Evaluation Method, Device, Computer Equipment, Medium and Product for Blood Flow Parameters

Abstract

The present application relates to a method, device, computer device, medium and product for evaluating blood flow parameters. The method includes: obtaining a medical image of a subject to be measured; the medical image includes a vascular structure; inputting the medical image into a preset recognition network for attribute recognition to obtain attribute parameters of the vascular structure; inputting the attribute parameters into a vascular fluid dynamics model for blood flow parameter evaluation to obtain blood flow parameters of the vascular structure. In this case, the attribute parameters of the vascular structure are recognized through a preset recognition network, and the blood flow parameters are evaluated based on the attribute parameters and the vascular fluid dynamics model; compared with the traditional method of evaluating blood pressure using hydraulic diameter, in this case, the vascular diameter at the vascular point is equivalently calculated through a neural network model, replacing the traditional hydraulic diameter, which can improve the accuracy of evaluating hemodynamic parameters.

Description

Technical Field

[0001] This application relates to the technical field of medical parameter analysis, and particularly to a method, device, computer device, medium, and product for evaluating blood flow parameters. Background Art

[0002] With the development of medical technology, as the medical imaging of the human vascular structure becomes more and more accurate, the blood flow evaluation of the human vascular structure based on the medical imaging of the human vascular structure has also become more and more important.

[0003] Traditionally, by constructing a three-dimensional vascular model and combining computational fluid dynamics (CFD), the blood flow parameters of the vascular structure are evaluated. Due to the complexity of constructing the three-dimensional model, a method of reducing the three-dimensional model to a one-dimensional model and combining CFD for blood flow evaluation has emerged.

[0004] However, the traditional method of combining a one-dimensional vascular model with CFD for blood flow evaluation has the problem of inaccurate blood flow parameter evaluation results. Summary of the Invention

[0005] Based on this, in view of the above technical problems, it is necessary to provide an evaluation method, device, computer device, computer-readable storage medium, and computer program product for blood flow parameters that can improve the accuracy of blood flow parameter evaluation results.

[0006] In a first aspect, this application provides an evaluation method for blood flow parameters, including:

[0007] Obtain a medical image of a subject to be measured; the medical image includes a vascular structure;

[0008] Input the medical image into a preset recognition network for attribute recognition to obtain attribute parameters of the vascular structure;

[0009] Input the attribute parameters into a vascular fluid dynamics model for blood flow parameter evaluation to obtain blood flow parameters of the vascular structure.

[0010] In one embodiment, the method further includes:

[0011] Obtain a sample medical image and corresponding gold standard blood flow parameters;

[0012] Input the sample medical image into an initial recognition network to obtain a recognition result;

[0013] Train the initial recognition network according to the gold standard blood flow parameters and the recognition result to obtain a preset recognition network.

[0014] In one embodiment, the initial recognition network is trained according to the gold standard blood flow parameters and the recognition results to obtain a preset recognition network, including:

[0015] Input the recognition result into the vascular fluid dynamics model to obtain intermediate blood flow parameters;

[0016] Determine the target loss according to the gold standard blood flow parameters and the intermediate blood flow parameters;

[0017] Adjust the parameters of the initial recognition network according to the target loss until the target loss meets the preset training conditions to obtain the preset recognition network.

[0018] In one embodiment, the method for obtaining the gold standard blood flow parameters includes:

[0019] Based on the vascular structure in the sample medical image, establish a three-dimensional vascular fluid dynamics model of the vascular structure;

[0020] Identify the center line of the vascular structure in the sample medical image to obtain the position information of the target center point of the vascular structure;

[0021] Input the position information of the target center point into the three-dimensional vascular fluid dynamics model for blood flow parameter calculation to obtain the gold standard blood flow parameters.

[0022] In one embodiment, inputting the position information of the target center point into the three-dimensional vascular fluid dynamics model for blood flow parameter calculation to obtain the gold standard blood flow parameters includes:

[0023] Input the position information of the target center point into the three-dimensional vascular fluid dynamics model for blood flow parameter calculation to obtain multiple candidate blood flow parameters around the target center point;

[0024] Process the multiple candidate blood flow parameters around the target center point to obtain the gold standard blood flow parameters.

[0025] In one embodiment, the method further includes:

[0026] Use the attribute parameters of the vascular structure in the sample medical image as the gold standard attribute parameters to train the preset neural network to obtain the initial recognition network.

[0027] In a second aspect, the present application further provides an evaluation device for blood flow parameters, including:

[0028] A first acquisition module, configured to acquire a medical image of a to-be-measured object; the medical image includes a vascular structure;

[0029] A first recognition module, configured to input the medical image into the preset recognition network for attribute recognition to obtain the attribute parameters of the vascular structure;

[0030] An evaluation module, configured to input attribute parameters into a vascular fluid dynamics model to evaluate blood flow parameters and obtain the blood flow parameters of a vascular structure.

[0031] In a third aspect, the present application further provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of the blood flow parameter evaluation method in the first aspect are implemented.

[0032] In a fourth aspect, the present application further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the blood flow parameter evaluation method in the first aspect are implemented.

[0033] In a fifth aspect, the present application further provides a computer program product, including a computer program. When the computer program is executed by a processor, the steps of the blood flow parameter evaluation method in the first aspect are implemented.

[0034] For the above-mentioned blood flow parameter evaluation method, device, computer device, storage medium, and computer program product, by acquiring a medical image of a to-be-measured object including a vascular structure, inputting the medical image into a preset recognition network for attribute recognition to obtain the attribute parameters of the vascular structure; then, inputting the attribute parameters into a vascular fluid dynamics model for blood flow parameter evaluation to obtain the blood flow parameters of the vascular structure. That is to say, the method proposed in the embodiments of the present application identifies the attribute parameters of the vascular structure through a preset recognition network, and evaluates the blood flow parameters based on the recognized attribute parameters and in combination with the vascular fluid dynamics model; based on this, for the traditional method of evaluating blood pressure based on hydraulic diameter and one-dimensional fluid dynamics model, in the present application, the neural network model is used to equivalently calculate the vascular diameter at a vascular point, and the calculated equivalent diameter is used to replace the traditional hydraulic diameter and substitute it into the one-dimensional fluid dynamics model to evaluate the blood pressure at this vascular point; since the neural network model can learn an equivalent diameter closer to the actual diameter, therefore, using the equivalent diameter in the present application can solve the drawbacks of the hydraulic diameter, and has more advantages in evaluating hemodynamics, and the hemodynamic parameters evaluated based on the equivalent diameter in the present application are closer to the hemodynamic parameters evaluated based on a complex three-dimensional model; therefore, using the method of the present application can reduce the computational complexity of the three-dimensional model while improving the accuracy of hemodynamic parameter evaluation. Description of the Drawings

[0035] To more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the related art. Obviously, the accompanying drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0036] Figure 1 It is an application environment diagram of the method for evaluating blood flow parameters in an embodiment;

