Vascular focus detection method, device and equipment, storage medium and program product

Abstract

The invention relates to the technical field of image detection, and discloses a vascular lesion detection method, device and equipment, a storage medium and a program product, and the method comprises the steps: obtaining a to-be-detected vascular image; determining a blood vessel center line based on the to-be-detected blood vessel image; determining feature points on the center line of the blood vessel based on the walking of the to-be-detected blood vessel; constructing feature vectors in one-to-one correspondence with the feature points; determining feature data of the blood vessel center line based on the feature vectors; and inputting the feature data into a pre-constructed vascular lesion detection model to obtain a detection result of each feature point. According to the method, the blood vessel center line and the feature points thereof in the to-be-detected blood vessel image are extracted, the feature vectors are constructed based on the feature points, so that continuous blood vessel feature data are formed, the feature data are trained in a model supervised or weakly supervised learning mode, the position of a blood vessel focus can be accurately recognized, and the accuracy of blood vessel detection is improved. Higher detection sensitivity is provided in a complex blood vessel structure, and the detection precision of a blood vessel focus is improved.

Description

Technical Field

[0001] This invention relates to the field of imaging detection technology, specifically to a method, apparatus, equipment, storage medium, and program product for detecting vascular lesions. Background Technology

[0002] With the continuous development of medical imaging technology, the detection of vascular lesions has gradually become an important research direction in clinical diagnosis. Especially in the imaging analysis of cerebrovascular diseases, vascular lesions usually appear as tiny lesions, occupying a very small volume in the image, and their shape and location vary greatly. Therefore, how to accurately identify and distinguish lesions from blood vessels and other tissue structures in complex medical images has become a key issue in improving detection performance.

[0003] While existing deep learning models have made some progress in detecting vascular lesions, the subtle differences between lesions and other vascular structures in images, coupled with the high noise and variability of images, make it difficult for networks to accurately distinguish the subtle differences between them. Summary of the Invention

[0004] This invention provides a method, apparatus, device, storage medium, and program product for detecting vascular lesions, thereby improving the ability to distinguish lesions from other vascular structures and further enhancing the accuracy of vascular lesion detection.

[0005] In a first aspect, the present invention provides a method for detecting vascular lesions, the method comprising: Acquire images of the blood vessels to be detected; Determine the centerline of the blood vessel based on the image of the vessel to be detected; Based on the course of the blood vessel to be detected, characteristic points on the center line of the blood vessel are determined; Construct feature vectors that correspond one-to-one with each feature point; Based on feature vectors, determine the feature data of the blood vessel centerline; The feature data is input into a pre-built vascular lesion detection model to obtain the detection result for each feature point.

[0006] The vascular lesion detection method provided by this invention extracts the vascular centerline and its feature points from the image of the vascular vessel to be detected, and constructs feature vectors based on these feature points, thereby forming continuous vascular feature data. This feature data is input into a neural network or other large-scale deep learning model, and trained through supervised or weakly supervised learning. This method can accurately distinguish lesions from other vascular structures, identify the location of vascular lesions, and provide higher detection sensitivity in complex vascular structures, effectively improving the detection accuracy and robustness of vascular lesions.

[0007] In one optional implementation, constructing a feature vector corresponding to each feature point includes: The eigenvector of a feature point is a one-dimensional vector composed of multiple eigenvalues; the eigenvalues ​​include: The density value of the voxel where the feature point is located; The median of the neighborhood of the voxel containing the feature point; The eigenvalues ​​of the Hessian matrix of the voxel containing the feature point; The gradient vector of the density along the diameter line with a preset angular spacing in the spherical coordinate system where the feature point is located; The curvature, deflection angle, and radius of the vessel centerline where the feature point is located; The location of the feature point on the contour of the blood vessel and the voxel value within a preset range perpendicular to the contour.

[0008] In this embodiment, the above-mentioned feature values ​​are selected as feature vectors, which can comprehensively and meticulously describe the multidimensional information of vascular structure. They reflect the geometric shape, local density, curvature changes and relationship with surrounding tissues of blood vessels from multiple perspectives, which helps to identify the differences between lesions and healthy blood vessels, thereby improving the accuracy of lesion detection.

