Fluid parameter prediction method, device, equipment, medium and program product

Abstract

The invention provides a fluid parameter prediction method and device, equipment, a medium and a program product, and relates to the technical field of data processing. The method comprises the steps that medical image data and physiological information data of a to-be-predicted object are acquired, the medical image data are used for representing structural features of a target part of the to-be-predicted object, and the physiological information data are used for representing physiological state features of the to-be-predicted object; the medical image data and the physiological information data are input into a trained fluid parameter prediction model, fluid parameters at different moments of the whole cardiac cycle of the target part are output, and the fluid parameter prediction model is constructed based on a deep learning model and is obtained through training of a data driving loss function with physical constraints. The problem of prediction deviation when pure image data or pure physiological parameters are singly used for fluid parameter prediction is solved, and accurate and rapid prediction of the fluid parameters at different moments in the whole cardiac cycle of the target part is achieved.

Description

Technical Field

[0001] This disclosure relates to the field of data processing technology, and in particular to a method, apparatus, equipment, medium, and program product for predicting fluid parameters. Background Technology

[0002] In the fields of biomedical engineering and clinical medicine, in-depth research on the fluid dynamics within the human body is crucial for disease diagnosis, treatment planning, and medical device development. Fluid parameters, as key indicators reflecting the physiological state and fluid dynamics characteristics of the human body, play an irreplaceable role in understanding physiological processes such as those in the cardiovascular system. For example, in cardiovascular disease research, fluid parameters within the aorta provide crucial information for doctors to assess vascular disease risk and evaluate cardiac function; in medical device development, precise fluid parameters help optimize the design of products such as ventricular assist devices and artificial heart valves, improving their performance and safety.

[0003] However, current technologies for acquiring fluid parameters have significant shortcomings. Existing measurement and prediction methods struggle to meet the needs of clinical practice and research, facing the dual challenges of efficiency and accuracy. On one hand, traditional measurement techniques are often complex and time-consuming, failing to provide rapid detection. In emergency medical situations, they cannot promptly provide doctors with crucial fluid parameter information, potentially delaying optimal diagnosis and treatment. On the other hand, existing prediction models vary in accuracy and are susceptible to interference from various factors, resulting in significant deviations between predicted and actual outcomes, greatly reducing their reliability in clinical applications and research analysis. Therefore, how to acquire fluid parameters quickly and accurately has become a critical technical problem that urgently needs to be solved.

[0004] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] This disclosure provides a method, apparatus, equipment, medium, and program product for predicting fluid parameters, which at least to some extent overcomes the problems of low efficiency and low accuracy in fluid parameter prediction.

[0006] Other features and advantages of this disclosure will become apparent from the following detailed description, or may be learned in part from practice of this disclosure.

[0007] According to one aspect of this disclosure, a fluid parameter prediction method is provided, comprising: acquiring medical imaging data and physiological information data of an object to be predicted, wherein the medical imaging data is used to characterize the structural features of a target site of the object to be predicted, and the physiological information data is used to characterize the physiological state features of the object to be predicted; inputting the medical imaging data and physiological information data into a pre-trained fluid parameter prediction model, and outputting fluid parameters of the target site at different times during the entire cardiac cycle, wherein the fluid parameter prediction model is constructed based on a deep learning model and trained using a data-driven loss function with physical constraints.

[0008] In some embodiments, the medical imaging data includes aortic geometry data obtained from computed tomography angiography (CTA), and the physiological information data includes at least one of blood pressure, heart rate, cardiac output, and systolic percentage.

[0009] In some embodiments, the target site is the aorta, and the fluid parameters include at least one of wall shear stress WSS, time-averaged wall shear stress TAWSS, oscillating shear index OSI, relative residence time RRT, and helicity Psi.

[0010] In some embodiments, acquiring medical image data and physiological information data of the object to be predicted includes: performing three-dimensional modeling and mesh generation on the medical image data to generate a three-dimensional model containing the geometric boundaries of the target area; and inputting the physiological information data as boundary conditions into a computational fluid dynamics (CFD) simulation to generate fluid parameter reference data corresponding to the medical image data.

[0011] In some embodiments, the training process of the fluid parameter prediction model includes: converting acquired medical image data into 3D point cloud data and constructing a training dataset by combining the physiological information data; inputting the training dataset into a deep learning-based model framework, the model framework including a Kpconv encoder-decoder structure capable of learning multi-scale features, and a graph convolutional neural network for mining point cloud context information; configuring a data-driven loss function with physical constraints on the model framework, the physical constraints being constructed based on the continuity equation and the Navier-Stokes equation, the data-driven loss function being used to measure the error between the model prediction result and the reference data obtained from computational fluid dynamics simulation; and training the model framework using the adaptive moment estimation optimizer Adam to obtain the fluid parameter prediction model.

[0012] In some embodiments, before inputting the training dataset into the deep learning-based model framework, one or more of the following methods are further included: processing the 3D point cloud data using the farthest point sampling (FPS) algorithm; applying a random rigid transformation to the 3D point cloud data, including random rotation and random translation; and randomizing the arrangement order of points in the 3D point cloud data.

[0013] In some embodiments, outputting fluid parameters at different times during the entire cardiac cycle of the target site includes: predicting point-by-point flow field parameters using the fluid parameter prediction model, wherein the flow field parameters include velocity components in three directions and pressure scalars; and calculating the hemodynamic parameter distribution of the target site at different times based on the flow field parameters.