[0037] Figure 2 It is a schematic flowchart of the method for evaluating blood flow parameters in an embodiment;

[0038] Figure 3 It is a schematic flowchart of the method for evaluating blood flow parameters in another embodiment;

[0039] Figure 4 It is a schematic flowchart of the method for evaluating blood flow parameters in another embodiment;

[0040] Figure 5 It is a schematic flowchart of the evaluation process of blood flow parameters in an embodiment;

[0041] Figure 6 It is a schematic flowchart of the training process of a deep learning network in an embodiment;

[0042] Figure 7 It is a structural block diagram of the device for evaluating blood flow parameters in an embodiment;

[0043] Figure 8 It is an internal structure diagram of a computer device in an embodiment. Detailed implementation manners

[0044] In order to make the objectives, technical solutions and advantages of the present application more clear and understandable, the following further details the present application in combination with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0045] In recent years, with the development of imaging technologies such as Computed Tomography (CT), Magnetic Resonance (MR), and Digital Subtraction Angiography (DSA), the acquisition of the human cardiovascular structure has become more and more accurate, and the functional research on this structure has become more and more important, especially the research on blood flow evaluation based on the structure of the human cardiovascular system.

[0046] Traditionally, blood flow assessment based on a three-dimensional model is based on the method of Computational Fluid Dynamics (CFD). By establishing a three-dimensional vascular model, meshing the model, determining boundary conditions such as pressure and flow rate, and then solving the discretized differential equation system based on the Navier-Stokes (N-S) equation, blood flow information such as blood pressure and flow velocity within a three-dimensional range is finally obtained.

[0047] However, due to the complexity of constructing the three-dimensional model and the calculation of CFD, blood flow assessment based on a reduced-order model has emerged. The three-dimensional vascular structure is replaced by a one-dimensional vascular model. By calculating some parameters, including the vessel diameter, based on the structure included in the three-dimensional vascular model, and applying these parameters to the one-dimensional model, the one-dimensional model indirectly obtains some information about the three-dimensional vascular structure. Then, calculations are performed according to the hydrodynamics formula of the one-dimensional model to obtain one-dimensional blood flow information.

[0048] In the method of blood flow assessment using a one-dimensional model, there is a model based on the hydraulic diameter, that is, when calculating the three-dimensional vascular diameter, the hydraulic diameter is used instead of the actual diameter; thus ensuring that for vessels with the same hydraulic diameter, their Reynolds numbers are the same. Among them, the hydraulic diameter can be defined as the ratio of the cross-sectional area of the vessel at a certain cross-section to the cross-sectional perimeter of that cross-section; for the vessels at the stenosis, it is impossible to directly identify the actual diameter of the vessel through an image. In this case, the hydraulic diameter can be used instead, and then substituted into the one-dimensional hydrodynamics formula to calculate the blood flow information of the vessels at the stenosis, such as blood pressure.

[0049] However, since the principle of the hydraulic diameter can only ensure that the frictional forces, that is, the resistances, experienced by vessels with the same hydraulic diameter are the same, but it cannot ensure that the pressure drops are the same. Therefore, the blood pressure evaluated based on the hydraulic diameter is not accurate. In addition, for the cross-section of the vessels in the lesion / stenosis area, the calculation of the cross-sectional area and cross-sectional perimeter of the vessel required for calculating the hydraulic diameter is not easy, and it is also difficult to ensure the accuracy of the calculation of the hydraulic diameter.

[0050] Furthermore, in related technologies, there is also a method of blood flow assessment based on deep neural network technology, that is, directly calculating hydrodynamics parameters through deep neural network technology; this type of method completely does not require a hydrodynamics model. However, hydrodynamics is a complex physical process, and the vascular structure forms and case characteristics are very diverse. Whether it can adapt to various complex vascular structures by completely abandoning the hydrodynamics model and using neural network training is still unknown.

[0051] Based on this, in the embodiments of the present application, a blood flow evaluation method with a simple calculation method and accurate evaluation results is proposed; that is, a one-dimensional hydrodynamic model based on the equivalent diameter is adopted, and the equivalent diameter of the blood vessel is calculated through deep learning technology to replace the traditional hydraulic diameter, and the three-dimensional blood vessel structures such as the cardiovascular and cerebrovascular vessels are equivalent to a one-dimensional blood vessel hydrodynamic model, and then the equivalent diameter is input into the one-dimensional blood vessel hydrodynamic model to calculate the blood flow parameters; this can simplify the calculation process of blood flow, improve the calculation speed, and has important significance in the software of the image-based blood vessel evaluation system. Among them, the equivalent diameter in this embodiment is an equivalent diameter obtained by learning through deep learning technology and is closer to the actual diameter of a certain cross-section of the blood vessel.

[0052] The blood flow parameter evaluation method provided by the embodiments of the present application can be applied to, for example, Figure 1 the application environment shown. Among them, the computer device 102 can be a terminal or a server; the terminal can include, but is not limited to, various personal computers, laptop computers, smart phones, tablet computers, and portable wearable devices, etc.; among them, the portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, etc. The server can be implemented by an independent server or a server cluster composed of multiple servers.

[0053] In an exemplary embodiment, as Figure 2 shown, a blood flow parameter evaluation method is provided. Taking the method applied to the Figure 1 computer device as an example, the following steps 202 to 206 are included. Among them:

[0054] Step 202, obtain the medical image of the object to be measured; the medical image includes the blood vessel structure.

[0055] Exemplarily, the medical image of the object to be measured can include, but is not limited to, CT images, MR images, DSA images, computer tomography angiography (CTA) images, etc.; the type of the medical image is not specifically limited in the embodiments of the present application.

[0056] Exemplarily, the medical image may include the blood vessel tissue of the object to be measured, that is, the blood vessel structure; among them, the blood vessel tissue may include at least one of artery blood vessels, coronary blood vessels, vein blood vessels, etc.

[0057] Exemplarily, the computer device can obtain the medical image of the object to be measured from a medical imaging scanning device, or can also obtain the medical image of the object to be measured from a Picture Archiving and Communication System (PACS); of course, the computer device can also obtain the medical image of the object to be measured from other storage devices or storage areas, or the computer device can also receive the medical image of the object to be measured sent by other devices, etc.

[0058] Step 204, input the medical image into a preset recognition network for attribute recognition to obtain the attribute parameters of the vascular structure.

[0059] Among them, the preset recognition network can be obtained by training an initial recognition network based on gold standard blood flow parameters; exemplarily, the gold standard blood flow parameters can be hemodynamic parameters determined based on a three-dimensional vascular fluid dynamics model. The attribute parameters of the vascular structure can include but are not limited to at least one of diameter, radius, cross-sectional area, and cross-sectional perimeter.

