A data processing method, apparatus, device and medium

Abstract

Embodiments of the present application disclose a data processing method, apparatus, device, and medium. The method includes: obtaining an initial sample angiography image set; performing data augmentation processing on a first sample angiography image based on the physical characteristics of intracranial blood vessels to obtain a processed sample angiography image, and performing label conversion processing on a first label based on the physical characteristics of intracranial blood vessels to obtain a second label for the processed sample angiography image, and adding the processed sample angiography image and the second label to a target sample angiography image set; pre-training an angiography image recognition model using the initial sample angiography image set and the target sample angiography image set to obtain a trained angiography image recognition model. Using the embodiments of the present application can solve the problem of insufficient sample quantity, ensure the training process of the angiography image recognition model, and ensure the performance of the trained angiography image recognition model.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular, to a data processing method, a data processing device, a data processing device, and a computer-readable storage medium. Background Art

[0002] An intracranial aneurysm is a foreign object with a shape similar to a tumor formed by the long-term impact of blood flow on the intracranial arterial wall. Since latent asymptomatic intracranial aneurysms are not uncommon in the general population, and the fatality rate of subarachnoid hemorrhage caused by the rupture of intracranial aneurysms is relatively high (about 50%), and most survivors (up to 46%) may suffer from long-term cognitive impairment, which greatly affects the living ability and quality. Therefore, predicting the rupture risk of intracranial aneurysms is of great significance.

[0003] Moreover, since the rupture risk of intracranial aneurysms is correlated with intracranial blood vessels, for example, the rupture risk of an intracranial aneurysm in a certain intracranial blood vessel is relatively high. Therefore, the rupture risk of the intracranial aneurysm in the intracranial blood vessel can be predicted by identifying the intracranial blood vessels in the DSA (Digital Subtraction Angiography) image of the intracranial aneurysm. And how to identify the intracranial blood vessels in the DSA image of the intracranial aneurysm has become an important topic in the medical field. Summary of the Invention

[0004] The embodiments of this application provide a data processing method, device, device, and medium, which can solve the problem of insufficient sample quantity, ensure the training process of the contrast image recognition model, and ensure the performance of the trained contrast image recognition model.

[0005] On the one hand, the embodiments of this application provide a data processing method, which includes:

[0006] Obtain an initial sample angiography image set, where the initial sample angiography image set includes a first sample angiography image and a first label of the first sample angiography image, and the first label of the first sample angiography image indicates that the first sample angiography image is obtained by performing angiography on the first intracranial blood vessel;

[0007] Perform data augmentation processing on the first sample angiography image based on the physical characteristics of the intracranial blood vessel to obtain a processed sample angiography image, and perform label conversion processing on the first label based on the physical characteristics of the intracranial blood vessel to obtain a second label of the processed sample angiography image, and add the processed sample angiography image and the second label to the target sample angiography image set; the second label indicates that the processed sample angiography image is obtained by performing angiography on the second intracranial blood vessel, and the first intracranial blood vessel and the second intracranial blood vessel satisfy the physical characteristics of the intracranial blood vessel;

[0008] Pre-train a contrast image recognition model using an initial sample angiography image set and a target sample angiography image set to obtain a trained contrast image recognition model, where the trained contrast image recognition model is used to predict any angiography image, and any blood vessel image is obtained by performing angiography on an intracranial target blood vessel.

[0009] On the other hand, an embodiment of the present application provides a data processing device, which includes:

[0010] An acquisition unit, configured to acquire an initial sample angiography image set, where the initial sample angiography image set includes a first sample angiography image and a first label of the first sample angiography image, and the first label of the first sample angiography image indicates that the first sample angiography image is obtained by performing angiography on a first blood vessel in the brain;

[0011] A processing unit, configured to perform data augmentation processing on the first sample angiography image based on the physical characteristics of the intracranial blood vessels to obtain a processed sample angiography image, and perform label transformation processing on the first label based on the physical characteristics of the intracranial blood vessels to obtain a second label of the processed sample angiography image, and add the processed sample angiography image and the second label to the target sample angiography image set; the second label indicates that the processed sample angiography image is obtained by performing angiography on a second blood vessel in the brain, and the first blood vessel and the second blood vessel in the brain satisfy the physical characteristics of the intracranial blood vessels;

[0012] The processing unit is further configured to pre-train a contrast image recognition model using the initial sample angiography image set and the target sample angiography image set to obtain a trained contrast image recognition model, where the trained contrast image recognition model is used to predict any angiography image, and any blood vessel image is obtained by performing angiography on an intracranial target blood vessel.

[0013] In one implementation, the physical characteristics of the intracranial blood vessels include symmetry; when the processing unit is configured to perform data augmentation processing on the first sample angiography image based on the physical characteristics of the intracranial blood vessels to obtain a processed sample angiography image, it is specifically configured to:

[0014] Based on the symmetry included in the physical characteristics of the intracranial blood vessels, perform left-right mirror processing on the first sample angiography image to obtain a processed sample angiography image symmetric to the first sample angiography image, and add the processed sample angiography image to the target sample angiography image set.

[0015] In one implementation, the first intracranial blood vessel can be any one of the intracranial blood vessels, and the intracranial blood vessels include: the left carotid artery, the right carotid artery, the left vertebral artery, and the right vertebral artery; the label of any angiography image is a vector including multiple elements, and each element corresponds to an intracranial blood vessel;

[0016] In the first label corresponding to the first sample angiography image, the element value corresponding to the first intracranial blood vessel is the first element value, and the element values corresponding to the other intracranial blood vessels except the first intracranial blood vessel are the second element values.

[0017] In one implementation, when the processing unit is used to perform label conversion processing on the first label based on the physical characteristics of the intracranial blood vessels to obtain the second label of the processed sample angiography image, it is specifically used for:

[0018] Adjust the element value corresponding to the first intracranial blood vessel in the first label from the first element value to the second element value; and, adjust the element value corresponding to the second intracranial blood vessel in the first label from the second element value to the first element value;

[0019] Use the adjusted first label as the second label of the processed sample angiography image.

[0020] In one implementation, the initial sample angiography image set further includes a second sample angiography image, which is obtained by performing angiography on the third intracranial blood vessel, and the first intracranial blood vessel is different from the third intracranial blood vessel; the physical characteristics of the intracranial blood vessels include fusibility; when the processing unit is used to perform data augmentation processing on the first sample angiography image based on the physical characteristics of the intracranial blood vessels to obtain the processed sample angiography image, it is specifically used for:

[0021] Based on the fusibility included in the physical characteristics of the intracranial blood vessels, fuse the first sample angiography image and the second sample angiography image to obtain the processed sample angiography image.

[0022] In one implementation, when the processing unit is used to perform label conversion processing on the first label based on the physical characteristics of the intracranial blood vessels to obtain the second label of the processed sample angiography image, it is specifically used for:

[0023] Adjust the element value corresponding to the third intracranial blood vessel in the first label to the first element value;

[0024] Determine the adjusted first label as the second label of the processed sample angiography image.

[0025] In one implementation, when the processing unit is used to pre-train the angiography image recognition model with the initial sample angiography image set and the target sample angiography image set to obtain the trained angiography image recognition model, it is specifically used for:

[0026] Call the angiography image recognition model to perform recognition processing on multiple sample angiography images in the initial sample angiography image set, and obtain multiple first recognition results corresponding to the multiple sample angiography images in the initial sample angiography image set;

[0027] Call the angiography image recognition model to perform recognition processing on multiple sample angiography images in the target sample angiography image set, and obtain multiple second recognition results corresponding to the multiple sample angiography images in the target sample angiography image set;

[0028] Based on the difference between the multiple first recognition results and the first labels corresponding to the corresponding sample angiography images, and the difference between the multiple second recognition results and the second labels corresponding to the corresponding sample angiography images, obtain the first loss function of the angiography image recognition model;

[0029] Update the parameters of the angiography image recognition model based on the first loss function to train the angiography image recognition model.

[0030] In one implementation, when the processing unit is used to update the parameters of the angiography image recognition model based on the first loss function, it is specifically used for:

[0031] Obtain an annotated sample angiography image set, which includes multiple annotated sample angiography images and the annotation information of each annotated sample angiography image. The annotation information of any annotated sample angiography image indicates the fragmentation information of the target object included in the fourth intracranial blood vessel of any annotated sample angiography image. The fragmentation information includes the fragmentation risk information of the target object and the fragmentation time information of the target object;

[0032] Call the angiography image recognition model to perform prediction processing on multiple annotated sample angiography images in the annotated sample angiography image set, and obtain the predicted fragmentation information of the target object in the multiple annotated sample angiography images;

[0033] According to the difference between the predicted fragmentation information of the target object in each annotated sample angiography image and the annotation information of the corresponding annotated sample angiography image, obtain the second loss function of the angiography image recognition model;

[0034] Optimize the angiography image recognition model based on the first loss function and the second loss function. The trained angiography image recognition model is used to predict the intracranial target blood vessel in any angiography image, and to predict the fragmentation information of the target object of the intracranial target blood vessel in any angiography image.

[0035] In one implementation, when the processing unit is used to optimize the contrast image recognition model based on the first loss function and the second loss function, it is specifically used for:

[0036] Perform a weighted process on the first loss function and the second loss function to obtain a third loss function after the weighted process;

[0037] Update the parameters of the contrast image recognition model according to the third loss function.

[0038] In one implementation, the processing unit is further used for:

[0039] Obtain an annotated sample angiography image set, where the annotated sample angiography image set includes a plurality of annotated sample angiography images and the annotation information of each annotated sample angiography image. The annotation information of any one of the annotated sample angiography images indicates the fragmentation information of the target object included in the fourth intracranial blood vessel in any one of the annotated sample angiography images. The fragmentation information includes the fragmentation risk information of the target object and the fragmentation time information of the target object;

[0040] Call the annotated sample angiography image set to train the trained contrast image recognition model to obtain a target contrast image recognition model, and the target contrast image recognition model is used to predict the intracranial target blood vessel included in any angiography image and the fragmentation information of the target object in the intracranial target blood vessel.

