Monitoring method, system and storage medium for dynamically monitoring lower extremity venous blood flow status

Abstract

The present invention relates to the field of orthopedic diagnosis and processing technology, and in particular to a monitoring method, system and storage medium for dynamically monitoring the blood flow status of lower extremity veins. The method includes the following steps: acquiring a sequence of lower extremity vein ultrasound images of a target patient, and establishing a vascular space probability distribution map based on anatomical prior knowledge; encoding the vascular space probability distribution map into conditional prior information to construct a vascular ontology knowledge base; constructing a multi-level diffusion probability model based on the vascular ontology knowledge base; and using the multi-level diffusion probability model to reconstruct the vascular structure and extract blood flow features from the lower extremity vein ultrasound image sequence, thereby obtaining vascular fluid feature data. The present invention uses a three-level diffusion model to gradually complete vascular positioning, branch identification and morphological analysis from macro to micro, while extracting blood flow dynamic information in combination with time series features, thereby achieving a comprehensive characterization of the vascular system.

Description

Technical Field

[0001] The present invention relates to the technical field of orthopedic diagnosis, and in particular to a monitoring method, system and storage medium for dynamically monitoring the venous blood flow status of lower extremities. Background Art

[0002] After orthopedic surgery (especially major lower limb surgery such as hip and knee replacement), patients often need to stay in bed for a long time. Due to long-term bed rest and surgical trauma, blood flow slows down, which greatly increases the risk of deep vein thrombosis. Dynamic monitoring of lower limb venous blood flow status can timely assess the risk of DVT in patients, which is of great significance for the prevention and early diagnosis of DVT. Orthopedic surgery patients need to recover in a specific position after surgery, and different positions have different effects on lower limb venous hemodynamics.

[0003] At present, the prevention and treatment of perioperative DVT in orthopedic surgery mainly rely on empirical drug prevention and intermittent lower limb pressure pumps, lacking targeted real-time monitoring methods. However, traditional methods for dynamic monitoring of lower limb venous blood flow often have the following problems: due to the complex course of lower limb veins and the interweaving of deep and superficial veins, it is very challenging to accurately locate the target blood vessels and keep the probe in the optimal sampling position; especially when observing deep veins, interference from different anatomical levels will increase the difficulty of positioning. When processing deep venous blood flow data, diffusion models often have difficulty in accurately characterizing multi-level vascular structures; especially when observing early signs of deep vein thrombosis, the model may not be able to effectively distinguish between normal blood flow variations and pathological changes. Summary of the invention

[0004] Based on this, it is necessary for the present invention to provide a monitoring method, system and storage medium for dynamically monitoring the venous blood flow status of the lower extremities to solve at least one of the above-mentioned technical problems.

[0005] To achieve the above object, a method for dynamically monitoring the blood flow status of lower extremity veins comprises the following steps:

[0006] Step S1: Obtain a sequence of lower limb venous ultrasound images of a target patient, and establish a vascular space probability distribution map based on anatomical prior knowledge; encode the vascular space probability distribution map into conditional prior information to construct a vascular ontology knowledge base;

[0007] Step S2: constructing a multi-level diffusion probability model according to the vascular ontology knowledge base, wherein the multi-level diffusion probability model specifically includes a first-level diffusion model for spatially locating the main vascular directions in the ultrasound image, a second-level diffusion model for identifying the vascular branching structure and interweaving relationship, and a third-level diffusion model for analyzing the morphological characteristics of a single vascular segment;

[0008] Step S3: using a multi-level diffusion probability model to reconstruct the vascular structure and extract blood flow characteristics from the lower limb venous ultrasound image sequence, thereby obtaining vascular fluid characteristic data;

[0009] Step S4: establishing a deep vein thrombosis risk assessment model based on vascular structure and blood flow characteristics according to the vascular fluid characteristic data, and performing deep vein thrombosis risk assessment on the lower extremity venous ultrasound image sequence, thereby obtaining thrombosis risk assessment data;

[0010] Step S5: Determine the monitoring warning level based on the thrombosis risk assessment data, generate a monitoring report containing vascular structure reconstruction images, blood flow dynamic characteristics and risk assessment analysis, and feed the monitoring report back to the vascular ontology knowledge base to continuously optimize the accuracy of postoperative vascular identification and risk assessment in orthopedics.

[0011] The present invention can provide important prior information for the identification of vascular structure by acquiring the target patient's lower limb venous ultrasound image sequence and establishing a vascular space probability distribution map based on anatomical prior knowledge. As conditional prior information, the map helps to enhance the accuracy of the vascular ontology knowledge base, thereby providing strong support for subsequent vascular structure reconstruction and blood flow feature extraction. By establishing a vascular space probability distribution map, the misidentification caused by image noise or changes in acquisition conditions can be greatly reduced. The construction of a multi-level diffusion probability model makes vascular identification more accurate. The first-level diffusion model focuses on the spatial positioning of the main vascular direction, ensuring the accurate detection of large blood vessels; the second-level diffusion model focuses on the identification of vascular branch structures and intertwined relationships, which helps to more carefully identify complex vascular networks and reduce structural errors; the third-level model focuses on the morphological feature analysis of a single vascular segment, further improving the recognition accuracy of vascular details. This hierarchical and layer-by-layer optimized model construction method can provide information at different levels, so that the spatial distribution and morphological characteristics of the blood vessels are effectively captured. Using a multi-level diffusion probability model to reconstruct vascular structure and extract blood flow characteristics can deeply explore the movement law and pressure distribution of intravascular fluid, and provide the necessary physical basis for the potential risk of thrombosis. Vascular fluid characteristic data not only helps the accuracy of thrombosis risk assessment, but also provides a scientific basis for further clinical decision-making, especially in early clinical detection and intervention. Establishing a deep vein thrombosis risk assessment model based on vascular structure and blood flow characteristics can comprehensively consider the morphological characteristics and hemodynamic parameters of the blood vessels, so as to more comprehensively and accurately assess the risk of thrombosis. Thrombosis is a complex process. Considering the multiple influencing factors of vascular structure and flow state, it can greatly improve the accuracy of prediction, identify potential high-risk patients in advance, and help doctors develop more accurate treatment and prevention plans. According to the thrombosis risk assessment data, the monitoring warning level is determined, and a detailed monitoring report is generated. This report not only contains vascular structure reconstruction images, but also includes blood flow dynamic characteristics and risk assessment analysis, providing doctors with a comprehensive clinical reference. Feeding the monitoring report back to the vascular ontology knowledge base will help to continuously optimize the recognition accuracy of vascular morphology and blood flow characteristics, forming a self-iterative and gradually optimized knowledge system to ensure the accuracy and adaptability of models and methods in practical applications. In summary, the above steps can effectively improve the accuracy and practicality of thrombosis risk assessment through the introduction of multi-level models and comprehensive analysis of vascular and blood flow characteristics, which is of great significance for the early detection of lower extremity venous thrombosis, risk warning and the formulation of personalized treatment plans.

[0012] The diffusion model simulates the process of gradually adding noise to a data set until the data is completely transformed into noise, and then recovers the original data from the noise through the reverse process. The core idea of ​​this model is to gradually transform complex data distribution into simple distribution, and then gradually restore the simple distribution to complex distribution, thereby generating new data. The working process of the diffusion model includes two main stages: forward diffusion process and reverse diffusion process. In the forward diffusion process, the model gradually introduces noise into the data until the data is completely transformed into noise; in the reverse diffusion process, the model gradually recovers the original data from the noise.

