Cardiac three-dimensional reconstruction method, system and program product based on improved GRAF algorithm

Abstract

The application discloses a kind of based on improved GRAF algorithm heart three-dimensional reconstruction method, system and program product.The three-dimensional reconstruction method includes acquisition and pre-processing generates different patients, different contrast types and normal and abnormal heart anatomical structure digital reconstruction radiogram;Improved GRAF algorithm model is constructed and trained;By inputting a small amount of CT image data to the trained GRAF algorithm model, continuous image frame is rendered, and is used for three-dimensional reconstruction and the like steps.The improved GRAF algorithm model obtained by the method based on data enhancement transformation and the combination of multiple training heads can be trained under limited CT image input, to realize the accurate reconstruction of heart three-dimensional structure.

Description

Technical Field

[0001] This invention relates to the field of robot-assisted surgical systems, and more specifically, to a method, system, and program product for three-dimensional cardiac reconstruction based on an improved GRAF algorithm. Background Technology

[0002] In the application of robotic cardiac surgery, precise 3D reconstruction of the heart before surgery is a crucial step. Traditional techniques typically rely on computed tomography (CT) scans to obtain detailed structural data of the heart. CT scans acquire a series of 2D images, which are then used to reconstruct a 3D model through image processing algorithms. However, traditional 3D reconstruction methods depend on continuous CT image input, which is not only time-consuming but also carries a high risk of radiation exposure for patients, posing a potential threat to their health. Therefore, it is essential to find a high-precision method for reconstructing 3D models of the heart using low-radiation, non-continuous CT images.

[0003] Generative Radiance Field (GRAF) is a generative model of a radiation field trained using an adversarial framework. It achieves high-resolution 3D-aware image synthesis by introducing a multi-scale patch-based discriminator (see K Schwarz, Y Liao, M Niemeyer and A Geiger, “GRAF: Generative Radiance Fields for 3D-Aware Image Synthesis”, NeurIPS 2020). However, medical imaging systems differ significantly from conventional radiographic equipment, and the medical field demands much higher accuracy from generative models. These factors pose challenges to traditional neural radiation field models and their variants when applied to 3D reconstruction of CT images. To address these issues, it is necessary to develop more advanced cardiac 3D reconstruction methods based on improved GRAF algorithms to overcome these limitations and further improve the safety and effectiveness of preoperative cardiac 3D reconstruction. Summary of the Invention

[0004] This invention addresses the problem of reducing CT data input in CT reconstruction of the heart by proposing a method, system, and program product for three-dimensional reconstruction of the heart based on an improved GRAF algorithm. This method enables accurate reconstruction of the three-dimensional structure of the heart with limited CT data input through advanced deep learning technology.

[0005] To achieve the above objectives, in a first aspect, the present invention provides a method for three-dimensional reconstruction of the heart based on an improved GRAF algorithm, characterized in that it includes:

[0006] Step S1: Acquire information from cardiac CT scan data: Acquire images containing different patients, different contrast types, and normal and abnormal cardiac anatomy structures;

[0007] Step S2: Perform comprehensive and data processing on the original CT image data, including denoising, standardization and digital reconstruction of radiographic images, and perform several data enhancement transformations on the processed digital reconstruction radiographic images to construct training and test sets.

[0008] Step S3: Construct and train the improved GRAF algorithm model: The improved GRAF algorithm model includes a generator network and a discriminator network based on the generative radiation field principle; the generator network samples rays from the digitally reconstructed radiation image obtained in step S2 through ray sampling, and uses localization encoding to map the 3D position and viewing direction to a high-dimensional feature space, further utilizing shape encoding and appearance encoding to determine the shape and appearance of objects, then predicts the density and color of any point in the scene through a fully connected layer, and finally synthesizes continuous image frames through volume rendering technology; the discriminator network applies multiple discriminators corresponding to different data augmentation transformations, each discriminator focusing on extracting features under a specific data augmentation transformation and making predictions, and optimizes the model parameters by combining the output of the discriminators and the results of data augmentation through a comprehensive loss function; training iterates until the loss function value of the discriminator network tends to stabilize or reaches the predetermined number of iterations;

[0009] Step S4: Perform three-dimensional reconstruction of the heart structure using the trained GRAF algorithm model and discontinuous cardiac CT images.