[0014] According to another aspect of this disclosure, a fluid parameter prediction device is also provided, comprising: a data acquisition module for acquiring medical imaging data and physiological information data of an object to be predicted, wherein the medical imaging data is used to characterize the structural features of a target site of the object to be predicted, and the physiological information data is used to characterize the physiological state features of the object to be predicted; and a parameter output module for inputting the medical imaging data and physiological information data into a pre-trained fluid parameter prediction model and outputting fluid parameters of the target site at different times during the entire cardiac cycle, wherein the fluid parameter prediction model is constructed based on a deep learning model and trained using a data-driven loss function with physical constraints.

[0015] According to another aspect of this disclosure, an electronic device is also provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform the fluid parameter prediction method described in any one of the preceding claims by executing the executable instructions.

[0016] According to another aspect of this disclosure, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the fluid parameter prediction method described in any of the preceding claims.

[0017] According to another aspect of this disclosure, a computer program product is also provided, including a computer program that, when executed by a processor, implements the fluid parameter prediction method of any of the above.

[0018] The fluid parameter prediction methods, apparatus, devices, media, and program products provided in the embodiments of this disclosure acquire medical imaging data and physiological information data of the object to be predicted, and fuse the two into a fluid parameter prediction model based on deep learning and with physical constraints. This solves the prediction bias problem caused by the one-sided information when using pure imaging data or pure physiological parameters alone for fluid parameter prediction. It breaks through the technical bottleneck of traditional methods that cannot fully represent the real physiological environment of the target site, and realizes accurate and rapid prediction of fluid parameters at different times during the entire cardiac cycle of the target site, providing clinicians with multi-dimensional clinical indicators for diagnosing aortic diseases.

[0019] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0020] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure. It is obvious that the drawings described below are merely some embodiments of this disclosure, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0021] Figure 1 This diagram illustrates an exemplary application system architecture in an embodiment of the present disclosure.

[0022] Figure 2 This diagram illustrates a flow chart of a fluid parameter prediction method according to an embodiment of the present disclosure.

[0023] Figure 3 This diagram illustrates a flowchart of yet another fluid parameter prediction method according to an embodiment of the present disclosure.

[0024] Figure 4 This diagram illustrates a flowchart of yet another fluid parameter prediction method according to an embodiment of the present disclosure.

[0025] Figure 5 This diagram illustrates a flowchart of yet another fluid parameter prediction method according to an embodiment of the present disclosure.

[0026] Figure 6 This diagram illustrates a fluid parameter prediction device according to an embodiment of the present disclosure.

[0027] Figure 7 A structural block diagram of an electronic device according to an embodiment of the present disclosure is shown. Detailed Implementation

[0028] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that this disclosure will be more comprehensive and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0029] Furthermore, the accompanying drawings are merely illustrative of this disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0030] The specific implementation methods of the embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0031] Figure 1 A schematic diagram of an exemplary application system architecture to which the fluid parameter prediction method of the embodiments of this disclosure can be applied is shown. For example... Figure 1 As shown, the system architecture may include terminal device 101, network 102 and server 103.

[0032] Network 102 is a medium used to provide a communication link between terminal device 101 and server 103, and can be a wired network or a wireless network.

[0033] Optionally, the aforementioned wireless or wired networks use standard communication technologies and / or protocols. The network is typically the Internet, but can also be any network, including but not limited to Local Area Networks (LANs), Metropolitan Area Networks (MANs), Wide Area Networks (WANs), mobile, wired or wireless networks, private networks, or any combination of virtual private networks. In some embodiments, technologies and / or formats including Hyper Text Markup Language (HTML), Extensible Markup Language (XML), etc., are used to represent data exchanged over the network. Furthermore, conventional encryption technologies such as Secure Socket Layer (SSL), Transport Layer Security (TLS), Virtual Private Networks (VPNs), and Internet Protocol Security (IPSec) can be used to encrypt all or some links. In other embodiments, custom and / or dedicated data communication technologies can be used to replace or supplement the aforementioned data communication technologies.

[0034] Terminal device 101 can be various electronic devices, including but not limited to smartphones, tablets, laptops, desktop computers, smart speakers, smartwatches, wearable devices, augmented reality devices, virtual reality devices, etc.

[0035] Optionally, the client of the application installed on different terminal devices 101 may be the same, or the client of the same type of application based on different operating systems. Depending on the terminal platform, the specific form of the application client may also be different; for example, the application client may be a mobile client, a PC client, etc.

[0036] Server 103 can be a server that provides various services, such as a backend management server that supports the device operated by the user using terminal device 101. The backend management server can analyze and process received requests and other data, and feed the processing results back to the terminal device.

[0037] Optionally, the server can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0038] Those skilled in the art will know that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative; any number of terminal devices, networks, and servers can be included depending on actual needs. This disclosure does not limit the scope of the embodiments.

[0039] Under the above system architecture, this disclosure provides a fluid parameter prediction method, which can be executed by any electronic device with computing power.

[0040] In some embodiments, the fluid parameter prediction method provided in this disclosure can be executed by a terminal device of the system architecture described above; in other embodiments, the fluid parameter prediction method provided in this disclosure can be executed by a server in the system architecture described above; in still other embodiments, the fluid parameter prediction method provided in this disclosure can be implemented by the terminal device and the server in the system architecture described above through interaction.