[0013] The present invention also provides a monitoring system for dynamically monitoring the blood flow state of the lower limb veins, which is used to execute the monitoring method for dynamically monitoring the blood flow state of the lower limb veins as described above. The monitoring system for dynamically monitoring the blood flow state of the lower limb veins includes:

[0014] The vascular knowledge construction module is used to obtain the target patient's lower limb venous ultrasound image sequence and establish a vascular space probability distribution map based on anatomical prior knowledge; the vascular space probability distribution map is encoded into conditional prior information to construct a vascular ontology knowledge base;

[0015] A multi-level diffusion modeling module is used to construct a multi-level diffusion probability model based on a vascular ontology knowledge base, wherein the multi-level diffusion probability model specifically includes a first-level diffusion model for spatially locating the main vascular orientation in an ultrasound image, a second-level diffusion model for identifying vascular branch structures and interweaving relationships, and a third-level diffusion model for analyzing the morphological characteristics of a single vascular segment;

[0016] The vascular reconstruction and feature extraction module is used to reconstruct the vascular structure and extract the blood flow features of the lower limb venous ultrasound image sequence using a multi-level diffusion probability model, thereby obtaining vascular fluid feature data;

[0017] A risk assessment model building module is used to establish a deep vein thrombosis risk assessment model based on vascular structure and blood flow characteristics according to vascular fluid characteristic data, and to perform deep vein thrombosis risk assessment on lower extremity venous ultrasound image sequences, thereby obtaining thrombosis risk assessment data;

[0018] The dynamic monitoring module is used to determine the monitoring warning level based on the thrombosis risk assessment data, generate a monitoring report containing vascular structure reconstruction images, blood flow dynamic characteristics and risk assessment analysis, and feed the monitoring report back to the vascular ontology knowledge base to continuously optimize the accuracy of vascular identification and risk assessment after orthopedic surgery.

[0019] The present invention also provides a storage medium, comprising:

[0020] Memory for storing computer programs;

[0021] The processor is used to implement the above-mentioned method for dynamically monitoring the lower extremity venous blood flow status when executing the computer program.

[0022] The present invention can improve the accuracy of vascular detection by acquiring the lower limb venous ultrasound image sequence of the target patient and establishing a vascular space probability distribution map in combination with anatomical prior knowledge. Anatomical prior knowledge enables the spatial position, direction, size and other characteristics of the vascular structure to be more effectively encoded and expressed, providing a solid foundation for subsequent vascular identification and feature extraction. Encoding the vascular space probability distribution map as conditional prior information and constructing a vascular ontology knowledge base can help the system to deeply understand and analyze the vascular structure. The construction of the knowledge base can not only integrate the vascular data of different cases, but also provide rich reference materials for subsequent image analysis, modeling, risk assessment, etc., thereby improving the system's ability to recognize and understand complex vascular morphology. The role of this module is to construct a multi-level diffusion probability model based on the vascular ontology knowledge base, so as to realize the hierarchical analysis of the vascular structure in the lower limb venous ultrasound image. The multi-level model can accurately model the different scales and complex structures of the blood vessels, which helps to improve the recognition ability of characteristics such as vascular direction, vascular branching structure and blood flow dynamics. The first-level diffusion model locates the main blood vessel direction in the ultrasound image, which can help the system accurately determine the basic direction and distribution of the blood vessels, laying the foundation for the subsequent analysis of the branch structure and interweaving relationship of the blood vessels. The second-level model can identify the branch structure of the blood vessels and their interweaving relationship, improve the recognition ability of the complex structure of the blood vessels, especially in the vascular network involving multiple branches, and can effectively distinguish the relative relationship between the blood vessels. The third-level model analyzes the morphological characteristics of a single blood vessel segment to provide more detailed structural feature data for subsequent vascular health assessment, and help identify abnormalities such as changes in the vascular wall, vascular stenosis, and bending. By using a multi-level diffusion probability model to reconstruct the vascular structure and extract blood flow features from the ultrasound image sequence of the lower limb veins, accurate vascular fluid feature data can be obtained. The core advantage of this module is that it can comprehensively model the morphology and function of the blood vessels, and provide important basic data for subsequent risk assessment and diagnosis by extracting the dynamic characteristics of the blood flow. This step can reflect the changing trend of blood flow in the blood vessels in real time, reveal the abnormality of hemodynamics, and help assess the risk of thrombosis. According to the vascular fluid characteristic data, a deep vein thrombosis risk assessment model based on vascular structure and blood flow characteristics is established. Through the comprehensive analysis of hemodynamic characteristics, vascular morphological characteristics and other data, the risk of thrombosis in patients can be quantitatively assessed. Predicting thrombosis risk through the model can provide doctors with quantitative risk assessment, help doctors make more accurate diagnosis and treatment decisions, and provide early warning and intervention measures. Risk assessment data provides a basis for subsequent treatment and monitoring, can help identify high-risk patients, and provide support for the formulation of personalized treatment plans. Through risk assessment, doctors can better plan patient monitoring plans and treatment strategies.This module determines the monitoring and warning level based on the thrombosis risk assessment data, and generates a monitoring report containing vascular structure reconstruction images, blood flow dynamic characteristics and risk assessment analysis. These reports not only help doctors to monitor patients regularly, but also issue early warning information in time when the patient's thrombosis risk changes to ensure the health of the patient. The report integrates the vascular health status, hemodynamic characteristics and potential risks, providing doctors with a detailed and intuitive decision support tool. Feeding back the monitoring report to the vascular ontology knowledge base can provide data support for the system to continuously optimize the recognition accuracy of vascular morphology and blood flow characteristics. With the continuous accumulation of monitoring data, the knowledge base can self-update and optimize to improve the system's recognition and prediction capabilities. This feedback mechanism enables the system to evolve continuously, thereby improving the diagnosis and treatment support capabilities for vascular diseases. These modules combine vascular anatomical knowledge, image analysis, fluid dynamics modeling, risk assessment and dynamic monitoring. Through the synergistic effect of each module, a comprehensive thrombosis risk assessment and intervention system is formed. The system can achieve efficient and accurate vascular health assessment, thrombosis risk prediction and monitoring and early warning, and comprehensively improve the decision support capabilities of clinicians. At the same time, the system can continuously optimize its recognition capabilities through feedback mechanisms and knowledge base updates, and gradually achieve accurate management of vascular health. This process not only helps reduce the occurrence of thrombosis-related diseases, but also improves the treatment effect and survival rate of patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments thereof made with reference to the following drawings:

[0024] Figure 1 It is a schematic diagram of the steps of the monitoring method for dynamically monitoring the blood flow status of the lower extremity veins of the present invention;

[0025] Figure 2 for Figure 1 Detailed step flow diagram of step S1;

[0026] Figure 3 for Figure 1 Detailed step flow chart of step S2 in FIG. DETAILED DESCRIPTION

[0027] The technical method of the present invention is described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by technicians in this field without creative work are within the scope of protection of the present invention.

[0028] In addition, the accompanying drawings are only schematic illustrations of the present invention and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor methods and / or microcontroller methods.

[0029] It should be understood that, although the terms "first", "second", etc. may be used herein to describe various units, these units should not be limited by these terms. These terms are used only to distinguish one unit from another unit. For example, without departing from the scope of the exemplary embodiments, the first unit may be referred to as the second unit, and similarly the second unit may be referred to as the first unit. The term "and / or" used herein includes any and all combinations of one or more of the listed associated items.