[0060] Exemplarily, the vascular structure can be the structure of a section of blood vessel, or can also be the vascular structure at a certain position or a certain blood vessel point; for example: the attribute parameters of the vascular structure can include the attribute parameters corresponding to one or more blood vessel points on the vascular structure, where the attribute parameters corresponding to the blood vessel point can include at least one of the blood vessel diameter, blood vessel radius, cross-sectional area, and cross-sectional perimeter of the blood vessel at the blood vessel point, etc.

[0061] Exemplarily, the computer device can input the medical image into a preset recognition network for attribute recognition of blood vessel points to obtain the attribute parameters of at least one blood vessel point on the vascular structure.

[0062] In one implementation, when the vascular structure in the medical image includes the structure of a section of blood vessel, the computer device can determine multiple blood vessel points on the vascular structure from the medical image, and for each blood vessel point, extract the Volume of Interest (VOI) image corresponding to the blood vessel point from the medical image; then, input the VOI image of the volume of interest corresponding to the blood vessel point into a preset recognition network for attribute recognition, so as to obtain the attribute parameters corresponding to the blood vessel point.

[0063] Exemplarily, when a computer device extracts multiple vascular points on a vascular structure in a medical image, it can first determine the vascular centerline corresponding to the vascular structure from the medical image; then, determine multiple vascular points on the vascular centerline. Among them, when the computer device determines the vascular centerline, it can input the medical image into a vascular segmentation model for vascular segmentation to obtain a vascular segmentation mask; then, by performing image processing on the vascular segmentation mask, the vascular centerline corresponding to the vascular structure can be obtained. Of course, the computer device can also use other image processing methods to process the medical image to obtain the vascular centerline corresponding to the vascular structure.

[0064] Exemplarily, for a preset recognition network, it can be trained based on at least one of a convolutional neural network (CNN), a fully connected network (FCN), a self-attention network, etc. For example: the preset recognition network can be composed of a convolutional neural network and a fully connected network, or it can also be composed of a self-attention network model based on a transformer network architecture.

[0065] Step 206, input the attribute parameters into a vascular hemodynamics model for blood flow parameter evaluation to obtain the blood flow parameters of the vascular structure.

[0066] Among them, the vascular hemodynamics model can be a vascular evaluation model based on computational fluid dynamics (CFD). The vascular hemodynamics model can be a three-dimensional CFD model or any reduced-dimensional CFD model, such as a one-dimensional CFD model, etc.

[0067] Exemplarily, the vascular hemodynamics model can be used to evaluate the blood flow parameters of the vascular structure. The blood flow parameters can include but are not limited to blood pressure, flow rate, wall shear stress, circumferential stress, etc. Exemplarily, the vascular hemodynamics model can include sub-hemodynamics models for evaluating different blood flow parameters, such as a sub-hemodynamics model for blood pressure evaluation, a sub-hemodynamics model for flow rate evaluation, etc. It should be noted that when different sub-hemodynamics models evaluate the corresponding blood flow parameters, the attribute parameters of the vascular structure required can be the same or different.

[0068] Exemplarily, in the case where the vascular fluid dynamics model includes one or more sub-fluid dynamics models, the computer device can determine the attribute parameters corresponding to each sub-fluid dynamics model, and input the determined attribute parameters into the corresponding sub-fluid dynamics model to evaluate the corresponding blood flow parameters, obtaining the corresponding blood flow parameter evaluation results.

[0069] In addition, it should be noted that in the case where the attribute parameters of the vascular structure include the attribute parameters of multiple vascular points, the computer device can, for each vascular point, determine the attribute parameters corresponding to different sub-fluid dynamics models from the attribute parameters of the vascular point; then, for each sub-fluid dynamics model, input the attribute parameters of the vascular point corresponding to the sub-fluid dynamics model into the sub-fluid dynamics model to evaluate the corresponding blood flow parameters, obtaining the evaluation results of the blood flow parameters of the vascular point.

[0070] Exemplarily, for blood pressure, in the case where the attribute parameters required for the target sub-fluid dynamics model for blood pressure evaluation include the vascular diameter at the vascular point, for each vascular point, the computer device can determine the vascular diameter corresponding to each vascular point from the attribute parameters of the vascular structure, and input the vascular diameter corresponding to each vascular point into the target sub-fluid dynamics model for blood pressure evaluation, thereby obtaining the blood pressure values at each vascular point.

[0071] In the above method for evaluating blood flow parameters, a medical image including a vascular structure of a subject to be measured is obtained, and the medical image is input into a preset recognition network for attribute recognition to obtain attribute parameters of the vascular structure. Among them, the preset recognition network is obtained by training an initial recognition network based on gold standard blood flow parameters. Then, the attribute parameters are input into a vascular fluid dynamics model for blood flow parameter evaluation to obtain blood flow parameters of the vascular structure. That is to say, the method proposed in the embodiments of the present application identifies the attribute parameters of the vascular structure through a preset recognition network, and evaluates the blood flow parameters based on the recognized attribute parameters and in combination with the vascular fluid dynamics model. Based on this, for the traditional method of blood pressure evaluation based on hydraulic diameter and one-dimensional fluid dynamics model, in the present application, the vascular diameter at a vascular point is equivalently calculated through a neural network model, and the calculated equivalent diameter is used to replace the traditional hydraulic diameter and substituted into the one-dimensional fluid dynamics model to evaluate the blood pressure at the vascular point. Since the neural network model can learn an equivalent diameter that is closer to the actual diameter, and moreover, the neural network model in the present application is trained based on gold standard blood flow parameters, which can further improve the accuracy of the equivalent diameter. Therefore, using the equivalent diameter in the present application can solve the drawbacks of the hydraulic diameter, is more advantageous for the evaluation of hemodynamics, and the hemodynamic parameters evaluated based on the equivalent diameter in the present application are closer to the hemodynamic parameters evaluated based on a complex three-dimensional model. Therefore, using the method of the present application can reduce the computational complexity of the three-dimensional model while improving the accuracy of hemodynamic parameter evaluation.

[0072] In an exemplary embodiment, a method for training a preset recognition network is provided. As Figure 3 shown, the above method may further include steps 302 to 306. Among them:

[0073] Step 302, obtain a sample medical image and corresponding gold standard blood flow parameters.

[0074] Exemplarily, the sample medical image may be a three-dimensional medical image including a vascular structure, such as a CT, MR, or DSA images at multiple different angles, etc. The gold standard blood flow parameters may include blood flow parameters corresponding to multiple vascular points on the vascular structure in the sample image.

[0075] Exemplarily, when the vascular fluid dynamics model in the above step 206 is a one-dimensional vascular fluid dynamics model, that is to say, in the embodiments of the present application, at least one attribute parameter related to the vascular morphology may be identified based on a preset recognition network, and based on the attribute parameter and the one-dimensional vascular fluid dynamics model, the blood flow parameters of the vascular points on the vascular structure are evaluated.

