Coronary vessel naming model training method and device, electronic equipment and storage medium
By obtaining the image characteristics of coronary blood vessels and heart chambers, the coronary blood vessel naming model was trained, and the problem of poor naming of mutant blood vessels was solved, and the rapid and accurate naming of coronary blood vessels was achieved.
Patent Information
- Application Number
- CN202310658058.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-06-02
- Publication Date
- 2024-12-03
AI Technical Summary
The existing coronary vascular naming technology is poorly robust when dealing with mutated blood vessels, and cannot achieve accurate naming, which affects the accuracy of the coronary vascular model.
By obtaining the image characteristics of the coronary blood vessels and the heart chamber, the characteristics of the distance between the center line and the heart chamber are calculated, the coronary blood vessel naming model is trained, and the model parameters are adjusted to achieve accurate naming of mutated blood vessels.
The rapid and automated naming of mutated blood vessels in coronary blood vessels is achieved, the accuracy of blood vessel naming is improved, and the problem of low naming accuracy in the prior art is solved.
Smart Images

Figure CN119068282A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to coronary vessel image processing technology, and more particularly to a coronary vessel naming model training method, a coronary vessel naming method, a device, an electronic device and a storage medium. Background Art
[0002] In the medical field, stents, as important interventional devices for treating vascular stenosis, have been widely used in the field of cardiovascular diseases. Coronary stent implantation involves dilating the narrowed part of the coronary artery through interventional means, and then inserting a metal stent to support the narrowed part, so that the narrowed blood vessel wall expands outward. After the stent is implanted, the supporting blood vessel remains in a continuously open state, so that the blood flow of the coronary artery is smooth. However, the process of implanting the stent requires first accurately modeling the coronary vessels, and at the same time accurately naming the coronary vessels appearing in the coronary vessel model. With the development of computer technology, the application scenarios of using DSA (Digital Subtraction Angiography) equipment to assist in vascular research are becoming more and more extensive. For example, a coronary vessel model for simulating vascular morphology can be constructed based on the vascular image output by the DSA device. However, in the related technology for naming coronary vessels, for coronary vessels with variations, when naming the coronary vessels in the coronary vessel model, the centerline data will be first extracted as the input for naming. Then, according to the characteristics of the centerline data, conditions are set to determine the name of the blood vessel. However, this method has poor robustness and cannot achieve accurate naming of the coronary vessels, thus affecting the accuracy of the coronary vessel model. Summary of the invention
[0003] In view of this, the embodiments of the present application provide a coronary vessel naming model training method, a coronary vessel naming method, an apparatus, an electronic device and a storage medium, which can determine the model parameters of the coronary vessel naming model by training the coronary vessel naming model, so that the trained coronary vessel naming model can not only quickly and automatically name the vessels, but also realize the naming of the variant vessels in the coronary vessels, thereby solving the defect of low accuracy of vessel naming in the related technology.
[0004] The technical solution of the embodiment of the present application is implemented as follows:
[0005] The present application embodiment provides a coronary vessel naming model training method, the method comprising:
[0006] Obtain angiographic images of the coronary vessels and cardiac chambers;
[0007] Acquiring a centerline feature of the coronary artery according to the angiography image of the coronary artery;
[0008] Acquiring the cardiac chamber feature according to the cardiac chamber image;
[0009] Calculating the distance between the centerline of the coronary blood vessel and the heart chamber to obtain a distance feature;
[0010] Training a coronary vessel naming model according to the centerline feature, the heart chamber feature, and the distance feature to obtain a prediction effect of the coronary vessel naming model;
[0011] According to the prediction effect, the model parameters of the coronary vessel naming model are adjusted to achieve the naming of the coronary vessels in the angiography image of the coronary vessels through the coronary vessel naming model.
[0012] The present application also provides a method for naming a coronary vessel, the method comprising:
[0013] acquiring angiographic images of target coronary vessels;
[0014] extracting the centerline of the target coronary vessel;
[0015] The node name of the center line of the target coronary vessel is predicted by using a coronary vessel naming model to obtain a predicted name of the target coronary vessel, wherein the coronary vessel naming model is trained according to the above embodiment.
[0016] The embodiment of the present application also provides a coronary vessel naming model training device, characterized in that the device comprises:
[0017] An information transmission module, used for acquiring angiographic images of coronary vessels and images of cardiac chambers;
[0018] An information processing module, used for acquiring the centerline feature of the coronary blood vessel according to the angiography image of the coronary blood vessel;
[0019] The information processing module is used to obtain the cardiac chamber characteristics according to the cardiac chamber image;
[0020] The information processing module is used to calculate the distance between the center line of the coronary blood vessel and the heart chamber to obtain a distance feature;
[0021] A training module, used for training the coronary vessel naming model according to the centerline feature, the heart chamber feature and the distance feature, to obtain a prediction effect of the coronary vessel naming model;
[0022] The training module is used to adjust the model parameters of the coronary vessel naming model according to the prediction effect, so as to realize the naming of the coronary vessels in the angiography image of the coronary vessels through the coronary vessel naming model.
[0023] In some embodiments, the training module is used to denoise the centerline features to obtain a first training sample set that matches the coronary vessel naming model, wherein the first training sample set includes at least one group of training samples;
[0024] The training module is used to encode the first training sample through the encoder block of the coronary vessel naming model to obtain an encoding result of the first training sample;
[0025] The training module is used to iteratively process the encoding result of the first training sample through the core block of the coronary vessel naming model to obtain the hidden information of the first training sample;
[0026] The training module is used to decode the encoding result of the first training sample through the encoder block of the coronary vessel naming model when a preset number of iterations is reached, so as to obtain a first prediction effect of the coronary vessel naming model.
[0027] In some embodiments, the training module is used to decode the encoding result of the first training sample through the encoder block of the coronary vessel naming model to obtain the predicted name of the training sample;
[0028] The training module is used to obtain label information matching the first training sample;
[0029] The training module is used to detect the predicted name according to the label information to obtain a first prediction effect of the coronary vessel naming model.
[0030] In some embodiments, the training module is used to combine the centerline feature, the heart chamber feature and the distance feature to obtain a second training sample set that matches the coronary vessel naming model, wherein the second training sample set includes at least one group of training samples;
[0031] The training module is used to adjust the depth of the fully connected layer and the width of the fully connected layer in the perceptron module of the coronary vessel naming model, and to add classification labels corresponding to the second training sample set;
[0032] The training module is used to encode the second training sample through the encoder block of the coronary vessel naming model to obtain an encoding result of the second training sample;
[0033] The training module is used to iteratively process the encoding result of the second training sample through the core block of the coronary vessel naming model to obtain the hidden information of the second training sample;
[0034] The training module is used to decode the encoding result of the second training sample through the encoder block of the coronary vessel naming model when a preset number of iterations is reached to obtain a second prediction result of the coronary vessel naming model.
[0035] In some embodiments, the training module is used to obtain a first accuracy rate, a first precision rate, and a first recall rate in the first prediction result;
[0036] The training module is used to obtain a second accuracy, a second precision and a second recall in the second prediction result;
[0037] The training module is used to adjust the model parameters of the coronary vessel naming model when the second accuracy rate is greater than the first accuracy rate, the second precision rate is greater than the first precision rate, and the second recall rate is greater than the first recall rate, wherein the model parameters of the coronary vessel naming model include:
[0038] The depth of the fully connected layers in the perceptron module, the width of the fully connected layers in the perceptron module, the number of core block iterations, and the batch parameter.
[0039] The present application also provides a coronary vessel naming device, the device comprising:
[0040] A signal transmission module, used for acquiring angiographic images of target coronary vessels;
[0041] A signal processing module, used for extracting the centerline of the target coronary vessel;
[0042] The signal processing module is used to predict the node name of the center line of the target coronary vessel through the coronary vessel naming model to obtain the predicted name of the target coronary vessel.
[0043] The embodiment of the present application further provides an electronic device, characterized in that the electronic device includes:
[0044] A memory for storing executable instructions;
[0045] The processor is used to implement the aforementioned coronary vessel naming model training method or coronary vessel naming method when running the executable instructions stored in the memory.