[0030] To achieve this, please refer to Figures 1 to 3 The present invention provides a method for dynamically monitoring the blood flow status of the lower extremity veins, the method comprising the following steps:

[0031] In the embodiment of the present invention, reference Figure 1 FIG. 1 is a schematic diagram of a step flow chart of a monitoring method for dynamically monitoring the blood flow status of the lower extremity veins of the present invention. In this example, the monitoring method for dynamically monitoring the blood flow status of the lower extremity veins includes the following steps:

[0032] Step S1: Obtain a sequence of lower limb venous ultrasound images of a target patient, and establish a vascular space probability distribution map based on anatomical prior knowledge; encode the vascular space probability distribution map into conditional prior information to construct a vascular ontology knowledge base;

[0033] The embodiment of the present invention obtains a sequence of ultrasound images of the lower extremity veins of a target patient, which are collected by standard ultrasound equipment to ensure coverage of all relevant vein areas from the thigh root to the calf area. Then, based on anatomical prior knowledge, a vascular space probability distribution atlas is established by using automated image segmentation technology and deep learning methods. The atlas combines anatomical data, such as the common location and size of veins, to generate a high-precision estimate of the vascular distribution of the target area. The atlas is encoded as conditional prior information as the basis for vascular structure recognition in subsequent steps, and a vascular ontology knowledge base is constructed. The knowledge base is used to store the anatomical information, typical vascular paths, and spatial relationships of the veins to form a vascular ontology database that can be used for subsequent reasoning and analysis.

[0034] Step S2: constructing a multi-level diffusion probability model according to the vascular ontology knowledge base, wherein the multi-level diffusion probability model specifically includes a first-level diffusion model for spatially locating the main vascular directions in the ultrasound image, a second-level diffusion model for identifying the vascular branching structure and interweaving relationship, and a third-level diffusion model for analyzing the morphological characteristics of a single vascular segment;

[0035] The embodiment of the present invention constructs a multi-level diffusion probability model based on the vascular ontology knowledge base. First, a first-level diffusion model is constructed to spatially locate the main vascular directions in ultrasound images. This level uses convolutional neural networks (CNNs) and spatial filtering technology to identify the main vascular paths in the image and accurately locate the vascular directions from the thigh to the calf. Next, a second-level diffusion model is constructed to identify the branching structure and interweaving relationship of the blood vessels. By analyzing the branching angles, diameter changes and other characteristics of the blood vessels, a graph neural network (GNN) is used to analyze the connection pattern and branching relationship of the blood vessels. Finally, a third-level diffusion model is constructed, focusing on analyzing the morphological characteristics of a single vascular segment, and using deep learning morphological analysis methods (such as U-Net) to extract the shape, width and curvature of the blood vessels to ensure accurate description and identification of each vascular segment.

[0036] Step S3: using a multi-level diffusion probability model to reconstruct the vascular structure and extract blood flow characteristics from the lower limb venous ultrasound image sequence, thereby obtaining vascular fluid characteristic data;

[0037] The embodiment of the present invention utilizes a constructed multi-level diffusion probability model to reconstruct the vascular structure and extract blood flow features from a sequence of lower extremity venous ultrasound images. In terms of vascular structure reconstruction, the model uses the vascular pixel information in the image to connect and infer the branches of the vascular tree to ensure that all key vascular structures are completely reconstructed. In terms of blood flow feature extraction, multimodal data (such as ultrasound spectrum analysis) is used to extract dynamic information such as blood flow velocity, direction, and blood flow, and the blood flow characteristics are analyzed in detail in combination with acoustic characteristics. This process combines spatiotemporal information with multi-dimensional information of ultrasound data to accurately obtain vascular fluid feature data.

[0038] Step S4: establishing a deep vein thrombosis risk assessment model based on vascular structure and blood flow characteristics according to the vascular fluid characteristic data, and performing deep vein thrombosis risk assessment on the lower extremity venous ultrasound image sequence, thereby obtaining thrombosis risk assessment data;

[0039] The embodiment of the present invention constructs a deep vein thrombosis risk assessment model based on the obtained vascular fluid characteristic data. The model integrates information such as vascular structure, blood flow velocity, blood flow direction, etc., and is trained through a machine learning algorithm (such as support vector machine SVM or random forest RF) to assess the risk of thrombosis. During the training process, a historical data set containing known thrombosis cases is used to adjust the model parameters to optimize its accuracy. The evaluation result output by the model is the thrombosis risk level of each vascular segment, and a personalized risk prediction is given in combination with the individual differences of each patient.

[0040] Step S5: Determine the monitoring warning level based on the thrombosis risk assessment data, generate a monitoring report containing vascular structure reconstruction images, blood flow dynamic characteristics and risk assessment analysis, and feed the monitoring report back to the vascular ontology knowledge base to continuously optimize the accuracy of postoperative vascular identification and risk assessment in orthopedics.

[0041] The embodiment of the present invention determines the monitoring and early warning level based on the thrombosis risk assessment data. Based on the assessment results, the system sets different monitoring levels, such as low risk, medium risk, and high risk. Each level corresponds to a different early warning strategy, such as regular inspections, close monitoring, or immediate intervention. The monitoring report generated by the system includes detailed contents such as vascular structure reconstruction images, blood flow dynamic characteristics, and risk assessment analysis. Finally, the report is returned to the vascular ontology knowledge base through a feedback mechanism for continuous optimization, improving the recognition accuracy of vascular morphology and blood flow characteristics, and ensuring efficient and accurate analysis of future patient images.

[0042] As an embodiment of the present invention, refer to Figure 2 As shown, Figure 1 Detailed step flow diagram of step S1 in the embodiment of the present invention, step S1 includes the following steps:

[0043] Step S11: Scanning the lower extremity veins of the target patient using a high-resolution ultrasound device to collect continuous cross-sectional images and longitudinal images of the venous system including the femoral vein, popliteal vein, and small saphenous vein, thereby obtaining a lower extremity venous ultrasound image sequence;

[0044] The embodiment of the present invention uses a high-resolution ultrasound device (such as GE LOGIQ E10 or Philips EPIQ 7) to scan the lower extremity veins of the target patient. During the scanning process, the main veins such as the femoral vein, popliteal vein, small saphenous vein and their branches are covered to ensure that the image covers the entire lower extremity venous system. The ultrasound equipment uses color Doppler technology to obtain image information of venous blood flow in real time, and continuously collects cross-sectional images and longitudinal images to ensure that the complete morphology and blood flow dynamic data of the vein are obtained. The acquisition of the image sequence is divided into multiple slices to ensure the accurate recording of each layer of the venous structure during the scanning process, and finally obtain a clear lower extremity venous ultrasound image sequence, providing high-quality input data for subsequent analysis.

[0045] Step S12: acquiring standard anatomical structure data based on human lower limb veins; constructing a standardized knowledge base of vein direction, position and morphological characteristics according to the standard anatomical structure data;

[0046] When acquiring standard anatomical data, the embodiment of the present invention uses a human anatomy database (such as VisibleBody or Zygote Body) as a reference to obtain standard anatomical structure data of lower extremity veins. These data include information such as the spatial position, direction, branch points, intersection points, and diameter changes of the major veins. Based on the standard anatomical structure data and combined with anatomical principles, a standardized knowledge base of the direction, position, and morphological characteristics of the veins is constructed. This knowledge base stores the anatomical details of each major blood vessel in the venous system, such as the path of the femoral vein from the thigh root to the knee, the tortuous shape of the popliteal vein, and the cross-direction of the small saphenous vein, providing a basis for subsequent registration and vascular identification.