[0009] In one optional implementation, feature points on the centerline of the blood vessel to be detected are determined based on its course, including: Determine the bifurcation point of the blood vessel's centerline; Determine the endpoint of the vessel centerline based on the course of the vessel to be tested; Divide the distance between the branching point and the endpoint to obtain multiple intermediate points; The bifurcation point, the midpoint, and the endpoint are used as feature points.

[0010] In one alternative implementation, the vascular lesion detection model is established through the following steps: Acquire a training image set, which includes images of healthy blood vessels and images of lesions in blood vessels; Annotate healthy vascular images and lesion vascular images to obtain annotation data; Determine the characteristic data that correspond one-to-one between healthy vascular images and lesion vascular images; The labeled data is associated with the feature data of the corresponding feature points to obtain training data; A pre-set initial model is trained based on the training image set and training data to obtain a vascular lesion detection model.

[0011] In one optional implementation, a preset initial model is trained based on a training image set, labeled data, and feature data to obtain a vascular lesion detection model, including: The training image set and training data are input into the initial neural network model for training to obtain a vascular lesion detection model.

[0012] The vascular lesion detection model constructed in this implementation combines labeled healthy vascular images with lesion vascular images, associates image annotations with vascular feature point data to form a training dataset with rich structural information, and uses a small model network, such as LSTM network training, to perform deep learning on the continuity and structural changes of blood vessels by utilizing the time series processing capabilities of LSTM network. This can effectively identify lesion areas in blood vessels and significantly improve the accuracy of vascular lesion detection.

[0013] In one optional implementation, a preset initial model is trained based on a training image set, labeled data, and feature data to obtain a vascular lesion detection model, including: Get the pre-built prompt text; The training image set and training data are input into the pre-trained text model, and prompt words are input to correct the text model and obtain the vascular lesion detection model.

[0014] The vascular lesion detection model constructed in this embodiment combines labeled data and feature point data, and integrates a pre-trained large text model with prompt text. The prompt text provides semantic guidance on the lesion region, enabling the model to make more refined adjustments and corrections based on the feature data. This allows the model to accurately identify lesion areas in vascular images. Furthermore, the input prompt text can further optimize the detection effect of vascular lesions, enhance the perception of changes in vascular structure, and make the model more flexible and adaptable when dealing with different types of lesions.

[0015] In a second aspect, the present invention provides a vascular lesion detection device, the device comprising: The acquisition module is used to acquire images of the blood vessels to be detected. The vessel centerline determination module is used to determine the vessel centerline based on the image of the vessel to be detected. The feature point determination module is used to determine feature points on the centerline of the blood vessel based on its course. The construction module is used to construct feature vectors that correspond one-to-one with feature points; The feature data determination module is used to determine the feature data of the blood vessel centerline based on the feature vector; The detection module is used to input feature data into a pre-built vascular lesion detection model to obtain the detection result for each feature point.

[0016] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the vascular lesion detection method of the first aspect or any corresponding embodiment described above.

[0017] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to perform the vascular lesion detection method of the first aspect or any corresponding embodiment described above.

[0018] Fifthly, the present invention provides a computer program product, including computer instructions, which are used to cause a computer to execute the vascular lesion detection method of the first aspect or any corresponding embodiment described above. Attached Figure Description

[0019] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a schematic flowchart of the first method for detecting vascular lesions according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the blood vessel centerline and feature points according to an embodiment of the present invention; Figure 3 This is a structural block diagram of a vascular lesion detection device according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0023] Detection of vascular lesions in medical imaging is a typical task involving small targets, with lesions typically occupying less than 0.5‰ of the image volume. Therefore, improving the network's ability to accurately distinguish lesions from other vascular structures is crucial for enhancing detection performance. Given the close connection between lesions and blood vessels, and the importance of vascular course in manual image interpretation, enabling neural networks to learn features such as the course of cerebral arteries and textural changes in images from a vascular perspective would allow for a comprehensive understanding of vascular semantics, leading to precise extraction of lesion semantics and significantly improving the performance of lesion detection networks.