[0041] In one implementation, the processing unit is further used for:

[0042] Obtain the angiography image to be analyzed;

[0043] Call the trained contrast image recognition model to perform recognition processing on the angiography image to be analyzed to obtain the intracranial target blood vessel in the angiography image and the fragmentation information of the target object in the intracranial target blood vessel.

[0044] On the other hand, an embodiment of the present application provides a data processing device, and the data processing device includes:

[0045] A processor, adapted to execute a computer program;

[0046] A computer-readable storage medium, in which a computer program is stored. When the computer program is executed by the processor, the data processing method as described above is implemented.

[0047] On the other hand, an embodiment of the present application provides a computer-readable storage medium, and the computer-readable storage medium stores a computer program, and the computer program is adapted to be loaded and executed by the processor to implement the data processing method as described above.

[0048] On the other hand, an embodiment of the present application provides a computer program product or a computer program. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. A processor of a data processing device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the data processing device executes the above-mentioned data processing method.

[0049] In an embodiment of the present application, which intracranial blood vessel is specifically included in the first sample angiogram image is determined before DSA scanning. Then, the annotation information of the first sample angiogram image (that is, which intracranial artery is included in the DSA image of the intracranial aneurysm) is already contained in the first label of the first sample angiogram image, eliminating the need for manual annotation of data, improving the training efficiency, and reducing the workload. Moreover, based on the physical characteristics of the intracranial blood vessels, data augmentation processing is performed on the sample angiogram images in the initial sample angiogram image set to obtain relatively rich sample angiogram images for training the angiogram image recognition model, ensuring the training process of the angiogram image recognition model and obtaining a relatively excellent angiogram image recognition model. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0051] Figure 1a FIG. 1 shows a schematic diagram of an intracranial aneurysm provided by an exemplary embodiment of the present application;

[0052] Figure 1b FIG. 2 shows a schematic diagram of a DSA image of an intracranial aneurysm provided by an exemplary embodiment of the present application;

[0053] Figure 2 FIG. 3 shows a schematic flowchart of a data processing method provided by an exemplary embodiment of the present application;

[0054] Figure 3 FIG. 4 shows a schematic diagram of data augmentation processing based on symmetry provided by an exemplary embodiment of the present application;

[0055] Figure 4 FIG. 5 shows a schematic diagram of data augmentation processing based on fusibility provided by an exemplary embodiment of the present application;

[0056] Figure 5 FIG. 6 shows a schematic diagram of an angiogram image recognition model provided by an exemplary embodiment of the present application;

[0057] Figure 6a Shows a schematic diagram of another contrast image recognition model provided by an exemplary embodiment of the present application;

[0058] Figure 6b Shows a schematic flowchart of another data processing method provided by an exemplary embodiment of the present application;

[0059] Figure 7 Shows a schematic flowchart of another data processing method provided by an exemplary embodiment of the present application;

[0060] Figure 8 Shows a schematic flowchart of another data processing method provided by an exemplary embodiment of the present application;

[0061] Figure 9 Shows a schematic diagram of the structure of a data processing device provided by an exemplary embodiment of the present application;

[0062] Figure 10 Shows a schematic diagram of the structure of a data processing device provided by an exemplary embodiment of the present application. Detailed implementation manners

[0063] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0064] The embodiments of the present application propose a data processing solution, and this data processing solution involves the following concepts and terms:

[0065] 1. Technologies such as artificial intelligence.

[0066] Artificial Intelligence (AI) uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, including theories, methods, technologies, and application systems that can perceive the environment, acquire knowledge, and use knowledge to achieve the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a way similar to human intelligence. Artificial intelligence also studies the design principles and implementation methods of various intelligent machines to enable them to have functions of perception, reasoning, and decision-making. Artificial intelligence technology is an interdisciplinary subject with a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics. The software technologies of artificial intelligence mainly include several major directions such as computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0067] Among them, Machine Learning (ML) is an interdisciplinary subject that involves multiple disciplines such as probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve their own performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent, and its applications cover all fields of artificial intelligence. Machine learning and deep learning usually include technologies such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and rote learning. Machine learning can be regarded as a task, and the goal of this task is to enable machines (computers in a broad sense) to obtain human-like intelligence through learning. For example, humans can play Go, and computer programs (AlphaGo or AlphaGo Zero) are designed to master Go knowledge and be able to play Go. Among them, multiple methods can be used to achieve the task of machine learning, such as neural networks, linear regression, decision trees, support vector machines, Bayesian classifiers, reinforcement learning, probabilistic graphical models, clustering, and many other methods.

[0068] II. Technologies such as self-supervised pre-training.

[0069] It has been found through practice that the DSA images of intracranial aneurysms are different from natural images, and the DSA images of intracranial aneurysms are three-dimensional medical image data. Since the network models used to process three-dimensional images (such as three-dimensional deep convolutional neural networks) have far more parameters to learn than two-dimensional network models, and according to the characteristic that the DSA images of intracranial aneurysms are three-dimensional data, a large amount of labeled data is required to train the network model in order to obtain a network model for predicting the DSA images of intracranial aneurysms. However, in clinical practice, it is often difficult to obtain a large amount of labeled data due to various objective limitations. Based on this, the data processing solution involved in the embodiments of the present application also involves transfer learning and self-supervised pretraining, where:

[0070] Transfer learning is a commonly used technique in the field of natural images to solve the problem of insufficient training data and / or labels. Transfer learning transfers (copies) the pre-trained model parameters (using data and labels for non-current tasks) to a new network model to assist in training. However, the parameters of the network model pre-trained on two-dimensional images or natural video data are not applicable or ideal for three-dimensional medical image data such as DSA images of intracranial aneurysms.

[0071] Self-supervised pretraining can be considered a variant of transfer learning. Different from transfer learning that uses data and labels for non-current tasks to pre-train the parameters of the network model, self-supervised pretraining relies on proxy tasks directly defined on the current target task data for pre-training, and the proxy tasks themselves provide supervision signals (without the need to provide additional labels). Research shows that the network model pre-trained through self-supervised proxy tasks can achieve satisfactory results with only less labels.

[0072] In short, the main principle of the above self-supervised pretraining includes: pre-training the network model based on proxy tasks (or self-supervised tasks) with a relatively sufficient number of samples to obtain a pre-trained network model; and providing supervision information by the proxy tasks, training the pre-trained network model based on the target task with a small amount of labeled data (that is, transferring the parameters of the pre-trained network model to a new network model to facilitate training the new network model based on the target task with a small amount of labeled data) to obtain a trained network model (or new network model); where the self-supervised task and the target task have a high degree of relevance. This can better solve the problem of insufficient labels, ensure the training process of the network model, and ensure the performance of the trained network model.

[0073] III. Techniques such as angiography.

[0074] Angiography is an auxiliary examination technique for blood vessels; by injecting a contrast agent (or called a radiographic contrast agent) into the blood vessels and taking advantage of the property that X-rays (i.e., a type of electromagnetic wave commonly used as an auxiliary examination in medicine) cannot penetrate the contrast agent, angiographic images of blood vessels can be obtained. Common angiography techniques include Digital Subtraction Angiography (DSA) technique. DSA technique is the gold standard for the diagnosis of intracranial aneurysms. By scanning the human brain with DSA, DSA images of intracranial aneurysms in the human brain can be obtained. The so-called intracranial aneurysm (or simply called aneurysm) is a foreign object formed by the long-term impact of blood flow on the wall of intracranial artery blood vessels, and its shape is similar to that of a tumor. For a schematic diagram of an intracranial aneurysm, see Figure 1a .

[0075] Among them, the method of collecting angiographic images of intracranial blood vessels using DSA technique, that is, performing angiography on cerebral artery blood vessels, may include: First, perform local anesthesia on a specific area of the groin; Second, insert a microcatheter into the femoral artery, and then the catheter ascends through large arteries such as the abdominal aorta and reaches the cerebral artery blood vessels in the brain; Finally, inject the contrast agent through the catheter into the cerebral artery to enhance its visualization, and then obtain angiographic images through X-ray transmission. The angiographic images here are DSA images of intracranial aneurysms. Also see Figure 1b , and the intracranial blood vessels in the angiographic images (i.e., DSA images of intracranial aneurysms) include 2 carotid artery blood vessels (i.e., the left carotid artery blood vessel and the right carotid artery blood vessel) and 2 vertebral artery blood vessels (i.e., the left vertebral artery blood vessel and the right vertebral artery blood vessel).

[0076] It should be noted that intracranial blood vessels have unique physical properties (i.e., clinical imaging characteristics), including but not limited to: symmetry (i.e., the left-right symmetry of the human brain), fusibility (i.e., according to multiple angiographic images containing different intracranial blood vessels, angiographic images containing multiple intracranial blood vessels can be fused), only 1 or 2 arteries are included in the DSA images of intracranial aneurysms, and there is a preference for the predilection sites of intracranial aneurysms (such as aneurysms are prone to occur in the left carotid artery blood vessel, etc.).

[0077] Since latent asymptomatic intracranial aneurysms are not uncommon in the general population - the overall prevalence of unruptured intracranial aneurysms is estimated to be 3.2%. Unless accidentally discovered or compressing adjacent nerves, most intracranial aneurysms have no obvious symptoms before rupture; moreover, the fatality rate of subarachnoid hemorrhage caused by the rupture of intracranial aneurysms is relatively high (about 50%), and most survivors (up to 46%) may suffer from long-term cognitive impairment, which greatly affects the ability to live and quality of life. Therefore, in view of the universality and high risk of intracranial aneurysms when ruptured, early screening of intracranial aneurysms and the formulation of appropriate treatment plans have important clinical value.

