An imaging recognition method and system for carotid artery dissection related to elongated styloid process

Abstract

The present invention relates to the field of medical assistance technology, and discloses a method and system for image recognition of carotid artery dissection related to elongated styloid process. The recognition method includes the following steps: obtaining historical CTA image data and clinical parameters of a user; using three-dimensional reconstruction technology according to the CTA image data to construct a three-dimensional image of the styloid process, and constructing a graph structure based on the three-dimensional image of the styloid process; encoding the CTA image data and inputting it into a CNN model to output the identified abnormal area of the styloid process and carotid artery dissection features; updating the node features of the graph structure; constructing a styloid process lesion prediction model, fusing the updated node features and the clinical parameters of the user and inputting them into the styloid process lesion prediction model, and the styloid process lesion prediction model outputs the risk probability of the user suffering from carotid artery dissection. The recognition method of the present invention combines three-dimensional reconstruction technology and convolutional neural network model, which helps to more accurately diagnose elongated styloid process and carotid artery dissection caused by it.

Description

Technical Field

[0001] The present invention relates to the field of medical assistance, and more specifically, to a method and system for image recognition of carotid artery dissection related to elongated styloid process. Background Art

[0002] Carotid artery dissection caused by an elongated styloid process is an important cause of stroke in young people. However, currently, the clinical understanding of carotid artery dissection caused by an elongated styloid process is insufficient, and radiologists do not pay attention to its formation cause when reporting carotid artery dissection, resulting in repeated occurrences in the prognosis of many patients with carotid artery dissection caused by an elongated styloid process or stent fracture after stenting. At the same time, due to work nature and other reasons, some patients with an elongated styloid process need to turn their heads repeatedly, which inevitably leads to repeated stimulation and compression of the adjacent carotid artery, ultimately resulting in the formation of dissection. Therefore, early detection and prevention are crucial.

[0003] In recent years, the development of multimodal medical imaging technology and artificial intelligence methods has provided a new way for the detection of the cause of carotid artery dissection related to an elongated styloid process. Some studies have attempted to fuse the image information of different imaging modalities and establish a recognition and prediction model for the cause of carotid artery dissection through machine learning algorithms, achieving preliminary results. However, the existing methods lack effective multimodal image registration and fusion methods: Since the association between an elongated styloid process and carotid artery dissection involves complex anatomical structures, the existing technologies are difficult to achieve accurate multimodal image registration and fusion, and cannot make full use of the complementary information provided by different imaging modalities such as CT and MRI to comprehensively evaluate the styloid process and its impact on blood vessels. Summary of the Invention

[0004] The present invention provides a method and system for image recognition of carotid artery dissection related to an elongated styloid process to solve the technical problems raised in the above background art.

[0005] The present invention provides a method for image recognition of carotid artery dissection related to an elongated styloid process, including the following steps:

[0006] Step 101, obtaining the historical CTA image data and clinical parameters of the user, where the CTA image data includes the user's anatomical structure information, blood vessel condition information, pathological change information, and spatial relationship information, and the clinical parameters include the user's age, gender, and medical history information;

[0007] Step 102, using three-dimensional reconstruction technology according to the CTA image data to construct a three-dimensional image of the styloid process, constructing a graph structure based on the three-dimensional image of the styloid process, where the nodes include equally spaced carotid arteries, different carotid arteries are marked as different nodes, one node represents a section of a carotid artery, and the condition for there to be an edge between two nodes is that these two sections of carotid arteries are connected;

[0008] Step 103: Encode the CTA image data and input it into the CNN model to output the identified abnormal styloid process area and carotid artery dissection features. The CNN model includes a U-Net network and an output layer. The U-Net network includes an encoder and a decoder;

[0009] Step 104: Update the node features of the graph structure constructed in Step 102;

[0010] Step 105: Construct a styloid process lesion prediction model. After feature fusion of the updated node features and the user's clinical parameters, generate fused features and input the fused features into the styloid process lesion prediction model. The styloid process lesion prediction model outputs the risk probability of the user having carotid artery dissection.

[0011] In a preferred embodiment, the anatomical structure information includes: the tip position of the styloid process, the length measurement value, and the internal inclination angle and anterior inclination angle of the styloid process; the vascular condition information includes: the diameter of the carotid artery, the thickness of the blood vessel wall, and the patency of the blood vessel lumen; the pathological change information includes: the location, scope, calcification condition, and plaque distribution of the arterial dissection; the spatial relationship information includes: the distance between the styloid process and the adjacent blood vessels.