[0047] Step S13: registering the standardized knowledge base with the lower extremity venous ultrasound image sequence, and correcting the angle deviation or size scaling error in the scanning process in real time, thereby obtaining a venous anatomy priori knowledge base, wherein the venous anatomy priori knowledge base includes the spatial distribution and anatomical hierarchical relationship of each major vein;

[0048] The embodiment of the present invention uses a lower extremity venous ultrasound image sequence and venous anatomical data in a standardized knowledge base for registration processing. The registration method adopts a feature point-based registration algorithm, such as using manually calibrated main blood vessel position points, combined with a deep learning image registration algorithm (for example, a network structure based on U-Net) to automatically adjust the matching degree of the ultrasound image and the anatomical data. At the same time, in order to deal with the angle deviation or size scaling error that may occur during the scanning process, real-time dynamic correction technology is used, such as adjusting the scale and orientation of the scanning data through the displacement tracking system of the ultrasound probe to ensure accurate pairing of the image and the anatomical data. Ultimately, the obtained venous anatomy prior knowledge base contains the spatial distribution and anatomical hierarchical relationship of each major vein, which serves as the core reference for subsequent vascular identification and analysis.

[0049] Step S14: Calculate the probability value of each pixel belonging to the main blood vessel area according to the venous anatomy prior knowledge base and the lower extremity venous ultrasound image sequence, thereby generating a vascular space probability distribution map;

[0050] The embodiment of the present invention uses a probability graph model to calculate the probability value of each image pixel belonging to the main vascular area based on the constructed venous anatomy prior knowledge base and the lower limb venous ultrasound image sequence. The specific method includes using a convolutional neural network (CNN) to perform pixel-level classification on the ultrasound image to obtain the probability of each pixel belonging to the venous area. By calculating the contrast, texture features and hemodynamic information (such as the Doppler frequency shift of blood echoes) between the vascular area and the non-vascular area, the model output is further optimized to generate a high-precision vascular space probability distribution map. The map reflects the spatial distribution of blood vessels in ultrasound images, which helps to accurately locate and analyze target veins.

[0051] Step S15: hierarchically representing the vascular space probability distribution map into spatial position features, structural features, and association features, and performing standardization on each layer of features to obtain conditional prior information encoding data;

[0052] The embodiment of the present invention hierarchically represents the spatial probability distribution map of blood vessels into spatial position features, structural features, and association features. The spatial position features mainly describe the spatial coordinate distribution of blood vessels in the image, the structural features focus on the morphology and direction of the blood vessels, and the association features analyze the relative position and interweaving relationship between the blood vessels. Each layer of features is standardized, and data standardization methods (such as Z-score standardization) are used to uniformly process data of different scales and units to ensure the dimensional consistency of each feature layer, thereby generating conditional prior information encoding data. This data set can provide high-quality prior information for subsequent model reasoning to ensure accurate identification and analysis of blood vessels.

[0053] Step S16: Fusing the conditional prior information encoding data with the venous anatomy prior knowledge base to construct a vascular ontology knowledge base.

[0054] The embodiment of the present invention fuses the generated conditional prior information encoding data with the venous anatomy prior knowledge base to construct a vascular ontology knowledge base. During the fusion process, the knowledge graph technology is used to associate the conditional prior information encoding data with the known venous anatomy data to form a structured vascular ontology knowledge base. The knowledge base contains the spatial distribution of each major vein, the anatomical hierarchical relationship and the way they are interconnected, and combines the dynamic blood flow characteristics in the ultrasound image data to provide comprehensive knowledge support for the multi-level analysis of blood vessels. Through the construction of the ontology knowledge base, more accurate training data and more efficient reasoning capabilities are provided for subsequent deep learning algorithms, thereby improving the accuracy of vascular structure recognition and the reliability of thrombosis risk assessment.

[0055] As an embodiment of the present invention, refer to Figure 3 As shown, Figure 1 Detailed step flow diagram of step S2 in the embodiment of the present invention, step S2 includes the following steps:

[0056] Step S21: extracting standard morphological features and spatial distribution features of blood vessels according to the blood vessel ontology knowledge base, thereby constructing a feature priori matrix;

[0057] The embodiment of the present invention extracts the standard morphological features and spatial distribution features of blood vessels based on the vascular ontology knowledge base. First, the standard morphological data (such as blood vessel diameter, curvature, direction, etc.) and spatial distribution data (such as relative position of blood vessels, branch point coordinates, etc.) of the veins contained in the vascular ontology knowledge base are analyzed, and key parameters reflecting the anatomical characteristics of blood vessels are extracted therefrom through data mining and pattern recognition technology. These parameters include the standard length, number of branches, branch angles, etc. of each blood vessel. Then, the extracted features are uniformly processed using data normalization methods (such as minimum-maximum normalization) to construct a feature prior matrix, which contains the standard morphological features and spatial distribution features of all blood vessels, providing normalized data for subsequent model input.

[0058] Step S22: embedding the feature prior matrix into the conditional control module of the preset diffusion model, thereby obtaining basic diffusion model framework data;

[0059] The embodiment of the present invention embeds the extracted feature prior matrix into the conditional control module of the preset diffusion model. First, a conditional control module based on a graph neural network (GNN) is constructed, which can receive the vascular morphology features and spatial distribution features in the feature prior matrix and use these features as input for further processing. Through the conditional control module, the feature prior matrix can be used as the initialization information of the diffusion model, thereby providing the model with initial vascular morphology and distribution information. This step realizes the close integration of the model input data and the vascular anatomical knowledge through the embedding process, and generates the basic diffusion model framework data containing the vascular morphology prior knowledge.

[0060] Step S23: establishing a noise prediction network of the first-level diffusion model based on the basic diffusion model framework data, and using a progressive noise reduction strategy to process the ultrasound image sequence in the vascular ontology knowledge base, thereby obtaining the main vascular contour positioning data;

[0061] The embodiment of the present invention establishes a noise prediction network of the first-level diffusion model based on the basic diffusion model framework data. The goal of this network is to accurately locate the main blood vessel contours by identifying and removing noise in the image. First, a noise prediction network is constructed, and a deep convolutional neural network (CNN) combined with residual learning is used to predict the noise area in the ultrasound image, and the quality of the vascular image is enhanced by gradually eliminating the noise signal. In order to further improve the image quality, a progressive noise reduction strategy is adopted. First, the noise of the low-resolution image is predicted, and then the image resolution is improved layer by layer and the noise is removed, and finally clear main blood vessel contour positioning data is obtained. These positioning data will be used as the benchmark for blood vessel identification and analysis in the next steps.

[0062] Step S24: Based on the main blood vessel contour positioning data, a region of interest is set around the main blood vessel contour, and a local feature enhancement network of the second-level diffusion model is constructed based on the region of interest, so as to obtain blood vessel branch structure recognition data, wherein the range of the region of interest is 3 times the diameter of the main blood vessel;

[0063] Based on the positioning data of the main blood vessel contour, the embodiment of the present invention sets a region of interest (ROI) around the main blood vessel contour. The range of the region of interest is set to three times the diameter of the main blood vessel to ensure that the blood vessel branches and surrounding structures can be covered. Then, based on these regions of interest, a local feature enhancement network of the second-level diffusion model is constructed. The network focuses on enhancing the features of the blood vessel branch structure within the region of interest, and enhances the detection capabilities of branch points and interweaving points through local feature extraction methods (such as local response normalization). The U-Net network structure is adopted to combine the feature information of different layers through jump connections, further refine the recognition process of blood vessel branches, and finally obtain high-precision blood vessel branch structure recognition data.