[0046] An embodiment of the present application also provides a computer-readable storage medium storing executable instructions, characterized in that when the executable instructions are executed by a processor, the aforementioned coronary vessel naming model training method or coronary vessel naming method is implemented. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1A schematic diagram of a usage scenario of the coronary vessel naming model training method provided in an embodiment of the present application;
[0048] Figure 2 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application;
[0049] Figure 3 is a schematic diagram of the structure of an electronic device for executing the coronary vessel naming method provided in the present application, provided in an embodiment of the present application;
[0050] Figure 4 An optional flowchart of the coronary vessel naming model training method provided in an embodiment of the present application;
[0051] Figure 5a This is an optional schematic diagram of a coronary tree in an embodiment of the present application;
[0052] Figure 5b It is an optional schematic diagram of a centerline tree in an embodiment of the present application;
[0053] Figure 6 Schematic diagram of atrium and ventricle in the embodiment of the present application;
[0054] Figure 7 This is a schematic diagram of the intersection line of the interventricular groove in the embodiment of the present application;
[0055] Figure 8 Schematic diagram of different center lines in the coronary blood vessels in the embodiments of the present application;
[0056] Fig. 9 This is a schematic diagram of the structure of the coronary vessel naming model in the embodiment of the present application;
[0057] Fig.10 Schematic diagram of the training process of the coronary vessel naming model in the embodiment of the present application;
[0058] Fig.11 Schematic diagram of the training process of the coronary vessel naming model in the embodiment of the present application;
[0059] Fig.12 It is a schematic diagram of the prediction effect in the first prediction result;
[0060] Fig.13 It is a schematic diagram of the prediction effect in the first prediction result;
[0061] Fig.14 This is a schematic diagram of the coronary vessel naming process in the embodiment of the present application;
[0062] Fig.15 It is a schematic diagram showing the result of naming coronary vessels according to an embodiment of the present application. DETAILED DESCRIPTION
[0063] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The described embodiments should not be regarded as limiting the present application. All other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of this application.
[0064] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.
[0065] Before further describing the embodiments of the present application in detail, the nouns and terms involved in the embodiments of the present application are explained. The nouns and terms involved in the embodiments of the present application are subject to the following interpretations.
[0066] 1) Computed Tomographic Angiography (CTA) is a non-invasive method for evaluating the vascular system. Coronary CTA is used to check whether the coronary arteries are narrowed and thus affect the blood supply to the heart to cause coronary heart disease. During the test, the contrast agent is injected into the coronary artery. The image of the contrast agent under X-ray is used to diagnose the coronary blood flow by taking advantage of the fact that X-rays cannot penetrate the contrast agent. Traditional linear target extraction methods are difficult to extract vascular information coherently and accurately from the complex background of coronary CTA images.
[0067] 2) Pass is used to indicate the conditions or states on which the executed operations depend. When the dependent conditions or states are met, one or more operations executed may be in real time or have a set delay. Unless otherwise specified, there is no restriction on the order in which the multiple operations executed may be executed.
[0068] 3) Cloud technology refers to a hosting technology that unifies hardware, software, network and other resources in a wide area network or local area network to achieve data computing, storage, processing and sharing. The general term for network technology, information technology, integration technology, management platform technology, application technology, etc. applied through the cloud computing business model can form a resource pool, which can be used on demand and is flexible and convenient. Cloud computing technology will become an important support. The background services of technical network systems require a large amount of computing and storage resources, such as video websites, picture websites and more portal websites. With the high development and application of the Internet industry, in the future, each item may have its own identification mark, and all need to be transmitted to the background system for logical processing. Data of different levels will be processed separately. All kinds of industry data require strong system backing support, which can only be achieved through cloud computing.
[0069] 4) Coronary angiography: If systolic stenosis of the coronary artery or delayed diastolic relaxation is found, it indicates the presence of a myocardial bridge. However, coronary angiography can only detect those myocardial bridges that have a significant impact on coronary blood flow. The detection of a myocardial bridge is related to its length, the direction of the bridge fibers, and the tissue between the bridge and the related arteries. Some myocardial bridges are difficult to detect by angiography because the proximal coronary artery is almost completely occluded or the fixed stenosis caused by atherosclerosis limits the blood perfusion of the coronary artery, thereby masking its systolic stenosis, or because of the presence of vasospasm. Coronary angiography often cannot detect atherosclerotic stenosis at the myocardial bridge.
[0070] 5) Intracoronary Doppler examination found that the blood flow velocity of the coronary artery in the myocardial bridge increased significantly in the early diastole, forming a peak value, which quickly decreased and then formed a platform until it decreased again in the systole. The peak value appeared due to the existence of the maximum perfusion pressure in the coronary artery, and the corresponding continuous decrease in the vascular area, resulting in a significant pressure gradient at both ends of the myocardial bridge. When the myocardial bridge relaxes in diastole, the pressure gradient at both ends disappears, the vascular area expands rapidly, and the blood flow velocity also decreases rapidly.
[0071] Before introducing the coronary vessel naming model training method provided by this application, Figure 1 The implementation environment shown is briefly introduced to briefly introduce the defects in the related technology. The inventor found in the research that the related technology has the following defects when analyzing the coronary arteries of the heart:
[0072] 1) Solution 1: Extract the centerline data of the coronary vessels as the input for naming. According to the centerline data characteristics of the coronary vessels, set conditions to determine the vessel name of the coronary vessels. However, this method has poor robustness and cannot handle the variant centerline tree of the coronary vessels, making the variant coronary vessels impossible to name.
[0073] 2) Solution 2: Using the naming method of three-dimensional template matching. The drawback of this method is that a suitable template needs to be provided before naming the coronary vessels. The choice of template has a great impact on the final result. At the same time, the matching accuracy is low, and it cannot cover large-scale data and cannot adapt to the clinical use requirements of coronary vessel naming.
[0074] 3) Solution 3: Use a two-dimensional tree structure to name the coronary vessels. This method relies on accurately abstracting the three-dimensional coronary vascular tree into a two-dimensional structure and modeling it. However, it is impossible to accurately name the vessels in the case of vascular variations (congenital vascular malformations or deformation caused by external force damage).
[0075] In order to solve the above-mentioned defects, the present application provides a coronary vessel naming model training method, which can use the computed tomography angiography images of the coronary vessels to accurately name all blood vessels (including normal coronary vessels and atypical coronary vessels) appearing in the coronary vessel model.
[0076] refer to Figure 1 , Figure 1 The schematic diagram of the use scenario of the coronary vessel naming model training method provided in the embodiment of the present application, the implementation environment of the present application may include: a computer device 11 and a vascular image acquisition device 12. The computer device 11 and the vascular image acquisition device 12 may be directly or indirectly connected to obtain the image of the coronary vessel through wired or wireless communication, and the coronary vessel image may be stored in the server 200 for easy access and use by medical personnel.
[0077] The vascular image acquisition device 12 is used to acquire images of the blood vessels, and then send the acquired vascular images to the computer device 11, which constructs a coronary vascular model based on the vascular images and names all blood vessels appearing in the coronary vascular model. Exemplarily, the image acquisition device 12 can be a DSA device, which is used to subtract the non-contrast vascular image from the contrast-enhanced vascular image to obtain a vascular image with structures other than the blood vessels eliminated. It can also be a coronary CT image processing system with a projection image calculation function, which is used to determine the coronary vascular model of the target object. The computer device 11 can also deploy a trained coronary vascular naming model to name the normal coronary vessels and anomalous coronary vessels in the coronary vascular model of the target object.
[0078] Exemplarily, the computer device 11 may be a terminal or a server. Optionally, the terminal may be any electronic product that can interact with a user through one or more methods such as a keyboard, a touch pad, a touch screen, a remote control, voice interaction or a handwriting device, such as a PC (Personal Computer), a mobile phone, a smart phone, a PDA (Personal Digital Assistant), a wearable device, a PPC (Pocket PC), a tablet computer, a smart car machine, a smart TV, a smart speaker, a car terminal, etc. The server may be a single server, a server cluster consisting of multiple servers, or a cloud computing service center.
[0079] In some embodiments, the computer device 11 and the vascular image acquisition device 12 can be integrated into one, so that the vascular image acquisition and coronary vessel naming model training can be realized on one device, and the coronary vessel naming method provided in the present application can be executed.