[0024] Furthermore, due to the highly individual heterogeneity of blood vessels, the vascular structure, morphology, and even image texture vary significantly among different patients. The location of lesions is also highly variable, with lesions in different locations exhibiting different imaging manifestations. Limited training data alone cannot cover all possible detection scenarios. Therefore, vascular lesion detection requires a certain degree of few-shot or even zero-shot learning capability at the lesion level. The key to overcoming the performance bottleneck of existing supervised learning lesion detection methods lies in enabling neural networks, beyond supervised learning, to autonomously expand their learning to areas such as regions of interest, feature spaces, relationships between different regions of the blood vessel, and the relationship between lesions and blood vessels, and to autonomously summarize relevant knowledge and features. This allows the network to autonomously explore blood vessels and perceive aneurysms. Ultimately, this is crucial to enabling vascular lesion detection methods to truly meet clinical needs.

[0025] In view of this, according to an embodiment of the present invention, a method for detecting vascular lesions is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0026] This embodiment provides a method for detecting vascular lesions, which can be used on servers, terminals, and mobile terminals, such as mobile phones and tablets. Figure 1 This is a flowchart of a method for detecting vascular lesions according to an embodiment of the present invention, such as... Figure 1 As shown, the process includes the following steps: Step S101: Obtain the image of the blood vessel to be detected.

[0027] The images of the blood vessels to be detected can be obtained using imaging techniques such as magnetic resonance imaging (MRI) and computed tomography (CT). In this embodiment, the images of the blood vessels to be detected are preferably obtained using TOF-MRA (Time-of-Flight Magnetic Resonance Angiography). TOF-MRA is a magnetic resonance imaging technique used for vascular imaging. It generates images of blood vessels by utilizing the signal difference between blood flow and stationary tissue. It does not require the use of contrast agents, which can reduce patient trauma. The blood vessels to be detected can be cerebral arteries, carotid arteries, pulmonary arteries, great saphenous veins, microvessels, etc.

[0028] Step S102: Determine the centerline of the blood vessel based on the image of the blood vessel to be detected.

[0029] Before extracting the vascular centerline, the image of the blood vessel to be detected is preprocessed to remove noise and enhance vascular contrast, making the vascular structure clearer. Then, image segmentation can be used to accurately segment the blood vessel. Further, the vascular region is converted into a thin linear structure, i.e., skeletonization. Key point processing is performed on the skeletonized vascular image to determine bifurcation points, endpoints, etc. The centerline of each branch is determined by finding the shortest path for each vascular branch. In this embodiment, cerebral arteries are used as an example, and the vascular centerline is referenced... Figure 2 As shown.

[0030] Step S103: Based on the course of the blood vessel to be detected, determine the feature points on the center line of the blood vessel.

[0031] The course of blood vessels refers to the direction or path of blood vessels in an image, that is, the distribution and extension direction of blood vessels from their origin to their terminal point. For example, points on the central line of a blood vessel that have three characteristics—a point on a vascular segment, a bifurcation point, and a terminal point—and that are connected to each other in a way that conforms to the course of the blood vessel, can be identified as feature points.

[0032] Step S104: Construct feature vectors corresponding to each feature point.

[0033] The feature vector includes features that can characterize the information of the feature point, such as: the density value of the voxel where the feature point is located, the median of the 6-neighborhood in three-dimensional space, the curvature of the blood vessel centerline where the feature point is located, the deflection angle, etc.

[0034] Step S105: Determine the feature data of the blood vessel centerline based on the feature vector.

[0035] Each feature point is arranged in the order of the points, i.e. the course of the blood vessel to be detected, and the feature vectors corresponding to the feature points are concatenated to form the feature data of the blood vessel centerline.

[0036] Step S106: Input the feature data into the pre-built vascular lesion detection model to obtain the detection result of each feature point.

[0037] The vascular lesion detection model in this embodiment can be based on a neural network structure, such as LSTM (Long Short-Term Memory), or on a large language text model, such as DeepSeekR1. Inputting feature data into the pre-built vascular lesion detection model outputs the classification label for each feature point, i.e., the detection result, such as aneurysm vessel, healthy vessel, etc. After obtaining the classification label for each feature point, it can be automatically mapped back to the corresponding location on the image, forming a rectangular or circular prompt box to obtain a visual detection result.

[0038] The vascular lesion detection method provided by this invention extracts the vascular centerline and its feature points from the image of the vascular vessel to be detected, and constructs feature vectors based on these feature points, thereby forming continuous vascular feature data. This feature data is input into a neural network or other large-scale deep learning model, and trained through supervised or weakly supervised learning. This method can accurately distinguish lesions from other vascular structures, identify the location of vascular lesions, and provide higher detection sensitivity in complex vascular structures, effectively improving the detection accuracy and robustness of vascular lesions.