[0064] Step S25: establishing a morphological feature extraction network of the third-level diffusion model according to the vascular branch structure recognition data, calculating the vascular cross-sectional features and wall thickness variation features, thereby obtaining vascular morphological feature data;

[0065] The embodiment of the present invention establishes a morphological feature extraction network of the third-level diffusion model based on the vascular branch structure recognition data. The goal of this network is to extract the cross-sectional features and wall thickness variation features of the blood vessels to further analyze the morphological characteristics of the blood vessels. First, a two-dimensional convolutional neural network (CNN) is used to perform cross-sectional analysis on the ultrasound image of the blood vessels to identify the cross-sectional shape, wall thickness and its changes of the blood vessels. Then, the health status of the blood vessels is evaluated by analyzing the changes in wall thickness of different blood vessel segments. The network automatically extracts the morphological feature data of the blood vessels through a deep feature learning method to ensure that the cross-sectional and wall thickness information of the blood vessels can be stably obtained under various vascular morphologies.

[0066] Step S26: multi-level feature fusion is performed on the main blood vessel contour positioning data, the blood vessel branch structure recognition data and the blood vessel morphology feature data, and a feature association mechanism between levels is constructed, thereby obtaining a multi-level diffusion probability model.

[0067] The embodiment of the present invention performs multi-level feature fusion on the main blood vessel contour positioning data, blood vessel branch structure identification data and blood vessel morphological feature data. First, the above three types of data are combined through a multi-level feature fusion network, and feature-level fusion methods (such as weighted average or feature splicing) are used to merge data at different levels into a high-dimensional feature vector. Then, a feature association mechanism between levels is constructed, and the graph convolutional network (GCN) is used to establish the correlation between hierarchical features, further strengthening the connection between blood vessel contours, branches and morphological features. Through this multi-level feature fusion, the generated multi-level diffusion probability model can comprehensively consider the global and local characteristics of blood vessels, and improve the accuracy of blood vessel identification and blood flow dynamic analysis.

[0068] Preferably, step S23 includes the following steps:

[0069] Step S231: constructing a noise prediction network of a U-Net structure based on the basic diffusion model framework data, wherein the U-Net includes an encoder and a decoder, the encoder is used for feature extraction, and the decoder is used for blood vessel contour reconstruction;

[0070] Step S232: Initializing the training parameters of the noise prediction network to obtain network training parameters, wherein the network training parameters specifically include an initial noise ratio of 0.5, a noise reduction step of 0.1, and a feature retention threshold of 0.8;

[0071] Step S233: performing initial sampling of the ultrasound image sequence in the vascular ontology knowledge base according to the network training parameters, and performing a first round of noise prediction, thereby obtaining initial vascular contour prediction data;

[0072] Step S234: iteratively processing the initial blood vessel contour prediction data based on a progressive noise reduction strategy to obtain optimized blood vessel contour data, wherein the iterative processing specifically includes gradually reducing the noise ratio in each iteration and retaining blood vessel contour points with a confidence level greater than a preset feature retention threshold;

[0073] Step S235: When the noise ratio drops to 0.1, the optimized blood vessel contour data is subjected to isolated point removal, contour line smoothing, and contour continuity optimization, thereby obtaining the main blood vessel contour positioning data.

[0074] The embodiment of the present invention constructs a noise prediction network of U-Net structure based on the basic diffusion model framework data. First, the basic diffusion model framework data is used as input to design and construct a noise prediction network of U-Net structure. U-Net includes two main parts: an encoder and a decoder. The encoder is responsible for extracting high-dimensional features from the input image, compressing the spatial information of the image through a series of convolution operations and extracting the key features of the blood vessels; the decoder gradually maps the compressed features back to the original image space, and restores the fine structure of the blood vessel contour through upsampling operations. In order to reduce noise interference, a noise learning module is added to the encoder, which can identify and remove unnecessary noise in the image and optimize the performance of the blood vessel contour. Through this structure, the U-Net network can accurately extract the blood vessel contour and reconstruct it, providing high-quality blood vessel image output. The training parameters of the noise prediction network are initialized. When initializing the network training parameters, the initial noise ratio is first set to 0.5, which means that at the beginning of training, the input image has 50% noise influence. The noise reduction step is set to 0.1, which means that after each iteration, the noise ratio will gradually decrease by 0.1, thereby gradually reducing noise interference and improving the accuracy of the model. The feature retention threshold is 0.8, which is used to control the network to retain the most significant feature parts of the image during the denoising process. Specifically, the network will retain those vascular contour points that are higher than the threshold according to the confidence value of the feature, ensuring that these feature points are optimized during the denoising process, thereby improving the accuracy and stability of the vascular contour. The ultrasound image sequence in the vascular ontology knowledge base is initially sampled according to the network training parameters, and the first round of noise prediction is performed. First, the ultrasound image sequence in the vascular ontology knowledge base is used as input data, and the initial sampling is performed in combination with the set network training parameters. Each sampling extracts the preliminary features of the vascular area from the ultrasound image, and uses the U-Net network to perform the first round of noise prediction, and outputs preliminary vascular contour prediction data. At this stage, the noise ratio is high, and the prediction results of the network will contain a certain degree of noise, but the approximate vascular contour can still be identified. The preliminary vascular contour data provides basic information for subsequent optimization. The initial vascular contour prediction data is iteratively processed based on the progressive denoising strategy. The progressive denoising strategy is adopted to optimize the accuracy of the vascular contour by gradually reducing the noise ratio (for example, reducing it by 0.1 each iteration). In each iteration, the network will update the initial vascular contour prediction data and filter out vascular contour points with high confidence based on the feature retention threshold (0.8). These points will be retained and further optimized in subsequent iterations. The core purpose of iterative processing is to gradually improve the accuracy of vascular contours through multiple rounds of denoising and feature enhancement, so that the final output vascular contour is clearer and free of noise interference. When the noise ratio drops to 0.1, the optimized vascular contour data is isolated and removed, the contour lines are smoothed, and the contour continuity is optimized.In this step, when the noise ratio drops to 0.1, it means that most of the noise has been removed and the vascular contour is clear. At this point, image processing technology is applied to further process the optimized vascular contour data. First, the isolated point removal algorithm identifies and deletes small noise points that cannot be connected to the surrounding vascular structure. Then, the vascular contour lines are smoothed by edge smoothing algorithms (such as Gaussian smoothing or median filtering) to reduce sharp angles and unnatural jagged shapes. Finally, the contour connection algorithm is used to repair the broken vascular contours to ensure the continuity of the vascular contour lines, and finally obtain high-quality main vascular contour positioning data, which provides an accurate basis for subsequent vascular analysis and blood flow feature extraction.