[0012] In a preferred embodiment, the method for updating node features includes the following steps:

[0013] Step 201: Perform a linear transformation on all the node features in the graph structure to generate a node transformation feature vector;

[0014] The calculation formula for the linear transformation is as follows:

[0015] ;

[0016] where, represents the node transformation feature vector of node i, represents the initial feature vector of node i, represents the weight matrix;

[0017] Step 202: Calculate the attention coefficient between node i and node j;

[0018] The calculation formula for the attention coefficient is as follows:

[0019] ;

[0020] where, represents the attention coefficient between node i and node j, represents the node transformation feature vector of node i, represents the node transformation feature vector of node j, represents the weight matrix, represents the attention vector, denotes vector transpose, denotes function, denotes vector concatenation;

[0021] Step 203, normalize the attention coefficients;

[0022] The calculation formula for normalizing the attention coefficients is as follows:

[0023] ;

[0024] where, denotes the normalized attention coefficient, denotes the set of neighbor nodes of node i, denotes the attention coefficient between node i and node j, denotes the natural exponential function;

[0025] Step 204, based on the normalized attention coefficients, perform weighted aggregation on the features of neighbor nodes to generate updated node features;

[0026] The calculation formula for weighted aggregation of neighbor features is as follows:

[0027] ;

[0028] where, denotes the updated node feature vector of node i, denotes the initial feature vector of node j, denotes the ReLU activation function, denotes the normalized attention coefficient, denotes the set of neighbor nodes of node i, denotes the weight matrix.

[0029] In a preferred embodiment, the calculation formulas for the internal inclination angle and the anterior inclination angle of the styloid process are as follows:

[0030] The calculation formula for the internal inclination angle is:

[0031] ;

[0032] where, denotes the internal inclination angle of the styloid process in the coronal plane, , denotes the projection vector of the styloid process in the coronal plane, , denotes the reference vector, usually selected as the horizontal direction of the coronal plane;

[0033] The calculation formula for the anterior inclination angle is:

[0034] ;

[0035] Among them, represents the anterior inclination angle of the styloid process in the sagittal plane, , represents the projection vector of the styloid process in the sagittal plane, , represents the reference vector, which is usually selected as the horizontal direction of the sagittal plane.

[0036] In a preferred embodiment, in step 102, the node features include the three-dimensional coordinates of the carotid artery corresponding to each node, the vascular condition information of the carotid artery, the pathological change information of the carotid artery, and the spatial relationship information of the carotid artery.

[0037] In a preferred embodiment, a method for constructing a three-dimensional image of the styloid process by using three-dimensional reconstruction technology according to CTA image data includes the following steps:

[0038] Step 1: First, perform standardized preprocessing on the original CTA image data;

[0039] Step 2: Generate two-dimensional slice images in three perspectives of coronal, sagittal, and axial by reorganizing the original axial images in different directions through the MPR technology;

[0040] Step 3: Generate a visualized three-dimensional image of the styloid process by using the three-dimensional volume rendering technology according to the two-dimensional slice images in the three perspectives of coronal, sagittal, and axial in step 2;

[0041] Step 4: Generate two-dimensional image slices of the continuous styloid process according to the surface reconstruction technology, and obtain the anatomical structure information of the user according to the generated three-dimensional image of the styloid process.

[0042] In a preferred embodiment, the fusion method of the fusion features is to splice the feature vectors and clinical parameters of all nodes together in sequence to form the fusion features.

[0043] In a preferred embodiment, the calculation formula of the styloid process lesion prediction model is:

[0044] ;

[0045] Among them, represents the probability that the patient has carotid artery disease, represents the Sigmoid function, represents the weight vector, represents the fusion features, represents the bias vector.

[0046] A diagnostic and identification system for carotid artery dissection related to elongated styloid process includes the following modules:

[0047] A data acquisition and preprocessing module for acquiring and preprocessing the CTA image data and clinical parameters of users.

[0048] A three-dimensional reconstruction and graph structure construction module for generating a visualized three-dimensional image of the styloid process based on three-dimensional reconstruction technology and constructing a graph structure based on this.

[0049] A CNN model feature extraction module for automatically identifying and extracting key features in the CTA image data.