[0076] To improve the accuracy of the one-dimensional vascular fluid dynamics model, it is necessary to ensure the accuracy of the attribute parameters output by the preset recognition network. Therefore, the accuracy of the preset recognition network is crucial. When training the preset recognition network, although it is relatively easy to obtain the attribute parameters of normal vascular structures, it is more difficult to determine the attribute parameters of abnormal vascular structures such as lesions and stenosis. The focus of this preset recognition network is to accurately identify the attribute parameters for abnormal vascular structures. That is to say, if the theoretical attribute parameters of the vascular structure are used as the gold standard for network training to iteratively train the network, it will be more difficult to implement; and the recognition effect of the trained recognition network on the attribute parameters of abnormal vascular structures may not be good.

[0077] Therefore, in this example, a scheme is proposed to guide the training of the preset recognition network based on blood flow parameters, that is, based on hemodynamic parameters. That is to say, during the network training process, based on the attribute parameters output by the recognition network and the blood flow parameters calculated by the one-dimensional vascular fluid dynamics model, the loss value is calculated with the corresponding gold standard blood flow parameters, so as to adjust the parameters of the recognition network, thereby completing the network training and obtaining the preset recognition network.

[0078] Exemplarily, for the gold standard blood flow parameters, they can be calculated based on a three-dimensional vascular fluid dynamics model, that is, based on the blood flow assessment of the three-dimensional model to obtain the blood flow parameters of each vascular point in the sample medical image as the gold standard blood flow parameters.

[0079] Since this example is based on the three-dimensional CFD results to guide the training of the preset recognition network, and the vascular fluid dynamics model in step 206 above is a one-dimensional vascular fluid dynamics model, when the computer device obtains the three-dimensional CFD results based on the three-dimensional vascular fluid dynamics model, it is also necessary to map the three-dimensional CFD results to one dimension to obtain the blood flow parameters under the one-dimensional vascular fluid dynamics model, and use the blood flow parameters under the one-dimensional vascular fluid dynamics model as the gold standard blood flow parameters for training the preset recognition network. Among them, the blood flow parameters under the one-dimensional vascular fluid dynamics model include the blood flow parameters of each vascular point on the vascular center line in the sample medical image.

[0080] Step 304, input the sample medical image into the initial recognition network to obtain a recognition result.

[0081] Exemplarily, the computer device can respectively obtain the region-of-interest images corresponding to each blood vessel point on the blood vessel centerline from the sample medical image, and for each blood vessel point, input the region-of-interest image corresponding to the blood vessel point into the initial recognition network for recognition to obtain the attribute parameters corresponding to the blood vessel point. Similarly, the computer device can thus obtain the attribute parameters corresponding to each blood vessel point on the blood vessel centerline in the sample medical image, that is, the recognition result can include the attribute parameters corresponding to each blood vessel point on the blood vessel centerline in the sample medical image.

[0082] Step 306: Train the initial recognition network according to the gold standard blood flow parameters and the recognition result to obtain a preset recognition network.

[0083] Exemplarily, the computer device can input the recognition result into the blood vessel hydrodynamics model to obtain intermediate blood flow parameters, and determine the target loss according to the gold standard blood flow parameters and the intermediate blood flow parameters; then, adjust the parameters of the initial recognition network according to the target loss until the target loss meets the preset training conditions to obtain a preset recognition network.

[0084] Exemplarily, in the case where the recognition result includes the attribute parameters of multiple blood vessel points on the blood vessel centerline, the computer device can respectively input the attribute parameters of each blood vessel point into the blood vessel hydrodynamics model for blood flow evaluation to obtain the intermediate blood flow parameters corresponding to each blood vessel point; then, the computer device can input the intermediate blood flow parameters corresponding to each blood vessel point and the gold standard blood flow parameters corresponding to each blood vessel point into a preset loss function to calculate the target loss, and perform iterative training on the initial recognition network based on the target loss, and finally obtain a preset recognition network.

[0085] Among them, the intermediate blood flow parameters can be the blood flow parameters calculated based on the attribute parameters output by the recognition network during the training process; after each iteration of optimizing the recognition network, the attribute parameters at the blood vessel point will be re-recognized based on the optimized recognition network, and based on the newly recognized attribute parameters corresponding to the blood vessel point, the new blood flow parameters corresponding to the blood vessel point will be obtained through the calculation of the blood vessel hydrodynamics model; then, compare the new blood flow parameters corresponding to the blood vessel point with the gold standard blood flow parameters of the blood vessel point to obtain a new target loss, and optimize the optimized recognition network again based on the new target loss, so as to realize the continuous iterative training of the recognition network.

[0086] In this embodiment, the computer device obtains a sample medical image and the corresponding gold standard blood flow parameters, and inputs the sample medical image into the initial recognition network to obtain a recognition result; then, based on the gold standard blood flow parameters and the recognition result, the initial recognition network is trained to obtain a preset recognition network. That is, in this embodiment, hemodynamic parameters are used to guide the training of the recognition network, rather than using the gold standard corresponding to the attribute parameters directly output by the recognition network for network training. On the one hand, it can solve the problems of difficult acquisition of the gold standard corresponding to the attribute parameters of abnormal vascular structures, large network training difficulty and poor network training effect. On the other hand, by using the three-dimensional vascular hydrodynamics model to guide the training of the recognition network, the recognition network obtained by training can be combined with the one-dimensional vascular hydrodynamics model to achieve higher accuracy in evaluating blood flow parameters; so that in later applications, using the high-precision preset recognition network and the one-dimensional vascular hydrodynamics model can accurately perform hemodynamic evaluation on the blood vessels with any abnormal structure, which can not only simplify the computational complexity of blood flow evaluation, but also improve the accuracy and evaluation efficiency of blood flow evaluation.

[0087] In an exemplary embodiment, for the step of obtaining the gold standard blood flow parameters in step 302 above, the computer device can construct a three-dimensional vascular hydrodynamic model based on the sample medical image, and determine the blood flow parameters of each vascular point on the center line of the blood vessels in the sample medical image based on this three-dimensional vascular hydrodynamic model as the gold standard blood flow parameters. Based on this, as Figure 4 shown, the method of obtaining the gold standard blood flow parameters in step 302 above may include steps 402 to 406. Among them:

[0088] Step 402, based on the vascular structure in the sample medical image, establish a three-dimensional vascular hydrodynamics model of the vascular structure.

[0089] Exemplarily, the computer device can extract the vascular structure from the sample medical image and establish a three-dimensional vascular model corresponding to the vascular structure; then, perform mesh division on the three-dimensional vascular model, and combine the preset boundary conditions and relevant parameter settings to establish a three-dimensional vascular hydrodynamics model of the vascular structure; based on this three-dimensional vascular hydrodynamics model, perform mesh calculation to obtain the hydrodynamic parameters corresponding to each three-dimensional grid point, that is, the blood flow parameters corresponding to each three-dimensional grid point; among them, when performing mesh calculation based on this three-dimensional vascular hydrodynamics model, CFD technology can be used, and according to different input boundary conditions and settings, hydrodynamic parameters including pressure, flow velocity, wall shear stress, circumferential stress, etc. can be calculated.