[0039] This embodiment provides a method for detecting vascular lesions, which can be used on servers, terminals, and mobile terminals, such as mobile phones and tablets. Figure 2 This is a flowchart of a method for detecting vascular lesions according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps: Step S201: Acquire the image of the blood vessel to be detected. For details, please refer to [link to relevant documentation]. Figure 1 Step S101 of the illustrated embodiment will not be described again here.

[0040] Step S202: Determine the centerline of the blood vessel based on the image of the vessel to be detected. For details, please refer to [link to relevant documentation]. Figure 1 Step S102 of the illustrated embodiment will not be described again here.

[0041] Step S203: Based on the course of the blood vessel to be detected, determine the feature points on the center line of the blood vessel.

[0042] Specifically, step S203 includes: Determine the bifurcation point of the blood vessel's centerline; Determine the endpoint of the vessel centerline based on the course of the vessel to be tested; Divide the distance between the branching point and the endpoint to obtain multiple intermediate points; The bifurcation point, the midpoint, and the endpoint are used as feature points.

[0043] In this embodiment, cerebral arteries are still used as an example. During the segmentation of cerebral arteries in the image and the determination of the vessel centerline, points on the vessel centerline with three properties can be identified: points on the vessel segment, bifurcation points, and endpoints. These points are connected in a manner consistent with the course of the blood vessel. (Refer to...) Figure 2 As shown, purple dots represent bifurcation points, red dots represent endpoints, and green dots represent intermediate points, i.e., points on the vessel segment. These feature points are constructed into a centerline point set [A1, A2, ..., An]. In the centerline point set, points on a vessel segment are arranged in order of direction starting from the basilar artery. A breadth-first search is used to search the vessel segment. Endpoints and bifurcation points are represented by point labels 01 and 02, respectively, and points on the vessel segment are represented by 00.

[0044] Step S204: Construct feature vectors corresponding to each feature point.

[0045] Specifically, step S204 includes: The eigenvector of a feature point is a one-dimensional vector composed of multiple eigenvalues; the eigenvalues ​​include: The density value of the voxel where the feature point is located.

[0046] The median of the neighborhood of the voxel containing the feature point; for example, the median of the 6-neighborhood of the voxel containing the feature point in 3D space.

[0047] The eigenvalues ​​of the Hessian matrix of the voxel containing the feature point; for example, λ1, λ3, and λ3 of the 5×5 Hessian matrix (also known as the Hessian matrix, which is a square matrix composed of the second-order partial derivatives of a multivariate function and is used to describe the local curvature characteristics of a function at a certain point) of the voxel containing the feature point are taken as eigenvalues.

[0048] The gradient vector of the density along a diameter line with a preset angular spacing in the spherical coordinate system where the feature points are located; for example, the gradient vector of the density calculated by linear interpolation of 0.5 voxels along a diameter line with 60° spacing in the spherical coordinate system where the feature points are located. The gradient vectors of the density of all feature points are arranged in order.

[0049] The curvature, deflection angle, and radius of the vessel centerline where the feature point is located; The feature point's position on the blood vessel contour and its voxel value within a preset range perpendicular to the contour; for example, the feature point's position on the blood vessel Sobel contour (the blood vessel contour extracted using the Sobel operator) and its voxel value within ±10 voxels perpendicular to the contour. The feature point's position on the contour and its voxel value are arranged in order of 0-360° of the cross section.

[0050] In this embodiment, the above-mentioned feature values ​​are selected as feature vectors, which can comprehensively and meticulously describe the multidimensional information of vascular structure. They reflect the geometric shape, local density, curvature changes and relationship with surrounding tissues of blood vessels from multiple perspectives, which helps to identify the differences between lesions and healthy blood vessels, thereby improving the accuracy of lesion detection.

[0051] Step S205: Determine the feature data of the blood vessel centerline based on the feature vector.