[0050] A graph structure update module for updating the node features in the graph structure according to the results output by the CNN model.

[0051] A fusion and risk assessment module for fusing the updated node features with the clinical parameters of the user, inputting them into a styloid process lesion prediction model, and outputting a risk probability.

[0052] A storage medium stores non-temporary computer-readable instructions, which can execute the steps in a method for identifying images related to carotid artery dissection caused by an elongated styloid process as described above when executed by a computer.

[0053] The beneficial effects of the present invention are as follows:

[0054] The identification method of the present invention can more accurately measure geometric parameters such as the length, angle of the styloid process, and the distance between the tip of the styloid process and adjacent blood vessels by combining three-dimensional reconstruction technology and a convolutional neural network model. These detailed anatomical structure information provides an intuitive view for doctors, helping to more accurately diagnose an elongated styloid process and the carotid artery dissection caused by it.

[0055] The present invention combines the image features extracted by CNN with the geometric parameters obtained by three-dimensional reconstruction to construct a more comprehensive styloid process lesion prediction model. The integration of such multiple information sources makes the finally output risk probability more reliable, providing strong support for clinical decision-making. Description of the Drawings

[0056] Figure 1 is a flowchart of the method for identifying images related to carotid artery dissection caused by an elongated styloid process of the present invention.

[0057] Figure 2 is a flowchart of the method for updating node features in the identification method of the present invention.

[0058] Figure 3 is a block diagram of the system for identifying images related to carotid artery dissection caused by an elongated styloid process of the present invention. Detailed Embodiments

[0059] The subject matter described herein will now be discussed with reference to example embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein, and the functions and arrangements of the elements discussed can be changed without departing from the scope of protection of the content of this specification. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described relative to some examples can also be combined in other examples.

[0060] As Figure 1 shown, a method for image recognition of carotid artery dissection related to elongated styloid process includes the following steps:

[0061] Step 101, obtain the historical CTA image data and clinical parameters of the user. The CTA image data includes the user's anatomical structure information, vascular condition information, pathological change information, and spatial relationship information. The clinical parameters include the user's age, gender, and medical history information;

[0062] It should be noted that the CTA image data is obtained from the CTA image films of the coronal, sagittal, and axial views obtained by the patient using coronary CTA technology in the hospital; that is, the CTA image data also includes axial images, and the CTA image films of the coronal, sagittal, and axial views are obtained from the axial images; finally, Gaussian filtering is used to remove noise and the preprocessing method of histogram equalization is used to process the user's historical CTA image data and clinical parameters.

[0063] In an embodiment of the present invention, the anatomical structure information includes: the tip position of the styloid process, the length measurement value, and the internal inclination angle and anterior inclination angle of the styloid process;

[0064] In an embodiment of the present invention, the calculation formula for the length of the styloid process is as follows:

[0065] ;

[0066] Wherein, represents the length of the styloid process, and respectively represent the coordinates of the starting point and the midpoint of the styloid process;

[0067] In an embodiment of the present invention, the calculation formula for the distance from the tip of the styloid process to the center of the adjacent carotid artery is as follows:

[0068] ;

[0069] Wherein, represents the minimum Euclidean distance from the tip of the styloid process to the center point of the adjacent carotid artery, represents the three-dimensional coordinate vector of the tip of the styloid process, Indicates a three-dimensional coordinate vector near the center point of the carotid artery, represents the minimum value function;

[0070] In an embodiment of the present invention, the calculation formulas for the internal inclination angle and the anterior inclination angle of the styloid process are as follows:

[0071] The calculation formula for the internal inclination angle is:

[0072] ;

[0073] where, represents the internal inclination angle of the styloid process on the coronal plane, , represents the projection vector of the styloid process on the coronal plane, , represents the reference vector, usually selected as the horizontal direction of the coronal plane;

[0074] The calculation formula for the anterior inclination angle is:

[0075] ;

[0076] where, represents the anterior inclination angle of the styloid process on the sagittal plane, , represents the projection vector of the styloid process on the sagittal plane, , represents the reference vector, usually selected as the horizontal direction of the sagittal plane;

[0077] The position and morphology of the carotid artery include the paths of the internal carotid artery and the external carotid artery, the position of the bifurcation point, and the morphological characteristics;