[0078] In actual research, it is found that the splitting risks of aneurysms on different intracranial blood vessels are different, that is to say, the rupture risk of an aneurysm has a certain correlation with the intracranial artery on which it is located. Based on this, in order to assist doctors in better analyzing information such as the rupture risk and rupture time of intracranial aneurysms to help patients treat diseases, currently, the prediction is mainly carried out on DSA images of intracranial aneurysms, that is, to predict which intracranial blood vessel is contained in the DSA image of the intracranial aneurysm to obtain the intracranial blood vessel recognition result; then, according to the correlation between the intracranial blood vessel and the intracranial aneurysm, such as the larger rupture risk of the intracranial aneurysm in a certain intracranial blood vessel, etc., the intracranial blood vessel recognition result is used to assist doctors in analyzing information such as the rupture risk of the intracranial aneurysm. Through the above process, it can be seen that accurately identifying the intracranial blood vessel from the DSA image of the intracranial aneurysm is of great significance for subsequent analysis of the rupture information of the intracranial aneurysm. Based on this, the embodiment of the present application proposes a self-supervised training task based on the clinical imaging characteristics of DSA images of intracranial aneurysms, and this self-supervised training task proposes a prediction task for predicting the intracranial target blood vessel in the DSA image of the intracranial aneurysm; the data processing solution proposed by the embodiment of the present application is implemented by training a contrast image recognition model based on this self-supervised training task.

[0079] In addition, the data processing solution proposed by the embodiment of the present application can be executed by a computer device (or data processing device). The computer device here may include, but is not limited to: terminal devices such as smart phones, tablet computers, laptop computers, desktop computers, etc.; or service devices such as data processing servers, web servers, application servers, etc. Among them, the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing 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. Here, the server can be a node server on the blockchain. The terminal device and the service device can be directly or indirectly connected through wired or wireless communication methods, and the present application does not make any restrictions here. For the convenience of description, in the following, the example of a computer device executing the data processing solution proposed by the embodiment of the present application is used for illustration. Here, the execution subject of the data processing solution is described, which does not limit the embodiment of the present application.

[0080] The specific implementation of the data processing solution may include: obtaining an initial sample angiography image set, where the initial sample angiography image set includes a first sample angiography image and a first label of the first sample angiography image; performing data augmentation processing on the first sample angiography image based on the physical characteristics of the intracranial blood vessels to obtain a processed sample angiography image, and performing label transformation processing on the first label based on the physical characteristics of the intracranial blood vessels to obtain a second label of the processed sample angiography image, and adding the processed sample angiography image and the second label to the target sample angiography image set; pre-training the angiography image recognition model using the initial sample angiography image set and the target sample angiography image set to obtain a trained angiography image recognition model.

[0081] In the above process, which specific intracranial blood vessel is included in the first sample angiography image is determined before the DSA scan. Then, the annotation information of the first sample angiography image (i.e., which intracranial artery is included in the DSA image of the intracranial aneurysm) is already contained in the first label of the first sample angiography image. There is no need for manual annotation data, which can improve the training efficiency and reduce the workload. Moreover, based on the physical characteristics of the intracranial blood vessels, data augmentation processing is performed on the sample angiography images in the initial sample angiography image set to obtain relatively rich sample angiography images for training the angiography image recognition model, which can ensure the training process of the angiography image recognition model and obtain a angiography image recognition model with better performance.

[0082] The following combines the attached Figure 2 Introduce in detail the data processing solution proposed in this application. Figure 2 The flowchart of a data processing method provided by an exemplary embodiment of this application is shown. This data processing method can be executed by the computer device mentioned above. This solution includes but is not limited to steps S201 - S203, where:

[0083] S201. Obtain an initial sample angiography image set.

[0084] The initial sample angiography image set includes a plurality of sample angiography images and first labels for each sample angiography image. Among them, the plurality of sample angiography images includes a first sample angiography image and the first label of the first sample angiography image; the first label of the first sample angiography image can be used to indicate that the first sample angiography image is obtained by performing angiography on the first intracranial blood vessel; in other words, by analyzing the first label of the first sample angiography image, it can be determined which specific intracranial blood vessel the first sample angiography image contains. Among them, the first intracranial blood vessel is any one of the intracranial blood vessels. As described above, the intracranial blood vessels include: the left carotid artery, the right carotid artery, the left vertebral artery, and the right vertebral artery. Then the first intracranial blood vessel can be any one of the left carotid artery, the right carotid artery, the left vertebral artery, and the right vertebral artery.

[0085] It should be noted that: (1) In addition to the first intracranial blood vessel, the first sample angiography image may also include other intracranial blood vessels; in other words, a frame of angiography image may include multiple intracranial blood vessels. (2) If the angiography image is obtained by performing angiography on other parts of the human body (such as the chest), then the number and type of blood vessels contained in the angiography image at this time will be different from the above-mentioned intracranial blood vessels, and no detailed description will be given here.

[0086] The labels of any angiography image are introduced in detail below. Optionally, the label of any angiography image can be represented as a vector including multiple elements, and each element corresponds to an intracranial blood vessel. In a specific implementation, the label of any angiography image is represented as a vector y = [y1, y2, y3, y4], where the element y1 represents the left carotid artery, the element y2 represents the right carotid artery, the element y3 represents the left vertebral artery, and the element y4 represents the right vertebral artery; when a certain intracranial blood vessel is included in the angiography image, the element value corresponding to the certain intracranial blood vessel is the first element value (such as the value 1), otherwise, the element value corresponding to the certain intracranial blood vessel is the second element value (such as the value 0). For example: assuming that the first intracranial blood vessel included in the first sample angiography image is the left carotid artery, in the first label corresponding to the first sample angiography image, the element value corresponding to the first intracranial blood vessel (i.e., the left carotid artery, which is the element value of the element y1) is set to the first element value, and the element values corresponding to the other intracranial blood vessels except the first intracranial blood vessel (i.e., the elements y2, y3, y4) are set to the second element value; further, the first label of the first sample angiography image is obtained as y = [0, 1, 0, 0], indicating that the first sample angiography image is obtained by performing angiography on the right carotid artery. Among them, the positions of the elements in the vector y = [y1, y2, y3, y4] in the vector can be changed. For example, the vector y = [y1, y2, y3, y4] is represented as the vector y = [y3, y1, y3, y4], etc. The embodiments of the present application do not limit this.

[0087] S202. Perform data augmentation processing on the first sample angiography image based on the physical characteristics of the intracranial blood vessels to obtain a processed sample angiography image, and perform label conversion processing on the first label based on the physical characteristics of the intracranial blood vessels to obtain a second label of the processed sample angiography image, and add the processed sample angiography image and the second label to the target sample angiography image set.

[0088] As described above, intracranial blood vessels have the physical characteristics of specific intracranial blood vessels. Therefore, in addition to the self-supervised training tasks mentioned above, the embodiments of the present application also support enriching the tasks and data for pre-training the angiography image recognition model based on the physical characteristics of intracranial blood vessels by operating the first sample angiography image and the first label corresponding to the first sample angiography image. Specifically, the first sample angiography image is processed for data augmentation based on the physical characteristics of intracranial blood vessels to obtain a processed sample angiography image, and the first label of the first sample angiography image is processed for label transformation to obtain a second label of the processed sample angiography image; wherein, the second label of the processed sample angiography image can be used to indicate that the processed sample angiography image is obtained by angiography of the second intracranial blood vessel, so that the first intracranial blood vessel and the second intracranial blood vessel satisfy the physical characteristics of intracranial blood vessels.

[0089] According to different physical characteristics of intracranial blood vessels, the methods of data augmentation processing for the first sample angiography image and label transformation processing for the first label will be different. Taking two physical characteristics of intracranial blood vessels as examples below, the methods of data augmentation processing for the first sample angiography image and label transformation processing for the first label will be elaborated in detail, where:

[0090] (1) The physical characteristics of intracranial blood vessels include symmetry.

[0091] Based on the knowledge in the medical field, the human brain is generally symmetric left and right, and the left-right symmetry of the human brain includes the left-right symmetry of the arterial blood vessels contained in the human brain. For example, there is symmetry between the left carotid artery and the right carotid artery, and for another example, there is symmetry between the left vertebral artery and the right vertebral artery. The left-right symmetry here can be understood as: the angiography image of the right brain of the same human brain can be obtained through the angiography image of the left brain. At this time, there is left-right mirror symmetry between the angiography image of the left brain and the angiography image of the right brain.

[0092] Taking the initial sample angiography image set including the first sample angiography image and the first label of the first sample angiography image as an example below, the implementation methods of data augmentation processing for the first sample angiography image and label transformation processing for the first label based on symmetry will be introduced, where:

[0093] 1) The implementation process of data augmentation processing for the first sample angiography image based on symmetry includes: performing left-right mirror processing on the first sample angiography image based on the symmetry included in the physical characteristics of intracranial blood vessels to obtain a processed sample angiography image symmetric to the first sample angiography image; and adding the processed sample angiography image to the target sample angiography image set. For example, assume that the angiography image of the left brain of a certain user isFigure 3 For the first sample angiography image 301 in, based on the symmetry included in the physical characteristics of the intracranial blood vessels, after performing a left-right mirror processing on the first sample angiography image 301, the angiography image of the right brain of a certain user can be obtained as Figure 3 the processed sample angiography image 302 in. Based on Figure 3 it can be known that the first sample angiography image 301 and the processed sample angiography image 302 present left-right mirror symmetry.

[0094] 2) The implementation process of performing label transformation processing on the first label of the first sample angiography image based on symmetry includes: First, adjust the element value corresponding to the first intracranial blood vessel in the first label of the first sample angiography image from the first element value to the second element value; Second, adjust the element corresponding to the second intracranial blood vessel in the first label from the second element value to the first element value; Finally, use the adjusted first label as the second label of the processed sample angiography image. Referring again to Figure 3 , if the first intracranial blood vessel included in the first sample angiography image 301 is the right carotid artery blood vessel, then the label of the first sample angiography image 301 is y = [0, 1, 0, 0]; by using the above steps to adjust the label of the first sample angiography image 301, the second label of the processed sample angiography image can be obtained as y = [1, 0, 0, 0], that is, the processed sample angiography image is obtained by performing angiography on the left carotid artery blood vessel (i.e., the second intracranial blood vessel).