[0080] Those skilled in the art should understand that the above-mentioned computer device 11 and vascular image acquisition device 12 are only examples, and other existing or future computer devices and vascular image acquisition devices, if applicable to the present application, should also be included in the protection scope of the present application, and the present application does not impose any specific restrictions on this.
[0081] The coronary vessel naming model training method provided in the embodiment of the present application can be applied to any scenario where the vessel name in the coronary vessel model needs to be determined, including but not limited to:
[0082] 1) Applied to the coronary CT image processing system, image reconstruction and analysis are performed through coronary CTA images, and the established coronary vascular model is used to automatically locate coronary stenosis, myocardial bridges and stent positions and their data, thereby assisting doctors in simulating coronary stent implantation and implanting coronary stents into designated locations of coronary vessels, saving users time in manually reconstructing coronary arteries and improving the modeling efficiency of the coronary model.
[0083] 2) Applied to the coronary CT image processing system, the coronary vessels are reconstructed and the blood flow reserve fraction is calculated through coronary CTA images. The determined coronary vessel model can be used to quickly and accurately reconstruct the three-dimensional image and accurately calculate the FFR value of the entire coronary tree.
[0084] 3) Applied to the coronary CT image processing system, during coronary angiography or PCI, the determined coronary vascular model can be used to quickly and accurately reconstruct the coronary DSA coronary tree in three dimensions and accurately locate the stenosis, providing doctors with coronary morphological data and a reference for the coronary lumen condition; the determined coronary vascular model can be used to provide a variety of coronary functional data and give a quantitative analysis of the degree of myocardial ischemia.
[0085] 4) Using the established coronary artery model, through data processing of CT images, multi-dimensional vascular reconstruction images are provided for clinical diagnosis and treatment decisions, and quantifiable vascular morphology data analysis is output, such as automatic stenosis judgment or the location of specific lesions in the coronary artery, thereby assisting doctors in diagnosis and making the determination of lesion location more visual.
[0086] In some embodiments, for complex coronary vessels or distorted coronary vessels, the coronary vessel naming model training method provided in the present application can be used to collect images and model the coronary vessels at the same position at different angles through the DSA device, obtain two coronary vessel models, and accurately name the normal coronary vessels and the abnormal coronary vessels in each coronary vessel model, so that the doctor can know the position and name of each blood vessel in the coronary vessel model at different angles through the visualized coronary vessel model display, so as to facilitate the further study of the distortion or complex structure of the coronary vessel based on the reconstructed coronary vessel model and the coronary vessel name. DSA is a technology that obtains obvious vascular images through subtraction, enhancement and other technologies. This technology is of great significance for the diagnosis of lesions in blood vessels and is often used to diagnose cardiovascular and cerebrovascular stenosis and coronary artery variation.
[0087] Among them, the embodiments of the present application can be implemented in combination with cloud technology. Cloud technology refers to a hosting technology that unifies a series of resources such as hardware, software and network in a wide area network or a local area network to realize data calculation, storage, processing and sharing. It can also be understood as a general term for network technology, information technology, integration technology, management platform technology and application technology applied through a cloud computing business model. The background services of the technical network system require a large amount of computing and storage resources, such as video websites, picture websites and more portal websites, so cloud technology needs to be supported by cloud computing.
[0088] It should be noted that cloud computing is a computing model that distributes computing tasks on a resource pool composed of a large number of computers, so that various application systems can obtain computing power, storage space and information services as needed. The network that provides resources is called the "cloud". From the user's point of view, the resources in the "cloud" can be infinitely expanded, and can be obtained at any time, used on demand, expanded at any time, and paid for by use. As a basic capability provider for cloud computing, a cloud computing resource pool platform will be established, referred to as a cloud platform, generally referred to as Infrastructure as a Service (IaaS), and various types of virtual resources will be deployed in the resource pool for external customers to choose to use. The cloud computing resource pool mainly includes: computing devices (which can be virtualized machines, including operating systems), storage devices and network devices.
[0089] Combined with the embodiment Figure 1 As shown, the coronary vascular modeling method provided in the embodiment of the present application can be implemented through corresponding cloud devices, for example: the computer device 11 and the vascular image acquisition device 12 can be integrated into one, so that the acquisition of vascular images and coronary vascular modeling can be realized on one device, and can be connected to the server 200 through the network 300, and the network 300 can be a wide area network or a local area network, or a combination of the two. It is worth noting that the server 200 can be a physical device or a virtualized device, that is, the server storing the acquired vascular images and coronary vascular models can be either an independent medical data server or a corresponding server in a cloud server cluster that processes medical data.
[0090] Specifically, combined Figure 1 In the illustrated embodiment, the server 200 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers for processing medical data, 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 communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The terminal may be a smart phone, a tablet computer, a laptop computer, a desktop computer, and a medical device that can perform parameter recording, etc., but is not limited thereto. The terminal and the server may be directly or indirectly connected via wired or wireless communication, and this application is not limited thereto.
[0091] The coronary artery model and corresponding coronary artery name of the target object established by the coronary artery naming model training method provided in the present application can be saved in the cloud server, so that they can be called at any time when performing auxiliary diagnosis on the target object, thereby reducing the time-consuming re-labeling of the coronary artery names.
[0092] The structure of the electronic device in the embodiment of the present application is described in detail below. The electronic device can be implemented in various forms, such as a medical image processing server with coronary vessel image processing, in which a trained coronary vessel naming model is deployed, such as the server 200 in the aforementioned embodiment. Figure 2 The schematic diagram of the composition structure of the electronic device provided in the embodiment of the present application can be understood as follows: Figure 2 Only exemplary structures of electronic devices are shown, not all structures, and may be implemented as needed. Figure 2 Partial or complete structure shown.
[0093] The electronic device provided in the embodiment of the present application includes: at least one processor 201, a memory 202, a user interface 203 and at least one network interface 204. The various components in the electronic device 20 are coupled together through a bus system 205. It can be understood that the bus system 205 is used to realize the connection and communication between these components. In addition to the data bus, the bus system 205 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, in Figure 2 Various buses are labeled as bus system 205 .
[0094] The user interface 203 may include a display, a keyboard, a mouse, a trackball, a click wheel, keys, buttons, a touch pad or a touch screen.
[0095] It is understood that the memory 202 may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. The memory 202 in the embodiment of the present application can store data to support the following: Figure 1 The operation of the computer device 11 and the vascular image acquisition device 12 shown. Examples of such data include: any computer program used to operate on the computer device 11 and the vascular image acquisition device 12, such as an operating system and an application program. Among them, the operating system includes various system programs, such as a framework layer, a core library layer, a driver layer, etc., which are used to implement various basic services and process tasks through hardware. The application program can include various application programs.
[0096] In some embodiments, the coronary vessel naming model training device provided in the embodiments of the present application can be implemented in a combination of software and hardware. As an example, the coronary vessel naming model training device provided in the embodiments of the present application can be a processor in the form of a hardware decoding processor, which is programmed to execute the coronary vessel naming model training method provided in the embodiments of the present application. For example, the processor in the form of a hardware decoding processor can adopt one or more application specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field programmable gate arrays (FPGAs), or other electronic components.
[0097] As an example of a coronary vessel naming model training device provided in an embodiment of the present application being implemented by a combination of software and hardware, the coronary vessel naming model training device provided in an embodiment of the present application can be directly embodied as a combination of software modules executed by a processor 201, and the software module can be located in a storage medium, and the storage medium is located in a memory 202. The processor 201 reads the executable instructions included in the software module in the memory 202, and combines with necessary hardware (for example, including the processor 201 and other components connected to the bus 205) to complete the coronary vessel naming model training method provided in an embodiment of the present application.
[0098] As an example, the processor 201 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., where the general-purpose processor can be a microprocessor or any conventional processor, etc.
[0099] As an example of hardware implementation of the coronary vessel naming model training device provided in the embodiment of the present application, the device provided in the embodiment of the present application can be directly executed by a processor 201 in the form of a hardware decoding processor, for example, one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field programmable gate arrays (FPGAs), or other electronic components to implement the coronary vessel naming model training method provided in the embodiment of the present application.