[0075] The U-Net in the present invention is a deep learning network structure, which is widely used in medical image processing, especially in image segmentation tasks. In this step, the encoder part of the U-Net is used to extract the features of the blood vessels, while the decoder part is used to reconstruct the contour of the blood vessels. This structure can retain detail information when extracting image features, thereby restoring the precise shape of the contour of the blood vessels. By applying the U-Net structure, the contour of the blood vessels is reconstructed more accurately, especially in ultrasound images with noise or loss of details. By initializing training parameters (such as noise ratio, noise reduction step size and feature retention threshold), reasonable starting conditions can be set for the training process. The initial noise ratio is set to 0.5, which helps the model to start training from image data with more noise, and gradually learn how to identify vascular structures and suppress noise. The noise reduction step size is 0.1, which ensures that noise interference is gradually reduced during the training process, making the model learning smoother and more stable. The feature retention threshold is 0.8, which sets that in each round of training, only features with a confidence greater than the threshold will be retained, thereby ensuring the accurate positioning of the vascular contour. The ultrasound image sequence is initially sampled according to the network training parameters, and the first round of noise prediction is performed. The key to this step is to make a preliminary prediction of the vascular image through the model and obtain the initial vascular contour data. This provides a basis for subsequent optimization processing, can identify the approximate location of the vascular contour in the preliminary stage, and lay the foundation for subsequent noise processing and contour refinement. The initial vascular contour data is iteratively processed using a progressive denoising strategy. Each iteration gradually reduces the noise ratio and retains the vascular contour points with a confidence level greater than a specific threshold. In this way, the network can gradually recover more accurate vascular contours from noisy images, avoiding the loss of valuable structural information that may be lost in a single process. This strategy can effectively reduce noise interference, improve the quality of vascular contours, and ensure that the details and morphology of the blood vessels are retained during the denoising process. When the noise ratio drops to 0.1, the optimized vascular contour data is subjected to isolated point removal, contour line smoothing, and continuity optimization. Isolated point removal can effectively remove isolated points that may be generated during noise prediction. These isolated points may be the result of noise or misidentification, which is not conducive to accurately describing the vascular structure. Through smoothing, unnecessary jagged curves can be removed, making the vascular contour smoother and more natural. Optimizing the continuity of the contour ensures that the vascular boundary will not be broken and maintains the integrity and consistency of the vascular structure. This step ultimately helps to obtain accurate vascular contour positioning data, providing high-quality input data for subsequent vascular analysis and risk assessment. Through the design and implementation of these steps, the accuracy and stability of vascular contours in lower extremity venous ultrasound images can be significantly improved. The noise prediction network with a U-Net structure can effectively remove noise interference while retaining vascular details when processing complex ultrasound images.The progressive noise reduction strategy ensures that the quality of the vascular contour is gradually optimized at each step, and finally the coherence and accuracy of the vascular contour are further improved through isolated point removal and smoothing. These effects help improve the accuracy of venous thrombosis detection and provide more reliable image data support for risk assessment.

[0076] Preferably, step S24 comprises the following steps:

[0077] Step S241: determining the range of the region of interest according to the main blood vessel contour positioning data, setting the radius of the region of interest to three times the diameter of the main blood vessel, and smoothing the region boundary to obtain the mask data of the region of interest;

[0078] Step S242: constructing a local feature enhancement network based on the mask data of the region of interest, thereby obtaining a feature enhancement network model, wherein the local feature enhancement network adopts a residual connection structure, including a feature extraction module, an attention module and a feature fusion module;

[0079] Step S243: performing local contrast enhancement and detail enhancement processing on the image in the region of interest, thereby obtaining enhanced image data;

[0080] Step S244: using the feature enhancement network model to perform vascular branch structure recognition on the enhanced image data, thereby obtaining initial branch recognition data, wherein the vascular branch structure recognition specifically sets the minimum diameter threshold of branch recognition to 1 / 5 of the main blood vessel diameter, and recognizes branch structures of different scales;

[0081] Step S245: performing post-processing optimization on the initial branch identification data and calculating the spatial interleaving degree between branches, thereby obtaining blood vessel branch structure identification data, wherein the post-processing optimization includes removing false branches, repairing branch breakpoints, and optimizing branch connection relationships.

[0082] The embodiment of the present invention determines the range of the region of interest according to the main blood vessel contour positioning data. First, by analyzing the main blood vessel contour positioning data, the core area of ​​the blood vessel and its diameter are determined. Then, the radius of the region of interest is set to 3 times the diameter of the main blood vessel according to the diameter of the main blood vessel. This area will include small blood vessel branches around the blood vessel and some possible abnormal areas, so expanding to 3 times the diameter can ensure that more blood vessel areas are covered. Next, the boundary of the area is smoothed to eliminate the jagged edges caused by image noise or scanning errors, and the edges are smoothed using Gaussian smoothing or Bézier smoothing algorithms to ensure that the contour of the region of interest is smooth and accurate. Through this processing method, the mask data of the region of interest is finally obtained, and this mask data can be effectively used for subsequent local feature extraction and enhancement processing. A local feature enhancement network is constructed based on the mask data of the region of interest. The mask data of the region of interest obtained in step S241 is used to construct a local feature enhancement network. The network adopts a residual connection structure to effectively avoid the gradient vanishing problem in the deep network and promote the flow of information in the network. The network structure includes three main modules: a feature extraction module, an attention module, and a feature fusion module. The feature extraction module extracts key vascular features from the image of the region of interest through the convolution layer. The attention module performs weighted processing on the features according to the importance information in the image to highlight the significant feature parts in the image. The feature fusion module fuses the extracted features of different levels to generate the final enhanced feature map. This local feature enhancement network can refine the feature expression of the vascular region and provide high-quality input data for subsequent branch structure recognition. The image in the region of interest is subjected to local contrast enhancement and detail enhancement processing. Using the local feature enhancement network constructed in step S242, the image in the region of interest is first subjected to local contrast enhancement. This process makes the difference between the blood vessels and the surrounding tissues more obvious by adjusting the brightness and contrast of the image, thereby facilitating further discrimination of the blood vessels. The detail enhancement processing highlights the tiny features and branch structures of the blood vessels by improving the clarity and edge sharpness of the image. The specific method can use local adaptive histogram equalization (CLAHE) or Laplace pyramid enhancement algorithm to improve the details of the local area, so that the vascular details in the ultrasound image are more obvious, and the enhanced image data is easier to be used by the subsequent network for feature extraction and branch structure recognition. The enhanced image data is subjected to vascular branch structure recognition using the feature enhancement network model. The enhanced image data obtained in step S243 is input into the feature enhancement network model for vascular branch structure recognition. During the recognition process, the vascular branch structure is first detected, and the presence of branches is determined by analyzing the vascular network pattern in the image. In order to ensure the accuracy of the detection, the minimum diameter threshold of branch recognition is set to 1 / 5 of the main vessel diameter, so that the branch structure can be at least 1 / 5 of the main vessel size, avoiding the error of identifying too small details.Branch recognition considers vascular branches of different scales at the same time, processes images by multi-scale convolution kernels or pyramid structures, captures branch structures at different scales, and ensures that both large and small branches can be accurately identified. In this way, the initial branch recognition data obtained can accurately reflect the existence of vascular branches and their spatial distribution. The initial branch recognition data is post-processed and optimized, and the spatial interweaving degree between branches is calculated. After the initial identification of vascular branches, post-processing optimization is performed to improve the accuracy of recognition and reduce false positive branches. First, the wrong branch recognition results are cleaned up by removing false branches (such as artifacts or noise interference), which can be achieved by morphological filtering or connected domain analysis. Then, the branch breakpoints are repaired, and the interrupted parts of the blood vessels are completed using curve interpolation or linear interpolation techniques. The branch connection relationship is optimized to ensure that the connection of vascular branches is more natural and reasonable, and possible connection errors are repaired by analyzing the connectivity and spatial relationship between blood vessels. Finally, the spatial interweaving degree between branches is calculated, that is, the degree of interweaving between branch structures is evaluated by spatial distance and connection relationship, which helps to further optimize the overall accuracy of branch structure. Through these optimization steps, the final vascular branch structure recognition data accurately reflects the true morphology and distribution of blood vessels.