[0052] The obtained feature vectors are integrated into the centerline point set [A1, A2, ..., An]. The length of the feature vector for each feature point is not guaranteed to be consistent. Using endpoints and bifurcation points as markers, points on line segments with clear connections are selected, and their feature values ​​are concatenated in the order of the points. Points are separated by a uniform identifier, such as "B", or other identifiers like " / #". For endpoints, the identifier "C" is added after the feature, or other identifiers such as " / &". For bifurcation points, the identifier "D" is added after the feature, or other identifiers such as " / @". The points are then concatenated in the order of the point set to form one-dimensional data, in the form "A1`-B-A2`...A7`-C-A8`-B-A9`...A20`-D-A21`-B-A22`...", where An` refers to the feature vector of that point, and this one-dimensional data represents the entire feature data M.

[0053] Step S206: Input the feature data into the pre-built vascular lesion detection model to obtain the detection result of each feature point.

[0054] In some alternative implementations, the vascular lesion detection model is established through the following steps: Acquire a training image set, which includes images of healthy blood vessels and images of lesions in blood vessels; Annotate healthy vascular images and lesion vascular images to obtain annotation data; Determine the characteristic data that correspond one-to-one between healthy vascular images and lesion vascular images; The labeled data is associated with the feature data of the corresponding feature points to obtain training data; The training image set and training data are input into the initial neural network model for training to obtain a vascular lesion detection model.

[0055] Specifically, a small model network structure, such as an LSTM network, can be used. First, a training set is built, and images are labeled to form a mask. Lesion regions are represented by specific values; for example, aneurysms are represented by a value of 1, vascular stenosis by a value of 2, and non-lesion regions by a value of 0. The mask is constructed as labeled data with the same dimensions as the feature data. That is, for a feature point, when the value of that feature point in the mask is x, the labeled values ​​of all feature vectors corresponding to that feature point are set to x, but the connection signs are the same as in the feature data. The data is then input into the LSTM network for training to obtain a vascular lesion detection model.

[0056] The vascular lesion detection model constructed in this embodiment combines labeled healthy vascular images with lesion vascular images, associates image annotations with vascular feature point data to form a training dataset with rich structural information, and uses a small model network, such as LSTM network training, to perform deep learning on the continuity and structural changes of blood vessels by utilizing the time series processing capabilities of LSTM network. This can effectively identify lesion areas in blood vessels and significantly improve the accuracy of vascular lesion detection.

[0057] In some alternative implementations, the vascular lesion detection model is established through the following steps: Acquire a training image set, which includes images of healthy blood vessels and images of lesions in blood vessels; Annotate healthy vascular images and lesion vascular images to obtain annotation data; Determine the characteristic data that correspond one-to-one between healthy vascular images and lesion vascular images; The labeled data is associated with the feature data of the corresponding feature points to obtain training data; Get the pre-built prompt text; The training image set and training data are input into the pre-trained text model, and prompt words are input to correct the text model and obtain the vascular lesion detection model.

[0058] In this embodiment, a large model structure is used for training. That is, the pre-trained text large model DeepSeekR1 is imported, and text labels for each point are generated using lesion annotation (i.e., the labeled data in this embodiment). Input M and text labels N, and use prompt words to allow the large model to learn and summarize the feature data of different text label parts.

[0059] For example: The input prompt is: "I have entered the main text feature data M and its text label content N. M and N have the same length. In M, / #, / &, and / @ indicate the end of a paragraph. Among them, / # is closely related to the paragraphs before and after it, / & is unrelated to the paragraphs before and after it, and / @ has a one-to-many relationship. In N, 0 indicates that the paragraph is normal, and 1 and 2 indicate two different abnormal paragraphs. Based on the labels in N, please learn the differences between the data of different paragraphs in M, and verify the learned rules on a small number of data. Use the rule that best fits the verification as the output and update your model."

[0060] Simultaneously, the large model performs unsupervised summarization of the feature types of M, forming three major categories: healthy blood vessels, suspected lesions, and confirmed lesions. Two of these lesion categories can further be subdivided into smaller categories. For example, the input prompt might be: "I have entered the main text feature data content M and its text label content N. M and N have the same length. / #, / &, and / @ in M ​​indicate the end of a paragraph. / # indicates a close connection between the preceding and following paragraphs, / & indicates no connection between the preceding and following paragraphs, and / @ indicates a one-to-many relationship. Can you summarize which patterns constitute the absolute majority of these paragraphs? Beyond these main patterns, summarize two patterns from other paragraphs, using a hierarchical clustering approach. For the other two patterns, further classification is possible. If further classification is not possible, this subclass is not further classified. Ultimately, at least one main pattern and two other patterns should be obtained, corresponding to different paragraphs. The set of these paragraphs should be the same as the entire dataset."