[0100] The memory 202 in the embodiment of the present application is used to store various types of data to support the operation of the electronic device 20. Examples of such data include: any executable instructions for operating on the electronic device 20, such as executable instructions, and the program for implementing the coronary vessel naming model training method of the embodiment of the present application can be included in the executable instructions.
[0101] In other embodiments, the coronary vessel naming model training device provided in the embodiments of the present application can be implemented in a software manner. Figure 2 The coronary vessel naming model training device 2020 stored in the memory 202 is shown, which may be software in the form of a program and a plug-in, and includes a series of modules. As an example of a program stored in the memory 202, the coronary vessel naming model training device 2020 may be included. The coronary vessel naming model training device 2020 includes the following software modules:
[0102] Information transmission module 2081, used to obtain angiographic images of coronary vessels and images of cardiac chambers;
[0103] An information processing module 2082 is used to obtain the centerline characteristics of the coronary vessels according to the angiography images of the coronary vessels;
[0104] An information processing module 2082 is used to obtain cardiac chamber features according to the cardiac chamber image;
[0105] An information processing module 2082 is used to calculate the distance between the center line of the coronary blood vessel and the heart chamber to obtain a distance feature;
[0106] A training module 2083 is used to train the coronary vessel naming model according to the centerline feature, the heart chamber feature and the distance feature to obtain the prediction effect of the coronary vessel naming model;
[0107] The training module 2083 is used to adjust the model parameters of the coronary vessel naming model according to the prediction effect, so as to realize the naming of the coronary vessels in the angiography image of the coronary vessels through the coronary vessel naming model.
[0108] In the above solution, the training module 2083 is used to denoise the centerline features to obtain a first training sample set that matches the coronary vessel naming model, wherein the first training sample set includes at least one group of training samples;
[0109] The training module 2083 is used to encode the first training sample through the encoder block of the coronary vessel naming model to obtain the encoding result of the first training sample;
[0110] The training module 2083 is used to iteratively process the encoding result of the first training sample through the core block of the coronary vessel naming model to obtain the hidden information of the first training sample;
[0111] The training module 2083 is used to decode the encoding result of the first training sample through the encoder block of the coronary vessel naming model when a preset number of iterations is reached, so as to obtain a first prediction effect of the coronary vessel naming model.
[0112] In the above solution, the training module 2083 is used to decode the encoding result of the first training sample through the encoder block of the coronary vessel naming model to obtain the predicted name of the training sample;
[0113] The training module 2083 is used to obtain label information matching the first training sample;
[0114] The training module 2083 is used to detect the predicted name according to the label information to obtain the first prediction effect of the coronary vessel naming model.
[0115] In the above solution, the training module 2083 is used to combine the centerline feature, the heart chamber feature and the distance feature to obtain a second training sample set that matches the coronary vessel naming model, wherein the second training sample set includes at least one group of training samples;
[0116] The training module 2083 is used to adjust the depth and width of the fully connected layer in the perceptron module of the coronary vessel naming model, and to add classification labels corresponding to the second training sample set;
[0117] The training module 2083 is used to encode the second training sample through the encoder block of the coronary vessel naming model to obtain the encoding result of the second training sample;
[0118] The training module 2083 is used to iteratively process the encoding result of the second training sample through the core block of the coronary vessel naming model to obtain the hidden information of the second training sample;
[0119] The training module 2083 is used to decode the encoding result of the second training sample through the encoder block of the coronary vessel naming model when a preset number of iterations is reached, so as to obtain a second prediction result of the coronary vessel naming model.
[0120] In the above solution, the training module 2083 is used to obtain the first accuracy, the first precision and the first recall in the first prediction result;
[0121] The training module 2083 is used to obtain a second accuracy rate, a second precision rate, and a second recall rate in the second prediction result;
[0122] The training module 2083 is used to adjust the model parameters of the coronary vessel naming model when the second accuracy rate is greater than the first accuracy rate, the second precision rate is greater than the first precision rate, and the second recall rate is greater than the first recall rate, wherein the model parameters of the coronary vessel naming model include:
[0123] The depth of the fully connected layers in the perceptron module, the width of the fully connected layers in the perceptron module, the number of core block iterations, and the batch parameter.
[0124] In some embodiments of the present application, the electronic device provided in the embodiment of the present application may also deploy the coronary vessel naming device provided in the present application, which is used to name the coronary vessels appearing in the angiography image of the target coronary vessels through the trained coronary vessel naming model. Figure 3 , Figure 3 is a schematic diagram of the structure of an electronic device for executing the coronary vessel naming method provided in the present application, provided in an embodiment of the present application, Figure 3 The electronic device shown may be a medical imaging server, or may be a medical electronic device encapsulated with a coronary vessel naming device, and this application does not impose any specific restrictions on this. Figure 3 The electronic device shown includes: at least one processor 410, a memory 450, at least one network interface 420 and a user interface 430. The various components in the electronic device are coupled together via a bus system 440. It is understood that the bus system 440 is used to achieve connection and communication between these components. In addition to the data bus, the bus system 440 also includes a power bus, a control bus and a status signal bus. However, for the sake of clarity, the bus system 440 is not shown in FIG. Figure 3 Various buses are labeled as bus system 440 .
[0125] The processor 410 can be an integrated circuit chip with signal processing capabilities, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., where the general-purpose processor can be a microprocessor or any conventional processor, etc.
[0126] The user interface 430 includes one or more output devices 431 capable of presenting media content, including one or more speakers and / or one or more visual display screens. The user interface 430 also includes one or more input devices 432, including user interface components that facilitate user input, such as a keyboard, mouse, microphone, touch screen display, camera, other input buttons and controls.
[0127] The memory 450 may be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state memory, hard disk drives, optical disk drives, etc. The memory 450 may optionally include one or more storage devices that are physically remote from the processor 410.
[0128] The memory 450 includes a volatile memory or a non-volatile memory, and may also include both volatile and non-volatile memories. The non-volatile memory may be a read-only memory (ROM), and the volatile memory may be a random access memory (RAM). The memory 450 described in the embodiments of the present application is intended to include any suitable type of memory.
[0129] In some embodiments, memory 450 can store data to support various operations, examples of which include programs, modules, and data structures, or a subset or superset thereof, as exemplarily described below.
[0130] Operating system 451, including system programs for processing various basic system services and performing hardware-related tasks, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks;
[0131] A network communication module 452, used to reach other electronic devices via one or more (wired or wireless) network interfaces 420, exemplary network interfaces 420 include: Bluetooth, Wireless Compatibility Certification (WiFi), and Universal Serial Bus (USB), etc.;
[0132] A presentation module 453 for presenting information via one or more output devices 431 (e.g., a display screen, a speaker, etc.) associated with the user interface 430 (e.g., a user interface for operating peripheral devices and displaying content and information);
[0133] The input processing module 454 is used to detect one or more user inputs or interactions from one of the one or more input devices 432 and translate the detected inputs or interactions.
[0134] In some embodiments, the device provided in the embodiments of the present application can be implemented in software. Figure 3 The coronary vessel naming device 455 stored in the memory 450 is shown, which can be software in the form of a program and a plug-in, etc., including the following software modules: a signal transmission module 4081 and a signal processing module 4082. These modules are logical, so they can be arbitrarily combined or further split according to the functions implemented. The functions of each module will be explained below.
[0135] The functions of the various software modules in the coronary vessel naming device 455 are described below, wherein:
[0136] A signal transmission module 4081 is used to obtain angiographic images of target coronary vessels;
[0137] A signal processing module 4082 is used to extract the centerline of the target coronary vessel;
[0138] The signal processing module 4082 is used to predict the node name of the center line of the target coronary vessel by using the coronary vessel naming model to obtain the predicted name of the target coronary vessel.
[0139] Combination Figure 2 The coronary vessel naming model training device shown in the figure illustrates the coronary vessel naming model training method provided in the embodiment of the present application, see Figure 4 , Figure 4 This is an optional flow chart of the coronary vessel naming model training method provided in the embodiment of the present application. It can be understood that: Figure 4 The coronary vessel naming model training method shown can be applied to the research of computer-aided diagnosis of coronary artery disease, or to the naming of variant coronary vessels. Figure 4 The steps shown can be executed by various electronic devices running the coronary vessel naming model training device, for example, a server with a coronary vessel image processing function or a terminal with a coronary vessel image processing function. Figure 4 The steps shown are explained.
