Cardiovascular atlas calculation and visualization system based on generative model

Abstract

The present invention discloses a cardiovascular atlas calculation and visualization system based on a generative model, which lies at the intersection of biomedical engineering and computer vision technology. The system includes a data acquisition module, a data preprocessing module, a generative model training module, a cardiovascular atlas calculation module, a visualization module, and a system control module. The present invention utilizes the aforementioned generative model-based cardiovascular atlas calculation and visualization system, introducing a generative model to achieve intelligent calculation and generation of cardiovascular atlases. Simultaneously, it combines advanced visualization technology to provide users with intuitive and highly interactive cardiovascular atlas displays, thereby improving the accuracy of cardiovascular disease diagnosis and the efficiency of medical research.

Description

Technical Field

[0001] The present invention relates to the interdisciplinary field of biomedical engineering and computer vision technology, and in particular to a cardiovascular atlas calculation and visualization system based on a generative model. Background Art

[0002] The cardiovascular system is a vital circulatory system in the human body, and accurate analysis of its structure and function is crucial for the early diagnosis, treatment planning, and prognosis assessment of cardiovascular diseases. Traditional cardiovascular atlas construction methods rely primarily on manual annotation of medical imaging data or registration algorithms based on fixed templates. Manual annotation is not only time-consuming and labor-intensive, but also highly influenced by subjective factors, making it difficult to ensure consistency. Registration algorithms based on fixed templates struggle to accurately match data when faced with large differences in individual cardiovascular structures, resulting in insufficient accuracy in atlas construction.

[0003] With the development of deep learning technology, several deep learning-based cardiovascular atlas construction methods have been proposed. However, most of these methods employ discriminative models, which can only classify and predict existing data and are unable to generate new, diverse cardiovascular atlas data. Furthermore, existing cardiovascular atlas visualization methods are relatively simple, lacking intuitive and interactive presentation of complex cardiovascular structure and function information, making it difficult to meet the needs of clinicians and researchers for in-depth analysis of the cardiovascular system. Therefore, a system that can utilize generative models to achieve high-precision cardiovascular atlas calculations and provide efficient visualization is urgently needed. Summary of the Invention

[0004] The purpose of this invention is to provide a cardiovascular atlas calculation and visualization system based on a generative model. By introducing a generative model, the intelligent calculation and generation of cardiovascular atlases can be realized. At the same time, combined with advanced visualization technology, it provides users with an intuitive and highly interactive cardiovascular atlas display, thereby improving the accuracy of cardiovascular disease diagnosis and the efficiency of medical research.

[0005] To achieve the above objectives, the present invention provides a cardiovascular atlas calculation and visualization system based on a generative model, comprising a data acquisition module, a data preprocessing module, a generative model training module, a cardiovascular atlas calculation module, a visualization module, and a system control module;

[0006] The data acquisition module is used to collect multimodal cardiovascular medical imaging data and clinical information data;

[0007] The data preprocessing module is used to perform preprocessing operations on the collected multimodal cardiovascular medical imaging data and clinical information data;

[0008] The generative model training module is used to construct a generative adversarial network (GAN). The pre-processed multimodal cardiovascular medical imaging data and clinical information data are used as training data. The multimodal cardiovascular medical imaging data and clinical information data are jointly encoded and input into the generative adversarial network (GAN) for training.

[0009] The cardiovascular atlas calculation module uses the average template as a priori condition input into the generative model to constrain the generated pathological structure to conform to anatomical rationality. Based on the specific condition information input by the user, the generative model generates the corresponding cardiovascular atlas, including a three-dimensional structural atlas and a cardiovascular function atlas. It uses high-dimensional dynamic modeling to dynamically simulate the cardiovascular function atlas and optimize and adjust the cardiovascular atlas.

[0010] The visualization module is used for visualization and rapid rendering of three-dimensional structural maps, as well as visualization of dynamically simulated cardiovascular function maps;

[0011] The system control module is used to coordinate the workflow and data transmission between modules.

