the three-dimensional CT image fusion features, training an image restoration network to learn to upsample the low-resolution three-dimensional CT image fusion features to high resolution while maintaining details and clarity, thereby obtaining a trained image restoration network, wherein the image restoration network includes an upsampling module and prior knowledge in the field of two-dimensional image processing; The third three-dimensional CT image feature map is input into a trained image restoration network to obtain a predicted image of the third three-dimensional CT.
[0042] The predicted image of the third three-dimensional CT scan obtained by the method provided by the present invention is only a basis for reference in diagnosis and treatment, and the diagnosis and treatment results cannot be directly obtained. Example
[0043] Based on the same inventive concept as Example 1, this example introduces a three-dimensional CT image fusion prediction system, including the following modules: A data acquisition module, used to acquire two three-dimensional CT image data of the patient's part to be detected at intervals of a preset inspection cycle; An impact prediction module, used for making predictions based on the two three-dimensional CT image data and a pre-trained three-dimensional CT image fusion prediction model; An image output module is used to obtain the predicted image of the third three-dimensional CT; The data processing process of the three-dimensional CT image fusion prediction model includes: Based on the residual neural network, all original features of the two three-dimensional CT image data are extracted; Using a fusion network based on an attention mechanism to fuse all the original features, a 3D CT image fusion feature is obtained; Using a prediction head including an attention module to predict the fusion features of the three-dimensional CT image, to obtain a third three-dimensional CT image feature map; Through transfer learning, upsampling is used to restore the third three-dimensional CT image feature map to obtain the third three-dimensional CT predicted image.
[0044] The specific functional implementation of each of the above modules can be found in the relevant contents of the method in Example 1 and will not be elaborated here. Example
[0045] Based on the same inventive concept as other embodiments, this embodiment introduces a computer-readable storage medium on which computer instructions are stored, characterized in that the computer instructions implement the steps of the method in the above-mentioned embodiment 1 when executed by a processor. Example
[0046] Based on the same inventive concept as other embodiments, this embodiment introduces a computer program product, including computer instructions, characterized in that the computer instructions implement the steps of the method in the above-mentioned embodiment 1 when executed by a processor.
[0047] In summary, the present invention uses the residual neural network to extract the two three-dimensional CT image data of the patient's part to be detected at intervals of a preset inspection cycle and the feature fusion performed by the fusion network using the attention mechanism, which ensures the high accuracy of the predicted image and improves the accuracy and generalization ability of medical image processing. It can help doctors obtain predictive image data without requiring patients to undergo multiple CT tests, thereby greatly reducing the burden of doctors processing a large number of images of suspected patients and improving diagnostic efficiency. Doctors can focus more on the analysis of the predicted CT images, providing a more accurate visualization basis for preventive treatment, while also reducing the need for patients to undergo multiple CT image irradiations, thereby reducing the risk and pressure on patients.
[0048] The present invention obtains two 3D CT image data of the patient's part to be tested at intervals of a preset inspection cycle, and uses a pre-trained 3D CT image fusion prediction model for prediction to obtain the predicted image of the third 3D CT, thereby optimizing the neural network model to process 3D CT image data, improving the accuracy and generalization ability of volume data analysis, and reducing the dependence on manual annotation and the burden of multiple tests on patients. It provides a new method for the medical field to conduct prospective and comprehensive research and analysis on patients based on a small number of test results. It not only helps to improve the quality and efficiency of medical services, but also provides new ideas and technical support for future medical research.
[0049] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented 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.
[0050] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0051] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.
[0052] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0053] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the enlightenment of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present invention and the claims, which all fall within the protection of the present invention.
Claims
1. A three-dimensional CT image fusion prediction method, characterized in that: include: Obtaining two three-dimensional CT image data of the patient's part to be tested at intervals of a preset inspection period; According to the two three-dimensional CT image data, based on the pre-trained three-dimensional CT image fusion prediction model, prediction is performed to obtain a predicted image of the third three-dimensional CT; The data processing process of the three-dimensional CT image fusion prediction model includes: Based on the residual neural network, all original features of the two three-dimensional CT image data are extracted; Using a fusion network based on an attention mechanism to fuse all the original features, a 3D CT image fusion feature is obtained; Using a prediction head including an attention module to predict the fusion features of the three-dimensional CT image, to obtain a third three-dimensional CT image feature map; Through transfer learning, upsampling is used to restore the third three-dimensional CT image feature map to obtain the third three-dimensional CT predicted image.