[0078] In an embodiment of the present invention, the blood vessel condition information includes: the diameter of the carotid artery, the thickness of the blood vessel wall, and the patency of the blood vessel lumen;

[0079] The diameter of the carotid artery: The diameter of the carotid artery is measured in segments at equal intervals to evaluate whether there is stenosis or dilation;

[0080] The thickness of the blood vessel wall: By measuring the thickness of the intima-media of the carotid artery wall, especially the thickness of the intima-media, it is mainly used to evaluate atherosclerosis;

[0081] The patency of the blood vessel lumen: Check whether there are plaque formations or other abnormal conditions that impede blood flow in the blood vessel lumen (expressed as a percentage);

[0082] In an embodiment of the present invention, the pathological change information includes: the location, scope, calcification condition, plaque distribution, etc. of the arterial dissection;

[0083] Calcification condition of aortic dissection: Detect the calcification points in the blood vessel wall, because calcification is one of the signs of arteriosclerosis;

[0084] Plaque distribution: Record in detail the location, size, and morphology of the plaques (the hardness of the plaques, such as soft plaques, hard plaques, etc.) and their impact on blood vessel hemodynamics;

[0085] In an embodiment of the present invention, the spatial relationship information includes: the distance between the styloid process and the adjacent blood vessels;

[0086] Distance between the styloid process and the adjacent blood vessels: The distance from the tip of the styloid process to the center of the nearest carotid artery.

[0087] Step 102, according to the CTA image data, adopt three-dimensional reconstruction technology to construct a three-dimensional image of the styloid process, construct a graph structure based on the three-dimensional image of the styloid process, the nodes include the carotid arteries divided at equal distances, and different carotid arteries are marked as different nodes. One node represents a section of a carotid artery, and the condition for there to be an edge between two nodes is that these two sections of carotid arteries are connected;

[0088] The node features include the three-dimensional coordinates of the carotid artery corresponding to each node, the blood vessel condition information of the carotid artery, the pathological change information of the carotid artery, and the spatial relationship information of the carotid artery;

[0089] It should be noted that the three-dimensional coordinates of the carotid artery corresponding to each node are the center point coordinates of this section of the carotid artery. The blood vessel condition information of the carotid artery includes the diameter of the carotid artery, the thickness of the blood vessel wall of the carotid artery, and the patency of the blood vessel lumen of the carotid artery. Among them, the diameter of the carotid artery is the average artery diameter of the section where each node is located, the thickness of the blood vessel wall of the carotid artery is the average artery wall thickness of the section where each node is located, and the patency of the blood vessel lumen of the carotid artery is the original value obtained from the CTA image data; the pathological change information of the carotid artery includes the location, scope, calcification condition, and plaque distribution of aortic dissection. Among them, the calcification condition and plaque distribution are represented by 0 or 1, and the location and scope of aortic dissection are represented by the midpoint coordinates and the area value of aortic dissection respectively; the spatial relationship information of the carotid artery includes the distance between the styloid process and the adjacent blood vessels and the spatial configuration at the blood vessel bifurcation. Among them, the distance between the styloid process and the adjacent blood vessels and the spatial configuration at the blood vessel bifurcation are both represented by the information obtained from the CTA image data.

[0090] In an embodiment of the present invention, the encoding method of the node features is carried out in the one-hot encoding method and spliced in sequence;

[0091] In an embodiment of the present invention, the method for constructing a three-dimensional image of the styloid process according to the CTA image data includes the following steps:

[0092] Step 1. First, perform standardized preprocessing on the original CTA image data;

[0093] It should be noted that the operations for standardizing and preprocessing the original CTA image data include operations such as denoising and contrast enhancement to improve the image quality and make subsequent analysis more accurate;

[0094] Step 2: Use the MPR technique to reconstruct the original axial images in different directions to generate two-dimensional slice images in three perspectives: coronal, sagittal, and axial;

[0095] Multi-planar reconstruction involves reconstructing the original axial images into two-dimensional slices in different perspectives. The specific steps are as follows:

[0096] Coronal plane: Cut along the anterior-posterior direction to generate a series of two-dimensional slices parallel to the anterior-posterior direction;

[0097] Sagittal plane: Cut along the left-right direction to generate a series of two-dimensional slices parallel to the left-right direction;

[0098] Axial plane: Cut along the up-down direction to generate a series of two-dimensional slices parallel to the ground;

[0099] It should be noted that the MPR technique is the multi-planar reconstruction technique. MPR is a technique that generates two-dimensional slice images in multiple perspectives such as coronal, sagittal, and axial by reconstructing the original axial images in different directions;

[0100] Coronal plane: Perpendicular to the anterior-posterior direction, scan from the front of the head to the back, showing the left and right structures;

[0101] Sagittal plane: Perpendicular to the left-right direction, observe from the side, showing the anterior and posterior structures;

[0102] Axial plane: Parallel to the ground, scan from the top of the head downwards, showing the up and down structures.