[0095] In summary, when performing data augmentation processing on the first sample angiography image through the left-right symmetry of the human brain, not only the processed sample angiography image corresponding to the first sample angiography image is generated, but also the second label of the processed sample angiography image is generated, which can enrich the tasks and data for pre-training and further improve the performance of the angiography image recognition model trained based on the sample angiography image.

[0096] (2) The physical characteristics of the intracranial blood vessels include fusibility.

[0097] The so-called fusibility means that at least two sample angiography images in the initial sample angiography image set are fused to obtain a more abundant sample angiography image; among them, at least two sample angiography images are obtained by performing angiography on different intracranial blood vessels. For the convenience of description, in the following, two sample angiography images are taken as an example for illustration. The two sample angiography images are respectively: the first sample angiography image and the second angiography image; among them, the second sample angiography image is obtained by performing angiography on the third intracranial blood vessel, and the first intracranial blood vessel is different from the third intracranial blood vessel; in other words, the first sample angiography image and the second sample angiography image are obtained by performing angiography on different intracranial blood vessels.

[0098] Taking the initial sample angiography image set including the first sample angiography image, the second sample angiography image, the first label of the first sample angiography image, and the first label of the second sample angiography image as an example, the implementation method of data augmentation processing for the first sample angiography image and the second sample angiography image based on fusibility, and label conversion processing for the first label of the first sample angiography image and the first label of the second sample angiography image is introduced, where:

[0099] 1) The implementation process of data augmentation processing for the first sample angiography image and the second sample angiography image based on fusibility includes: based on the fusibility included in the physical characteristics of the intracranial blood vessels, the first sample angiography image and the second sample angiography image are fused to obtain the processed sample angiography image. Combined with the appendix Figure 4 For illustration, assume that the initial sample angiography image set includes the first sample angiography image 401 and the second angiography image 402. Then, based on the fusibility included in the physical characteristics of the intracranial blood vessels, after fusing the first sample angiography image 401 and the second angiography image 402, the fused processed sample angiography image 403 can be obtained. Based on Figure 4 It can be known that the fused processed sample angiography image 403 includes the first sample angiography image 401 and the second sample angiography image 402.

[0100] 2) The implementation process of performing label conversion processing on the first label of the first sample angiography image and the first label of the second sample angiography image based on fusibility includes: adjusting the element value corresponding to the third intracranial blood vessel in the first label of the first sample angiography image to the first element value, and determining the adjusted first label as the second label of the processed sample angiography image. In other words, the above process can be elaborated in the following steps: First, determine the third intracranial blood vessel corresponding to the element with the element value of the first element value in the second sample angiography image; Second, adjust the element value corresponding to the third intracranial blood vessel in the first label of the first sample angiography image from the second element value to the first element value; Finally, use the adjusted first label of the first sample angiography image as the second label of the processed sample angiography image. For example, referring to Figure 4 , assume that the first intracranial blood vessel included in the first sample angiography image 401 is the right carotid artery, then the first label of the first sample angiography image 401 is y = [0, 1, 0, 0]. Assume that the third intracranial blood vessel included in the second sample angiography image 402 is the left carotid artery, then the first label of the second sample angiography image 402 is y = [1, 0, 0, 0]; After fusing the first label y = [0, 1, 0, 0] of the first sample angiography image 401 and the first label y = [1, 0, 0, 0] of the second sample angiography image 402, the second label of the processed sample angiography image 403 is obtained as y = [1, 1, 0, 0].

[0101] It should be noted that the above implementation method performs label conversion based on the first label of the first sample angiography image. In other implementation methods, label conversion can also be performed based on the first label of the second sample angiography image, or based on the original label with the element value of the second element value for all elements (such as the label y = [0, 0, 0, 0]). The embodiments of the present application do not limit this.

[0102] It should be noted that, in addition to the direct fusion described above, other fusion methods can also be used for fusing at least two frames of sample angiography images. For example, at least two frames of sample angiography images are first subjected to left-right mirror processing respectively, and then the at least two frames of sample angiography images after the mirror processing are subjected to fusion processing to obtain a processed sample angiography image after fusion. The specific implementation process of this implementation method may include: obtaining a blank original image; secondly, performing left-right mirror processing on the first sample angiography image so that the blank original image is changed into an image containing the processed sample angiography image corresponding to the first sample angiography image; finally, using the image containing the processed angiography image corresponding to the first sample angiography image as the original image, performing left-right mirror processing on the second sample angiography image to obtain an image containing both the processed angiography image corresponding to the first sample angiography image and the processed angiography image corresponding to the second sample angiography image, and using this image as the processed sample angiography image after fusion processing.

[0103] S203. Use the initial sample angiography image set and the target sample angiography image set to pre-train the angiography image recognition model to obtain a trained angiography image recognition model.

[0104] The angiography image recognition model can be various common lightweight classification network models, including but not limited to: network models such as ResNet18, EfficientNet-B0, etc.; among them, the ResNet18 classification network model is a CNN feature extraction network; the network model of the EfficientNet-B0 classification network model is a baseline network model. By optimizing the depth, width, and input resolution of the EfficientNet-B0 classification network model, the EfficientNet-B0 classification network model can obtain higher accuracy, thereby ensuring the performance of the EfficientNet-B0 classification network model. It should be noted that the angiography image recognition model can be any one of the above two classification network models, or other lightweight classification network models. The embodiments of the present application do not limit which specific classification network model the angiography image recognition model is.

[0105] In a specific implementation, the general principle of pre-training a contrast image recognition model using an initial sample angiography image set and a target sample angiography image set is as follows: First, call the contrast image recognition model to perform recognition processing on multiple sample angiography images in the initial sample angiography image set to obtain multiple first recognition results corresponding to the multiple sample angiography images in the initial sample angiography image; Second, call the contrast image recognition model to perform recognition processing on multiple sample angiography images in the target sample angiography image set to obtain multiple second recognition results corresponding to the multiple sample angiography images in the target sample angiography image; Second, based on the difference between the multiple first recognition results and the first labels corresponding to the corresponding sample angiography images, and the difference values between the multiple second recognition results and the second labels corresponding to the corresponding sample angiography images, obtain the first loss function of the contrast image recognition model; Finally, update the parameters of the contrast image recognition model based on the first loss function to train the contrast image recognition model.

[0106] It is worth mentioning that the embodiments of the present application support using binary cross-entropy loss to supervise the training of the contrast image recognition model. Among them, the loss function of binary cross-entropy is:

[0107]

[0108] Among them, is the loss function, y k is the element value of the kth intracranial blood vessel in the first label of a certain sample angiography image, and p k is the probability of predicting that a certain sample angiography image of the sample contains the kth intracranial blood vessel.

[0109] It should be noted that to determine whether the trained contrast angiography image recognition model meets the training requirements is achieved by determining whether the loss function meets the training requirements. In one implementation, if in a certain round of training, the loss value of the loss function is less than the loss threshold, it is determined that the trained contrast angiography image recognition model meets the training requirements, and at this time, the training can be ended; if in a certain round of training, the loss value of the loss function is equal to or greater than the loss threshold, it is determined that the trained contrast angiography image recognition model does not meet the training requirements, and it is not accurate enough to use the currently trained contrast angiography image recognition model for prediction. Then, the parameters of the contrast angiography image recognition model need to be adjusted according to the loss value to optimize the contrast angiography image recognition model, and the sample angiography image is input again to train the contrast angiography image recognition model after the previous parameter adjustment until the contrast angiography image recognition model meets the training requirements. It should be noted that the loss thresholds of different network models may be different, and no detailed description is given here. In another implementation, after a certain round of training, the loss of the loss function is compared with the historical loss value. If the difference between the loss value and the historical loss value is small, it is determined that the trained contrast angiography image recognition model meets the training requirements, and at this time, the training can be ended; otherwise, the sample angiography image is input again to train the contrast angiography image recognition model after the previous parameter adjustment until the contrast angiography image recognition model meets the training requirements.

[0110] The following combines Figure 5 , taking the training of a contrast angiography image recognition model with a specific sample angiography image as an example, to introduce the implementation process of the above pre-training. In the actual pre-training process, the following steps are included: ① Take any sample angiography image in the initial sample angiography image set (or target sample angiography image set) as the input of the contrast angiography image recognition model. ② The contrast recognition image model recognizes the probability of each intracranial blood vessel contained in the any sample angiography image to obtain the recognition result. Among them, a vector can be used to represent the probability of each intracranial blood vessel contained in the any sample angiography image; for example, the vector used to represent the probability of each intracranial blood vessel contained in the any sample angiography image is p = [p3, p1, p3, p4], p x corresponds to an intracranial blood vessel. For example, p1 corresponds to the left carotid artery, p2 corresponds to the right carotid artery, p3 corresponds to the left vertebral artery, and p4 corresponds to the right vertebral artery, p xThe value can be any percentage greater than or equal to 0 and less than or equal to 100%, where x is 1, 2, 3, or 4. For example, the probability that any sample angiography image includes each intracranial blood vessel is expressed as p = [0, 100%, 80%, 40%], which means: the probability that the any sample angiography image includes the left carotid artery is 0, the probability that the any sample angiography image includes the right carotid artery is 100%, the probability that the any sample angiography image includes the left vertebral artery is 80%, and the probability that the any sample angiography image includes the right vertebral artery is 40%. ③ Based on the difference between the recognition result of the any sample angiography image and the label corresponding to the any sample angiography image, the first loss function of the angiography image recognition model is obtained. ④ If the first loss function meets the training requirements, the training is terminated, and the trained angiography image recognition model is obtained. If the first loss function does not meet the training requirements, the parameters of the angiography image recognition model are adjusted based on the first loss function, and the above steps ①-④ are repeated until the first loss function meets the training requirements, and the trained angiography image recognition model is obtained.