[0140] Step 401: Acquire angiographic images of coronary vessels and cardiac chamber images.
[0141] In some embodiments, the angiographic images of the coronary vessels and the cardiac chamber images may be acquired in the following manners:
[0142] 1) Retrieve the angiography images of the coronary vessels and the cardiac chamber images stored in the server / cloud server.
[0143] 2) According to the real-time image data captured by the coronary vessel imaging device, the corresponding coronary vessel angiography image and heart chamber image are obtained.
[0144] 3) In response to the operation instruction of the target object, the angiography image of the coronary blood vessels and the heart chamber image transmitted by the external device are received through the communication port.
[0145] In some embodiments, the coronary vascular model training method provided in the present application does not specifically limit the source of the computed tomography angiography image of the coronary vessels. For example, it can be a complete computed tomography angiography image of the coronary artery, or a computed tomography angiography image of a specified position in the coronary vessel. The acquired computed tomography angiography image of the coronary vessels needs to comply with medical digital imaging and communication standards, such as the international image standard (ISO 12052). It defines a medical image format that can be used for data exchange and has a quality that meets clinical needs. Therefore, the coronary vascular model established by the computed tomography angiography image of the coronary vessels can be read and used by medical equipment from different manufacturers, and can be stored in a medical cloud server for easy retrieval and use by doctors.
[0146] Step 402: Acquire the centerline features of the coronary vessels based on the angiography images of the coronary vessels.
[0147] In some embodiments, the centerline feature of the coronary vessel may be obtained by:
[0148] First, the vascular tree model of the coronary vessels is extracted. After obtaining the vascular tree model of the coronary vessels, the centerline tree model is constructed using the centerline of the vascular tree model. After obtaining the centerline tree, the skeleton method can be used to extract the centerline in the centerline tree to obtain the centerline features of the coronary vessels of the heart. In this process, it should be noted that when extracting the centerline features of the coronary vessels of different target objects, no specific limitation is made on the threshold segmentation method. At the same time, the extraction of the centerline is not limited to the skeleton method. The appropriate method can be selected according to actual needs to extract the centerline tree and the centerline.
[0149] refer to Figure 5a , Figure 5aFIG. 1 is an optional schematic diagram of a coronary tree in an embodiment of the present application. The coronary vessels processed in the embodiment of the present application may include: coronary arteries and / or coronary veins, Figure 5a Taking the coronary artery as an example, the coronary artery is divided into two branches, the left and right branches, whose openings are located in the left and right coronary sinuses respectively. The left coronary artery is divided into the left main trunk (LM), the left anterior descending branch (LAD), and the circumflex branch (LCX). The left main trunk is a trunk about 1 to 3 cm long that originates from the root of the aorta. The left anterior descending branch runs in the interventricular groove, and the circumflex branch runs in the left atrioventricular groove. The right coronary artery (RCA) runs in the right atrioventricular groove. By obtaining the binary mask image of the coronary artery, all the pixels with a pixel value of 1 are combined to obtain Figure 5a Coronary tree model of the coronary arteries shown.
[0150] refer to Figure 5b , Figure 5b This is an optional schematic diagram of the centerline tree in the embodiments of the present application. In some embodiments, since the coronary vessels of the target object may have vascular malformations (including congenital vascular malformations and pathological malformations), when centerline features are extracted for the coronary vessels in these cases, the centerline features may be affected by the vascular malformations. Therefore, the coronary vessel naming model needs to add heart chamber features to the training samples during training to ensure the accuracy of the coronary vessel naming.
[0151] Step 403: Acquire the cardiac chamber features according to the cardiac chamber image.
[0152] In some embodiments, the cardiac chamber characteristics may include: characteristics of the atrioventricular groove plane and characteristics of the interventricular groove plane, wherein the atrioventricular groove may include: left atrioventricular groove, right atrioventricular groove, and the interventricular groove may include: anterior interventricular groove and posterior interventricular groove.
[0153] refer to Figure 6 , Figure 6 The schematic diagram of the atrium and ventricle in the embodiment of the present application is as follows: Figure 6 As shown, 601 is the right atrium of the target object, 602 is the right ventricle of the target object, and the junction of 601 and 602 is the intersection of the right atrioventricular groove, as shown in FIG. Figure 6 As shown, the intersection line of the right atrioventricular groove is an irregular closed curve.
[0154] In some embodiments, in some embodiments, the characteristic atrioventricular groove plane and interventricular groove plane of the heart chamber in step 403 can be obtained by plane fitting the surface model of the chamber respectively to obtain the characteristic planes of the atrium and ventricle images. Taking the extraction of the left atrium surface model as an example, the pixel points with a pixel value of 1 in the left atrium mask image can be first extracted, and the pixel points with a pixel value of 0 (background and noise pixels) can be discarded. Then, all the pixel points with a pixel value of 1 are combined to obtain the left atrium surface model of the target object. By analogy, the left ventricle surface model, the right atrium surface model, and the right ventricle surface model of the heart can be obtained using the mask images at the corresponding positions.
[0155] In some embodiments, the characteristic atrioventricular groove plane and interventricular groove plane of the heart chamber in step 403 can be obtained by plane fitting the surface model of the chamber to obtain the characteristic planes of the atrium and ventricle image, which is specifically expressed as the following acquisition method:
[0156] 1) Enlarging the right atrial surface model and the right ventricular surface model until the right atrial surface model and the right ventricular surface model generate a first intersection; determining the right atrioventricular groove intersection line according to the first intersection; performing plane fitting according to the points in the right atrioventricular groove intersection line to obtain the right atrioventricular groove plane; wherein, when the right atrial surface model and the right ventricular surface model are enlarged, they can be gradually enlarged according to a preset magnification factor, and the preset magnification factor range is preferably [1.1, 2]. Taking the magnification factor of 2 as an example, firstly, the right atrial surface model and the right ventricular surface model are enlarged to 2 times, so that the right atrial surface model and the right ventricular surface model generate an intersection, i.e., the first intersection in the above embodiment, taking the first intersection of the right atrial surface model and the right ventricular surface model to obtain the right atrioventricular groove intersection line.
[0157] Since the right atrioventricular groove intersection line is an irregular closed curve composed of all points on the atrioventricular groove intersection line, the right atrioventricular groove plane is determined by fitting through all points on the right atrioventricular groove intersection line using the least squares method, wherein the relevant algorithm can use the least squares method, and the same method is applied to the left atrium and left ventricle, expanding the left atrium and right ventricle model to 1.1 times, and determining the left atrioventricular groove intersection line based on the second intersection obtained, and determining the left atrioventricular groove plane by fitting.
[0158] 2) Enlarging the left atrial surface model and the left ventricular surface model until the left atrial surface model and the left ventricular surface model generate a second intersection; determining the left atrioventricular groove intersection line according to the second intersection; performing plane fitting according to the points in the left atrioventricular groove intersection line to obtain the left atrioventricular groove plane. Wherein, when the left atrial surface model and the left ventricular surface model are enlarged, they can be gradually enlarged according to a preset magnification factor, and the preset magnification factor range is preferably [1.1, 2]. Taking the magnification factor of 2 as an example, the left atrial surface model and the left ventricular surface model are first enlarged to 2 times, so that the left atrial surface model and the left ventricular surface model generate an intersection, that is, the second intersection in the above embodiment, and the second intersection of the left atrial surface model and the left ventricular surface model is taken to obtain the left atrioventricular groove intersection line.
[0159] Since the left atrioventricular groove intersection line is also an irregular closed curve, it is composed of all points on the left atrioventricular groove intersection line. Therefore, the left atrioventricular groove plane is determined by fitting all points on the left atrioventricular groove intersection line using the least squares method. Among them, the relevant algorithm can use the least squares method. The same method is applied to the left atrium and left ventricle, and the left atrium and left ventricle model is enlarged to 1.1 times. The left atrioventricular groove intersection line is determined based on the second intersection obtained, and the left atrioventricular groove plane is determined by fitting.