[0103] Step 3: Based on the two-dimensional slice images in the three perspectives of coronal, sagittal, and axial in Step 2, use the three-dimensional volume rendering technique to generate a visualized three-dimensional image of the styloid process;

[0104] Three-dimensional volume rendering is achieved by combining multiple two-dimensional slices into a complete three-dimensional model. The following are the basic steps:

[0105] Data preparation: Obtain the preprocessed CTA image data;

[0106] Volume rendering: Use a volume rendering algorithm (such as the ray casting method) to convert the two-dimensional slice data into a three-dimensional stereoscopic view;

[0107] Step 4: Generate two-dimensional image slices of the continuous styloid process according to the surface reconstruction technique, and obtain the user's anatomical structure information based on the generated three-dimensional image of the styloid process.

[0108] The recognition method of the present invention uses three-dimensional reconstruction technology to provide a more intuitive view of the anatomical structure, helping doctors better understand the specific location, shape, and relationship with surrounding tissues of the lesion. This visualization helps doctors make more accurate diagnoses. And through three-dimensional reconstruction, geometric parameters such as styloid process length, angle (such as internal inclination angle and anterior inclination angle), and the distance between the tip of the styloid process and adjacent blood vessels can be measured more precisely. These parameters are crucial for evaluating the impact of an elongated styloid process on the carotid artery. If surgical intervention is required, the detailed information provided by three-dimensional reconstruction can help surgeons formulate a more precise surgical plan.

[0109] The surface reconstruction technology can generate continuous two-dimensional slices along the surface of the anatomical structure of interest. The specific steps are as follows:

[0110] Define the path: Select a path along the structure of interest (such as the styloid process).

[0111] Generate slices: Generate a series of continuous two-dimensional slices along this path, and each slice represents a cross-section of the structure.

[0112] Step 103: After encoding the CTA image data, input it into the CNN model to output the identified abnormal area of the styloid process and the characteristics of carotid artery dissection. The CNN model includes a U-Net network and an output layer. The U-Net network includes an encoder and a decoder. The loss function of the CNN model is binary cross-entropy loss, and the Adam optimizer is used.

[0113] Step 104: Update the node features of the graph structure constructed in step 102.

[0114] In an embodiment of the present invention, as Figure 2 shown, the method for updating node features includes the following steps:

[0115] Step 201: Perform a linear transformation on all node features in the graph structure to generate node transformation feature vectors.

[0116] The calculation formula for the linear transformation is as follows:

[0117] ;

[0118] Among them, represents the node transformation feature vector of node i, represents the initial feature vector of node i, represents the weight matrix;

[0119] Step 202: Calculate the attention coefficient between node i and node j.

[0120] The calculation formula for the attention coefficient is as follows:

[0121] ;

[0122] Among them, represents the attention coefficient between node i and node j, represents the node transformation feature vector of node i, represents the node transformation feature vector of node j, represents the weight matrix, represents the attention vector, represents the vector transpose, represents function, represents vector concatenation;

[0123] Step 203, normalize the attention coefficient;

[0124] The calculation formula for normalizing the attention coefficient is as follows:

[0125] ;

[0126] Among them, represents the normalized attention coefficient, represents the set of neighbor nodes of node i, represents the attention coefficient between node i and node j, represents the natural exponential function;

[0127] Step 204, according to the normalized attention coefficient, perform weighted aggregation on the features of neighbor nodes to generate updated node features;

[0128] The calculation formula for weighted aggregation of neighbor features is as follows:

[0129] ;

[0130] Among them, represents the updated node feature vector of node i, represents the initial feature vector of node j, represents the ReLU activation function, represents the normalized attention coefficient, represents the set of neighbor nodes of node i, represents the weight matrix.