[0041] Figure 2 A flowchart of a fluid parameter prediction method according to an embodiment of this disclosure is shown, as follows: Figure 2 As shown, the fluid parameter prediction method provided in this embodiment includes the following steps:

[0042] S202, acquire medical imaging data and physiological information data of the object to be predicted. The medical imaging data is used to characterize the structural features of the target site of the object to be predicted, and the physiological information data is used to characterize the physiological state features of the object to be predicted.

[0043] It should be noted that the object to be predicted can be any subject requiring fluid parameter analysis and prediction, including but not limited to human bodies (such as patients with aortic dissection or aneurysm) and animal bodies (such as laboratory animals used for medical research). The target site is the specific physiological site to be predicted, especially those involving fluids, such as blood or body fluid flow. Medical imaging data is used to characterize the structural features of the target site of the object to be predicted. Specific medical imaging data includes three-dimensional image data obtained through medical imaging techniques such as computed tomography angiography (CTA) and magnetic resonance imaging (MRI), used to reconstruct the geometric morphology, spatial structure, and boundary features of the target site (such as the aorta). Physiological information data is used to characterize the physiological state characteristics of the object to be predicted. Physiological information data includes parameters reflecting an individual's physiological state, such as blood pressure, heart rate, cardiac output, and systolic percentage. For example, cardiac output can be measured by echocardiography, and systolic percentage can be calculated from electrocardiogram waveforms.

[0044] In some embodiments, the medical imaging data includes aortic geometry data obtained from computed tomography angiography (CTA), and the physiological information data includes at least one of blood pressure, heart rate, cardiac output, and systolic percentage.

[0045] Medical imaging data provides spatial structural features, while physiological information data provides functional state features. Together, they constitute the input feature space of deep learning models, solving the prediction bias problem caused by single data types, such as pure images or pure physiological parameters.

[0046] S204 inputs medical imaging data and physiological information data into a pre-trained fluid parameter prediction model, and outputs fluid parameters at different times of the whole cardiac cycle of the target site. The fluid parameter prediction model is built based on a deep learning model and trained using a data-driven loss function with physical constraints.

[0047] It should be noted that the fluid parameter prediction model is a pre-trained model for predicting fluid parameters. In this embodiment, the fluid parameter prediction model is a computational model built based on deep learning technology. Its core function is to accurately predict the fluid dynamic parameters of target parts (such as the aorta) in the human body or animal at different times during the entire cardiac cycle by analyzing medical imaging data and physiological information data.

[0048] After standardized preprocessing, medical imaging data and physiological information data are input into a pre-trained fluid parameter prediction model according to a preset format. In some embodiments, the fluid parameter prediction model is based on a hybrid architecture of convolutional neural network (CNN) and recurrent neural network (RNN), where the CNN is responsible for extracting spatial features from medical images, and the RNN parses the time-series features of physiological information data; and it is trained using a data-driven loss function with physical constraints to ensure that the fluid parameter prediction model is subjected to physical regularization during deep learning, ensuring that the prediction results conform to the basic laws of fluid dynamics.

[0049] The trained fluid parameter prediction model extracts spatial features and captures temporal dynamics from input medical imaging and physiological data. After calculation, the model outputs time-by-time fluid parameters for the target site within a complete cardiac cycle (including systole and diastole). Because the model is trained with physical constraints, its predictive reliability is ensured. The output fluid parameters are 4D (three-dimensional space + time dimension), such as hemodynamic parameters and fluid morphology parameters. The results are stored as a spatiotemporally aligned four-dimensional data cube ("three-dimensional space + time dimension"), which can be directly interfaced with medical image post-processing systems, providing temporal dynamic data support for the accurate quantitative analysis of hemodynamic abnormalities.

[0050] The fluid parameter prediction method provided in this disclosure significantly improves the accuracy and clinical application value of fluid parameter prediction by integrating multi-dimensional data and deep learning technology. Simultaneously, it utilizes the three-dimensional structural features and physiological information data of the target site obtained from medical imaging data to construct composite input data containing spatial structure and dynamic functional characteristics. This overcomes the prediction bias caused by single imaging data or pure physiological parameters, enabling the model to more comprehensively represent the real physiological environment of the target site. The fluid parameter prediction model introduces a loss function with physical constraints to ensure that the prediction results conform to real physical laws. The model can output fluid parameters of the target site at different times throughout the cardiac cycle, providing refined quantitative data support for the pathological mechanism analysis, rupture risk assessment, and personalized treatment plan formulation of cardiovascular diseases such as aortic dissection and aneurysm, thus promoting the technological upgrade of medical fluid dynamics analysis from static assessment to dynamic prediction.

[0051] In some embodiments of this disclosure, the target site is the aorta, and the fluid parameters include at least one of wall shear stress WSS, time-averaged wall shear stress TAWSS, oscillating shear index OSI, relative residence time RRT, and helicity Psi.

[0052] The target site is the human aorta, including segments such as the ascending aorta, aortic arch, and descending aorta. As the main trunk of the systemic circulation, it is subjected to the impact of high-pressure blood flow and is a common site for vascular diseases such as aortic dissection and aneurysm. The fluid parameters of the aorta are used to quantitatively characterize the hemodynamic features within the aorta.