[0111] In the embodiment of the present application, which intracranial blood vessel is specifically included in the first sample angiography image is determined before the DSA scan. Then, the annotation information of the first sample angiography image (i.e., which intracranial artery is included in the DSA image of the intracranial aneurysm) is already contained in the first label of the first sample angiography image, without the need for manual annotation data, which can improve the training efficiency and reduce the workload. Moreover, based on the physical characteristics of the intracranial blood vessels, the sample angiography images in the initial sample angiography image set are processed for data augmentation to obtain relatively rich sample angiography images for training the angiography image recognition model, which can ensure the training process of the angiography image recognition model and obtain an angiography image recognition model with better performance.

[0112] As described above, the pre-training process of the above-mentioned contrast image recognition model is implemented based on a self-supervised training task. The trained contrast image recognition model can accurately identify the intracranial target blood vessels contained in any angiography image. According to the physical characteristics of intracranial blood vessels (such as the preferential occurrence location of intracranial aneurysms, the significant differences in the incidence of intracranial aneurysms on different intracranial blood vessels, the correlation between the rupture information of intracranial aneurysms and each intracranial blood vessel, etc.), this application also supports proposing a target task that has a high correlation with the self-supervised training task based on the self-supervised training task; the target task may include, but is not limited to: predicting the rupture information of intracranial aneurysms (i.e., target objects) in intracranial blood vessels, where the rupture information includes the rupture risk information of the target object and the rupture time information of the target object, etc. Then, based on the target task, the contrast image recognition model trained based on the self-supervised training task is trained again to obtain a trained contrast image recognition model; this trained contrast image recognition model can be used to predict the intracranial target blood vessels in any angiography image and the rupture information of the target objects of the intracranial target blood vessels in any angiography image. In this way, a contrast image recognition model for predicting the rupture information of the target objects of the intracranial target blood vessels in any angiography image can be trained through a small amount of labeled information (i.e., information for labeling the rupture information of the target objects), improving the performance of the contrast image recognition model in the angiography image analysis task under the condition of insufficient labeled information, and obtaining a contrast image recognition model with better prediction performance.

[0113] It should be noted that the embodiments of this application support training the contrast image recognition model based on the self-supervised training task and the target task together to obtain a trained contrast image recognition model. The embodiments of this application also support first training the contrast image recognition model based on the self-supervised training task to obtain a trained contrast image recognition model, and then training the trained contrast image recognition model based on the target task to obtain a target contrast image recognition model. The following introduces the above training methods respectively, where:

[0114] (1) Training the contrast image recognition model based on the self-supervised training task and the target task together to obtain a trained contrast image recognition model.

[0115] The following combines the attached Figure 6a and the attached Figure 6b to elaborate on this implementation method in detail. Among them, Figure 6a shows a contrast image recognition model for multi-task training provided by an exemplary embodiment of this application; Figure 6bThe flowchart shows a data processing method provided by an exemplary embodiment of the present application; this data processing method can be executed by the computer device mentioned above. This solution includes but is not limited to steps S601 - S608, where:

[0116] S601. Obtain an initial sample angiography image set, where the initial sample angiography image set includes a first sample angiography image and a first label of the first sample angiography image.

[0117] S602. Perform data augmentation processing on the first sample angiography image based on the physical characteristics of the intracranial blood vessels to obtain a processed sample angiography image, and perform label transformation processing on the first label based on the physical characteristics of the intracranial blood vessels to obtain a second label of the processed sample angiography image, and add the processed sample angiography image and the second label to the target sample angiography image set.

[0118] S603. Invoke the angiography image recognition model to perform recognition processing on multiple sample angiography images in the initial sample angiography image set to obtain multiple first recognition results corresponding to the multiple sample angiography images in the initial sample angiography image; invoke the angiography image recognition model to perform recognition processing on multiple sample angiography images in the target sample angiography image set to obtain multiple second recognition results corresponding to the multiple sample angiography images in the target sample angiography image.

[0119] S604. Based on the differences between the multiple first recognition results and the first labels corresponding to the respective sample angiography images, and the differences between the multiple second recognition results and the second labels corresponding to the respective sample angiography images, obtain the first loss function of the angiography image recognition model.

[0120] It should be noted that for the specific implementation processes of steps S601 - S604, reference can be made to Figure 2 the relevant descriptions of the specific implementation processes of steps S201 - S203 in the shown embodiment, which will not be elaborated here.

[0121] S605. Obtain an annotated sample angiography image set, where the annotated sample angiography image set includes multiple annotated sample angiography images and the annotation information of each annotated sample angiography image.

[0122] Among them, the annotation information of any one annotated sample angiography image indicates the fragmentation information of the target object (such as an intracranial aneurysm) contained in the fourth intracranial blood vessel of the any one annotated sample angiography image.

[0123] S606. Call the contrast image recognition model to perform prediction processing on multiple labeled sample angiography images in the labeled sample angiography image set, and obtain the predicted fragmentation information of the target object in the multiple labeled sample angiography images.

[0124] Among them, according to the different target tasks, the manifestation forms of the predicted fragmentation information of the target object may be different. For example, if the target task is to predict the fragmentation risk information of the target object in the intracranial blood vessels, then the predicted fragmentation information of the target object may be expressed in forms such as percentages and decimals; another example is that if the target task is to predict the fragmentation time of the target object in the intracranial blood vessels, then the predicted fragmentation information of the target object can be expressed in the form of a time period (such as 24 hours) or a moment form (such as 12:00), etc.

[0125] S607. According to the difference between the predicted fragmentation information of the target object in each labeled sample angiography image and the labeled information of the corresponding sample angiography image, obtain the second loss function of the contrast image recognition model.

[0126] It should be noted that similar to the first loss function, in the embodiments of the present application, binary cross-entropy loss can also be used to supervise the training of the contrast image recognition model based on the target task, which will not be elaborated here.

[0127] S608. Optimize the contrast image recognition model based on the first loss function and the second loss function to obtain the trained contrast image recognition model.

[0128] Among them, the contrast image recognition model optimized based on the first loss function and the second loss function can be used to predict the intracranial target blood vessels in any angiography image, and to predict the fragmentation information of the target object in the any angiography image.

[0129] As an alternative implementation manner, the implementation manner of optimizing the contrast image recognition model based on the first loss function and the second loss function may include: performing weighted processing on the first loss function and the second loss function to obtain a third loss function after weighted processing; updating the parameters of the contrast image recognition model according to the third loss function to train the contrast image recognition model. For example, assume that the first loss function is L1, the second loss function is L2, the weighting coefficient of the first loss function L1 is a, and the weighting coefficient of the second loss function L2 is b, then the third loss function L3 = a * L1 + b * L 2。

[0130] (2) First, train the contrast image recognition model based on the self-supervised training task to obtain the trained contrast image recognition model, and then train the trained contrast image recognition model based on the target task to obtain the target contrast image recognition model.

[0131] The following will elaborate on this implementation manner in conjunction with the accompanying drawings. Figure 7 This will be elaborated in detail below. Figure 7 The flowchart of a data processing method provided by an exemplary embodiment of the present application is shown; this data processing method can be executed by the computer device mentioned above. This solution includes but is not limited to steps S701 - S705, where:

[0132] S701. Obtain an initial sample angiography image set, where the initial sample angiography image set includes a first sample angiography image and a first label of the first sample angiography image.

[0133] S702. Perform data augmentation processing on the first sample angiography image based on the physical characteristics of intracranial blood vessels to obtain a processed sample angiography image, and perform label conversion processing on the first label based on the physical characteristics of intracranial blood vessels to obtain a second label of the processed sample angiography image, and add the processed sample angiography image and the second label to the target sample angiography image set.

[0134] S703. Use the initial sample angiography image set and the target sample angiography image set to pre-train the angiography image recognition model to obtain a trained angiography image recognition model.

[0135] S704. Obtain an annotated sample angiography image set, where the annotated sample angiography image set includes multiple annotated sample angiography images and annotation information for each annotated sample angiography image.

[0136] It should be noted that for the specific implementation processes of steps S701 - S703, reference can be made to the relevant descriptions of the specific implementation processes shown in steps S201 - S203 in the embodiment shown in Figure 2 For the specific implementation process of step S704, reference can be made to the relevant descriptions of the specific implementation process shown in step 605 in the embodiment shown in Figure 6b and will not be elaborated here.

[0137] S705. Call the annotated sample angiography image set to train the trained angiography image recognition model to obtain a target angiography image recognition model.

[0138] Among them, the process of training the trained angiography image recognition model (i.e., the angiography image recognition model trained based on the self-supervised training task) by calling the labeled sample angiography image set is similar to the process of training the angiography image recognition model by calling the initial sample angiography image set and the target sample angiography image set described above. A brief description is as follows: Call the trained angiography image recognition model to perform recognition processing on multiple labeled sample angiography images in the labeled sample angiography image set, and obtain the predicted fragmentation results corresponding to the multiple labeled sample angiography images in the labeled sample angiography image set; Based on the difference between the predicted fragmentation results corresponding to the multiple labeled sample angiography images and the labeled information corresponding to the corresponding labeled sample angiography images, obtain the fourth loss function of the angiography image recognition model; Based on the fourth loss function, adjust the parameters of the trained angiography image recognition model to train the angiography image recognition model; The trained target angiography image recognition model can be used to predict the intracranial target blood vessels in any angiography image, and to predict the fragmentation information of the target object on the intracranial target blood vessels in any angiography image.

[0139] In summary, since the labeled information of the sample angiography images during self-supervised pre-training is already contained in the labels of the sample angiography images, manual labeling is not required to obtain an angiography image recognition model that meets the self-supervised training task. Also, since the self-supervised training task has a high correlation with the target task, by using a small number of labeled sample angiography images of the target task, a target angiography image recognition model that meets the target task can be obtained, ensuring the performance of the angiography image recognition model trained with a small number of labels. In addition, by training the angiography image recognition model based on both the self-supervised training task and the target task simultaneously, a target angiography image recognition model that can both identify the intracranial target blood vessels in the angiography image and predict the fragmentation information of the target object on the intracranial target blood vessels can be obtained in one training, improving the training efficiency and speed.