[0131] Step 105, construct a styloid process lesion prediction model, perform feature fusion on the updated node features and the user's clinical parameters to generate fusion features, and input the fusion features into the styloid process lesion prediction model. The styloid process lesion prediction model outputs the risk probability of the user suffering from carotid artery dissection;

[0132] In one embodiment of the present invention, the calculation formula of the styloid process lesion prediction model is as follows:

[0133] ;

[0134] wherein, represents the probability that the patient has carotid artery disease, represents the Sigmoid function, represents the weight vector, represents the fused feature, represents the bias vector;

[0135] In one embodiment of the present invention, the method for generating the fused feature by fusing the updated node features and the clinical parameters of the user is to splice the feature vectors of all nodes and the clinical parameters together to form a new feature vector;

[0136] For example, all the node feature vectors updated after step 104 , wherein, represents the number of nodes; assuming that the clinically set parameters of the user are encoded to form a feature vector , then the fused feature is: .

[0137] By designing a CNN model, the present invention can automatically identify and extract key features, and can automatically learn and extract key features from a large amount of CTA image data, such as styloid process abnormalities and carotid artery dissection, which greatly reduces the workload of manual annotation and improves the processing speed; through three-dimensional reconstruction, geometric parameters such as the length, angle (such as the internal inclination angle and the anterior inclination angle) of the styloid process, and the distance between the tip of the styloid process and the adjacent blood vessels can be measured more accurately, which can assist doctors in making more accurate diagnostic and treatment decisions. Especially in complex cases, by combining the features extracted by the CNN with the geometric parameters of the three-dimensional reconstruction, the present invention can provide more comprehensive information, thereby improving the accuracy of risk assessment, combining multiple information sources, and establishing a more powerful prediction model, making the final risk probability more reliable.

[0138] By combining the CNN and three-dimensional reconstruction methods, the present invention can make full use of the advantages of both, realize efficient and accurate key feature identification and extraction, and at the same time provide an intuitive anatomical structure view and accurate geometric parameter measurement. This method not only improves the accuracy of diagnosis, but also finally constructs a styloid process lesion prediction model that can output a more reliable risk probability by integrating multi-dimensional information, providing strong support for clinical decision-making.

[0139] In one embodiment of the present invention, as Figure 3 shown, a diagnostic and recognition system for carotid artery dissection related to elongated styloid process includes the following modules:

[0140] A data acquisition and preprocessing module for acquiring and preprocessing the user's CTA image data and clinical parameters;

[0141] It should be noted that the historical CTA image data of the user (including anatomical structure information, vascular condition information, pathological change information, and spatial relationship information) and clinical parameters (such as age, gender, medical history, etc.) are acquired; the original CTA image data is preprocessed standardly (denoising, contrast enhancement, etc.), and two-dimensional slice images in three perspectives of coronal, sagittal, and axial are generated;

[0142] A three-dimensional reconstruction and graph structure construction module for generating a visualized three-dimensional image of the styloid process according to three-dimensional reconstruction technology and constructing a graph structure based on this;

[0143] It should be noted that the MPR technology and three-dimensional volume rendering technology are used to generate a three-dimensional image of the styloid process; according to the generated three-dimensional image, geometric parameters such as the length, internal inclination angle, anterior inclination angle of the styloid process, and the distance from the tip of the styloid process to the center of the adjacent carotid artery are calculated; a graph structure is constructed, where the nodes represent different segments of the carotid artery, and the edges represent the connection relationships between adjacent carotid artery segments. Each node contains the three-dimensional coordinates of the corresponding carotid artery, vascular condition information, pathological change information, and spatial relationship information;

[0144] A CNN model feature extraction module for automatically identifying and extracting key features in the CTA image data;

[0145] It should be noted that the preprocessed CTA image data is input into a CNN model (such as a U-Net network), and the identified abnormal regions of the styloid process and carotid artery dissection features are output;

[0146] The loss function uses binary cross-entropy loss, and the optimizer uses the Adam optimizer;

[0147] A graph structure update module for updating the node features in the graph structure according to the results output by the CNN model;

[0148] It should be noted that linear transformation is performed on all node features to generate node transformation feature vectors; the attention coefficients between nodes are calculated and normalized; the features of neighbor nodes are weighted and aggregated according to the normalized attention coefficients to generate updated node features;

[0149] A fusion and risk assessment module fuses the updated node features with the user's clinical parameters, inputs them into a styloid process lesion prediction model, and outputs the risk probability.