[0053] Among them, wall shear stress (WSS) is the tangential stress generated between blood flow and the aortic wall due to viscosity, measured in Pascals, reflecting the intensity of mechanical stimulation of vascular endothelial cells by blood flow; time-averaged wall shear stress (TAWSS) is the time integral average of WSS over the entire cardiac cycle, reflecting the chronic mechanical load on the vascular wall under long-term blood flow action, and is an important indicator for assessing arterial remodeling; oscillatory shear index (OSI) is a dimensionless parameter characterizing the degree of oscillation in the direction of wall shear stress; relative residence time (RRT) is the ratio of the residence time of blood particles in the target region to the duration of the entire cardiac cycle, dimensionless, reflecting the degree of blood flow stagnation in a local area; and helicity (Ψ) Psi is a physical quantity that measures the intensity of helical blood flow, characterizing the dot product integral of the velocity vector and vorticity vector. The fluid parameter prediction method disclosed herein can obtain the spatiotemporal distribution of multiple parameters throughout the entire cardiac cycle, providing an analytical tool for the precise diagnosis and treatment of aortic diseases.

[0054] In some embodiments of this disclosure, such as Figure 3 As shown, obtaining medical imaging data and physiological information data of the object to be predicted also includes the following steps:

[0055] S302 performs 3D modeling and mesh generation on medical image data to generate a 3D model containing the geometric boundaries of the target area;

[0056] S304 uses physiological information data as boundary conditions to input computational fluid dynamics (CFD) simulations, generating fluid parameter reference data corresponding to medical imaging data.

[0057] It's important to note that Computational Fluid Dynamics (CFD) is a discipline that studies physical phenomena such as fluid flow, heat transfer, and mass transfer through numerical calculations and computer simulations. CFD simulation refers to the process of numerically simulating fluid flow using computer programs based on CFD theories and methods. By inputting physiological information data as boundary conditions into CFD simulations, and generating fluid parameter reference data corresponding to medical imaging data, it's possible to ensure that the simulation results conform to the real physiological environment. By using physiological information data as boundary conditions in CFD simulations, the simulated flow field can dynamically match the patient's actual physiological rhythms, avoiding the problem of geometric realism but functional distortion.

[0058] Furthermore, by using physiological information data as boundary conditions in computational fluid dynamics (CFD) simulations, the generated fluid parameter reference data corresponding to medical imaging data includes both the spatial structural features contained in the medical images and the dynamic functional features contained in the physiological information data. This provides more complex and comprehensive input data for the fluid parameter prediction model, resulting in more accurate and reliable output results.

[0059] In some embodiments of this disclosure, such as Figure 4 As shown, the training process of the above fluid parameter prediction model includes the following steps:

[0060] S402, the acquired medical image data is converted into 3D point cloud data, and a training dataset is constructed by combining the physiological information data.

[0061] It should be noted that 3D point cloud data is medical image data that has been processed through grid sampling or voxelization to transform it into a three-dimensional point cloud composed of discrete points. Each point contains spatial coordinates and geometric features (such as curvature and normal vector). Physiological information data includes time-series parameters such as blood pressure, heart rate, and cardiac output. After normalization processing, it is associated with 3D point cloud data to form a composite training sample containing spatial structure and functional state.

[0062] S404 inputs the training dataset into a deep learning-based model framework, which includes a Kpconv encoder-decoder structure capable of learning multi-scale features, and a graph convolutional neural network for mining point cloud contextual information.

[0063] It should be noted that in the Kpconv encoder-decoder structure, the encoder uses deformable convolution Kpconv to extract multi-scale features from the 3D point cloud. By dynamically adjusting the position of the convolution kernel center point, it adapts to the non-uniform distribution characteristics of the point cloud and captures geometric features at different scales. The decoder maps the high-dimensional features output by the encoder back to the point cloud space to predict the fluid parameters corresponding to each point. In this embodiment, the graph convolutional neural network constructs a graph structure with the point cloud as nodes. Through graph convolution operations, it mines the contextual relationships between adjacent points, enhancing the model's ability to learn the spatial distribution patterns of fluid parameters.

[0064] S406 configures a data-driven loss function with physical constraints for the model framework. The physical constraints are constructed based on the continuity equation and the Navier-Stokes equation. The data-driven loss function is used to measure the error between the model's prediction results and the reference data obtained from computational fluid dynamics simulation.

[0065] It should be noted that the physical constraints are constructed based on the fundamental equations of fluid mechanics (continuity equation, Navier-Stokes equations) to ensure more accurate model predictions. The data-driven loss function uses methods such as mean squared error to calculate the error between the fluid parameters predicted by the model and the reference data generated by CFD simulation, driving the trained fluid parameter prediction model to optimize towards closer approximation of real physical results.

[0066] S408 uses the adaptive moment estimator optimizer Adam to train the model framework and obtain the fluid parameter prediction model.

[0067] It should be noted that the Adam optimizer is a gradient descent algorithm with an adaptive learning rate. It accelerates training convergence by automatically adjusting the learning rate, while reducing the variance of parameter updates. It is suitable for optimizing the complex parameter space of deep learning models.