2. The three-dimensional CT image fusion prediction method according to claim 1, characterized in that: After obtaining two three-dimensional CT image data of the patient's part to be detected at intervals of a preset inspection period, the method further includes: Normalizing the two three-dimensional CT image data using a standard deviation and a mean value to obtain the two normalized three-dimensional CT image data; The Gaussian filtering method is used to remove noise from each transverse frame data in the normalized two-dimensional CT image data.
3. The three-dimensional CT image fusion prediction method according to claim 1, characterized in that: The residual neural network is used to extract all original features of the two three-dimensional CT image data, including: Using a three-dimensional convolution kernel in each layer of the residual neural network, the spatial structure features and hierarchical features of the two three-dimensional CT image data for processing are extracted to obtain multi-scale information of the two three-dimensional CT image data; Using the upsampling process of the UNet structure, the multi-scale information of the two 3D CT image data is fused to obtain all the original features of the two 3D CT image data; All original features of the two 3D CT image data are output from the output layer of the residual neural network.
4. The three-dimensional CT image fusion prediction method according to claim 1, characterized in that: All the original features are fused using a fusion network based on an attention mechanism to obtain 3D CT image fusion features, including: An attention mechanism is introduced to perform weighted processing on all original features of the two 3D CT image data to obtain a weighted fused feature map; Use the Encoder-Decoder module to process the weighted fused feature map to obtain the structural and semantic associations between adjacent features; The attention mechanism is introduced again to make fusion decisions on the structural and semantic associations between the adjacent features to obtain the fusion features of the three-dimensional CT image.
5. The three-dimensional CT image fusion prediction method according to claim 4, characterized in that: The Encoder-Decoder module includes: The Encoder module is used to gradually extract the high-level feature representation in the weighted fused feature map to obtain the encoded feature map; The Decoder module is used to decode the encoded feature map back to the original feature space to obtain the structure and semantic associations between adjacent features.
6. The three-dimensional CT image fusion prediction method according to claim 1, characterized in that: The three-dimensional CT image fusion features are predicted using a prediction head including an attention module to obtain a third three-dimensional CT image feature map, including: One or more attention modules are embedded in the network architecture of the prediction head; Inputting the three-dimensional CT image fusion features into a prediction head including an attention module; During the prediction process, the attention module dynamically adjusts the weights of different features or positions according to the content of the 3D CT image fusion features to obtain features processed by the attention module; The prediction head is used to process and predict the features processed by the attention module to obtain the third 3D CT image feature map.
7. The three-dimensional CT image fusion prediction method according to claim 1, characterized in that: The third three-dimensional CT image feature map is restored by upsampling through transfer learning to obtain a predicted image of the third three-dimensional CT, including: Based on the transfer learning strategy, the prior knowledge in the field of 2D image processing is transferred to 3D image processing; Using the three-dimensional CT image fusion features, training an image restoration network to learn to upsample the low-resolution three-dimensional CT image fusion features to high resolution while maintaining details and clarity, thereby obtaining a trained image restoration network, wherein the image restoration network includes an upsampling module and prior knowledge in the field of two-dimensional image processing; The third three-dimensional CT image feature map is input into a trained image restoration network to obtain a predicted image of the third three-dimensional CT.
8. A three-dimensional CT image fusion prediction system, characterized in that: Includes the following modules: A data acquisition module, used to acquire two three-dimensional CT image data of the patient's part to be detected at intervals of a preset inspection cycle; An impact prediction module, used for making predictions based on the two three-dimensional CT image data and a pre-trained three-dimensional CT image fusion prediction model; An image output module is used to obtain the predicted image of the third three-dimensional CT; The data processing process of the three-dimensional CT image fusion prediction model includes: Based on the residual neural network, all original features of the two three-dimensional CT image data are extracted; Using a fusion network based on an attention mechanism to fuse all the original features, a 3D CT image fusion feature is obtained; Using a prediction head including an attention module to predict the fusion features of the three-dimensional CT image, to obtain a third three-dimensional CT image feature map; Through transfer learning, upsampling is used to restore the third three-dimensional CT image feature map to obtain the third three-dimensional CT predicted image.
9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.
10. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.