[0150] It should be noted that the feature vectors of all nodes and the clinical parameters of the user are concatenated together to form a new feature vector; the fused feature is input into the styloid process lesion prediction model to calculate the risk probability of the patient having carotid artery dissection;

[0151] It should be further noted that an output and decision support module can also be connected after the fusion and risk assessment module to provide the final risk assessment result and assist the doctor in making a diagnosis and treatment decision;

[0152] For example, the risk probability of carotid artery dissection of the patient can be output, and a detailed anatomical structure view and geometric parameter measurement results can be provided to help the doctor formulate a more precise surgical plan or preventive measure.

[0153] Through the above modular system design, the present invention can effectively realize the full-process automatic processing from data acquisition to final risk assessment, improve the diagnostic efficiency and accuracy, and provide strong support for clinical practice.

[0154] A storage medium stores non-temporary computer-readable instructions, which can execute the steps in a method for identifying carotid artery dissection related to elongated styloid process as described above when executed by a computer.

[0155] The above describes the embodiments of this embodiment, but this embodiment is not limited to the above specific implementation manners. The above specific implementation manners are merely illustrative and not restrictive. Under the inspiration of this embodiment, those of ordinary skill in the art can also make many forms, all of which fall within the protection scope of this embodiment.

Claims

1. A method for image recognition of carotid artery dissection associated with excessive styloid process, characterized in that: The following steps are involved: Step 101, obtaining historical CTA image data and clinical parameters of the user, wherein the CTA image data includes the user's anatomical structure information, vascular condition information, pathological change information and spatial relationship information, and the clinical parameters include the user's age, gender and medical history information; Step 102, constructing a three-dimensional image of the styloid process using a three-dimensional reconstruction technology based on the CTA image data, and constructing a graph structure based on the three-dimensional image of the styloid process, wherein the nodes include carotid arteries divided at equal distances, and different carotid arteries are marked as different nodes, and a node represents a segment of a carotid artery. The condition for the existence of an edge between two nodes is that the two segments of the carotid arteries are connected; Step 103, encoding the CTA image data and inputting it into the CNN model, outputting the identified styloid process abnormal area and carotid artery dissection features, the CNN model includes a U-Net network and an output layer, and the U-Net network includes an encoder and a decoder; Step 104, updating the node features of the graph structure constructed in step 102; Step 105, constructing a styloid process lesion prediction model, fusing the updated node features and the user's clinical parameters to generate fusion features, inputting the fusion features into the styloid process lesion prediction model, and the styloid process lesion prediction model outputs the risk probability of the user suffering from carotid artery dissection; The method for updating node features comprises the following steps: Step 201, linearly transform all node features in the graph structure to generate a node transformation feature vector; The linear transformation is calculated as follows: ； in, represents the node transformation feature vector of node i, represents the initial eigenvector of node i, W represents the weight matrix; Step 202, calculating the attention coefficient between node i and node j; The calculation formula of attention coefficient is as follows: ； in, represents the attention coefficient between node i and node j, represents the node transformation feature vector of node i, represents the node transformation feature vector of node j, a represents the attention vector, represents the vector transpose, express function, Represents vector concatenation; Step 203, normalizing the attention coefficient; The calculation formula for the normalized attention coefficient is as follows: ； in, represents the normalized attention coefficient, represents the set of neighbor nodes of node i, represents the natural exponential function; Step 204, performing weighted aggregation on the features of neighbor nodes according to the normalized attention coefficient to generate updated node features; The calculation formula of weighted aggregated neighbor features is as follows: ； in, represents the updated node feature vector of node i, represents the initial eigenvector of node j, represents the ReLU activation function, Represents the normalized attention coefficient; The node features include the three-dimensional coordinates of the carotid artery corresponding to each node, the vascular condition information of the carotid artery, the pathological change information of the carotid artery, and the spatial relationship information of the carotid artery; The three-dimensional coordinates of the carotid artery corresponding to each node are the coordinates of the center point of the segment of the carotid artery. The vascular condition information of the carotid artery includes the diameter of the carotid artery, the vascular wall thickness of the carotid artery, and the vascular lumen patency of the carotid artery, where the diameter of the carotid artery is the average arterial diameter of the segment where each node is located, the vascular wall thickness of the carotid artery is the average arterial wall thickness of the segment where each node is located, and the vascular lumen patency of the carotid artery is the original value obtained from the CTA imaging data; the pathological change information of the carotid artery includes the location, range, calcification, and plaque distribution of the arterial dissection, where the calcification and plaque distribution are represented by 0 or 1, and the location and range of the arterial dissection are represented by the midpoint coordinates and area numerical size of the arterial dissection, respectively; the spatial relationship information of the carotid artery includes the distance between the styloid process and the adjacent blood vessels and the spatial configuration at the vascular bifurcation, where the distance between the styloid process and the adjacent blood vessels and the spatial configuration at the vascular bifurcation are both represented by the information obtained from the CTA imaging data.