[0010] This invention trains an improved GRAF algorithm model based on data augmentation transformation feature discrimination using specific types of discontinuous CT image data of specific patients, and obtains a method and system that can effectively generate continuous CT images from limited CT images to complete three-dimensional reconstruction of the heart. Therefore, it not only significantly reduces radiation dose and potential harm to the human body, but also improves the efficiency and accuracy of image reconstruction, making the entire reconstruction process more efficient and economical.

[0011] Furthermore, in step S1, patient data of different ages, genders, weights and ethnicities are collected, including X-ray images of normal and abnormal hearts. Different concentrations of iodine contrast agent are used during sampling to obtain images with different contrasts, so as to increase the generalization ability of the dataset. Good results can be achieved with a small amount of data for training.

[0012] Furthermore, in step S2, for each original CT image, a denoising algorithm is first used to reduce random noise in the image, then the image contrast is enhanced, and the image is standardized and segmented. Threshold segmentation, region growing, or deep learning methods are used to extract the heart region. The image segmentation results are then used to generate digitally reconstructed radiographic images (DRRs). Finally, the generated DRRs are resampled to have uniform resolution and size. Data augmentation techniques are applied to the sampled DRRs to obtain diverse DRRs. The enhanced DRRs are then organized into training and testing sets.

[0013] Furthermore, the data augmentation techniques include at least two of flipping, rotating, scaling, and translating to increase the diversity of DRRs.

[0014] Furthermore, the formula for calculating the total loss function of the generator and discriminator networks is as follows:

[0015]

[0016] In the above formula, L d To measure the discriminant head's accuracy against real data P p and its enhanced version P p,k Loss on predicted discrepancies, L r For generator G θ and discriminator The adversarial loss between them, k is the discriminant number, n is the number of discriminants, λ is used to measure the weight between different losses, and θ is the generator random weight or pre-trained weight parameter; For the k-th discriminator parameter.

[0017] Furthermore, in step S4, firstly, a small number or a single cardiac CT scan image is selected as the input to the improved GRAF algorithm model, and the input image covers the required anatomical structures and contrast types; secondly, the input CT image is preprocessed to match the data distribution during training; thirdly, the trained generator network is loaded again, and the generator is used to generate continuous image frames from the latent space vector; finally, a 3D reconstruction algorithm is used to reconstruct the 3D image of the heart from the continuous image frames.

[0018] Secondly, the present invention provides a cardiac three-dimensional reconstruction system based on an improved GRAF algorithm, characterized in that it includes the following components for implementing the cardiac three-dimensional reconstruction method as described above:

[0019] The image acquisition module acquires images of different patients, different contrast types, and normal and abnormal cardiac anatomy.

[0020] The X-ray image integration and data processing module performs integration and data processing on the raw CT data, including denoising, standardization, and digital reconstruction of radiographic images.

[0021] The improved GRAF algorithm model includes a generator module and a discriminator module based on the generative radiation field principle. The generator module contains ray sampling, scene representation, density prediction network, color prediction network, and volume rendering sub-modules. The discriminator module includes image patch extraction, a discriminator backbone network, and multiple data augmentation transformation discriminators. The model parameters are optimized by combining the output of the discriminators and the results of data augmentation through a comprehensive loss function. The GRAF algorithm model with optimized model parameters generates continuous image frames from a small number of images or a single view.

[0022] The 3D reconstruction module uses a 3D reconstruction algorithm to reconstruct a 3D image of the heart from consecutive image frames.

[0023] Furthermore, the data augmentation transformation discrimination head includes a flip discrimination head, a rotation discrimination head, and a scaling discrimination head.

[0024] Furthermore, the formula for calculating the total loss function of the generator and discriminator networks is as follows:

[0025]

[0026] In the above formula, L d To measure the discriminant head's accuracy against real data P p and its enhanced version P p,k Loss on predicted discrepancies, L r For generator G θ and discriminator The adversarial loss between them, k is the discriminant number, n is the number of discriminants, λ is used to measure the weight between different losses, and θ is the generator random weight or pre-trained weight parameter; For the k-th discriminator parameter.

[0027] Finally, the present invention provides a computer program product, characterized in that, when the computer program product is run on a computer, it causes the computer to perform the cardiac three-dimensional reconstruction method based on the improved GRAF algorithm as described above.