[0090] Step 404, identify the center line of the vascular structure in the sample medical image to obtain the position information of the target center point of the vascular structure.

[0091] Exemplarily, when mapping three-dimensional CFD results to one dimension, the blood flow parameters of each three-dimensional grid on the vascular structure are mapped to the blood flow parameters corresponding to each vascular point on the vascular centerline corresponding to the vascular structure. Based on this, while the computer device evaluates the blood flow parameters of the vascular structure in the sample medical image using a three-dimensional vascular hydrodynamics model, it also needs to determine each vascular point on the vascular centerline corresponding to the vascular structure in the sample medical image, and further determine the blood flow parameters corresponding to each vascular point on this vascular centerline based on the results of the blood flow evaluation, that is, the blood flow parameters of each three-dimensional grid on the vascular structure.

[0092] Exemplarily, the computer device can extract the vascular centerline corresponding to the vascular structure from the sample medical image and determine a plurality of vascular points on the vascular centerline, so as to obtain the position information of each vascular point on the vascular centerline.

[0093] It should be noted that when calculating the loss of the initial recognition network, the loss should often be calculated based on the gold standard blood flow parameters and the intermediate blood flow parameters of the same vascular point. Therefore, when determining the gold standard blood flow parameters, it should first be determined which vascular point the region of interest image input to the initial recognition network corresponds to. Then, the gold standard blood flow parameters corresponding to this vascular point should be determined accordingly; that is, the gold standard blood flow parameters of the vascular point corresponding to the current input are determined, and the vascular point corresponding to the current input can be used as the target center point of the vascular structure.

[0094] Step 406, input the position information of the target center point into the three-dimensional vascular hydrodynamics model for blood flow parameter calculation to obtain the gold standard blood flow parameters.

[0095] Exemplarily, the computer device can input the position information of the target center point into the three-dimensional vascular hydrodynamics model to calculate the gold standard blood flow parameters corresponding to this target center point.

[0096] Exemplarily, the computer device can input the position information of the target center point into the three-dimensional vascular hydrodynamics model for blood flow parameter calculation to obtain a plurality of candidate blood flow parameters around the target center point; then, the plurality of candidate blood flow parameters around the target center point can be processed to obtain the gold standard blood flow parameters corresponding to this target center point.

[0097] Based on the three-dimensional vascular fluid dynamics model, the blood flow parameters corresponding to each grid point on the vascular structure can be obtained. Based on this, when it is necessary to determine the blood flow parameters corresponding to the target center point, the computer device can determine the blood flow parameters of multiple grid points within the preset range of the target center point on the vascular structure based on the position information of the target center point, and use them as multiple candidate blood flow parameters around the target center point. Then, the blood flow parameters of these multiple grid points can be averaged, and the average blood flow parameter can be used as the gold standard blood flow parameter corresponding to the target center point. For example, the average blood pressure of multiple grid points within the preset range of the target center point can be used as the gold standard blood pressure corresponding to the target center point.

[0098] Exemplarily, the computer device can also determine the grid point closest to the target center point and use the blood flow parameter of this closest grid point as the gold standard blood flow parameter corresponding to the target center point.

[0099] In the same way, the gold standard blood flow parameters corresponding to each blood vessel point on the blood vessel center line can be determined. In this way, the three-dimensional CFD results corresponding to the three-dimensional blood vessel model can be mapped to the blood vessel evaluation results corresponding to the one-dimensional blood vessel center line model. Based on this, for each blood vessel point on the blood vessel center line, the image of the region of interest corresponding to the blood vessel point can be input into the initial recognition network for recognition to obtain the attribute parameters of the blood vessel point. Then, the attribute parameters of the blood vessel point can be input into the vascular fluid dynamics model for evaluation to obtain the intermediate blood flow parameters corresponding to the blood vessel point. According to the intermediate blood flow parameters corresponding to the blood vessel point and the gold standard blood flow parameters corresponding to the blood vessel point, the loss corresponding to the initial recognition network can be calculated, and based on this loss, the training of the initial recognition network can be realized.

[0100] It should be noted that when calculating the network loss in each iteration process, the loss can be calculated based on one or more blood vessel points on the blood vessel center line, and no specific limitation is made in this embodiment of the present application.

[0101] In this embodiment, when determining the gold standard blood flow parameters corresponding to the blood vessel points on the blood vessel center line, the computer device can establish a three-dimensional blood vessel hydrodynamics model based on the blood vessel structure in the sample medical image; identify the center line of the blood vessel structure in the sample medical image to obtain the position information of the target center point of the blood vessel structure; then, input the position information of the target center point into the three-dimensional blood vessel hydrodynamics model for blood flow parameter calculation to obtain the gold standard blood flow parameters corresponding to the target center point. That is, in this embodiment, the method of mapping the three-dimensional blood vessel hydrodynamics model to one dimension is used to determine the gold standard blood flow parameters of each blood vessel point; equivalently, the three-dimensional accurate blood flow evaluation result is used as the gold standard for training to train the preset recognition network and the one-dimensional blood vessel hydrodynamics model, so that after the one-dimensional blood vessel hydrodynamics model uses the attribute parameters identified by the preset recognition network to evaluate the blood flow parameters, a hemodynamic effect equivalent to the three-dimensional blood flow evaluation result can be obtained; therefore, by using this method, the training effect of the preset recognition network can be improved, the accuracy of the preset recognition network can be improved, and further the accuracy of blood flow evaluation can be improved.

[0102] In an exemplary embodiment, in the training process of the above-mentioned preset recognition network, the error between the evaluation result of the three-dimensional blood vessel hydrodynamics model and the evaluation result of the one-dimensional blood vessel hydrodynamics model is used as the loss of network training, or called the cost function; instead of using the gold standard of the attribute parameters directly calculated by the network to train the network, it can effectively avoid the problem of difficult network training caused by the difficulty of obtaining the gold standard of the attribute parameters of the blood vessel structure. However, there is also a problem. Compared with the method of directly training the network based on the result of the network and the corresponding gold standard, the loss calculation chain in the above training process is relatively long, increasing the training duration.

[0103] To further improve the training efficiency of the preset recognition network and make the network converge faster, in this example, the computer device can also pre-train the network first, and then, for the pre-trained network, use the method shown above Figure 4 to train again until the preset training conditions are met to obtain the preset recognition network.

[0104] Exemplarily, the computer device can use the attribute parameters of the blood vessel structure in the sample medical image as the gold standard attribute parameters to train the preset neural network to obtain the initial recognition network. That is, the computer device can obtain the attribute parameters manually marked by the user or marked according to experience as the gold standard attribute parameters; based on the gold standard attribute parameters, pre-train the preset neural network for a period of time to obtain the above-mentioned initial recognition network.