[0140] Please refer to Figure 8 , Figure 8 which shows a schematic flowchart of another data processing method provided by an exemplary embodiment of the present application; this data processing method can be executed by the computer device mentioned above. This data processing method may include steps S801 - S807, where:

[0141] S801. Obtain an initial sample angiography image set, where the initial sample angiography image set includes a first sample angiography image and a first label of the first sample angiography image.

[0142] S802. Perform data augmentation processing on the first sample angiography image based on the physical characteristics of intracranial blood vessels to obtain the processed sample angiography image, and perform label conversion processing on the first label based on the physical characteristics of intracranial blood vessels to obtain the second label of the processed sample angiography image, and add the processed sample angiography image and the second label to the target sample angiography image set.

[0143] S803. Use the initial sample angiography image set and the target sample angiography image set to pre-train the angiography image recognition model to obtain the trained angiography image recognition model.

[0144] S804. Obtain an annotated sample angiography image set, which includes multiple annotated sample angiography images and the annotation information of each annotated sample angiography image.

[0145] S805. Call the annotated sample angiography image set to train the trained angiography image recognition model to obtain the target angiography image recognition model.

[0146] It should be noted that for the specific implementation processes of steps S801 - S805, reference can be made to Figure 7 the relevant descriptions of the specific implementation processes shown in steps S701 - S705 in the embodiments shown, which will not be elaborated here.

[0147] S806. Obtain the angiography image to be analyzed.

[0148] S807. Call the trained angiography image recognition model to perform recognition processing on the angiography image to be analyzed to obtain the intracranial target blood vessels in the angiography image and the fragmentation information of the target object of the intracranial target blood vessels.

[0149] In the embodiments of the present application, data augmentation processing can be performed on the sample angiography images in the initial sample angiography image set based on the physical characteristics of intracranial blood vessels to obtain relatively rich sample angiography images for training the angiography image recognition model, and then obtain a angiography image recognition model with better performance; using this angiography image recognition model can effectively identify the intracranial target blood vessels in any angiography image (such as the DSA image of intracranial aneurysm), and predict the fragmentation information of the target object in the intracranial target blood vessels, which has important clinical significance.

[0150] The above has elaborated in detail the method of the embodiments of the present application. In order to facilitate better implementation of the above solutions of the embodiments of the present application, correspondingly, the device of the embodiments of the present application is provided below.

[0151] Figure 9The figure shows a schematic structural diagram of a data processing device provided by an exemplary embodiment of the present application. The data processing device can be a computer program (including program code) running on a terminal; the data processing device can be used to execute Figure 2 , Figure 6b , Figure 7 and Figure 8 shown in the method embodiments. Please refer to Figure 9 . The data processing device includes the following units:

[0152] An acquisition unit 901, configured to acquire an initial sample angiography image set, where the initial sample angiography image set includes a first sample angiography image and a first label of the first sample angiography image, and the first label of the first sample angiography image indicates that the first sample angiography image is obtained by performing angiography on a first intracranial blood vessel;

[0153] A processing unit 902, configured to perform data augmentation processing on the first sample angiography image based on the physical characteristics of intracranial blood vessels to obtain a processed sample angiography image, and perform label conversion processing on the first label based on the physical characteristics of intracranial blood vessels to obtain a second label of the processed sample angiography image, and add the processed sample angiography image and the second label to a target sample angiography image set; the second label indicates that the processed sample angiography image is obtained by performing angiography on a second intracranial blood vessel, and the first intracranial blood vessel and the second intracranial blood vessel satisfy the physical characteristics of intracranial blood vessels;

[0154] The processing unit 902 is further configured to pre-train an angiography image recognition model using the initial sample angiography image set and the target sample angiography image set to obtain a trained angiography image recognition model, and the trained angiography image recognition model is used to predict any angiography image, and any angiography is obtained by performing angiography on a target intracranial blood vessel.

[0155] In one implementation, the physical characteristics of intracranial blood vessels include symmetry; when the processing unit 902 is configured to perform data augmentation processing on the first sample angiography image based on the physical characteristics of intracranial blood vessels to obtain a processed sample angiography image, it is specifically configured to:

[0156] Based on the symmetry included in the physical characteristics of intracranial blood vessels, perform left-right mirror processing on the first sample angiography image to obtain a processed sample angiography image that is symmetric to the first sample angiography image, and add the processed sample angiography image to the target sample angiography image set.

[0157] In one implementation, the first intracranial blood vessel is any one of the intracranial blood vessels, and the intracranial blood vessels include: the left carotid artery, the right carotid artery, the left vertebral artery, and the right vertebral artery; the label of any angiography image is a vector including multiple elements, and each element corresponds to an intracranial blood vessel;

[0158] In the first label corresponding to the first sample angiography image, the element value corresponding to the first intracranial blood vessel is the first element value, and the element values corresponding to other intracranial blood vessels except the first intracranial blood vessel are the second element values.

[0159] In one implementation, when the processing unit 902 is used to perform label conversion processing on the first label based on the physical characteristics of the intracranial blood vessels to obtain the second label of the processed sample angiography image, it is specifically used for:

[0160] Adjust the element value corresponding to the first intracranial blood vessel in the first label from the first element value to the second element value; and adjust the element value corresponding to the second intracranial blood vessel in the first label from the second element value to the first element value;

[0161] Use the adjusted first label as the second label of the processed sample angiography image.

[0162] In one implementation, the initial sample angiography image set further includes a second sample angiography image, which is obtained by performing angiography on the third intracranial blood vessel, and the first intracranial blood vessel is different from the third intracranial blood vessel; the physical characteristics of the intracranial blood vessels include fusibility; when the processing unit 902 is used to perform data augmentation processing on the first sample angiography image based on the physical characteristics of the intracranial blood vessels to obtain the processed sample angiography image, it is specifically used for:

[0163] Based on the fusibility included in the physical characteristics of the intracranial blood vessels, fuse the first sample angiography image and the second sample angiography image to obtain the processed sample angiography image.

[0164] In one implementation, when the processing unit 902 is used to perform label conversion processing on the first label based on the physical characteristics of the intracranial blood vessels to obtain the second label of the processed sample angiography image, it is specifically used for:

[0165] Adjust the element value corresponding to the third intracranial blood vessel in the first label to the first element value;

[0166] Determine the adjusted first label as the second label of the processed sample angiography image.

[0167] In one implementation, when the processing unit 902 is used to pre-train the angiography image recognition model with the initial sample angiography image set and the target sample angiography image set to obtain the trained angiography image recognition model, it is specifically used for:

[0168] Call the angiography image recognition model to perform recognition processing on multiple sample angiography images in the initial sample angiography image set, and obtain multiple first recognition results corresponding to the multiple sample angiography images in the initial sample angiography image set;

[0169] Call the angiography image recognition model to perform recognition processing on multiple sample angiography images in the target sample angiography image set, and obtain multiple second recognition results corresponding to the multiple sample angiography images in the target sample angiography image set;

[0170] Based on the difference between the multiple first recognition results and the first labels corresponding to the respective sample angiography images, and the difference between the multiple second recognition results and the second labels corresponding to the respective sample angiography images, obtain the first loss function of the angiography image recognition model;

[0171] Update the parameters of the angiography image recognition model based on the first loss function to train the angiography image recognition model.

[0172] In one implementation, when the processing unit 902 is used to update the parameters of the angiography image recognition model based on the first loss function, it is specifically used for:

[0173] Obtain an annotated sample angiography image set, which includes multiple annotated sample angiography images and the annotation information of each annotated sample angiography image. The annotation information of any annotated sample angiography image indicates the fragmentation information of the target object included in the fourth intracranial blood vessel of any annotated sample angiography image. The fragmentation information includes the fragmentation risk information of the target object and the fragmentation time information of the target object;

[0174] Call the angiography image recognition model to perform prediction processing on multiple annotated sample angiography images in the annotated sample angiography image set, and obtain the predicted fragmentation information of the target object in the multiple annotated sample angiography images;

[0175] According to the difference between the predicted fragmentation information of the target object in each annotated sample angiography image and the annotation information of the corresponding annotated sample angiography image, obtain the second loss function of the angiography image recognition model;

[0176] Optimizing the contrast image recognition model based on the first loss function and the second loss function, the trained contrast image recognition model is used to predict the intracranial target blood vessels in any angiography image, and to predict the fragmentation information of the target object of the intracranial target blood vessels in any angiography image.

[0177] In one implementation, when the processing unit 902 is used to optimize the contrast image recognition model based on the first loss function and the second loss function, it is specifically used for:

[0178] Perform weighted processing on the first loss function and the second loss function to obtain a third loss function after weighted processing;

[0179] Update the parameters of the contrast image recognition model according to the third loss function.

[0180] In one implementation, the processing unit 902 is further used for:

[0181] Obtain an annotated sample angiography image set, which includes multiple annotated sample angiography images and the annotation information of each annotated sample angiography image. The annotation information of any annotated sample angiography image indicates the fragmentation information of the target object included in the intracranial fourth blood vessel in any annotated sample angiography image, and the fragmentation information includes the fragmentation risk information of the target object and the fragmentation time information of the target object;

[0182] Call the annotated sample angiography image set to train the trained contrast image recognition model to obtain a target contrast image recognition model, and the target contrast image recognition model is used to predict the intracranial target blood vessels included in any angiography image and the fragmentation information of the target object in the intracranial target blood vessels.

[0183] In one implementation, the processing unit 902 is further used for:

[0184] Obtain the angiography image to be analyzed;

[0185] Call the trained contrast image recognition model to perform recognition processing on the angiography image to be analyzed, and obtain the intracranial target blood vessels in the angiography image and the fragmentation information of the target object of the intracranial target blood vessels.