2. The method for image recognition of carotid artery dissection associated with excessive styloid process according to claim 1, characterized in that: The anatomical structure information includes: the tip position of the styloid process, the length measurement value, and the inclination and anteversion angles of the styloid process; the vascular condition information includes: the diameter of the carotid artery, the thickness of the vascular wall, and the patency of the vascular lumen; the pathological change information includes: the location, range, calcification, and plaque distribution of the arterial dissection; the spatial relationship information includes: the distance between the styloid process and the adjacent blood vessels.

3. The method for image recognition of carotid artery dissection associated with excessive styloid process according to claim 2, characterized in that: The calculation formulas for the medial and anteversion angles of the styloid process are as follows: The calculation formula for the inclination angle is: ； in, represents the inclination angle of the styloid process in the coronal plane, , represents the projection vector of the styloid process on the coronal plane, , represents the reference vector, which is usually chosen to be the horizontal direction of the coronal plane; The formula for calculating the forward inclination angle is: ； in, represents the anteversion angle of the styloid process in the sagittal plane. , represents the projection vector of the styloid process on the sagittal plane, , represents the reference vector, which is usually chosen to be the horizontal direction of the sagittal plane.

4. The method for image recognition of carotid artery dissection associated with excessive styloid process according to claim 3, characterized in that: The method for constructing a three-dimensional image of the styloid process using three-dimensional reconstruction technology based on CTA image data includes the following steps: Step 1: First, the original CTA image data is preprocessed in a standardized manner; Step 2: Recombining the original axial image in different directions through MPR technology to generate two-dimensional slice images of three viewing angles: coronal, sagittal and axial. Step 3: Based on the two-dimensional slice images of the coronal, sagittal and axial viewpoints in step 2, a three-dimensional volume rendering technology is used to generate a visualized three-dimensional image of the styloid process; Step 4: Generate continuous two-dimensional image slices of the styloid process based on the surface reconstruction technology, and obtain the user's anatomical structure information based on the generated three-dimensional image of the styloid process.

5. The method for image recognition of carotid artery dissection associated with excessive styloid process according to claim 4, characterized in that: The fusion method of fusion features is to sequentially concatenate the feature vectors and clinical parameters of all nodes to form fusion features.

6. The method for image recognition of carotid artery dissection associated with excessive styloid process according to claim 5, characterized in that: The calculation formula of the styloid process lesion prediction model is: ； in, represents the probability that the patient suffers from carotid artery disease, represents the Sigmoid function, represents the weight vector, represents the fusion feature, Represents the bias vector.

7. A system for image recognition of carotid artery dissection associated with excessive styloid process, used to execute the method for image recognition of carotid artery dissection associated with excessive styloid process as claimed in any one of claims 1 to 6, characterized in that: Includes the following modules: Data acquisition and preprocessing module, used to acquire and preprocess the user's CTA imaging data and clinical parameters; A three-dimensional reconstruction and graph structure building module is used to generate a visualized three-dimensional image of the styloid process according to the three-dimensional reconstruction technology and to build a graph structure based on the image; CNN model feature extraction module, used to automatically identify and extract key features from CTA image data; The graph structure update module is used to update the node features in the graph structure according to the output results of the CNN model; The fusion and risk assessment module fuses the updated node features with the user's clinical parameters, inputs them into the styloid process lesion prediction model, and outputs the risk probability.

8. A storage medium, characterized in that: It stores non-transitory computer-readable instructions, and when the non-transitory computer-readable instructions are executed by a computer, the steps in the image recognition method for carotid artery dissection associated with excessive styloid process as described in claim 6 can be executed.

Images