[0028] Compared with the prior art, the present invention has the following technical effects:

[0029] (1) This invention introduces the principle of generative radiation field into the problem of cardiac CT reconstruction. It trains an improved GRAF algorithm model based on feature discrimination of data enhancement transformation through non-continuous CT image data of specific patients and specific types. In the model application stage, continuous frame images can be obtained through a small number or single CT images, which effectively solves the disadvantage of needing long-term exposure to radiation in the process of CT image reconstruction.

[0030] (2) The improved GRAF algorithm model of the present invention provides a new discriminator structure, which improves the discriminator training performance in the case of insufficient data by combining data augmentation methods and multiple training heads.

[0031] (3) The image reconstruction method used in this invention is based on the principle of generative radiation field, which has advantages over traditional reconstruction methods such as multi-view consistency, high efficiency and high quality of image reconstruction. Attached Figure Description

[0032] Figure 1 This is a flowchart illustrating the execution of a cardiac three-dimensional reconstruction method according to an embodiment of the present invention.

[0033] Figure 2 This is a framework diagram of an improved GRAF algorithm model in one embodiment of the present invention.

[0034] Figure 3 This is a framework diagram of a cardiac three-dimensional reconstruction system according to an embodiment of the present invention. Detailed Implementation

[0035] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but these are not intended to limit the scope of the invention.

[0036] In the following detailed description, numerous specific details are set forth to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that well-known algorithms are not shown in detail to avoid obscuring the spirit of the invention.

[0037] Furthermore, the execution order of actions, steps, etc. in the apparatus and methods shown in the claims, specification, and drawings can be implemented in any order, unless a specific order is explicitly specified, and as long as the output of the preceding processing is not used in the subsequent processing.

[0038] Example 1

[0039] See Figure 1 This embodiment provides a method for three-dimensional reconstruction of the heart based on an improved GRAF algorithm, including the following steps:

[0040] Step S1 involves acquiring information from cardiac CT scan data to construct a training dataset for model training. This dataset contains images of different patients, different contrast types, and normal and abnormal anatomical structures. More specifically, it includes the following sub-steps:

[0041] a1: Choose a multi-detector CT scanner suitable for cardiac imaging, ensuring it can provide high-resolution and high-quality images, such as a high-end CT scanner like the GE Healthcare Discovery CT750 HD, and ensure the equipment is properly calibrated and maintained to guarantee image quality.

[0042] a2: Select an appropriate tube voltage to obtain sufficient X-ray penetration. For adult cardiac CT, 100-140kVp is usually selected. Adjust the tube current to ensure image brightness and clarity. Depending on the patient's size and the required image quality, select 200-400mA. Set the exposure time according to the required image quality and the patient's condition. Usually, a single rotation should be completed within 1 second. Select an appropriate slice thickness to obtain detailed anatomical information. A slice thickness of 0.5-1.0mm can be set.

[0043] A3: The patient is positioned on the scanning table. Adjust the scanner position to ensure the heart area is centered in the scanning field of view. Start the CT scan; the device will automatically rotate and acquire X-ray images.

[0044] a4: Collect patient data of different ages, sexes, weights, and ethnicities to increase the generalization ability of the dataset. Include X-ray images of normal and abnormal hearts, and consider using different concentrations of iodine contrast agents to obtain images with varying contrast. The model in this invention can achieve good results with a small amount of data; the collected patient data can be set to 10 groups.

[0045] Step S2 involves synthesizing and digitizing the collected raw X-ray images, including denoising, normalization, and generation of digitally reconstructed radiographic images (DRRs). Several data augmentation transformations are then applied to the processed DRRs to construct training and testing sets. More specifically, this includes the following sub-steps:

[0046] b1: A denoising algorithm is used for each original X-ray image to reduce random noise in the image, where I d This is the denoised image, I r This is the original image, (x,y) are pixel coordinates, (i x i y ) is the offset of the Gaussian kernel, and N is the kernel size:

[0047]

[0048] b2: Enhance the contrast of the denoised image to make anatomical structures clearer, using Contrast Limiting Adaptive Histogram Equalization (CLAHE) or other contrast enhancement techniques. e This is the enhanced image; α and β are parameters used to adjust the contrast.

[0049] I e (x,y)=α·I d (x,y)+β

[0050] b3: Standardize the contrast-enhanced image to ensure a consistent distribution of image intensity values, where I n It is a standardized image, I min and I max These are the minimum and maximum intensities of the enhanced image, respectively.