[0186] According to an embodiment of the present application, Figure 9Each unit in the data processing device shown can be separately or entirely combined into one or several other units to form, or some of them can be further split into multiple smaller units with more specific functions to form, which can achieve the same operations without affecting the realization of the technical effects of the embodiments of this application. The above units are divided based on logical functions. In practical applications, the function of one unit can also be realized by multiple units, or the functions of multiple units can be realized by one unit. In other embodiments of this application, the data processing device may also include other units. In practical applications, these functions can also be assisted by other units and can be realized through the cooperation of multiple units. According to another embodiment of this application, it can be achieved by running a computer program (including program code) that can execute the respective steps involved in the corresponding methods shown in Figure 2 , Figure 6b , Figure 7 and Figure 8 on a general computing device such as a computer that includes processing elements and storage elements such as a central processing unit (CPU), a random access storage medium (RAM), and a read-only storage medium (ROM), to construct a data processing device as shown in Figure 9 and to implement the data processing method of the embodiments of this application. The computer program can be recorded on, for example, a computer-readable recording medium, loaded into the above computing device through the computer-readable recording medium, and run therein.

[0187] In the embodiments of this application, which specific intracranial blood vessel is included in the first sample angiography image is determined before the DSA scan. Then, the annotation information of the first sample angiography image (that is, which intracranial artery is included in the DSA image of the intracranial aneurysm) is already contained in the first label of the first sample angiography image, eliminating the need for manual annotation of data, which can improve the training efficiency and reduce the workload. Moreover, the processing unit 902 performs data augmentation processing on the sample angiography images in the initial sample angiography image set based on the physical characteristics of the intracranial blood vessels to obtain relatively rich sample angiography images for training the angiography image recognition model, ensuring the training process of the angiography image recognition model and obtaining an angiography image recognition model with better performance.

[0188] Figure 10 shows a schematic structural diagram of a data processing device provided by an exemplary embodiment of this application. Please refer to Figure 10, the data processing device includes a processor 1001, a communication interface 1002, and a computer-readable storage medium 1003. Among them, the processor 1001, the communication interface 1002, and the computer-readable storage medium 1003 can be connected through a bus or other means. Among them, the communication interface 1002 is used to receive and send data. The computer-readable storage medium 1003 can be stored in the memory of the terminal. The computer-readable storage medium 1003 is used to store computer programs, and the computer programs include program instructions. The processor 801 is used to execute the program instructions stored in the computer-readable storage medium 1003. The processor 1001 (or CPU (Central Processing Unit, central processor)) is the computing core and control core of the terminal, and is adapted to implement one or more instructions. Specifically, it is adapted to load and execute one or more instructions to implement the corresponding method flow or corresponding function.

[0189] The embodiment of the present application also provides a computer-readable storage medium (Memory). The computer-readable storage medium is a memory device in the terminal and is used to store programs and data. It can be understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal and, of course, the extended storage medium supported by the terminal. The computer-readable storage medium provides a storage space, and this storage space stores the processing system of the terminal. And, one or more instructions suitable for being loaded and executed by the processor 1001 are also stored in this storage space. These instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory; optionally, it can also be at least one computer-readable storage medium located far from the aforementioned processor.

[0190] In one embodiment, one or more instructions are stored in the computer-readable storage medium; the processor 1001 loads and executes one or more instructions stored in the computer-readable storage medium to implement the corresponding steps in the above data processing method embodiment; specifically, one or more instructions in the computer-readable storage medium are loaded and executed by the processor 1001 as follows:

[0191] Obtain an initial sample angiography image set, where the initial sample angiography image set includes a first sample angiography image and a first label of the first sample angiography image. The first label of the first sample angiography image indicates that the first sample angiography image is obtained by performing angiography on the first blood vessel in the skull;

[0192] Perform data augmentation processing on the first sample angiography image based on the physical characteristics of the intracranial blood vessels to obtain the processed sample angiography image, and perform label conversion processing on the first label based on the physical characteristics of the intracranial blood vessels to obtain the second label of the processed sample angiography image, and add the processed sample angiography image and the second label to the target sample angiography image set; the second label indicates that the processed sample angiography image is obtained by performing angiography on the second intracranial blood vessel, and the intracranial first blood vessel and the intracranial second blood vessel satisfy the physical characteristics of the intracranial blood vessels.

[0193] Pre-train the angiography image recognition model using the initial sample angiography image set and the target sample angiography image set to obtain the trained angiography image recognition model, and the trained angiography image recognition model is used to predict any angiography image, and any angiography image is obtained by performing angiography on the target intracranial blood vessel.

[0194] In one implementation, the physical characteristics of the intracranial blood vessels include symmetry; when one or more instructions in the computer-readable storage medium are loaded by the processor 1001 and perform data augmentation processing on the first sample angiography image based on the physical characteristics of the intracranial blood vessels to obtain the processed sample angiography image, the following steps are specifically executed:

[0195] Based on the symmetry included in the physical characteristics of the intracranial blood vessels, perform left-right mirror processing on the first sample angiography image to obtain a processed sample angiography image symmetric to the first sample angiography image, and add the processed sample angiography image to the target sample angiography image set.

[0196] In one implementation, the intracranial first blood vessel is any one of the intracranial blood vessels, and the intracranial blood vessels include: left carotid artery, right carotid artery, left vertebral artery, and right vertebral artery; the label of any angiography image is a vector including multiple elements, and each element corresponds to an intracranial blood vessel;

[0197] In the first label corresponding to the first sample angiography image, the element value corresponding to the intracranial first blood vessel is the first element value, and the element values corresponding to the other intracranial blood vessels except the intracranial first blood vessel are the second element values.

[0198] In one implementation, when one or more instructions in the computer-readable storage medium are loaded by the processor 1001 and perform label conversion processing on the first label based on the physical characteristics of the intracranial blood vessels to obtain the second label of the processed sample angiography image, the following steps are specifically executed:

[0199] Adjust the element value corresponding to the first intracranial blood vessel in the first label from the first element value to the second element value; and, adjust the element value corresponding to the second intracranial blood vessel in the first label from the second element value to the first element value;

[0200] Use the adjusted first label as the second label of the processed sample angiogram image.

[0201] In one implementation, the initial sample angiogram image set further includes a second sample angiogram image, which is obtained by angiography of the third intracranial blood vessel, and the first intracranial blood vessel is different from the third intracranial blood vessel; the physical characteristics of intracranial blood vessels include fusibility; when one or more instructions in the computer-readable storage medium are loaded by the processor 1001 and execute data augmentation processing on the first sample angiogram image based on the physical characteristics of intracranial blood vessels to obtain the processed sample angiogram image, the following steps are specifically executed:

[0202] Based on the fusibility included in the physical characteristics of intracranial blood vessels, fuse the first sample angiogram image and the second sample angiogram image to obtain the processed sample angiogram image.

[0203] In one implementation, when one or more instructions in the computer-readable storage medium are loaded by the processor 1001 and execute label conversion processing on the first label based on the physical characteristics of intracranial blood vessels to obtain the second label of the processed sample angiogram image, the following steps are specifically executed:

[0204] Adjust the element value corresponding to the third intracranial blood vessel in the first label to the first element value;

[0205] Determine the adjusted first label as the second label of the processed sample angiogram image.

[0206] In one implementation, when one or more instructions in the computer-readable storage medium are loaded by the processor 1001 and execute pre-training on the angiogram recognition model using the initial sample angiogram image set and the target sample angiogram image set to obtain the trained angiogram recognition model, the following steps are specifically executed:

[0207] Call the angiogram recognition model to perform recognition processing on multiple sample angiogram images in the initial sample angiogram image set to obtain multiple first recognition results corresponding to the multiple sample angiogram images in the initial sample angiogram image set;

[0208] Call the angiogram recognition model to perform recognition processing on multiple sample angiogram images in the target sample angiogram image set to obtain multiple second recognition results corresponding to the multiple sample angiogram images in the target sample angiogram image set;

[0209] Based on the differences between multiple first recognition results and the corresponding first labels of sample angiography images, and the differences between multiple second recognition results and the corresponding second labels of sample angiography images, a first loss function of the angiography image recognition model is obtained;

[0210] Based on the first loss function, update the parameters of the angiography image recognition model to train the angiography image recognition model.

[0211] In one implementation, when one or more instructions in the computer-readable storage medium are loaded by the processor 1001 and execute to update the parameters of the angiography image recognition model based on the first loss function, the following steps are specifically executed:

[0212] Obtain an annotated sample angiography image set, which includes multiple annotated sample angiography images and the annotation information of each annotated sample angiography image. The annotation information of any annotated sample angiography image indicates the fragmentation information of the target object included in the fourth intracranial blood vessel of any annotated sample angiography image. The fragmentation information includes the fragmentation risk information of the target object and the fragmentation time information of the target object;

[0213] Call the angiography image recognition model to perform prediction processing on multiple annotated sample angiography images in the annotated sample angiography image set, and obtain the predicted fragmentation information of the target object in the multiple annotated sample angiography images;

[0214] Based on the differences between the predicted fragmentation information of the target object in each annotated sample angiography image and the annotation information of the corresponding annotated sample angiography image, obtain a second loss function of the angiography image recognition model;

[0215] Based on the first loss function and the second loss function, optimize the angiography image recognition model. The trained angiography image recognition model is used to predict the intracranial target blood vessel in any angiography image and the fragmentation information of the target object of the intracranial target blood vessel in any angiography image.

[0216] In one implementation, when one or more instructions in the computer-readable storage medium are loaded by the processor 1001 and execute to optimize the angiography image recognition model based on the first loss function and the second loss function, the following steps are specifically executed:

[0217] Perform weighted processing on the first loss function and the second loss function to obtain a third loss function after weighted processing;

[0218] Update the parameters of the angiography image recognition model according to the third loss function.

[0219] In one implementation, one or more instructions in the computer-readable storage medium are loaded by the processor 1001 and further perform the following steps:

[0220] Obtain an annotated sample angiography image set, where the annotated sample angiography image set includes multiple annotated sample angiography images and the annotation information of each annotated sample angiography image. The annotation information of any annotated sample angiography image indicates the fragmentation information of the target object included in the fourth intracranial blood vessel in any annotated sample angiography image, and the fragmentation information includes the fragmentation risk information of the target object and the fragmentation time information of the target object;

[0221] Call the annotated sample angiography image set to train the trained angiography image recognition model to obtain a target angiography image recognition model, where the target angiography image recognition model is used to predict the intracranial target blood vessel included in any angiography image and the fragmentation information of the target object in the intracranial target blood vessel.