[0105] Although the gold-standard attribute parameters manually labeled are not necessarily accurate, especially when it cannot be guaranteed that their hydrodynamic characteristics meet the requirements of one-dimensional vascular hydrodynamic model calculations, they can still provide a certain degree of approximation. After pre-training, the neural network can already fit the gold-standard attribute parameters manually labeled well, and then the neural network is migrated to the Figure 4 training process shown above, allowing the neural network to fit the network parameters according to the hydrodynamic results of the one-dimensional vascular hydrodynamic model; using this method, the neural network can converge to the target value faster and better, greatly improving the training speed of the preset recognition network and further reducing the training difficulty of the network.

[0106] In an exemplary embodiment, taking the blood flow parameters including blood pressure (or pressure) as an example, a specific embodiment of blood flow parameter evaluation is provided. In the process of blood pressure evaluation, a one-dimensional vascular hydrodynamic model based on the equivalent diameter is involved; among them, the equivalent diameter can be identified by a trained deep learning network. Referring to Figure 5 shown, the steps of blood flow parameter evaluation may include:

[0107] Step 1, obtain the medical image of the object to be measured, perform vascular segmentation on the medical image to obtain a vascular segmentation mask, and based on the vascular segmentation mask, obtain the vascular centerline, and multiple vascular points can be determined from the vascular centerline.

[0108] Step 2, extract the local region of interest (VOI) images of each vascular point from the medical image.

[0109] Step 3, input the local VOI images of each vascular point into the trained deep learning network for recognition to obtain the attribute parameters of each vascular point; among them, the attribute parameters may include the equivalent diameter at the vascular point.

[0110] Optionally, for each vascular point, the local VOI image corresponding to the vascular point and the corresponding local vascular segmentation mask can also be input into the deep learning network for recognition to obtain the equivalent diameter corresponding to the vascular point.

[0111] Step 4, for each vascular point, input the attribute parameters of the vascular point into the one-dimensional vascular hydrodynamic model for blood pressure evaluation, so as to obtain the blood pressure value at the vascular point.

[0112] Exemplarily, the one-dimensional vascular hydrodynamic model can be represented by the following formula (1) or a deformation of formula (1).

[0113] (1)

[0114] Among them, ΔP represents the pressure drop across a blood vessel segment, which can be the blood vessel segment between the initial blood vessel point and the target blood vessel point; generally, it is an abnormal blood vessel segment with stenosis. V represents the average blood flow velocity; c1 and c2 are the viscosity and expansion loss coefficient, respectively representing the blocking effect of the friction of the blood vessel wall on the blood flow and the blocking effect of stenosis on the blood flow.

[0115] Exemplarily, c1 can be expressed by the following formula (2) or a variant of formula (2).

[0116] (2)

[0117] Among them, represents the blood viscosity, L tot represents the length of the blood vessel segment, A in represents the area at the beginning of the lumen, D represents the diameter at the stenosis of the lumen; for the blood viscosity , it can be a fixed value; the length L tot of the blood vessel segment and the area A in at the beginning of the lumen are both parameters independent of the stenosis degree. For the length L tot of the blood vessel segment, it can be arbitrarily selected by the user. Since there is no stenosis at the beginning of the lumen, it can be approximated as circular, and its area can be calculated according to the area calculation formula of a circle.

[0118] Therefore, for the coefficient c1, the parameter related to the stenosis degree only includes the diameter D at the stenosis of the lumen; due to the existence of stenosis, the lumen here is not circular, and the diameter D here should be the diameter after equivalent circularization of the non-circular lumen.

[0119] In the traditional method, the hydraulic diameter is used as the diameter D at the stenosis of the lumen. The calculation of the hydraulic diameter is the ratio of the cross-sectional area A of the flow domain at the stenosis to the perimeter P of the cross-section of the blood vessel at the stenosis, that is, D = A / P. For the hydraulic diameter, it not only cannot accurately represent the actual diameter at the stenosis, but also the calculation method of the cross-sectional area A of the flow domain in the hydraulic diameter is relatively complex.

[0120] Therefore, in this example, a deep learning network is used to calculate the equivalent diameter at the stenosis.

[0121] Exemplarily, c2 can be expressed by the following formula (3) or a variant of formula (3).

[0122] (3)

[0123] Among them, represents the blood concentration, is an empirical parameter.

[0124] Therefore, when determining the one-dimensional vascular hydrodynamic model based on the above formulas (2) and (3), that is, obtaining formula (1), the attribute parameters required for the one-dimensional vascular hydrodynamic model may include blood viscosity , the length L of the blood vessel segment tot , the area A at the beginning of the lumen in , the equivalent diameter D at the blood vessel point, and the blood concentration ; among them, only the equivalent diameter D at the blood vessel point is not easy to determine and can be obtained by using the above deep learning network; for other attribute parameters, they can be determined based on the blood vessel morphology in the medical image; among them, for the empirical parameters, they can be determined according to the external measurement information of the object to be measured, and the external measurement information may include but is not limited to information such as the age, gender, blood pressure, and diabetes of the object to be measured.

[0125] In another example, c2 can also be expressed by the following formula (4) or a variant of formula (4).

[0126] (4)

[0127] Among them, represents blood density, Ke is an empirical parameter, A in represents the area at the beginning of the lumen, A sten represents the area at the stenosis of the lumen.

[0128] Assume that the one-dimensional vascular hydrodynamic model is determined by the above formulas (2) and (4) to obtain formula (1). Then, the parameters related to the stenosis degree of the blood vessel in the one-dimensional vascular hydrodynamic model include the diameter D at the stenosis of the lumen and the area A at the stenosis of the lumen sten . For this, the above deep learning network can be used to simultaneously identify the equivalent diameter D at the blood vessel point and the cross-sectional area A at the blood vessel point sten .

[0129] Exemplarily, as shown in Figure 6 , it shows a training flow chart of a deep learning network. Taking the training of the equivalent diameter as an example, in this example, the equivalent diameter D is trained by coupling the deep learning network and the one-dimensional vascular hydrodynamic model. The equivalent diameter corresponding to each blood vessel point on the blood vessel center line can be calculated by the trained deep learning network. Exemplarily, the deep learning network can be composed of a convolutional neural network and a fully connected network, or can be composed of a transformer self-attention network model, etc. The input of the deep learning network is the local region of interest image within a preset range of the blood vessel point, and the output is the attribute parameter corresponding to the blood vessel point, including but not limited to the equivalent diameter D.

[0130] Refer to Figure 6As described above, training a deep learning network involves three parts, namely, a three-dimensional vascular fluid dynamics model part, a one-dimensional reduced-order vascular fluid dynamics model part, and a deep learning network part.