[0012] Preferably, the data acquisition module establishes a data interface with the hospital's image archiving and communication system PACS and electronic medical record system EMR, and is connected to the data storage unit.

[0013] Preferably, the data preprocessing module calculates the average template based on the collected multimodal cardiovascular medical imaging data, uses a deep learning-based denoising network combined with GPU acceleration to perform image denoising, then performs image registration based on a hybrid registration algorithm based on feature points and deep learning combined with GPU acceleration, and finally performs image standardization.

[0014] Preferably, according to the collected For example, we extract the contours or voxel data of key anatomical structures from multimodal cardiovascular medical image data and use a GPU-accelerated parallel registration algorithm to unify all samples into a standard coordinate system. The calculation formula for the average template is as follows:

[0015] ;

[0016] in, is the average template, is the sample size, For the The anatomical structure data of the sample;

[0017] The process of image noise reduction is as follows: input noisy image data , extract noise features through multi-layer convolutional layers, and use residual connections Preserve original image details, where is the residual network mapping function, and the mean square error MSE is used to measure the difference between the denoised image and the real clean image. The loss function is calculated as follows:

[0018] ;

[0019] in, is the height of the image, is the width of the image, is the number of channels of the image, To traverse the image height, To traverse the image width, is the number of image channels traversed, The image after denoising is at position ,aisle The pixel value at For clean images in position ,aisle The pixel value at ;

[0020] The process of image registration is: extracting floating images using deep learning models The feature point set and fixed images The feature point set ,in is the index of the feature point, and the rigid transformation matrix is ​​solved by the iterative closest point ICP algorithm , so that the distance between feature points is minimized, the calculation formula of the objective function is as follows:

[0021] ;

[0022] in, is the rotation matrix, and , is the translation vector, and , is the number of matching feature point pairs;

[0023] The process of standardization is to standardize the continuous clinical information data so that it obeys the standard normal distribution. The calculation formula is as follows:

[0024] ;

[0025] in, is the standardized data, The original clinical data, is the data mean, is the standard deviation of the data.

[0026] Preferably, the network architecture of the generative adversarial network GAN in the generative model training module is:

[0027] Generator : ,in, is the noise vector, , is the conditional vector , To generate an image, input a noise vector and a condition vector, and output the generated image;

[0028] Discriminator : ,in For a real image, input a real image or generate an image and a conditional vector, and output the probability of true or false;

[0029] The calculation formula of the adversarial loss function is as follows:

[0030] ;

[0031] in, is the real image data distribution, is the noise prior distribution, is the mathematical expectation operator.

[0032] Preferably, the cardiovascular atlas calculation module uses the conditional probability constraint to generate the pathological structure, and the calculation formula is as follows:

[0033] ;

[0034] in, To generate cardiovascular structural data, As pathological characteristics, To generate cardiovascular structural data When the pathological characteristics are given The conditional probability of For pathological characteristics and average template Generate cardiovascular structure data under the common constraints of The conditional probability of Average template The prior probability of

[0035] High-dimensional state space modeling of cardiovascular dynamics is performed, and pressure and flow velocity are mapped to high-dimensional space. The calculation formula of the high-dimensional state vector is as follows:

[0036] ;

[0037] in, for The high-dimensional state vector at time , For the The pressure at each position, For the The flow rate at each location, ;

[0038] The specific process of optimization is to minimize the editing area The difference loss optimizes the generator parameters, and the calculation formula is as follows:

[0039] ;

[0040] in, is the three-dimensional space coordinate, , The target map edited by the user. is the learning rate, The generator parameters.

[0041] Therefore, the present invention adopts the above-mentioned cardiovascular atlas calculation and visualization system based on the generative model, which has the following beneficial effects:

[0042] 1) By introducing a generative model, intelligent calculation and generation of cardiovascular maps can be achieved;

[0043] 2) Through data preprocessing and joint encoding, deep data fusion is achieved, providing richer information support for the accurate calculation of cardiovascular maps;

[0044] 3) At the same time, it combines advanced visualization technology to provide users with intuitive and interactive cardiovascular map displays, thereby improving the accuracy of cardiovascular disease diagnosis and the efficiency of medical research.