[0083] In the present invention, the scope of the region of interest is determined according to the main blood vessel contour positioning data and is set to 3 times the diameter of the main blood vessel, which ensures that the region of interest can cover the blood vessel branch structure and its surrounding area. By smoothing the region boundary, the irregular boundary phenomenon caused by scanning errors or noise in the image can be reduced, the accurate positioning of the blood vessel region can be ensured, and more accurate regional mask data can be provided for subsequent processing. The mask data of the region of interest finally generated provides an accurate image region for the subsequent local feature enhancement and branch structure recognition. This step constructs a local feature enhancement network based on the mask data of the region of interest. The network structure adopts residual connection to effectively transfer gradients and prevent information loss during training. The local feature enhancement network consists of a feature extraction module, an attention module and a feature fusion module, which can further mine detail features on the basis of enhanced images. The feature extraction module extracts low-level features of the image, the attention module adjusts the focus of the network according to the importance of different regions, and the feature fusion module combines multi-level features to further enhance the local detail information of the image. Through this network, the detail features of the blood vessels are optimized, and the branch structures of the blood vessels can be better identified in subsequent steps. In this step, the image in the region of interest is subjected to local contrast and detail enhancement processing. This processing can significantly improve the visibility of vascular structures in images, especially in areas with low contrast or unclear details. The enhanced image can more clearly present the morphology of blood vessels and their branches, reduce the impact of background noise on image analysis, and thus provide high-quality image input data for subsequent vascular branch structure recognition. Based on the enhanced image data, the feature enhancement network model is used to identify the vascular branch structure. By setting the minimum diameter threshold for branch recognition to 1 / 5 of the main vessel diameter, important branches related to the main vascular trunk can be effectively screened out to avoid irrelevant small vessels being mistakenly identified as branches. Branch structures of different scales are considered during the recognition process, so that the network can effectively identify vascular regions of different sizes and accurately distinguish the complex branch structure of blood vessels. In this way, the model can achieve higher-precision branch structure recognition, which is helpful for further analysis of the morphological characteristics of the vascular system. The initially identified vascular branch structure is optimized through post-processing, removing false branches, repairing branch breakpoints, and optimizing branch connection relationships. This process can eliminate false recognition caused by noise, scanning errors or algorithm deficiencies, and ensure the integrity and coherence of the branch structure. By optimizing the branch connection relationship, the model can restore the natural morphology of vascular branches and avoid branch discontinuity. Calculating the spatial interweaving between branches helps to quantify the relative positional relationship between branches and provide additional information for further vascular analysis, especially when judging the risk of vascular obstruction or thrombosis. The design and implementation of these steps significantly improve the recognition accuracy and stability of vascular branch structures.Through the application of the local feature enhancement network, not only the vascular details in the image are enhanced, but also the recognition process of the vascular branch structure is optimized. Subsequent optimization processing further improves the accuracy and robustness of the results and reduces the interference of misidentification and image noise. The optimized vascular branch structure recognition data provides accurate basic data for further vascular fluid feature analysis and deep vein thrombosis risk assessment.

[0084] Preferably, step S25 comprises the following steps:

[0085] Step S251: constructing a morphological feature extraction network according to the blood vessel branch structure recognition data, thereby obtaining a feature extraction network model, wherein the morphological feature extraction network adopts a multi-branch parallel structure, including a cross-sectional feature analysis branch and a wall thickness feature analysis branch;

[0086] Step S252: using the cross-sectional feature analysis branch in the feature extraction network model to calculate the morphological features of the blood vessel cross section, thereby obtaining cross-sectional feature data, wherein the morphological features of the blood vessel cross section include the lumen area change rate, the cross-sectional roundness index, and the cross-sectional eccentricity;

[0087] Step S253: extracting the vascular wall thickness variation characteristics based on the wall thickness feature analysis branch in the feature extraction network model, thereby obtaining wall thickness feature data;

[0088] Step S254: Fusing the cross-sectional feature data and the wall thickness feature data to obtain vascular morphology feature data.

[0089] The embodiment of the present invention constructs a morphological feature extraction network according to the vascular branch structure recognition data. The vascular branch structure recognition data obtained in step S245 is used to construct a morphological feature extraction network. The network adopts a multi-branch parallel structure, which includes a cross-sectional feature analysis branch and a wall thickness feature analysis branch. The cross-sectional feature analysis branch focuses on analyzing the cross-sectional morphological features of the blood vessel, and the wall thickness feature analysis branch is used to extract the variation characteristics of the vascular wall thickness. Each branch processes different features through different convolutional neural network (CNN) structures to ensure that the network can simultaneously extract the geometric shape and biomechanical properties of the blood vessel. During design, the cross-sectional feature analysis branch can use multi-layer convolution and pooling operations to enhance the extraction of geometric information of the cross-sectional area of ​​the blood vessel; while the wall thickness feature analysis branch processes the variation of the vascular wall thickness through a multi-scale convolution layer to help capture the slight variation of the wall thickness. Finally, the output data of the two branches will be integrated to generate the overall morphological features of the blood vessel through a fully connected layer. The cross-sectional feature analysis branch in the feature extraction network model is used to calculate the morphological features of the cross-sectional area of ​​the blood vessel. The cross-sectional feature analysis branch in the morphological feature extraction network constructed by step S251 first analyzes the cross-sectional area of ​​the blood vessel. The calculation of cross-sectional morphological features includes lumen area change rate, cross-sectional roundness index and cross-sectional eccentricity. Lumen area change rate evaluates the channel change of blood vessels by calculating the cross-sectional area change of blood vessels at different positions, using geometric shape extraction algorithms such as Laplace operator based on contour analysis. Cross-sectional roundness index evaluates the circularity of blood vessel morphology by calculating the ratio of the roundness of the blood vessel cross section to the perfect circle. The formula can be , where A is the cross-sectional area and P is the perimeter. The degree of cross-sectional eccentricity reflects the degree of deviation of the blood vessel from a circular shape, which is usually obtained by measuring the major axis direction and the minor axis direction of the cross-sectional area of ​​the blood vessel. Through these calculations, the morphological feature data of the cross-sectional area of ​​the blood vessel is obtained, which provides detailed parameters for further analysis of the morphology of the blood vessel. The variation characteristics of the vascular wall thickness are extracted based on the wall thickness feature analysis branch in the feature extraction network model. The variation characteristics of the vascular wall thickness are extracted using the wall thickness feature analysis branch in step S251. The variation characteristics of the vascular wall thickness mainly reflect the elasticity and health of the blood vessel. Usually, the cross-sections at different positions of the blood vessel are analyzed and the vascular wall thickness is accurately measured using a pixel-level segmentation algorithm. In practical applications, image segmentation techniques, such as a segmentation network based on U-Net, can be used to identify the boundary between the vascular wall and the lumen. Then, by calculating the change in wall thickness, the increase or decrease trend of the vascular wall thickness can be obtained, thereby inferring the health of the blood vessel and whether there are lesions such as atherosclerosis. The key parameters used in this process include the maximum value, minimum value and change rate of the vascular wall thickness, which can provide necessary features for subsequent vascular health assessment. The cross-sectional feature data and the wall thickness feature data are feature fused. The cross-sectional feature data obtained in step S252 and the wall thickness feature data obtained in step S253 are feature fused to obtain complete vascular morphology feature data. The feature fusion method can be implemented by weighted averaging or a fusion layer based on a neural network. In the weighted averaging method, different weights are assigned for fusion according to the relative importance of the cross-sectional features and the wall thickness features; in the neural network fusion layer, the two feature sets are merged through a fully connected layer to generate a unified vascular morphology feature vector. The fused data contains comprehensive information about the geometric shape, elastic properties and health status of the blood vessels, which can more accurately reflect the overall condition of the blood vessels and help subsequent disease prediction and risk assessment. In addition, the feature fusion process can also introduce dimensionality reduction techniques, such as principal component analysis (PCA), to reduce redundant information and enhance the representation ability of the data, laying the foundation for further vascular modeling and analysis.