[0068] The training process involves inputting the training dataset into the model framework, calculating prediction results through forward propagation, updating model parameters through backpropagation, and iterating repeatedly until the loss function converges to obtain the final fluid parameter prediction model. In some embodiments, an exponentially decaying learning rate scheduler is configured during training, with a decay factor γ = 0.99, and a gradient accumulation algorithm is used to handle the difference in the number of point clouds for different samples until the model converges.

[0069] In some embodiments of this disclosure, one or more of the following methods are included before the training dataset is input into the deep learning-based model framework, with the aim of performing augmentation operations on the training data.

[0070] 1. The 3D point cloud data is processed using the farthest point sampling (FPS) algorithm to make the point cloud data distribution more uniform.

[0071] It's important to note that FPS (Farthest Point Sampling) is a point cloud downsampling algorithm. It iteratively selects the point furthest from the already selected point set, generating a uniformly distributed subset of points. This ensures that key geometric features are preserved while reducing the amount of point cloud data. This reduces the amount of point cloud data input to the model, shortens training time, and lowers memory consumption. It also avoids the problem of localized point density and global feature loss caused by uneven mesh partitioning in the original point cloud, enabling the model to learn the overall structure of the target area in a balanced way.

[0072] 2. Apply random rigid transformations, including random rotation and random translation, to the 3D point cloud data to simulate the shape of the target part under different viewpoints and positions.

[0073] It should be noted that random rigid transformation applies geometric transformations to 3D point clouds without changing their shape, including random rotation and random translation. This can simulate changes in the spatial position of point clouds caused by patient positioning differences and equipment errors during medical image acquisition, expanding the sample space of training data; it forces the model to learn features independent of spatial position, avoiding prediction bias caused by absolute positional shifts in the point cloud, for example, enabling the model to accurately identify aortic point clouds at different scanning angles.

[0074] 3. Randomize the arrangement order of points in the 3D point cloud data to eliminate the influence of the data point arrangement order on model training.

[0075] It's important to note that randomizing the point cloud arrangement means shuffling the input order of points in a 3D point cloud, preventing the model from relying on a fixed point arrangement pattern during training. Traditional neural networks are sensitive to input order; random arrangement ensures that the model learns only based on the spatial coordinates and features of the points, rather than their arrangement order. This avoids the model memorizing irrelevant arrangement rules and improves its generalization ability to arbitrary point cloud arrangements.

[0076] The data augmentation operations described above expand the diversity of the training dataset and improve the generalization ability of the fluid parameter prediction model. These preprocessing steps significantly enhance the quality of the training dataset and the model's learning efficiency, laying a foundation for the accuracy of subsequent fluid parameter predictions.

[0077] In some embodiments of this disclosure, such as Figure 5 As shown, the output of fluid parameters at different times during the full cardiac cycle of the target location also includes the following steps:

[0078] S502 predicts point-by-point flow field parameters using a fluid parameter prediction model. The flow field parameters include velocity components and pressure scalars in three directions.

[0079] It should be noted that flow field parameters are physical quantities that describe the motion state of a fluid in space, including velocity components in three directions and a pressure scalar. The velocity components in the three directions are the velocity vector decomposition of the fluid in the Cartesian coordinate system, reflecting the flow velocity in different spatial directions; the pressure scalar is the static pressure value inside the fluid, which is directly related to blood flow resistance and vascular wall load.

[0080] By using a deep learning model to predict the three-dimensional velocity components and pressure values ​​of each discrete point within the target area, a high-resolution flow field data cube is constructed to capture the flow velocity differences in local areas. For different moments in the entire cardiac cycle, a snapshot of the flow field parameters at the corresponding moment is output to achieve a temporal characterization of the dynamic changes in blood flow.

[0081] S504 calculates the distribution of hemodynamic parameters at the target site at different times based on flow field parameters.

[0082] It should be noted that hemodynamic parameters, i.e., the fluid parameters to be output in this study, are specifically composite physical quantities derived from flow field parameters, used to quantitatively assess the interaction between blood flow state and vessel wall. Transforming raw flow field data into parameters with clear pathological significance directly serves clinical diagnosis. Calculating the distribution of hemodynamic parameters at different times will provide quantitative evidence for the subsequent understanding of the spatiotemporal evolution of hemodynamic abnormalities and for the study of pathological mechanisms. Outputting the parameter distribution at each time step and point throughout the entire cardiac cycle supports 4D dynamic analysis of hemodynamic characteristics.

[0083] (Three-dimensional space + time dimension), for example, observing the changing trend of false lumen blood flow retention time in a patient with aortic dissection within a cardiac cycle, providing dynamic evidence for assessing the risk of rupture.

[0084] In some embodiments of this disclosure, an example will be used to specifically illustrate how to predict fluid parameters. It should be noted that the fluid parameter prediction model in this embodiment establishes a mapping relationship between 3D point cloud data of the aorta and 4D fluid parameters of the aorta based on computational fluid dynamics. The prediction of aortic fluid parameters in the target area (human) includes the following steps:

[0085] S100: 3D modeling.

[0086] We collected preoperative and postoperative CTA imaging data from AD patients and then desensitized the data.

[0087] Mimics (version 21.0, Materialise, Leuven, Belgium) was used to perform 3D modeling of the desensitized CTA. After modeling, DetecFluid v1.2 was used for smoothing and mesh generation. All operations were automated. For the entire aorta, a mesh of about 3 million is usually used for calculation without much adjustment.