[0131] The three-dimensional vascular fluid dynamics model is used to generate the gold standard attribute parameters corresponding to the sample medical images, including the gold standard vascular diameter. Exemplarily, the computer device can extract the coronary artery from the input sample medical image, establish a three-dimensional vascular model of the coronary artery; then divide the grid according to the three-dimensional vascular model of the coronary artery to establish a three-dimensional vascular fluid dynamics model; perform grid calculations based on the three-dimensional vascular fluid dynamics model to obtain the fluid dynamics parameters of each grid point on the three-dimensional grid, that is, obtain the three-dimensional fluid dynamics results.

[0132] Among them, when evaluating blood flow based on the three-dimensional vascular fluid dynamics model, CFD technology can be used, and according to different input boundary conditions and different parameter settings, attribute parameters such as pressure (blood pressure), flow velocity, wall shear stress, and circumferential stress can be calculated.

[0133] Furthermore, in order to use the three-dimensional fluid dynamics results to guide the output of the one-dimensional vascular fluid dynamics model, it is also necessary to map the three-dimensional fluid dynamics results to one dimension. Exemplarily, the computer device can extract the vascular centerline from the sample medical image, for each vascular point on the vascular centerline, determine one or more grid points within the preset range of the vascular point, and based on the fluid dynamics parameters corresponding to the one or more grid points, determine the fluid dynamics parameters corresponding to the vascular point on the vascular centerline, so as to obtain the blood flow parameters on the one-dimensional centerline fluid dynamics model, that is, obtain the gold standard blood flow parameters.

[0134] Based on the above one-dimensional vascular fluid dynamics model, it can be seen that the pressure drop at the vascular point can be calculated based on the equivalent diameter, that is, the blood pressure (that is, pressure) at the vascular point can be calculated; therefore, for the gold standard blood flow parameters, the computer device can determine the pressure corresponding to each vascular point on the vascular centerline based on the pressure corresponding to each grid point as the gold standard pressure value corresponding to each vascular point.

[0135] In addition, local region of interest images corresponding to each vascular point are extracted from the sample medical images, and the local region of interest images corresponding to each vascular point are respectively input into the initial deep learning network for equivalent diameter recognition to obtain the initial equivalent diameter corresponding to each vascular point. Then, the initial equivalent diameter corresponding to each vascular point is respectively input into the above one-dimensional vascular fluid dynamics model for pressure calculation to obtain the pressure drop corresponding to each vascular point, and based on the pressure at the vascular starting point and each pressure drop, the intermediate pressure value corresponding to each vascular point is calculated. Furthermore, the loss is calculated based on the intermediate pressure value corresponding to each vascular point and the gold standard pressure value corresponding to each vascular point, and the gradient is backpropagated according to the loss to iteratively train the initial deep learning network until the preset training conditions are met, obtaining the trained deep learning network. Thus, this deep learning network has the function of recognizing the equivalent diameter.

[0136] It should be noted that when the deep learning network needs to recognize other attribute parameters, the above training process can also be adopted. For the deep learning network, it can output one or more different attribute parameters. Of course, different deep learning networks can also be used to respectively output different attribute parameters.

[0137] Exemplarily, after training is completed, the testing phase can be entered. In the testing phase, the test medical images can be preprocessed, including but not limited to extracting the local region of interest images corresponding to each vascular point on each vascular centerline. Then, the local region of interest images corresponding to each vascular point are input into the trained deep learning network for recognition to obtain the equivalent diameter corresponding to each vascular point. The equivalent diameter corresponding to each vascular point is respectively input into the above one-dimensional vascular fluid dynamics model for calculation, so as to obtain the pressure drop and pressure value corresponding to each vascular point.

[0138] In this example, first, a one-dimensional vascular fluid dynamics model simulating vascular fluid is established, and then the accurate results obtained from three-dimensional CFD calculations are used as the training set to train the equivalent diameter of the one-dimensional vascular fluid dynamics model, so that after the one-dimensional vascular fluid dynamics model uses this equivalent diameter for calculation, it has a hemodynamic effect equivalent to that of the three-dimensional CFD results. In addition, by combining the iterative process of deep learning and the calculation process of the one-dimensional vascular fluid dynamics model to train the equivalent diameter, the hemodynamic results of three-dimensional CFD are used as the training set, and the difference between the hemodynamic results of the one-dimensional vascular fluid dynamics model and the hemodynamic results of the three-dimensional vascular fluid dynamics model is used as the cost function to train the equivalent diameter of the one-dimensional vascular fluid dynamics model.

[0139] By adopting this method, it can not only retain the characteristics of simplicity and rapidity of the one-dimensional vascular fluid dynamics model, but also have the accuracy of the three-dimensional vascular fluid dynamics model. Compared with the calculation of the traditional hydraulic diameter, the calculation of the equivalent diameter has more advantages in calculating hemodynamic parameters such as the fractional flow reserve (FFR) of coronary arteries, and the blood flow assessment result is more accurate.

[0140] It should be understood that although each step in the flowcharts involved in the above-described embodiments is sequentially shown as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise clearly stated in this article, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least a part of other steps or steps or stages in other steps.

[0141] Based on the same inventive concept, an embodiment of the present application further provides an apparatus for evaluating blood flow parameters for implementing the above-described method for evaluating blood flow parameters. The solution provided by this apparatus for solving problems is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the apparatus for evaluating blood flow parameters provided below can refer to the limitations on the method for evaluating blood flow parameters in the above text, and will not be repeated here.

[0142] In an exemplary embodiment, as Figure 7 shown, an apparatus for evaluating blood flow parameters is provided, including: a first acquisition module 702, a first recognition module 704, and an evaluation module 706, wherein:

[0143] The first acquisition module 702 is configured to acquire a medical image of a to-be-tested object; the medical image includes a vascular structure.

[0144] The first recognition module 704 is configured to input the medical image into a preset recognition network for attribute recognition to obtain attribute parameters of the vascular structure.

[0145] The evaluation module 706 is configured to input the attribute parameters into a vascular fluid dynamics model for blood flow parameter evaluation to obtain blood flow parameters of the vascular structure.

[0146] In one of the embodiments, the apparatus further includes:

[0147] A second acquisition module, configured to acquire a sample medical image and corresponding gold standard blood flow parameters;

[0148] A second recognition module, configured to input a sample medical image into an initial recognition network to obtain a recognition result;

[0149] A first training module, configured to train the initial recognition network according to the gold standard blood flow parameters and the recognition result to obtain a preset recognition network.

[0150] In one embodiment, the first training module includes:

[0151] A processing sub-module, configured to input the recognition result into a vascular fluid dynamics model to obtain intermediate blood flow parameters;

[0152] A determination sub-module, configured to determine a target loss according to the gold standard blood flow parameters and the intermediate blood flow parameters;

[0153] An adjustment sub-module, configured to adjust the parameters of the initial recognition network according to the target loss until the target loss meets a preset training condition to obtain a preset recognition network.