[0160] In some embodiments, reference Figure 7 , Figure 7 The schematic diagram of the intersection line of the interventricular groove in the embodiment of the present application shows that the target object has a longitudinal groove in front and back of the ventricle. Figure 7 701 in the figure are called the anterior interventricular groove and the posterior interventricular groove, which are the boundary marks between the left and right ventricles. Since the characteristic plane of the atrium and ventricle image also includes the interventricular groove characteristic plane, in order to improve the accuracy of coronary vessel modeling, it is necessary to accurately obtain the interventricular groove characteristic plane. The determination of the interventricular groove characteristic plane can be achieved by the following methods:
[0161] The left ventricular surface model and the right ventricular surface model are enlarged until the left ventricular surface model and the right ventricular surface model generate a third intersection; the interventricular groove intersection line is determined according to the third intersection; plane fitting is performed according to the points in the interventricular groove intersection line to obtain the anterior interventricular groove characteristic plane and the posterior interventricular groove characteristic plane. Wherein, when the right ventricular surface model and the left ventricular surface model are enlarged, they can be gradually enlarged according to a preset magnification factor, and the preset magnification factor range is preferably [1.1, 2]. Taking the magnification factor of 2 as an example, the right ventricular surface model and the left ventricular surface model are first enlarged to 2 times, so that the right ventricular surface model and the left ventricular surface model generate an intersection, that is, the third intersection in the above embodiment, and the third intersection of the right ventricular surface model and the left ventricular surface model is taken to obtain the interventricular groove intersection line.
[0162] Since the intersection line of the interventricular groove is also an irregular closed curve composed of all points on the intersection line of the interventricular groove, the left atrioventricular groove plane is determined by fitting all points on the intersection line of the interventricular groove using the least squares method. Among them, the relevant algorithm can use the least squares method to fit the characteristic planes of the anterior interventricular groove and the posterior interventricular groove.
[0163] In the above embodiment, since there are fat tissue and blood vessels that nourish the myocardium in the right atrioventricular groove, the left atrioventricular groove, the anterior interventricular groove and the posterior interventricular groove, if the coronary vessels in the right atrioventricular groove, the left atrioventricular groove, the anterior interventricular groove and the posterior interventricular groove have narrowed diameters, the blood flow velocity of the coronary vessels will be affected, resulting in insufficient oxygen supply to the coronary vessels and coronary atherosclerotic heart disease. However, the relative positions of the interventricular groove and the atrioventricular groove and the coronary vessels will not change due to narrowed diameters of the coronary vessels (centerline deviation). Therefore, the coronary vessel naming model that uses the characteristics of the heart chambers as training samples can accurately name the mutated coronary vessels, thereby providing model support for doctors' auxiliary diagnosis. The target object can also visually obtain the coronary vessel model and the name of each coronary vessel in the coronary vessels through the display interface.
[0164] Step 404: Calculate the distance between the centerline of the coronary blood vessel and the heart chamber to obtain a distance feature.
[0165] In some embodiments, through step 404, the distance between the centerline of the coronary artery and the heart chamber can be expressed as the distance from the centerline of the coronary artery to the four chambers of the heart, that is, the distance from the centerline of the coronary artery to the left atrium, the distance from the centerline of the coronary artery to the left ventricle, the distance from the centerline of the coronary artery to the right atrium, and the distance from the centerline of the coronary artery to the right ventricle.
[0166] In some embodiments, since the coronary vessels are arteries that supply blood to the heart, taking the coronary arteries as an example, the coronary arteries originate from the aortic sinus at the root of the aorta, are divided into left and right branches, and grow on the surface of the heart. For example, the left coronary artery (LCA for short) is a coronary artery that originates from the aorta above the left cusp of the aortic valve and supplies blood to the left side of the myocardium. Therefore, in order to model the coronary vessels more accurately, it is necessary to determine the distance from the centerline of the coronary vessels to the surface of the heart chamber. Since the centerline of the coronary vessels is composed of different centerline points, different coronary vessel centerlines can be named through centerline features, heart chamber features and distance features, using the coronary vessel naming model, and the name of the coronary vessel centerline can correspond to the name of the corresponding coronary vessel.
[0167] Figure 8Schematic diagram of different center lines in the coronary vessels in the embodiment of the present application, wherein the center line of the coronary vessels can be defined as a line connecting points in the coronary vessels that are equidistant from the two ends of the coronary vessel wall, that is, a perpendicular line is drawn from any point on the center line of the coronary vessels to the coronary vessel wall, and the distance from the point to the two ends of the coronary vessel wall along the perpendicular direction is equal. Fig.10 As shown, taking coronary artery 401 as an example, coronary artery 401 includes two left and right coronary arteries, wherein the left coronary artery is 408 and the right coronary artery is 412; coronary artery 401 includes three coronary artery centerlines, namely centerlines 402 and 404 of the right coronary artery 412 and centerline 409 of the left coronary artery 408, wherein, taking the left coronary artery 408 as an example, 403 is the center point of the starting position of the coronary artery centerline 409, and 410 is the center point of the coronary artery centerline 409. The end point 410 is located at the center of the cross section 411 of the center line 409. The coronary artery center line 409 from 403 to 410 is composed of a number of center points (the number of center points is determined by different coronary artery center line measurement methods or different coronary artery center line measurement requirements). A number of center points can constitute a center line point set; 407 and 405 are the vessel walls at both ends of the coronary artery center line 409, and 406 is any point not on the center line 410. Figure 8 The nodes in the center lines 402 and 404 of the right coronary artery 412 and the center line 409 of the left coronary artery 408 are classified to obtain the names of the center lines, that is, the names of the coronary arteries corresponding to the corresponding center lines can be obtained.
[0168] Step 405: The coronary vessel naming model is trained according to the centerline features, the heart chamber features, and the distance features to obtain the prediction effect of the coronary vessel naming model.
[0169] refer to Fig. 9 , Fig. 9 Schematic diagram of the structure of the coronary vessel naming model in the embodiment of the present application. Fig. 9As shown in the figure, the framework of the coronary vessel naming model includes three GN blocks, the graph topology of the GN blocks is the same, but they have different neural network parameters; the three GN blocks include an encoding block Encode, a core block Core and a decoding block Decode, the input end of the encoding block is connected to the input training data, the output end of the encoding block is connected to the input end of the Core block, the output end of the Core block is connected to the input end of the decoding block, the output end of the decoding block obtains the training result, and the hidden information Hidden(t+1) output by the core block Core after t+1 iterations is the hidden information Hidden(t) of t iterations and the output of the encoding block as input; the encoding block Encode performs one calculation of the edge and node feature values in the attribute graph, which is equivalent to encoding; the core block Core performs N rounds of message passing processing, and the input of the core block Core is the output of the encoding block Encode and the output of the core block Core in the previous step; the decoding block Decode performs one calculation of the edge and node feature values in the attribute graph, which is equivalent to data decoding.
[0170] refer to Fig.10 , Fig.10 This is a schematic diagram of the training process of the coronary vessel naming model in the embodiment of the present application. Fig.10 The steps shown can be executed by various electronic devices running the coronary vessel naming model training device, for example, a server with a coronary vessel image processing function or a terminal with a coronary vessel image processing function. Fig.10 The steps shown are explained.
[0171] Step 1001: De-noise the centerline features to obtain a first training sample set that matches the coronary vessel naming model, wherein the first training sample set includes at least one group of training samples.
[0172] In some embodiments, since the coronary vessels of the target object may have vascular malformations (including congenital vascular malformations and pathological malformations), when processing the coronary vessels in these cases, the extracted centerline will usually produce problems such as uneven centerline points and inconsistent centerline starting points. The centerline points whose offset exceeds the offset threshold can be denoised to obtain a first training sample set that matches the coronary vessel naming model, thereby improving the accuracy of the coronary vessel naming model.
[0173] Step 1002: Encode the first training sample through the encoder block of the coronary vessel naming model to obtain the encoding result of the first training sample.
[0174] Step 1003: Iteratively process the encoding result of the first training sample through the core block of the coronary vessel naming model to obtain hidden information of the first training sample.