[0222] In one implementation, one or more instructions in the computer-readable storage medium are loaded by the processor 1001 and further perform the following steps:

[0223] Obtain the angiography image to be analyzed;

[0224] Call the trained angiography image recognition model to perform recognition processing on the angiography image to be analyzed, and obtain the intracranial target blood vessel in the angiography image and the fragmentation information of the target object in the intracranial target blood vessel.

[0225] In the embodiment of the present application, which intracranial blood vessel is specifically included in the first sample angiography image is determined before the DSA scan. Then, the annotation information of the first sample angiography image (that is, which intracranial artery is included in the DSA image of the intracranial aneurysm) is already contained in the first label of the first sample angiography image. There is no need for manual annotation data, which can improve the training efficiency and reduce the workload. Moreover, the processor 1001 performs data augmentation processing on the sample angiography images in the initial sample angiography image set based on the physical characteristics of the intracranial blood vessels to obtain relatively rich sample angiography images for training the angiography image recognition model, which can ensure the training process of the angiography image recognition model and obtain a relatively excellent angiography image recognition model.

[0226] The embodiment of the present application further provides a computer program product or a computer program. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the data processing device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the data processing device executes the above data processing method.

[0227] Those of ordinary skill in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed in this application can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0228] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present invention are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center in a wired manner (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or a wireless manner (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that the computer can access or a data storage device such as a server or data center that includes one or more integrated available media. The available medium can be a magnetic medium (such as a floppy disk, hard disk, magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid-state disk (SSD)).

[0229] As described above, the above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of changes or substitutions, which should all be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

Claims

1. A data processing method, characterized in that, Including: Obtain an initial sample angiography image set, where the initial sample angiography image set includes a first sample angiography image and a first label of the first sample angiography image, and the first label of the first sample angiography image indicates that the first sample angiography image is obtained by performing angiography on a first intracranial blood vessel; Based on the symmetry or fusibility of intracranial blood vessels, perform data augmentation processing on the first sample angiography image to obtain a processed sample angiography image, and based on the symmetry or fusibility of intracranial blood vessels, perform label transformation processing on the first label to obtain a second label of the processed sample angiography image, and add the processed sample angiography image and the second label to a target sample angiography image set; the second label indicates that the processed sample angiography image is obtained by performing angiography on a second intracranial blood vessel, and the first intracranial blood vessel and the second intracranial blood vessel satisfy the symmetry or fusibility of intracranial blood vessels; Use the initial sample angiography image set and the target sample angiography image set to pre-train an angiography image recognition model to obtain a trained angiography image recognition model, and the trained angiography image recognition model is used to predict any angiography image, where the any angiography image is obtained by performing angiography on a target intracranial blood vessel.

2. The method according to claim 1, characterized in that The physical properties of the intracranial blood vessels include symmetry; based on the symmetry or fusibility of intracranial blood vessels, performing data augmentation processing on the first sample angiography image to obtain a processed sample angiography image includes: Based on the symmetry included in the physical properties of the intracranial blood vessels, perform left-right mirror processing on the first sample angiography image to obtain a processed sample angiography image symmetric to the first sample angiography image, and add the processed sample angiography image to the target sample angiography image set.

3. The method according to claim 2, characterized in that, The first intracranial blood vessel is any one of the intracranial blood vessels, and the intracranial blood vessels include: left carotid artery, right carotid artery, left vertebral artery, and right vertebral artery; the label of any angiography image is a vector including multiple elements, and each element corresponds to an intracranial blood vessel; In the first label corresponding to the first sample angiography image, the element value corresponding to the first intracranial blood vessel is a first element value, and the element values corresponding to other intracranial blood vessels except the first intracranial blood vessel are second element values.

4. The method according to claim 3, characterized in that, Based on the symmetry or fusibility of the intracranial blood vessels, performing label transformation processing on the first label to obtain a second label of the processed sample angiography image includes: Adjust the element value corresponding to the first intracranial blood vessel in the first label from the first element value to the second element value; and adjust the element value corresponding to the second intracranial blood vessel in the first label from the second element value to the first element value; Use the adjusted first label as the second label of the processed sample angiography image.

5. The method according to claim 2, wherein The initial sample angiography image set further includes a second sample angiography image, which is obtained by performing angiography on the third intracranial blood vessel, and the first intracranial blood vessel is different from the third intracranial blood vessel; the physical properties of the intracranial blood vessels include fusibility; Based on the symmetry or fusibility of the intracranial blood vessels, performing data augmentation processing on the first sample angiography image to obtain a processed sample angiography image, including: Based on the fusibility included in the physical properties of the intracranial blood vessels, fusing the first sample angiography image and the second sample angiography image to obtain the processed sample angiography image.

6. The method according to claim 5, wherein Based on the symmetry or fusibility of the intracranial blood vessels, performing label transformation processing on the first label to obtain a second label of the processed sample angiography image, including: Adjusting the element value corresponding to the third intracranial blood vessel in the first label to a first element value; Determining the adjusted first label as the second label of the processed sample angiography image.

7. The method according to claim 2, wherein Using the initial sample angiography image set and the target sample angiography image set to pre-train an angiography image recognition model to obtain a trained angiography image recognition model, including: Invoking the angiography image recognition model to perform recognition processing on multiple sample angiography images in the initial sample angiography image set to obtain multiple first recognition results corresponding to the multiple sample angiography images in the initial sample angiography image set; Invoking the angiography image recognition model to perform recognition processing on multiple sample angiography images in the target sample angiography image set to obtain multiple second recognition results corresponding to the multiple sample angiography images in the target sample angiography image set; Based on the difference between the multiple first recognition results and the first labels corresponding to the corresponding sample angiography images, and the difference between the multiple second recognition results and the second labels corresponding to the corresponding sample angiography images, obtaining a first loss function of the angiography image recognition model; Updating the parameters of the angiography image recognition model based on the first loss function to train the angiography image recognition model.

8. The method according to claim 7, wherein Updating the parameters of the angiography image recognition model based on the first loss function, including: Obtaining an annotated sample angiography image set, which includes multiple annotated sample angiography images and the annotation information of each annotated sample angiography image. The annotation information of any one annotated sample angiography image indicates the fragmentation information of the target object included in the fourth intracranial blood vessel in the any one annotated sample angiography image, and the fragmentation information includes the fragmentation risk information of the target object and the fragmentation time information of the target object; Invoking the angiography image recognition model to perform prediction processing on multiple annotated sample angiography images in the annotated sample angiography image set to obtain the predicted fragmentation information of the target object in the multiple annotated sample angiography images; According to the difference between the predicted fragmentation information of the target object in each labeled sample angiography image and the labeling information of the corresponding labeled sample angiography image, the second loss function of the angiography image recognition model is obtained; Based on the first loss function and the second loss function, the angiography image recognition model is optimized. The trained angiography image recognition model is used to predict the intracranial target blood vessels in any angiography image and the fragmentation information of the target object of the intracranial target blood vessels in the any angiography image.

9. The method according to claim 8, wherein The optimizing the angiography image recognition model based on the first loss function and the second loss function includes: Performing a weighted process on the first loss function and the second loss function to obtain a third loss function after the weighted process; Updating the parameters of the angiography image recognition model according to the third loss function.

10. The method according to claim 1, characterized in that, After pre-training the angiography image recognition model by using the initial sample angiography image set and the target sample angiography image set to obtain a trained angiography image recognition model, it further includes: Obtaining a labeled sample angiography image set, which includes a plurality of labeled sample angiography images and the labeling information of each labeled sample angiography image. The labeling information of any labeled sample angiography image indicates the fragmentation information of the target object included in the intracranial fourth blood vessel in the any labeled sample angiography image. The fragmentation information includes the fragmentation risk information of the target object and the fragmentation time information of the target object; Invoking the labeled sample angiography image set to train the trained angiography image recognition model to obtain a target angiography image recognition model. The target angiography image recognition model is used to predict the intracranial target blood vessels included in any angiography image and the fragmentation information of the target object in the intracranial target blood vessels.

11. The method according to claim 8 or 10, characterized in that, The method further includes: Obtaining an angiography image to be analyzed; Invoking the trained angiography image recognition model to perform recognition processing on the angiography image to be analyzed to obtain the intracranial target blood vessels in the angiography image and the fragmentation information of the target object of the intracranial target blood vessels.

12. A data processing device, characterized in that, It includes: An obtaining unit, configured to obtain an initial sample angiography image set, where the initial sample angiography image set includes a first sample angiography image and a first label of the first sample angiography image, and the first label of the first sample angiography image indicates that the first sample angiography image is obtained by performing angiography on the intracranial first blood vessel; A processing unit, configured to perform data augmentation processing on the first sample angiography image based on the symmetry or fusibility of intracranial blood vessels to obtain a processed sample angiography image, and perform label transformation processing on the first label based on the symmetry or fusibility of the intracranial blood vessels to obtain a second label of the processed sample angiography image, and add the processed sample angiography image and the second label to a target sample angiography image set; the second label indicates that the processed sample angiography image is obtained by performing angiography on a second intracranial blood vessel, and the symmetry or fusibility of the intracranial blood vessels is satisfied between the first intracranial blood vessel and the second intracranial blood vessel; The processing unit is further configured to pre-train an angiography image recognition model by using the initial sample angiography image set and the target sample angiography image set to obtain a trained angiography image recognition model, and the trained angiography image recognition model is used to predict any angiography image, and the any angiography image is obtained by performing angiography on a target intracranial blood vessel.

13. A data processing device, characterized in that, Comprising: A processor, adapted to execute a computer program; A computer-readable storage medium storing a computer program, which when executed by the processor, implements the data processing method according to any one of claims 1-11.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which is adapted to be loaded and executed by a processor to implement the data processing method according to any one of claims 1-11.

15. A computer program product, characterized in that, The computer program product includes computer instructions stored in a computer-readable storage medium; the processor reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions to implement the data processing method according to any one of claims 1-11.

Images