[0045] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 It is a structural diagram of the cardiovascular atlas calculation and visualization system based on the generative model of the present invention. DETAILED DESCRIPTION

[0047] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0048] Unless otherwise defined, technical or scientific terms used in the present invention shall have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs.

[0049] Example 1

[0050] like Figure 1As shown, the present invention provides a cardiovascular atlas calculation and visualization system based on a generative model, including a data acquisition module, a data preprocessing module, a generative model training module, a cardiovascular atlas calculation module, a visualization module and a system control module.

[0051] The data acquisition module is used to collect multimodal cardiovascular medical imaging data and clinical information data;

[0052] The data acquisition module establishes a data interface with the hospital's image archiving and communication system (PACS) and electronic medical record system (EMR), and connects to the data storage unit to automatically acquire and transmit data, storing the acquired data in the system's data storage unit. Multimodal cardiovascular medical imaging data primarily includes, but is not limited to, echocardiography, CT angiography (CTA), magnetic resonance angiography (MRA), positron emission tomography (PET), and other medical imaging data.

[0053] The data preprocessing module is used to perform preprocessing operations on the collected multimodal cardiovascular medical imaging data and clinical information data;

[0054] The data preprocessing module calculates the average template based on the collected multimodal cardiovascular medical imaging data, uses a deep learning-based denoising network combined with GPU acceleration to perform image denoising, then performs image registration using a hybrid registration algorithm based on feature points and deep learning combined with GPU acceleration, and finally performs image standardization.

[0055] According to the collected For example, we extract the contours or voxel data of key anatomical structures from multimodal cardiovascular medical image data and use a GPU-accelerated parallel registration algorithm to unify all samples into a standard coordinate system. The calculation formula for the average template is as follows:

[0056] ;

[0057] in, is the average template, is the sample size, For the The anatomical structure data of the sample;

[0058] The process of image noise reduction is as follows: input noisy image data , extract noise features through multi-layer convolutional layers, and use residual connections Preserve original image details, where is the residual network mapping function, and the mean square error MSE is used to measure the difference between the denoised image and the real clean image. The loss function is calculated as follows:

[0059] ;

[0060] in, is the height of the image, is the width of the image, is the number of channels of the image, To traverse the image height, To traverse the image width, is the number of image channels traversed, The image after denoising is at position ,aisle The pixel value at For clean images in position ,aisle The pixel value at ;

[0061] The process of image registration is: extracting floating images using deep learning models The feature point set and fixed images The feature point set ,in is the index of the feature point, and the rigid transformation matrix is ​​solved by the iterative closest point ICP algorithm , so that the distance between feature points is minimized, the calculation formula of the objective function is as follows:

[0062] ;

[0063] in, is the rotation matrix, and , is the translation vector, and , is the number of matching feature point pairs;

[0064] The process of standardization is to standardize the continuous clinical information data so that it obeys the standard normal distribution. The calculation formula is as follows:

[0065] ;

[0066] in, is the standardized data, The original clinical data, is the data mean, is the standard deviation of the data.

[0067] The generative model training module is used to construct a generative adversarial network (GAN). The pre-processed multimodal cardiovascular medical imaging data and clinical information data are used as training data. The multimodal cardiovascular medical imaging data and clinical information data are jointly encoded and input into the generative adversarial network (GAN) for training.

[0068] The network architecture of the Generative Adversarial Network (GAN) in the Generative Model Training Module is:

[0069] Generator : ,in, is the noise vector, , is the conditional vector , To generate an image, input a noise vector and a condition vector, and output the generated image;

[0070] Discriminator : ,in For a real image, input a real image or generate an image and a conditional vector, and output the probability of true or false;

[0071] The calculation formula of the adversarial loss function is as follows:

[0072] ;

[0073] in, is the real image data distribution, is the noise prior distribution, is the mathematical expectation operator.