[0088] S200: CFD simulation.

[0089] CFD simulation of the entire aorta was performed using DetecFluid v1.2, with a full cardiac cycle assessment as the calculation mode. The calculation was performed using approximately 3 million grids. Specific physiological information of the patient (blood pressure, heart rate, stroke volume, systolic percentage, etc.) was input, and the simulation calculation was performed using a solver specifically designed for blood flow simulation scenarios. The calculation of multiple parameters can be completed at once, and a CSV file with vector data can be exported. All operations are automated and require minimal adjustments.

[0090] S300: Model building.

[0091] The point cloud data is downsampled using the Kpconv encoder, and contextual information is mined using a graph network. The point cloud data is then upsampled using the Kpconv decoder to output point-by-point features. Finally, 4D hemodynamic parameters are predicted using a multilayer perceptron.

[0092] Specifically, the model is built by first acquiring point cloud data of the patient's entire cardiac cycle, and then performing data augmentation to obtain point cloud data with a uniform number of points. Where t = 0.01 + Δt·m, t ∈ [0.01, T], Δt represents the time sampling step size, and T represents the total cardiac cycle.

[0093] Among them, the data augmentation strategies include the farthest point sampling algorithm FPS, which aims to ensure the uniformity of aortic sampling; and the application of a random rigid transformation T∈SE(3), where T is composed of the rotation matrix R∈SO(3) and the translation vector t∈R. 3 Composition; randomized arrangement of points.

[0094] Secondly, the Kpconv decoder is used to perform downsampling operations in the spatial dimension to generate super point cloud data. Simultaneously encoding multi-scale morphological features and temporal embeddings of the artery yields a high-dimensional feature matrix.

[0095] Next, a graph convolutional neural network is used to mine contextual information and output an enhanced feature matrix. Graph convolutional neural networks at each time step obtain the superpoint p from S302 in Euclidean space. i After connecting the neighborhood graph structure using the k-nn method, its features are updated as follows:

[0096]

[0097] Where h θThis represents a linear layer with instance normalization and LeakyReLU activation functions sequentially, cat[·,·] represents the concatenation operation, max(·) represents the max pooling operation on the neighbor dimension, and the final superpoint p i Features are represented as

[0098]

[0099] Then, the enhanced feature moments obtained in the above steps are... By combining NN-upsampling and linear layers with Kpconv downsampling skip connections, the output feature matrix has the same resolution as the input point cloud data.

[0100] Finally, the feature matrix The flow field parameters at each point are predicted using a multilayer perceptron. Where U contains vectors in the x, y, and z directions. x U y U z P is a pressure scalar.

[0101] S400: Construct the loss function and train the model.

[0102] Construct a data-driven loss function with physical constraints. Optimize model parameters using the backpropagation algorithm based on CFD simulation data results.

[0103] loss function in U x U y U z The average relative error of P, where y and These represent the CFD simulation results and the model prediction results, respectively. Represents the physical constraint loss function, where This represents the loss function corresponding to the continuity equation constraints. Let represent the loss function corresponding to the constraints of the Navier-Stocks equations. Here, ρ represents blood flow density, and v represents hemodynamic viscosity. All differential terms in the formula are implemented using automatic differentiation. By introducing a loss function with physical constraints, the model achieves accuracy while maintaining consistency with physics, reducing the number of samples required for training, and enhancing the model's generalization ability and robustness.

[0104] During model training, the Adam adaptive moment estimator optimizer was used as the training algorithm, with the initial learning rate set to 0.01 and a learning rate scheduler based on an exponential decay strategy configured.

[0105] (Attenuation factor γ = 0.99). Given the differences in the number of point clouds obtained by downsampling operations for each sample, which makes tensor stitching along the batch dimension impossible, this method introduces a gradient accumulation algorithm to achieve an expansion of the equivalent batch processing scale through a multi-step gradient summation mechanism.

[0106] S500: Hemodynamic parameter calculation.

[0107] The 4D flow field parameters predicted by S400 were used to calculate hemodynamic parameters, including wall shear stress (WSS), time-averaged wall shear stress (TAWSS), oscillating shear index (OSI), relative residence time (RRT), and helicity (Psi).

[0108] In the calculation of hemodynamic parameters, the formula for calculating WSS is as follows: Where v is the dynamic viscosity, and u is the viscosity. || - The component of blood flow velocity parallel to the wall, d ⊥ - The normal distance from the centroid of the boundary element to the wall. Where d ⊥ =d·n, where d represents the distance from the wall point to the nearest outermost wall point, and n = (n x ,n y ,n z ) is the point normal vector pointing outwards from that point (calculated using the open-source Open3D library). u || The calculation is as follows:

[0109]

[0110] WSS is approximately calculated using the following formula:

[0111]

[0112] The formula for calculating TAWSS is as follows: Where M represents the total time step. The OSI calculation formula is: The formula for calculating RRT is as follows: Based on the previous calculation result, direct algebraic operations are performed. The formula for calculating Psi is: in U - velocity vector, ω - eddy current vector.