[0051]

[0052] b4: Segment the standardized image to extract the heart region using thresholding, region growing, or deep learning methods, where S is the segmented binary image and T is the threshold.

[0053]

[0054] b5: Generate digitally reconstructed radiographic images (DRRs) using the image segmentation results, I 3D (x, y, z) is 3D image data, representing the volume data of the segmented heart region. δ represents the path of the X-ray, θ is the direction of the ray, t is the position along the ray path, and z is the coordinate along the ray direction.

[0055]

[0056] b6: Resample the generated DRRs to ensure they have uniform resolution and size.

[0057] b7: Apply data augmentation techniques, such as flipping, rotating, scaling, and translating, to increase the diversity of DRRs.

[0058] b8: Organize the enhanced DRRs into training and testing sets, and perform validation to ensure the quality of the dataset.

[0059] Step S3: In this embodiment, an improved GRAF algorithm model is constructed. See [link to relevant documentation]. Figure 2 This model can be effectively trained on small datasets without relying on complex pose information or 3D supervision. This model will be able to reconstruct consecutive image frames from a small number of images or a single view, providing a new solution for the field of medical imaging. More specifically, it includes:

[0060] c1: A generator module based on the principle of neural radiation fields. This module includes sub-modules such as ray sampling, scene representation, density prediction network, color prediction network, and volume rendering.

[0061] c1.1: The first step of the generator is ray sampling, which involves sampling rays from the image. It involves sampling rays from the image, and the sampled X-ray beam is defined as follows, where u and s represent the image coordinates and the sampling step size, respectively:

[0062] v = (u, s)

[0063] To numerically integrate the radiation field, we sample N points along each ray r.

[0064] c1.2: The second step is scene representation, which uses localization encoding to map the 3D position x and view direction d to a higher-dimensional feature representation. This encoding is applied element-wise to all three components of x and d, as shown in the following formula.

[0065] φ(p)=(sin(20πp),cos(20πp),sin(21πp),cos(21πp),sin(22πp),cos(22πp),…)

[0066] 3D position encoding and d r The formula is as follows:

[0067]

[0068]

[0069] Apart from and d r There are two more potential additional conditions for latent encoding, shape code z s and appearance code Z a This is a set of latent variables sampled from a Gaussian distribution, used to determine the shape of objects in the model. These four codes form a radiation field, mapping the 3D position and viewing direction d to the relationship between c and σ:

[0070]

[0071] c1.3: Shape Embedding Network: This network uses fully connected layers as its representation, representing the encoded information of the object's shape. The formula is as follows:

[0072]

[0073] c1.4: Density Prediction Network (DDoS Logit): This network uses fully connected layers as its representation and is responsible for predicting the density at any point in the scene. The formula is as follows:

[0074]

[0075] c1.5: Color Prediction Network (color logit): This network uses fully connected layers as its representation and predicts color values ​​based on density prediction results and viewing direction, as shown in the following formula:

[0076]

[0077] c1.6: Volume rendering: Uses the alpha compositing method to synthesize color and density along ray r to generate the final pixel color C. r :

[0078]

[0079] Where Δ j r is the adjacent sampling point x j and x j+1 The distance between them, exp(-σ j Δ j r) calculated the light from x j To x j+1 The degree of attenuation along the path.

[0080] c2: Construct the discriminator module, which includes image patch extraction, discriminator backbone network, loss function, etc.

[0081] c2.1: Image Patch Extraction

[0082] P = sample(I, θ)

[0083] Where θ = (u, s) are parameters extracted from the distribution pθ, u is the image coordinates, s is the sampling step size, and sample represents bilinear interpolation sampling on image I.

[0084] c2.2: Discriminator backbone network, which uses a pre-trained convolutional neural network to extract basic features of the image, as shown in the following formula:

[0085]

[0086] G b This is the basic feature extraction network, typically a pre-trained convolutional neural network such as VGG or ResNet. I is the input image, and f is the feature representation output by the network. These features capture key visual information of the image and can be used by subsequent discriminators for more detailed analysis.

[0087] c2.3: Data Augmentation Transformation. This network applies different data augmentation transformations to increase data diversity. The formula is as follows:

[0088]

[0089] Where k represents different enhancement transformation types, the following examples are flip, rotation, and scaling enhancement transformations.