[0154] In one embodiment, the second acquisition module includes:

[0155] An establishment sub-module, configured to establish a three-dimensional vascular fluid dynamics model of a vascular structure based on the vascular structure in the sample medical image;

[0156] A recognition sub-module, configured to recognize the center line of the vascular structure in the sample medical image to obtain the position information of the target center point of the vascular structure;

[0157] A calculation sub-module, configured to input the position information of the target center point into the three-dimensional vascular fluid dynamics model to calculate blood flow parameters to obtain gold standard blood flow parameters.

[0158] In one embodiment, the calculation sub-module includes:

[0159] A calculation unit, configured to input the position information of the target center point into the three-dimensional vascular fluid dynamics model to calculate blood flow parameters to obtain a plurality of candidate blood flow parameters around the target center point;

[0160] A processing unit, configured to process a plurality of candidate blood flow parameters around the target center point to obtain gold standard blood flow parameters.

[0161] In one embodiment, the device further includes:

[0162] A second training module, configured to use the attribute parameters of the vascular structure in the sample medical image as gold standard attribute parameters to train a preset neural network to obtain an initial recognition network.

[0163] Each module in the above-mentioned blood flow parameter evaluation device can be implemented in whole or in part by software, hardware, or a combination thereof. Each of the above modules can be embedded in the processor of the computer device in hardware form or be independent of it, or be stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each of the above modules.

[0164] In an exemplary embodiment, a computer device is provided. The computer device can be a terminal or a server. When the computer device is a server, its internal structure diagram can be as Figure 8 shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data such as a trained preset recognition network and a one-dimensional vascular fluid dynamics model. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a method for evaluating blood flow parameters.

[0165] Those skilled in the art can understand that Figure 8 the structure shown in

[0166] is only a block diagram of some structures related to the solution of this application, and does not constitute a limitation on the computer device to which the solution of this application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0167] In an embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the method for evaluating blood flow parameters in any of the above embodiments.

[0168] In one embodiment, a computer program product is provided, including a computer program which, when executed by a processor, implements the steps of the method for evaluating blood flow parameters in any of the above embodiments.

[0169] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0170] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memories can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in this application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., and are not limited thereto. The processors involved in the embodiments provided in this application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., and are not limited thereto.

[0171] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0172] The above-described embodiments merely represent several implementation manners of the present application. The description is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A method for evaluating blood flow parameters, characterized in that The method includes: Obtaining a medical image of an object to be measured; the medical image includes a vascular structure; Inputting the medical image into a preset recognition network for attribute recognition to obtain attribute parameters of the vascular structure; the attribute parameters of the vascular structure at least include at least one of the vascular diameter, vascular radius, cross-sectional area, and cross-sectional perimeter of the vascular structure; Inputting the attribute parameters into a vascular fluid dynamics model for blood flow parameter evaluation to obtain blood flow parameters of the vascular structure; the vascular fluid dynamics model is a one-dimensional vascular evaluation model based on computational fluid dynamics; The method further includes: Obtaining a sample medical image and corresponding gold standard blood flow parameters; Inputting the sample medical image into an initial recognition network to obtain a recognition result; Inputting the recognition result into the vascular fluid dynamics model to obtain intermediate blood flow parameters; Determining a target loss according to the gold standard blood flow parameters and the intermediate blood flow parameters; Adjusting the parameters of the initial recognition network according to the target loss until the target loss meets a preset training condition to obtain the preset recognition network.

2. The method according to claim 1, wherein The method for obtaining the gold standard blood flow parameters includes: Based on the vascular structure in the sample medical image, establishing a three-dimensional vascular fluid dynamics model of the vascular structure; Identifying the centerline of the vascular structure in the sample medical image to obtain the position information of the target center point of the vascular structure; Inputting the position information of the target center point into the three-dimensional vascular fluid dynamics model for blood flow parameter calculation to obtain the gold standard blood flow parameters.

3. The method according to claim 2, wherein The step of inputting the position information of the target center point into the three-dimensional vascular fluid dynamics model for blood flow parameter calculation to obtain the gold standard blood flow parameters includes: Inputting the position information of the target center point into the three-dimensional vascular fluid dynamics model for blood flow parameter calculation to obtain a plurality of candidate blood flow parameters around the target center point; Processing the plurality of candidate blood flow parameters around the target center point to obtain the gold standard blood flow parameters.

4. The method according to claim 1, wherein The method further includes: Using the attribute parameters of the vascular structure in the sample medical image as gold standard attribute parameters to train a preset neural network to obtain the initial recognition network.

5. An evaluation device for blood flow parameters, characterized in that, The device includes: A first acquisition module, configured to acquire a medical image of an object to be measured; the medical image includes a vascular structure; A first recognition module, configured to input the medical image into a preset recognition network for attribute recognition to obtain attribute parameters of the vascular structure; the attribute parameters of the vascular structure at least include at least one of the vascular diameter, vascular radius, cross-sectional area, and cross-sectional perimeter of the vascular structure; An evaluation module, configured to input the attribute parameters into a vascular fluid dynamics model for blood flow parameter evaluation to obtain blood flow parameters of the vascular structure; the vascular fluid dynamics model is a one-dimensional vascular evaluation model based on computational fluid dynamics; The device further includes: A second acquisition module, configured to acquire a sample medical image and corresponding gold standard blood flow parameters; A second recognition module, configured to input the sample medical image into an initial recognition network to obtain a recognition result; The first training module includes: A processing sub-module for inputting the recognition result into the vascular fluid dynamics model to obtain intermediate blood flow parameters; A determination sub-module for determining a target loss according to the gold standard blood flow parameters and the intermediate blood flow parameters; An adjustment sub-module for adjusting the parameters of the initial recognition network according to the target loss until the target loss meets a preset training condition to obtain the preset recognition network.

6. The device according to claim 5, characterized in that The second acquisition module includes: A construction sub-module for constructing a three-dimensional vascular fluid dynamics model of the vascular structure based on the vascular structure in the sample medical image; A recognition sub-module for recognizing the center line of the vascular structure in the sample medical image to obtain the position information of the target center point of the vascular structure; A calculation sub-module for inputting the position information of the target center point into the three-dimensional vascular fluid dynamics model for blood flow parameter calculation to obtain the gold standard blood flow parameters.

7. The device according to claim 6, characterized in that, The calculation sub-module includes: A calculation unit for inputting the position information of the target center point into the three-dimensional vascular fluid dynamics model for blood flow parameter calculation to obtain a plurality of candidate blood flow parameters around the target center point; A processing unit for processing the plurality of candidate blood flow parameters around the target center point to obtain the gold standard blood flow parameters.

8. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, the steps of the method according to any one of claims 1 to 4 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 4 are implemented.

10. A computer program product comprising a computer program, characterized in that, When the computer program is executed by the processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images