[0074] During training, the generative model learns the distribution characteristics and inherent laws of cardiovascular data through adversarial training or variational inference, enabling it to generate atlas data that conforms to the structural and functional characteristics of real cardiovascular systems. To improve the performance and generalization of the generative model, transfer learning and data augmentation techniques are employed. Pre-training is performed on a large, publicly available cardiovascular dataset, followed by fine-tuning on the dataset of this system. Furthermore, data augmentation operations such as rotation, scaling, and translation are performed on the training data to expand the training data volume.

[0075] The calculation formula for pre-training is as follows:

[0076] ;

[0077] in, is the generator parameter, Public datasets;

[0078] The cardiovascular atlas calculation module uses the average template as a priori condition input into the generative model to constrain the generated pathological structure to conform to anatomical rationality. Based on the specific condition information input by the user, the generative model generates the corresponding cardiovascular atlas, including a three-dimensional structural atlas and a cardiovascular function atlas. It uses high-dimensional dynamic modeling to dynamically simulate the cardiovascular function atlas and optimize and adjust the cardiovascular atlas.

[0079] Users can input specific criteria, such as clinical information like disease type, age range, and gender, or partial cardiovascular imaging data. The generative model then generates a corresponding cardiovascular atlas based on these input conditions. During the atlas calculation process, the generative model's generative capabilities enable the generation of not only complete 3D cardiovascular structural atlases but also cardiovascular functional atlases.

[0080] The cardiovascular atlas calculation module uses conditional probability constraints to generate pathological structures. The calculation formula is as follows:

[0081] ;

[0082] in, To generate cardiovascular structural data, As pathological characteristics, To generate cardiovascular structural data When the pathological characteristics are given The conditional probability of For pathological characteristics and average template Generate cardiovascular structure data under the common constraints of The conditional probability of Average template The prior probability of

[0083] High-dimensional state space modeling of cardiovascular dynamics is performed, and pressure and flow velocity are mapped to high-dimensional space. The calculation formula of the high-dimensional state vector is as follows:

[0084] ;

[0085] in, for The high-dimensional state vector at time , For the The pressure at each position, For the The flow rate at each location, .

[0086] The cardiovascular atlas calculation module can also edit and optimize the cardiovascular atlas. The specific process of optimization is to minimize the editing area. The difference loss optimizes the generator parameters, and the calculation formula is as follows:

[0087] ;

[0088] in, is the three-dimensional space coordinate, , The target map edited by the user. is the learning rate, The generator parameters.

[0089] The visualization module is used for visualization and rapid rendering of three-dimensional structural maps, as well as visualization of dynamically simulated cardiovascular function maps.

[0090] The system control module coordinates workflows and data transmission between modules. It receives user commands, invokes the corresponding modules to execute tasks based on the commands, and provides feedback to the user. It also monitors and manages the system's operational status, including data storage management, system performance monitoring, and error handling. When system anomalies occur, timely error diagnosis and recovery are performed to ensure stable system operation.

[0091] Therefore, the present invention adopts the above-mentioned cardiovascular atlas calculation and visualization system based on the generative model. By introducing the generative model, it realizes the intelligent calculation and generation of cardiovascular atlases. At the same time, it combines advanced visualization technology to provide users with intuitive and highly interactive cardiovascular atlas displays, thereby improving the accuracy of cardiovascular disease diagnosis and the efficiency of medical research.