[0090] c2.4 Flip Discriminator Head: This discriminator head focuses on extracting features that have undergone flipping enhancement transformation and then making predictions. The formula is as follows, P... p,1 It is the prediction of the flip discriminant head:

[0091]

[0092] c2.5: Rotation Discriminator Head. This discriminator head focuses on extracting features after rotation enhancement transformation and making predictions. The formula is as follows, P... p,2 It is the prediction of the rotating discriminant head:

[0093]

[0094] c2.6: Scale Discriminator Head. This discriminator head focuses on extracting features that have undergone scaling and enhancement transformations, and then making predictions. The formula is as follows: P p,3 It is the prediction of the scaling discriminant head:

[0095]

[0096] c2.7: Loss function, which combines the output of each discriminator with the results of data augmentation to calculate the total loss. The formula is as follows:

[0097]

[0098] In the above formula, L d To measure the discriminant head's accuracy against real data P p and its enhanced version P p,k Loss on predicted discrepancies, L r For generator G θ and discriminator The adversarial loss between them, k is the discriminant number, n is the number of discriminants, λ is used to measure the weight between different losses, and θ is the generator random weight or pre-trained weight parameter; For the k-th discriminator parameter.

[0099] c3: The GRAF model is trained using a preprocessed dataset. During training, the model learns to decouple 3D shape and view orientation from 2D images to generate high-fidelity images.

[0100] c3.1: Initialize the generator network G θ Random weights or pre-trained weights are typically used. Initialize the discriminator network. Random weights or pre-trained weights are typically used. Ensure that the network parameters are initialized reasonably, avoiding weight values ​​that are too large or too small.

[0101] c3.2: Select the Adam optimizer, an adaptive learning rate optimization algorithm suitable for most deep learning tasks. Set the initial learning rate to 0.0002, a commonly used starting learning rate that allows for fast convergence in the early stages of training. During training, adjust the learning rate based on model performance, for example, by using learning rate decay or warm-up strategies.

[0102] c3.3: In each iteration, the generator G θ Image generation from random noise or latent spatial vectors, discriminator Evaluate the generated and real images. The discriminator attempts to distinguish between generated and real images, and the generator's goal is to maximize the probability that its generated images are misclassified as real by the discriminator. Calculate the loss functions for the generator and discriminator, and update the network parameters through backpropagation. Train iteratively until the loss function values ​​of the generator and discriminator tend to stabilize or the predetermined number of iterations is reached.

[0103] Step S4: After model training is complete, inference is performed using the trained model. By inputting a small amount of CT data, the model can render consecutive image frames and use them for 3D reconstruction.

[0104] d1: Select a small number or single CT scan images as input for model inference. These images should cover the required anatomical structures and contrast types. Perform necessary preprocessing on the input CT images, such as denoising, contrast enhancement, and normalization, to match the data distribution during training, ensuring that the format and size of the input data are consistent with those during training, such as resolution and image depth.

[0105] d2: Load the trained generator network G θ This network is responsible for generating images from the latent space, ensuring that network weights are loaded from checkpoints saved during training.

[0106] d3: Generates random or predefined latent space vectors for the generator, which will guide the generation of the image. Using generator G... θGenerate consecutive image frames from latent spatial vectors. These image frames should cover a 360-degree viewpoint to facilitate subsequent 3D reconstruction. The images can be synthesized according to the required viewpoint by adjusting the generator's input parameters.

[0107] d4: Select a suitable 3D reconstruction algorithm, such as voxel reconstruction, surface reconstruction, or depth estimation, to reconstruct the 3D shape from consecutive image frames. Use the generated image frames as input and integrate them into the 3D reconstruction algorithm. Adjust the parameters of the generator and reconstruction algorithm based on the reconstruction results to improve the accuracy and quality of the reconstruction.

[0108] As can be seen from the above steps, by training an improved GRAF algorithm model based on feature discrimination using data enhancement transformation with specific types of discontinuous CT image data of specific patients, continuous CT images can be effectively generated from limited CT images to complete three-dimensional reconstruction of the heart. This method not only significantly reduces the radiation dose suffered by patients and minimizes potential harm to the human body, but also improves the efficiency and accuracy of image reconstruction, making the entire reconstruction process more efficient and economical.