[0061] Then, the unsupervised results are corrected using the text-labeled results to finally obtain the vascular lesion detection model. For example, the input prompt is: "You generated two models, the first with a clear label, the second without a clear label. However, the three modes of the second should include the first label. The main mode of the second should be close to the label 0 of the first. Generally speaking, the label 0 of the first is slightly larger than the second. The other two modes of the second should have a certain correspondence with the labels 1 and 2 of the first. According to this rule, the second model you generated is corrected to obtain the final model. The validation results on the data are also referenced during the correction."

[0062] In actual testing, you only need to briefly describe the task requirements, such as "the input data is vascular data, and we need to detect aneurysms", and you can obtain the classification labels of each feature point.

[0063] The vascular lesion detection model constructed in this embodiment combines labeled data and feature point data, and integrates a pre-trained large text model with prompt text. The prompt text provides semantic guidance on the lesion region, enabling the model to make more refined adjustments and corrections based on the feature data. This allows the model to accurately identify lesion areas in vascular images. Furthermore, the input prompt text can further optimize the detection effect of vascular lesions, enhance the perception of changes in vascular structure, and make the model more flexible and adaptable when dealing with different types of lesions.

[0064] This embodiment also provides a vascular lesion detection device, which is used to implement the above embodiments and preferred embodiments, and will not be repeated as described previously. As used below, the term "module" can be a combination of software and / or hardware that performs a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0065] This embodiment provides a device for detecting vascular lesions, such as... Figure 3 As shown, the device includes: Acquisition module 301 is used to acquire images of the blood vessels to be detected; The vessel centerline determination module 302 is used to determine the vessel centerline based on the image of the vessel to be detected. The feature point determination module 303 is used to determine feature points on the center line of the blood vessel based on the course of the blood vessel to be detected; specifically, it is used to determine the bifurcation point of the blood vessel center line; determine the endpoint of the blood vessel center line according to the course of the blood vessel to be detected; divide the distance between the bifurcation point and the endpoint to obtain multiple intermediate points; and use the bifurcation point, intermediate points and endpoint as feature points.

[0066] The construction module 304 is used to construct feature vectors corresponding to each feature point. The feature vector of each feature point is a one-dimensional vector composed of multiple feature values. The feature values ​​include: the density value of the voxel where the feature point is located; the median of the neighborhood of the voxel where the feature point is located; the eigenvalue of the Hessian matrix of the voxel where the feature point is located; the gradient vector of the density on the diameter line of the preset angular spacing in the spherical coordinate system where the feature point is located; the curvature, deflection angle, and radius of the vessel centerline where the feature point is located; the point position on the contour of the vessel contour where the feature point is located; and the voxel value perpendicular to the preset range of the contour.

[0067] The feature data determination module 305 is used to determine the feature data of the blood vessel centerline based on the feature vector; The detection module 306 is used to input feature data into a pre-built vascular lesion detection model to obtain the detection result of each feature point.

[0068] In some alternative embodiments, the apparatus further includes: The training module is used to acquire a training image set, which includes images of healthy blood vessels and images of lesions; to annotate the healthy and lesion images to obtain annotation data; to determine the feature data corresponding to each healthy and lesion image; to associate the annotation data with the feature data of the corresponding feature points to obtain training data; and to train a preset initial model based on the training image set and training data to obtain a vascular lesion detection model. This includes inputting the training image set and training data into an initial neural network model for training to obtain the vascular lesion detection model. Alternatively, it can acquire pre-constructed prompt text; input the training image set and training data into a pre-trained large text model, and input the prompt text to correct the large text model to obtain a vascular lesion detection model.