[0175] Among them, combined Fig. 9 In the model structure shown in the figure, the encoding block performs one calculation of the edge and node feature values in the attribute graph, which is equivalent to encoding. The core block performs N rounds of message passing. The input of the core block is the output of the encoding block and the output of the previous step of the core block, as shown in Fig. 9 Hidden(t+1) in is calculated by input Hidden(t) and the output of the encoding block. The decoding block performs the calculation of the edge and node feature values in the attribute graph once, which is equivalent to decoding the data.
[0176] Step 1004: When the preset number of iterations is reached, the encoding result of the first training sample is decoded by the encoder block of the coronary vessel naming model to obtain a first prediction effect of the coronary vessel naming model.
[0177] In some embodiments, since the training samples use a ternary structure including: centerline features, heart chamber features, and distance features, and the names of the corresponding centerline nodes have accurate labels, the encoding result of the first training sample is decoded by the encoder block of the coronary vessel naming model to obtain the first prediction effect of the coronary vessel naming model, which can be achieved in the following manner:
[0178] The encoding result of the first training sample is decoded by the encoder block of the coronary vessel naming model to obtain the predicted name of the training sample; the label information matching the first training sample is obtained; the predicted name is detected according to the label information to obtain the first prediction effect of the coronary vessel naming model. The label information matching the first training sample can be any one of the items in Table 1:
[0179] Table 1 Label information matching the first training sample
[0180]
[0181] In some embodiments, when training the coronary vessel naming model, since the first training sample only includes the centerline feature, the trained coronary vessel naming model may not be able to accurately name the coronary vessels with centerline variation. Therefore, the training sample of the coronary vessel naming model may be adjusted. Fig.11 , Fig.11 This is a schematic diagram of the training process of the coronary vessel naming model in the embodiment of the present application. Fig.11The steps shown can be executed by various electronic devices running the coronary vessel naming model training device, for example, a server with a coronary vessel image processing function or a terminal with a coronary vessel image processing function. Fig.11 The steps shown are explained.
[0182] Step 1101: combining centerline features, cardiac chamber features, and distance features to obtain a second training sample set that matches the coronary vessel naming model, wherein the second training sample set includes at least one group of training samples.
[0183] Step 1102: Adjust the depth and width of the fully connected layer in the perceptron module of the coronary vessel naming model, and add classification labels corresponding to the second training sample set.
[0184] Step 1103: Encode the second training sample through the encoder block of the coronary vessel naming model to obtain an encoding result of the second training sample.
[0185] Step 1104: Iteratively process the encoding result of the second training sample through the core block of the coronary vessel naming model to obtain hidden information of the second training sample, wherein the hidden information is used for iterative processing of the core block.
[0186] Step 1105: When the preset number of iterations is reached, the encoding result of the second training sample is decoded by the encoder block of the coronary vessel naming model to obtain a second prediction result of the coronary vessel naming model.
[0187] In some embodiments, after obtaining the prediction results of the training process corresponding to different training samples through the above embodiments, step 406 can be continued to adjust the model parameters of the coronary vessel naming model according to the prediction results to improve the accuracy of the entire coronary vessel naming model in naming coronary vessels.
[0188] Step 406: According to the prediction result, the model parameters of the coronary vessel naming model are adjusted to enable the coronary vessels in the angiography image of the coronary vessels to be named by the coronary vessel naming model.
[0189] In some embodiments, adjusting the model parameters of the coronary vessel naming model according to the prediction effect can be achieved in the following ways:
[0190] Obtain the first accuracy, first precision and first recall in the first prediction result; wherein, reference Fig.12 , Fig.12It is a schematic diagram of the prediction effect in the first prediction result, where 1201 shows the number of corresponding coronary vessels in the first training sample identified by using the label information, and 1202 shows the accuracy, recall and precision. Among them, the accuracy (Accuracy) is the ratio of all correctly predicted first training samples to the total number of first samples; the precision (Precision) is the ratio of the number of positive samples predicted as positive in the first training sample to the number of positive samples predicted as positive in the total samples; the recall rate (Recall) is the ratio of the number of positive samples predicted as positive in the first training sample to the number of positive samples in the total samples. The precision rate and the recall rate are negatively correlated. When the harmonic mean of the precision rate and the recall rate is 1, it means that the performance of the coronary vessel naming model reaches the best. According to Fig.12 As shown in the figure, the initial first accuracy, initial first precision and initial first recall in the first prediction result are 0.82, 0.7 and 0.82 respectively; when the precision is infinite (NaN), the recall is 0 and the accuracy is 0.99. Through iterative training of the coronary vessel naming model, the target first accuracy, target first precision and target first recall in the first prediction result are 0.87, 0.87 and 0.98 respectively. Fig.12 From the training results shown on the left, it can be seen that the coronary vessel naming model has a low accuracy rate when identifying and naming coronary vessels labeled LCx, RPLB, and RPDA, which affects the target first accuracy rate.
[0191] In some embodiments, after adjusting the depth and width of the fully connected layer in the perceptron module of the coronary vessel naming model and adding the classification labels corresponding to the second training sample set, the second accuracy, second precision and second recall rate in the second prediction result are obtained when the coronary vessel naming model is continuously trained; wherein, reference Fig.13 , Fig.13 1301 shows the number of corresponding coronary vessels in the second training sample identified by using label information, and 1302 shows the accuracy rate. Fig.13 As shown, the initial second accuracy, initial second precision and initial second recall in the second prediction result are 0.88, 0.79 and 0.88; when the precision is infinite (NaN), the recall is 0 and the accuracy is 1. Through iterative training of the coronary vessel naming model, the target second accuracy, target second precision and target second recall in the second prediction result are 0.98, 0.99 and 0.87, and the harmonic mean of the precision and recall is close to 1, indicating that the performance of the coronary vessel naming model is optimal.
[0192] In some embodiments, when the second accuracy rate is greater than the first accuracy rate, the second precision rate is greater than the first precision rate, and the second recall rate is greater than the first recall rate, the model parameters of the coronary vessel naming model are adjusted, wherein the model parameters of the coronary vessel naming model include:
[0193] The depth of the fully connected layer in the perceptron module, the width of the fully connected layer in the perceptron module, the number of core block iterations, and the batch processing parameters. In some embodiments, the model parameters of the whole coronary vessel naming model may include: 1) adjusting the width of the fully connected layer so that the width of the fully connected layer increases, 2) increasing the depth of the fully connected layer, and increasing the number of fully connected layers of the coronary vessel naming model, for example, adjusting the number of fully connected layers of the coronary vessel naming model from 2 layers to 5 layers.
[0194] In some embodiments, since the second accuracy is greater than the first accuracy, the second precision is greater than the first precision, the second recall is greater than the first recall, and the first training sample and the second training sample are not the same, it can be determined that the training effect of the second training sample obtained by combining the centerline feature, the heart chamber feature and the distance feature is greater than that of the first training sample including only the centerline feature. Therefore, the training of the coronary vessel model is preferably the second training sample.
[0195] In some embodiments, due to the addition of classification labels corresponding to the second training sample set, for example, adjusting the classification labels from the left coronary middle node leftNode to LADnode, LCXnode, LMnode, the second accuracy rate can be greater than the first accuracy rate, and the second precision rate can be greater than the first precision rate. Therefore, new classification labels can be added during the training process of the coronary vessel naming model to achieve accurate naming of the coronary vessel naming model.
[0196] After the coronary vessel naming model is trained, it can be deployed in different electronic devices to process coronary vessel images from different sources and obtain corresponding coronary vessel names. Fig.14 , Fig.14 This is a schematic diagram of the coronary vessel naming process in the embodiment of the present application. Fig.14 The steps shown can be executed by various electronic devices running the coronary vessel naming device, such as a server with a coronary vessel image processing function or a terminal with a coronary vessel image processing function. Fig.14 The steps shown are explained.
[0197] Step 1401: Acquire angiography images of target coronary vessels.
[0198] Step 1402: Extract the centerline of the target coronary vessel.
[0199] Step 1403: Predict the node name of the centerline of the target coronary vessel using the coronary vessel naming model to obtain the predicted name of the target coronary vessel.
[0200] Step 1404: Mark the target coronary vessel in the angiography image by using the predicted name, and display the marked coronary angiography image.