[0109] Example 2

[0110] See Figure 3 This embodiment provides a cardiac three-dimensional reconstruction system based on an improved GRAF algorithm, used to implement the cardiac three-dimensional reconstruction method as described in Embodiment 1, including:

[0111] Image acquisition module 100 acquires images containing different patients, different contrast types, and normal and abnormal cardiac anatomy structures;

[0112] The X-ray image integration and data processing module 200 performs integration and data processing on the raw CT data, including denoising, standardization, and digital reconstruction of radiographic images.

[0113] The improved GRAF algorithm model 300 includes a generator module and a discriminator module based on the generative radiation field principle. The generator module includes ray sampling, scene representation, density prediction network, color prediction network, and volume rendering sub-modules. The discriminator module includes image patch extraction, a discriminator backbone network, and multiple data augmentation transformation discriminators. The model parameters are optimized by combining the output of the discriminators and the results of data augmentation through a comprehensive loss function. The GRAF algorithm model with optimized model parameters generates continuous image frames from a small number of images or a single view.

[0114] The 3D reconstruction module 400 uses a 3D reconstruction algorithm to reconstruct a 3D image of the heart from consecutive image frames.

[0115] This cardiac 3D reconstruction system, based on an improved GRAF algorithm model, can accurately reconstruct the 3D structure of the heart using advanced deep learning technology with limited CT data input.

[0116] The aforementioned three-dimensional cardiac reconstruction method can be embodied in the form of a computer program product or a software functional unit. If the aforementioned three-dimensional cardiac reconstruction method is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Therefore, the essence of this technical solution, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an electronic system (which may be a personal computer, server, or network system, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0117] Those skilled in the art will recognize that the units, i.e., algorithm steps, of the various examples described in connection with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0118] In summary, this invention provides a method, system, and program product for three-dimensional cardiac reconstruction based on an improved GRAF algorithm. The three-dimensional reconstruction method includes acquiring and preprocessing digitally reconstructed radiographic images of different patients, different contrast types, and normal and abnormal cardiac anatomy; constructing and training an improved GRAF algorithm model; and rendering consecutive image frames by inputting a small amount of CT image data into the trained GRAF algorithm model, which are then used for three-dimensional reconstruction. This invention enables accurate reconstruction of the three-dimensional structure of the heart with limited CT image input, through an improved GRAF algorithm model trained using a combination of data augmentation transformation and multiple training heads.

[0119] Those skilled in the art should understand that variations can be implemented by combining existing technology with the above embodiments, which will not be elaborated here. Such variations do not affect the essence of the present invention, and will not be elaborated here either.

[0120] The preferred embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above. Systems and structures not described in detail should be understood as being implemented in a conventional manner in the art. Any person skilled in the art can make many possible variations and modifications to the technical solutions of the present invention using the methods and techniques disclosed above, or modify them into equivalent embodiments with equivalent changes, without departing from the scope of the present invention. This does not affect the essential content of the present invention. Therefore, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for three-dimensional reconstruction of heart based on improved GRAF algorithm, characterized in that, The method comprises the following steps: Step S1, information collection of cardiac CT scan data: collecting images containing different patients, different contrast types, and normal and abnormal cardiac anatomical structures; Step S2, comprehensive and data processing of the original CT image data, including denoising, standardization and digital reconstructed radiograph generation, and performing several data enhancement transformations on the processed digital reconstructed radiograph to construct a training set and a test set; Step S3, constructing and training an improved GRAF algorithm model: the improved GRAF algorithm model comprises a generator network and a discriminator network based on the principle of generated radiation field; the generator network samples rays from the digital reconstructed radiograph obtained in step S2 through ray sampling, and uses positional encoding to map 3D positions and observation directions to a high-dimensional feature space, further uses shape encoding and appearance encoding to determine the shape and appearance of the object, and then predicts the density and color of any point in the scene through a fully connected layer, and finally synthesizes continuous image frames through volume rendering technology; the discriminator network applies multiple discriminators corresponding to different data enhancement transformations, each discriminator focuses on extracting features under a specific data enhancement transformation and making a prediction, and combines the outputs of the discriminators and the data enhancement results through a comprehensive loss function to optimize the model parameters; The training is iterated until the loss function values of the generator network and the discriminator network tend to be stable or reach a predetermined number of iterations; Step S4, three-dimensional reconstruction of the cardiac structure using the trained GRAF algorithm model and non-continuous cardiac CT images.