[0113] In this embodiment, point cloud data compared to grid data can capture subtle changes in geometric boundaries; no data interpolation or extrapolation is used, thus preserving the accuracy of CFD data for neural network training; gridded data often faces challenges in selecting a suitable resolution due to differences in input data, resulting in high computational costs at high resolutions and difficulty in learning accurate representations of geometric boundaries at low resolutions. By applying randomized rigid transformations and point arrangement randomization, the network's learning ability for rigid transformation invariance and point arrangement invariance of point cloud outputs is enhanced; and the overfitting problem with limited samples is alleviated through farthest point downsampling technology.

[0114] This embodiment learns multi-scale morphological features of the aorta through a Kpconv-based decoder-encoder structure, enabling the network to perform region segmentation and local feature sampling, further learning the fine local features of small vessel structures. A graph convolutional neural network is constructed to mine contextual information from the point cloud and build neighborhood feature structures for superpoints. Coupled with the continuity equation and the Navier-Stokes equation into the loss function, both prediction and generalization performance are significantly improved. Other hemodynamic parameters are calculated using the predicted pressure and velocity fields, providing clinicians with multi-dimensional clinical indicators for diagnosing aortic diseases.

[0115] This embodiment ensures consistency between the model's predictions and real-world prior physical information by coupling physical constraints, improving the model's generalization ability and prediction accuracy. Furthermore, utilizing the time gradient term of the Navier-Stokes equations enables modeling in the time dimension, ensuring prediction accuracy across the entire cardiac cycle and overcoming the unavoidable information loss inherent in existing methods that only predict in low dimensions. While the PointNet-based network structure uses pooling operations to ensure the transformation invariance of learned features, it leads to significant feature loss. This network constructs a Kpconv-based point convolutional network to learn multi-scale features of the aorta, enabling region segmentation and local feature sampling, further learning fine local features of small vessel structures. By mining the contextual information of the point cloud through the constructed graph convolutional neural network and building a neighborhood feature structure for superpoints, it is expected to further improve the accuracy of flow field prediction.

[0116] It should be noted that the acquisition, storage, use, and processing of data in this disclosed technical solution comply with the relevant provisions of national laws and regulations. The various types of data, such as personal identity data, operational data, and behavioral data related to individuals, customers, and groups, obtained in the embodiments of this disclosure have all been authorized.

[0117] Based on the same inventive concept, this disclosure also provides a fluid parameter prediction device, as described in the following embodiments. Since the principle by which this device solves the problem is similar to that of the method embodiments described above, the implementation of this device embodiment can refer to the implementation of the method embodiments described above, and repeated details will not be repeated.

[0118] Figure 6 This diagram illustrates a fluid parameter prediction device according to an embodiment of the present disclosure, such as... Figure 6 As shown, the device includes:

[0119] The data acquisition module 601 is used to acquire medical imaging data and physiological information data of the object to be predicted. The medical imaging data is used to characterize the structural features of the target part of the object to be predicted, and the physiological information data is used to characterize the physiological state features of the object to be predicted.

[0120] The parameter output module 602 is used to input medical imaging data and physiological information data into the pre-trained fluid parameter prediction model and output the fluid parameters of the target site at different times of the whole cardiac cycle. The fluid parameter prediction model is built based on a deep learning model and trained by a data-driven loss function with physical constraints.

[0121] It should be noted that the examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in the above method embodiments. It should also be noted that the above modules, as part of an apparatus, can be executed in a computer system such as a set of computer-executable instructions.

[0122] Those skilled in the art will understand that various aspects of this disclosure can be implemented as a system, method, or program product. Therefore, various aspects of this disclosure can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or a combination of hardware and software aspects, collectively referred to herein as a "circuit," "module," or "system."

[0123] The following reference Figure 7 To describe an electronic device 700 according to such an embodiment of the present disclosure. Figure 7 The electronic device 700 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments disclosed herein.

[0124] like Figure 7 As shown, the electronic device 700 is manifested in the form of a general-purpose computing device. The components of the electronic device 700 may include, but are not limited to: at least one processing unit 710, at least one storage unit 720, and a bus 730 connecting different system components (including storage unit 720 and processing unit 710).

[0125] The storage unit stores program code that can be executed by the processing unit 710, causing the processing unit 710 to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure. For example, the processing unit 710 can perform the following steps of the above method embodiments: acquiring medical imaging data and physiological information data of the object to be predicted, wherein the medical imaging data is used to characterize the structural features of the target site of the object to be predicted, and the physiological information data is used to characterize the physiological state features of the object to be predicted; inputting the medical imaging data and physiological information data into a trained fluid parameter prediction model, and outputting the fluid parameters of the target site at different times of the whole cardiac cycle, wherein the fluid parameter prediction model is constructed based on a deep learning model and trained using a data-driven loss function with physical constraints.

[0126] Storage unit 720 may include a readable medium in the form of a volatile storage unit, such as random access memory (RAM) 7201 and / or cache memory 7202, and may further include a read-only memory (ROM) 7203.

[0127] The storage unit 720 may also include a program / utility 7204 having a set (at least one) program module 7205, such program module 7205 including but not limited to: an operating system, one or more application programs, other program modules and program data, each or some combination of these examples may include an implementation of a network environment.

[0128] Bus 730 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the various bus structures.