[0092] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A generative model-based cardiovascular atlas computation and visualization system, characterized by: It includes data acquisition module, data preprocessing module, generative model training module, cardiovascular atlas calculation module, visualization module and system control module; The data acquisition module is used to collect multimodal cardiovascular medical imaging data and clinical information data; The data preprocessing module is used to perform preprocessing operations on the collected multimodal cardiovascular medical imaging data and clinical information data; The generative model training module is used to construct a generative adversarial network (GAN). The pre-processed multimodal cardiovascular medical imaging data and clinical information data are used as training data. The multimodal cardiovascular medical imaging data and clinical information data are jointly encoded and input into the generative adversarial network (GAN) for training. The cardiovascular atlas calculation module uses the average template as a priori condition input into the generative model to constrain the generated pathological structure to conform to anatomical rationality. Based on the specific condition information input by the user, the generative model generates the corresponding cardiovascular atlas, including a three-dimensional structural atlas and a cardiovascular function atlas. It uses high-dimensional dynamic modeling to dynamically simulate the cardiovascular function atlas and optimize and adjust the cardiovascular atlas. The visualization module is used for visualization and rapid rendering of three-dimensional structural maps, as well as visualization of dynamically simulated cardiovascular function maps; The system control module is used to coordinate the workflow and data transmission between modules; The data preprocessing module calculates the average template based on the collected multimodal cardiovascular medical imaging data, uses a deep learning-based denoising network combined with GPU acceleration to perform image noise reduction, then performs image registration using a hybrid registration algorithm based on feature points and deep learning combined with GPU acceleration, and finally performs image normalization. According to the collected For example, we extract the contours or voxel data of key anatomical structures from multimodal cardiovascular medical image data and use a GPU-accelerated parallel registration algorithm to unify all samples into a standard coordinate system. The calculation formula for the average template is as follows: ； in, is the average template, is the sample size, For the The anatomical structure data of the sample; The process of image noise reduction is as follows: input noisy image data , extract noise features through multi-layer convolutional layers, and use residual connections Preserve original image details, where is the residual network mapping function, and the mean square error MSE is used to measure the difference between the denoised image and the real clean image. The loss function is calculated as follows: ； in, is the height of the image, is the width of the image, is the number of channels of the image, To traverse the image height, To traverse the image width, is the number of image channels traversed, The image after denoising is at position ,aisle The pixel value at For clean images in position ,aisle The pixel value at ; The process of image registration is: extracting floating images using deep learning models The feature point set and fixed images The feature point set ,in is the index of the feature point, and the rigid transformation matrix is ​​solved by the iterative closest point ICP algorithm , so that the distance between feature points is minimized, the calculation formula of the objective function is as follows: ； in, is the rotation matrix, and , is the translation vector, and , is the number of matching feature point pairs; The process of standardization is to standardize the continuous clinical information data so that it obeys the standard normal distribution. The calculation formula is as follows: ； in, is the standardized data, The original clinical data, is the data mean, is the standard deviation of the data.

2. The generative model-based cardiovascular atlas calculation and visualization system according to claim 1, characterized in that: The data acquisition module establishes a data interface with the hospital's image archiving and communication system PACS and electronic medical record system EMR, and is connected to the data storage unit.

3. The generative model-based cardiovascular atlas calculation and visualization system according to claim 1, characterized in that: The network architecture of the Generative Adversarial Network (GAN) in the Generative Model Training Module is: Generator : ,in, is the noise vector, , is the conditional vector , To generate an image, input a noise vector and a condition vector, and output the generated image; Discriminator : ,in For a real image, input a real image or generate an image and a conditional vector, and output the probability of true or false; The calculation formula of the adversarial loss function is as follows: ； in, is the real image data distribution, is the noise prior distribution, is the mathematical expectation operator.

4. The generative model-based cardiovascular atlas calculation and visualization system according to claim 1, characterized in that: The cardiovascular atlas calculation module uses conditional probability constraints to generate pathological structures. The calculation formula is as follows: ； in, To generate cardiovascular structural data, As pathological characteristics, To generate cardiovascular structural data When the pathological characteristics are given The conditional probability of For pathological characteristics and average template Generate cardiovascular structure data under the common constraints of The conditional probability of Average template The prior probability of High-dimensional state space modeling of cardiovascular dynamics is performed, and pressure and flow velocity are mapped to high-dimensional space. The calculation formula of the high-dimensional state vector is as follows: ； in, for The high-dimensional state vector at time , For the The pressure at each position, For the The flow rate at each location, ; The specific process of optimization is to minimize the editing area The difference loss optimizes the generator parameters, and the calculation formula is as follows: ； in, is the three-dimensional space coordinate, , The target map edited by the user. is the learning rate, The generator parameters.

Images