[0069] The vascular lesion detection device provided in this embodiment of the invention can execute the vascular lesion detection method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects for executing the method. Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0070] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0071] The following is a detailed reference. Figure 4 This diagram illustrates a structural schematic suitable for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 401, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 402 or a program loaded from memory 408 into random access memory (RAM) 403. The RAM 403 also stores various programs and data required for the operation of the electronic device. The processor 401, ROM 402, and RAM 403 are interconnected via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0072] Typically, the following devices can be connected to I / O interface 405: input devices 406 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 407 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 408 including, for example, magnetic tapes, hard disks, etc.; and communication devices 409. Communication device 409 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 4Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0073] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 409, or installed from a memory 408, or installed from a ROM 402. When the computer program is executed by the processor 401, it performs the functions defined in the vascular lesion detection method of the embodiments of the present invention.

[0074] Figure 4 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0075] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the vascular lesion detection method shown in the above embodiments is implemented.

[0076] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0077] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for detecting vascular lesions, characterized in that, The method includes: Acquire images of the blood vessels to be detected; Based on the image of the blood vessel to be detected, the centerline of the blood vessel is determined; Based on the course of the blood vessel to be detected, feature points on the center line of the blood vessel are determined; Construct feature vectors corresponding to each feature point; Based on the feature vector, the feature data of the blood vessel centerline are determined; The feature data is input into a pre-built vascular lesion detection model to obtain the detection result for each feature point.

2. The method according to claim 1, characterized in that, The construction of the feature vector corresponding to each feature point includes: The feature vector of the feature point is a one-dimensional vector composed of multiple feature values; the feature values ​​include: The density value of the voxel where the feature point is located; The median of the neighborhood of the voxel containing the feature point; The eigenvalues ​​of the Hessian matrix of the voxel containing the feature point; The gradient vector of the density along the diameter line of the preset angular spacing in the spherical coordinate system where the feature point is located; The curvature, deflection angle, and radius of the blood vessel centerline where the feature point is located; The feature point is located on the contour of the blood vessel and has a voxel value within a preset range perpendicular to the contour.

3. The method according to claim 1, characterized in that, The determination of feature points on the centerline of the blood vessel based on its course includes: Determine the bifurcation point of the blood vessel centerline; Based on the course of the blood vessel to be detected, determine the endpoint of the centerline of the blood vessel; Divide the distance between the bifurcation point and the endpoint to obtain multiple intermediate points; The bifurcation point, the midpoint, and the endpoint are used as the feature points.

4. The method according to claim 1, characterized in that, The vascular lesion detection model was established through the following steps: Acquire a training image set, which includes images of healthy blood vessels and images of lesions in blood vessels; The healthy vascular images and the lesion vascular images are annotated to obtain annotation data; Determine the feature data that corresponds one-to-one with the healthy vascular images and the lesion vascular images; The labeled data is associated with the feature data of the corresponding feature points to obtain training data; The vascular lesion detection model is obtained by training a preset initial model based on the training image set and the training data.

5. The method according to claim 4, characterized in that, The step of training a preset initial model based on the training image set, the labeled data, and the feature data to obtain the vascular lesion detection model includes: The training image set and the training data are input into the initial neural network model for training to obtain the vascular lesion detection model.

6. The method according to claim 4, characterized in that, The step of training a preset initial model based on the training image set, the labeled data, and the feature data to obtain the vascular lesion detection model includes: Retrieve pre-built prompt text; The training image set and the training data are input into the pre-trained text model, and the prompt text is input to correct the text model and obtain the vascular lesion detection model.

7. A device for detecting vascular lesions, characterized in that, The device includes: The acquisition module is used to acquire images of the blood vessels to be detected. A blood vessel centerline determination module is used to determine the blood vessel centerline based on the image of the blood vessel to be detected. The feature point determination module is used to determine feature points on the center line of the blood vessel based on the course of the blood vessel to be detected. The construction module is used to construct the feature vectors corresponding to the feature points one by one; A feature data determination module is used to determine the feature data of the blood vessel centerline based on the feature vector; The detection module is used to input the feature data into a pre-built vascular lesion detection model to obtain the detection result of each feature point.

8. An electronic device, characterized in that, include: The system includes a memory and a processor, which are interconnected and the memory stores computer instructions. The processor executes the computer instructions to perform the vascular lesion detection method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to perform the vascular lesion detection method according to any one of claims 1 to 6.

10. A computer program product, characterized in that, Includes computer instructions for causing a computer to perform the vascular lesion detection method according to any one of claims 1 to 6.

Images