[0201] In some embodiments, the angiography image of the coronary vessels can be labeled with the name of the coronary vessels, and the effect displayed by the coronary vessel model can also be labeled with the name. Users can flexibly adjust according to different usage scenarios, and this application does not impose specific restrictions on this.
[0202] refer to Fig.15 , Fig.15 FIG. 1 is a schematic diagram showing the result of naming coronary vessels according to an embodiment of the present application, as shown in FIG. Fig.15 As shown, through the coronary vessel naming method provided in the present application, the coronary vessel model established in the coronary vessel naming process can be displayed in 1501. It should be noted that Fig.15 In the illustrated embodiment, in order to more clearly show the connection relationship between different coronary vessels in the coronary vessel model, the coronary vessel model shown in 1501 does not include the corresponding myocardial tissue. Fig.15 As shown, the coronary artery model displayed in 1501 includes at least: an obtuse marginal artery (OM), a first diagonal artery (D1), a second diagonal artery (D2), a fourth diagonal artery (D4), a fifth diagonal artery (D5), and a left anterior descending artery (LAD); in some embodiments, since the coronary artery model displayed in 1501 is in a 3D state, when the observation angle of the coronary artery model displayed in 1501 is adjusted through a control instruction, an angiography image corresponding to the fourth observation angle of the coronary artery model can be displayed in 1503, an angiography image corresponding to the fifth observation angle of the coronary artery model can be displayed in 1504, and an angiography image corresponding to the sixth observation angle of the coronary artery model can be displayed in 1505. The present application does not impose any specific restrictions on the specific values of the fourth observation angle, the fifth observation angle, and the sixth observation angle. At the same time, when each coronary artery in the coronary artery model presented in 1501 is named through the coronary artery model provided by the present application, a locally enlarged visualization of the coronary artery can be performed in 1503, so as to facilitate doctors and patients to have a comprehensive understanding of coronary artery lesions and treatment plans.
[0203] Beneficial technical effects:
[0204] The present application obtains angiography images and heart chamber images of coronary vessels; obtains centerline features of the coronary vessels based on the angiography images of the coronary vessels; obtains heart chamber features based on the heart chamber images; calculates the distance between the centerline of the coronary vessels and the heart chamber to obtain distance features; trains a coronary vessel naming model based on the centerline features, heart chamber features and distance features to obtain a prediction effect of the coronary vessel naming model; and adjusts model parameters of the coronary vessel naming model based on the prediction effect to enable naming of the coronary vessels in the angiography images of the coronary vessels through the coronary vessel naming model. The trained coronary vessel naming model can not only quickly and automatically name vessels, but also name variant vessels in the coronary vessels, thereby solving the defect of low accuracy in vessel naming in related technologies.
[0205] The above are merely embodiments of the present application and are not intended to limit the scope of protection of the present application. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the scope of protection of the present application.
Claims
1. A coronary vessel naming model training method, characterized in that: The method comprises: Obtain angiographic images of the coronary vessels and cardiac chambers; Acquiring a centerline feature of the coronary artery according to the angiography image of the coronary artery; Acquiring the cardiac chamber feature according to the cardiac chamber image; Calculating the distance between the centerline of the coronary blood vessel and the heart chamber to obtain a distance feature; Training a coronary vessel naming model according to the centerline feature, the heart chamber feature, and the distance feature to obtain a prediction effect of the coronary vessel naming model; According to the prediction effect, the model parameters of the coronary vessel naming model are adjusted to achieve the naming of the coronary vessels in the angiography image of the coronary vessels through the coronary vessel naming model.
2. The method according to claim 1, characterized in that The coronary vessel naming model is trained according to the centerline feature, the cardiac chamber feature and the distance feature to obtain a prediction effect of the coronary vessel naming model, including: Denoising the centerline features to obtain a first training sample set that matches the coronary vessel naming model, wherein the first training sample set includes at least one group of training samples; encoding the first training sample by using an encoder block of the coronary vessel naming model to obtain an encoding result of the first training sample; Iteratively processing the encoding result of the first training sample through the core block of the coronary vessel naming model to obtain hidden information of the first training sample, wherein the hidden information is used for iterative processing of the core block; When the preset number of iterations is reached, the encoding result of the first training sample is decoded by the encoder block of the coronary vessel naming model to obtain a first prediction effect of the coronary vessel naming model.
3. The method according to claim 2, characterized in that The decoding of the encoding result of the first training sample by the encoder block of the coronary vessel naming model to obtain a first prediction effect of the coronary vessel naming model includes: Decoding the encoding result of the first training sample by the encoder block of the coronary vessel naming model to obtain a predicted name of the training sample; Acquire label information matching the first training sample; The predicted name is detected according to the label information to obtain a first prediction effect of the coronary vessel naming model.
4. The method according to claim 2, characterized in that: The method further comprises: Combining the centerline feature, the heart chamber feature, and the distance feature to obtain a second training sample set that matches the coronary vessel naming model, wherein the second training sample set includes at least one group of training samples; Adjust the depth of the fully connected layer and the width of the fully connected layer in the perceptron module of the coronary vessel naming model, and Adding classification labels corresponding to the second training sample set; encoding the second training sample by using an encoder block of the coronary vessel naming model to obtain an encoding result of the second training sample; Iteratively processing the encoding result of the second training sample through the core block of the coronary vessel naming model to obtain hidden information of the second training sample, wherein the hidden information is used for iterative processing of the core block; When the preset number of iterations is reached, the encoding result of the second training sample is decoded by the encoder block of the coronary vessel naming model to obtain a second prediction result of the coronary vessel naming model.
5. The method according to claims 3 and 4, characterized in that According to the prediction effect, adjusting the model parameters of the coronary vessel naming model includes: Obtaining a first accuracy rate, a first precision rate, and a first recall rate in the first prediction result; Obtaining a second accuracy, a second precision, and a second recall in the second prediction result; When the second accuracy rate is greater than the first accuracy rate, the second precision rate is greater than the first precision rate, and the second recall rate is greater than the first recall rate, adjusting the model parameters of the coronary vessel naming model, wherein the model parameters of the coronary vessel naming model include: The depth of the fully connected layers in the perceptron module, the width of the fully connected layers in the perceptron module, the number of core block iterations, and the batch parameter.
6. A method for naming coronary vessels, characterized in that: The method comprises: acquiring angiographic images of target coronary vessels; extracting the centerline of the target coronary vessel; The node name of the center line of the target coronary vessel is predicted by a coronary vessel naming model to obtain a predicted name of the target coronary vessel, wherein the coronary vessel naming model is trained according to any one of claims 1-5.
7. A coronary vessel naming model training device, characterized in that: The device comprises: An information transmission module, used for acquiring angiographic images of coronary vessels and images of cardiac chambers; An information processing module, used for acquiring the centerline feature of the coronary blood vessel according to the angiography image of the coronary blood vessel; The information processing module is used to obtain the cardiac chamber characteristics according to the cardiac chamber image; The information processing module is used to calculate the distance between the center line of the coronary blood vessel and the heart chamber to obtain a distance feature; A training module, used for training the coronary vessel naming model according to the centerline feature, the heart chamber feature and the distance feature, to obtain a prediction effect of the coronary vessel naming model; The training module is used to adjust the model parameters of the coronary vessel naming model according to the prediction effect, so as to realize the naming of the coronary vessels in the angiography image of the coronary vessels through the coronary vessel naming model.
8. A coronary vessel naming device, characterized in that: The device comprises: A signal transmission module, used for acquiring angiographic images of target coronary vessels; A signal processing module, used for extracting the centerline of the target coronary vessel; The signal processing module is used to predict the node name of the center line of the target coronary vessel through a coronary vessel naming model to obtain the predicted name of the target coronary vessel, wherein the coronary vessel naming model is trained according to any one of claims 1-5.
9. An electronic device, characterized in that: The electronic device comprises: A memory for storing executable instructions; The processor is used to implement the coronary vessel naming model training method described in any one of claims 1 to 5, or the coronary vessel naming method described in claim 6 when running the executable instructions stored in the memory.
10. A computer-readable storage medium storing executable instructions, characterized in that: When the executable instructions are executed by the processor, the coronary vessel naming model training method described in any one of claims 1 to 5 is implemented, or the coronary vessel naming method described in claim 6 is implemented.