2. The method for three-dimensional reconstruction of heart based on improved GRAF algorithm according to claim 1, characterized in that, In step S1, patient data of different ages, genders, weights and races are collected, including X-ray images of normal and abnormal hearts, and different concentrations of iodinated contrast agents are used during sampling to obtain images of different contrasts.

3. The method for three-dimensional reconstruction of heart based on improved GRAF algorithm according to claim 1, characterized in that, In step S2, for each original CT image, first, a denoising algorithm is used to reduce random noise in the image, second, the image contrast is enhanced, third, the image is standardized and segmented, the heart region is extracted using threshold segmentation, region growing or deep learning methods, further digital reconstructed radiographs (DRRs) are generated using the image segmentation results, and finally the generated DRRs are resampled to have uniform resolution and size; data enhancement techniques are applied to the sampled DRRs to obtain diversified DRRs; The enhanced DRRs are organized into a training set and a test set.

4. The method for three-dimensional reconstruction of heart based on improved GRAF algorithm according to claim 3, characterized in that, The data enhancement techniques include at least two of flipping, rotating, scaling and translating.

5. The method for three-dimensional reconstruction of heart based on improved GRAF algorithm according to claim 1, characterized in that, The total loss function calculation formula of the generator and the discriminator network is: ， In the above formula, is a loss for measuring the discriminative head on real data and its enhanced version is a loss of the difference of prediction values, is a generator and a discriminator is a loss of the confrontation between the generator and the discriminator, is the number of discriminative heads, is the number of discriminative heads, is used to measure the weight between different losses, is a random weight or a pre-training weight parameter of the generator; is the first discriminator weight parameter.

6. The method for three-dimensional reconstruction of heart based on improved GRAF algorithm according to claim 1, characterized in that, In step S4, a small number or a single cardiac CT scan image is first selected as the input of the improved GRAF algorithm model, and the input image covers the required anatomical structure and contrast type; secondly, the input CT image is preprocessed to match the data distribution during training; thirdly, the trained generator network is loaded, and the generator is used to generate continuous image frames from the latent space vector; finally, a three-dimensional reconstruction algorithm is used to reconstruct a cardiac 3D image from the continuous image frames.

7. A three-dimensional reconstruction system of heart based on improved GRAF algorithm, characterized in that, A method for implementing the three-dimensional reconstruction of a heart according to any one of claims 1 to 6, comprising: an image acquisition module for acquiring images containing different patients, different contrast types, and normal and abnormal heart anatomical structures; an X-ray image synthesis and data processing module for performing comprehensive and data processing of the original CT data, including denoising, standardization, and digital reconstructed radiograph generation; an improved GRAF algorithm model, including a generator module and a discriminator module based on the principle of generated radiation field; the generator module includes ray sampling, scene representation, density prediction network, color prediction network, and volume rendering submodule; the discriminator module includes image block extraction, discriminator backbone network, and multiple data enhancement transformation discriminators, and the model parameters are optimized by combining the output of the discriminators and the data enhancement results through a comprehensive loss function; the GRAF algorithm model after optimization of the model parameters generates continuous image frames from a small number or a single view of images; a three-dimensional reconstruction module for reconstructing the continuous image frames into a 3D image of the heart using a three-dimensional reconstruction algorithm.

8. The heart three-dimensional reconstruction system based on the improved GRAF algorithm of claim 7, characterized in that, The data enhancement transformation discriminators include a flipping discriminator, a rotation discriminator, and a scaling discriminator.

9. The heart three-dimensional reconstruction system based on the improved GRAF algorithm of claim 7, wherein, The total loss function calculation formula of the generator and discriminator network is: ， In the above formula, To measure the discriminant head against real data and its enhanced version Loss due to the difference in predicted values For generator and discriminator The losses in the confrontation between them For the identification head number, To determine the number of the first one, Used to measure the weights between different losses These are random weights or pre-trained weight parameters for the generator. For the first Discriminator parameters.

10. A computer program product, characterised in that, When the computer program product is running on the computer, the computer is caused to execute the three-dimensional reconstruction method of a heart based on the improved GRAF algorithm according to any one of claims 1 to 6.

Images