[0129] Electronic device 700 can also communicate with one or more external devices 740 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable a user to interact with electronic device 700, and / or with any device that enables electronic device 700 to communicate with one or more other computing devices (e.g., router, modem, etc.). This communication can be performed via input / output (I / O) interface 750. Furthermore, electronic device 700 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 760. As shown, network adapter 760 communicates with other modules of electronic device 700 via bus 730. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 700, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0130] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, terminal device, or network device, etc.) to execute the methods according to the embodiments of this disclosure.

[0131] In particular, according to embodiments of this disclosure, the process described above with reference to the flowchart can be implemented as a computer program product, which includes a computer program that, when executed by a processor, implements the methods described in the above embodiments.

[0132] In exemplary embodiments of this disclosure, a computer-readable storage medium is also provided, which may be a readable signal medium or a readable storage medium. A program product capable of implementing the methods described above is stored thereon. In some possible implementations, various aspects of this disclosure may also be implemented as a program product including program code, which, when run on a terminal device, causes the terminal device to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure.

[0133] More specific examples of computer-readable storage media in this disclosure may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0134] In this disclosure, a computer-readable storage medium may include a data signal propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of transmitting, propagating, or transmitting a program for use by or in connection with an instruction execution system, apparatus, or device.

[0135] Optionally, the program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0136] In practical implementation, program code for performing the operations of this disclosure can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0137] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0138] Furthermore, although the steps of the method in this disclosure are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.

[0139] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, mobile terminal, or network device, etc.) to execute the methods according to the embodiments of this disclosure.

[0140] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the appended claims.

Claims

1. A method for predicting fluid parameters, characterized in that, include: Acquire medical imaging data and physiological information data of the object to be predicted. The medical imaging data is used to characterize the structural features of the target site of the object to be predicted, and the physiological information data is used to characterize the physiological state features of the object to be predicted. The medical imaging data and physiological information data are input into the pre-trained fluid parameter prediction model, which outputs the fluid parameters of the target site at different times of the whole cardiac cycle. The fluid parameter prediction model is built based on a deep learning model and trained using a data-driven loss function with physical constraints.

2. The fluid parameter prediction method according to claim 1, characterized in that: The medical imaging data includes aortic geometric morphology data obtained based on computed tomography angiography (CTA), and the physiological information data includes at least one of blood pressure, heart rate, cardiac output, and systolic percentage.

3. The fluid parameter prediction method according to claim 1, characterized in that: The target site is the aorta, and the fluid parameters include at least one of wall shear stress WSS, time-averaged wall shear stress TAWSS, oscillating shear index OSI, relative residence time RRT, and helicity Psi.

4. The fluid parameter prediction method according to claim 1, characterized in that, Obtain medical imaging data and physiological information data of the object to be predicted, including: The medical image data is subjected to 3D modeling and mesh generation to generate a 3D model containing the geometric boundaries of the target area; The physiological information data is used as boundary conditions input into computational fluid dynamics (CFD) simulation to generate fluid parameter reference data corresponding to the medical imaging data.

5. The fluid parameter prediction method according to claim 1, characterized in that: The training process of the fluid parameter prediction model includes: The acquired medical image data is converted into 3D point cloud data, and a training dataset is constructed by combining it with the physiological information data. The training dataset is input into a deep learning-based model framework, which includes a Kpconv encoder-decoder structure capable of learning multi-scale features, and a graph convolutional neural network for mining point cloud contextual information. The model framework is configured with a data-driven loss function with physical constraints, which are constructed based on the continuity equation and the Navier-Stokes equation. The data-driven loss function is used to measure the error between the model prediction results and the reference data obtained from computational fluid dynamics simulation. The model framework is trained using the adaptive moment estimator optimizer Adam to obtain the fluid parameter prediction model.

6. The fluid parameter prediction method according to claim 5, characterized in that, Before inputting the training dataset into the deep learning-based model framework, one or more of the following are also included: The 3D point cloud data is processed using the farthest point sampling FPS algorithm; A random rigid transformation, including random rotation and random translation, is applied to the 3D point cloud data; Randomize the arrangement order of points in the 3D point cloud data.

7. The fluid parameter prediction method according to claim 1, characterized in that: Output the fluid parameters of the target site at different times during the full cardiac cycle, including: The fluid parameter prediction model predicts the flow field parameters corresponding to each point, and the flow field parameters include velocity components in three directions and pressure scalars. Based on the flow field parameters, the distribution of hemodynamic parameters of the target site at different times is calculated.

8. A fluid parameter prediction device, characterized in that, include: The data acquisition module is used to acquire medical imaging data and physiological information data of the object to be predicted. The medical imaging data is used to characterize the structural features of the target part of the object to be predicted, and the physiological information data is used to characterize the physiological state features of the object to be predicted. The parameter output module is used to input the medical imaging data and physiological information data into the pre-trained fluid parameter prediction model and output the fluid parameters of the target site at different times of the whole cardiac cycle. The fluid parameter prediction model is built based on a deep learning model and trained using a data-driven loss function with physical constraints.

9. An electronic device, characterized in that, include: processor; as well as Memory for storing the executable instructions of the processor; The processor is configured to execute the fluid parameter prediction method according to any one of claims 1 to 7 by executing the executable instructions.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the fluid parameter prediction method according to any one of claims 1 to 7.

11. A computer program product, comprising: A computer program or instruction, characterized in that, when executed by a processor, the computer program or instruction implements the fluid parameter prediction method according to any one of claims 1 to 7.

Images