[0090] The present invention can analyze the morphological characteristics of blood vessels from two important dimensions at the same time by constructing a morphological feature extraction network with a multi-branch parallel structure. The network includes a cross-sectional feature analysis branch and a wall thickness feature analysis branch, which complement each other and focus on the geometric morphology and wall thickness characteristics of the blood vessels respectively. The multi-branch structure can make full use of the specialized analysis capabilities of each branch to ensure that when extracting vascular morphological features, factors such as vascular morphology and wall thickness changes can be comprehensively considered, thereby improving the comprehensiveness and accuracy of vascular health analysis. Through this parallel structure, vascular features of multiple dimensions can be extracted simultaneously, avoiding the possible feature loss or analysis limitations of a single branch network. Through the cross-sectional feature analysis branch, this step calculates the cross-sectional morphological characteristics of the blood vessel, such as the lumen area change rate, the cross-sectional roundness index, and the degree of cross-sectional eccentricity. The lumen area change rate can reflect the change in the space inside the blood vessel lumen, and then evaluate the expansion or contraction of the blood vessel; the cross-sectional roundness index is used to measure the regularity of the blood vessel cross section. A low roundness may indicate abnormal blood vessel morphology (for example, blood vessel deformation or stenosis); the cross-sectional eccentricity is used to evaluate the asymmetry of the blood vessel cross section. A high degree of eccentricity may indicate problems such as local damage to the inner wall of the blood vessel or arteriosclerosis. By calculating these vascular cross-sectional morphological features, the state of the blood vessel can be understood more accurately, providing basic data for subsequent vascular health assessment. The wall thickness feature analysis branch extracts the characteristics of vascular wall thickness changes. This analysis can help identify the thickening or thinning of the vascular wall, which is usually directly related to the health of the blood vessel. For example, thickening of the vascular wall may be an early sign of diseases such as atherosclerosis and arteriosclerosis, while thinning of the wall thickness may be related to loss of vascular elasticity and degenerative changes in the disease. By extracting wall thickness feature data, the health of the vascular wall can be evaluated and support can be provided for early diagnosis. By fusing the cross-sectional feature data with the wall thickness feature data, the vascular morphological feature data is obtained. The effect of this fusion process is that it integrates information from two aspects, namely, vascular morphology (such as lumen area, roundness, and eccentricity) and wall thickness (such as wall thickness changes), which can fully reflect the health status and potential risks of blood vessels. By fusing data from these two dimensions, it is possible to more accurately analyze the structural changes of blood vessels and provide a more comprehensive assessment of vascular health. Clinically, the morphological characteristics of blood vessels and changes in wall thickness are important indicators for predicting vascular diseases (such as thrombosis, stenosis, sclerosis, etc.). Therefore, the fused feature data can help to more accurately determine whether there are lesions or abnormalities in the blood vessels, thereby providing a basis for individualized treatment and intervention. These steps provide rich data support for the morphological analysis of blood vessels by extracting and fusing the cross-sectional and wall thickness features of the blood vessels. The morphological feature extraction network with a multi-branch parallel structure ensures the accurate analysis of cross-sectional and wall thickness features, providing a more comprehensive perspective for vascular health assessment. The cross-sectional morphological features help to analyze the geometric changes of blood vessels, while the wall thickness features help to assess the health status of the blood vessel wall.By integrating these data, more accurate vascular morphological feature data can be obtained, thus providing strong support for the diagnosis, risk prediction and formulation of treatment plans for vascular diseases.

[0091] Preferably, step S4 comprises the following steps:

[0092] Step S41: analyzing the change rate of the vascular lumen cross-sectional area according to the vascular fluid characteristic data, thereby obtaining vascular stenosis index data, wherein the vascular structural parameters include the change rate of the vascular lumen cross-sectional area, the distribution of vascular wall thickness, and the vascular elasticity coefficient;

[0093] Step S42: extracting hemodynamic parameters according to the vascular fluid characteristic data, thereby obtaining blood flow resistance index data, wherein the hemodynamic parameters include blood flow velocity field distribution, blood flow direction change characteristics, and blood flow turbulence index;

[0094] Step S43: constructing a feature space of a deep vein thrombosis risk assessment model, mapping the vascular stenosis index data and the blood flow resistance index data to the feature space, and establishing a feature weight matrix, thereby obtaining risk feature representation data;

[0095] Step S44: training a deep vein thrombosis risk assessment model based on the risk feature representation data, thereby obtaining a risk prediction model;

[0096] Step S45: Use the risk prediction model to divide the risk level of the lower limb venous ultrasound image sequence to obtain thrombosis risk assessment data, where the risk level division specifically includes when the stenosis index is greater than 0.5 and the resistance index is greater than 0.7, it is judged as high risk; when the stenosis index is between 0.3 and 0.5 or the resistance index is between 0.5 and 0.7, it is judged as medium risk; when the stenosis index is less than 0.3 and the resistance index is less than 0.5, it is judged as low risk.

[0097] The embodiment of the present invention performs a blood vessel lumen cross-sectional area change rate analysis based on blood vessel fluid characteristic data. First, the blood vessel lumen cross-sectional area change rate is calculated using blood vessel fluid characteristic data, including blood flow velocity, pressure field distribution, and blood vessel geometry information. The specific operation includes evaluating the deformation and contraction of blood vessels at different positions by comparing the cross-sectional areas at different blood vessel sections and combining hemodynamic models (such as the Navier-Stokes equation). The blood vessel lumen cross-sectional area change rate reflects the degree of vascular stenosis. In the application, this data can be obtained by ultrasound or CT imaging of a two-dimensional cross-sectional image of the blood vessel, and the contour of the blood vessel is extracted by an image processing method to calculate the change in cross-sectional area. Through these data, combined with the distribution of vascular wall thickness and vascular elastic coefficient (for example, vascular wall hardness data obtained using ultrasound elastic imaging technology), the stenosis index of the blood vessel can be further inferred. In a specific application scenario, the calculation method of the vascular stenosis index can use a threshold setting to mark the severity of vascular stenosis, such as greater than 0.5 for severe stenosis, 0.3 to 0.5 for moderate stenosis, and less than 0.3 for mild stenosis. Extract hemodynamic parameters based on vascular fluid characteristic data. According to the characteristic data of vascular fluid, such as blood flow velocity field, blood flow direction and turbulence information, key hemodynamic parameters are extracted to obtain the blood flow resistance index. The blood flow velocity field distribution can be measured by Doppler ultrasound or CT blood flow imaging technology, and the blood flow direction change characteristics can be identified by analyzing the change trend of the blood flow vector field to identify the directional change of the blood flow. The blood flow turbulence index is usually evaluated by calculating the irregularity and pulsation of the blood flow velocity. The specific operation includes calculating the turbulence intensity using fluid dynamics simulation or a turbulence model based on blood flow images. Combined with the degree of stenosis of the vascular lumen, the blood flow velocity and turbulence information can help evaluate the resistance of the blood flow, and then calculate the blood flow resistance index. In specific applications, blood flow data can be collected by ultrasound imaging, and accurate velocity fields can be obtained using high-speed photography or imaging algorithms, and data processing can be performed to obtain hemodynamic parameters to calculate the blood flow resistance index. Construct a feature space for a deep vein thrombosis risk assessment model. First, the vascular stenosis index data obtained in step S41 and the blood flow resistance index data obtained in step S42 are mapped to the feature space. In order to effectively fuse these different types of data, machine learning methods, such as support vector machines (SVM) or random forest algorithms, can be used to construct a feature space model. The specific operation includes constructing a feature weight matrix, which is used to measure the relative contribution of vascular stenosis and blood flow resistance to thrombosis risk. The construction of the feature space can be achieved by standardizing parameters such as vascular stenosis index and blood flow resistance index, and reducing feature redundancy through data dimensionality reduction methods (such as principal component analysis PCA), and finally obtaining a data matrix containing key risk features. The deep vein thrombosis risk assessment model is trained based on the risk feature representation data. The deep vein thrombosis risk assessment model is trained using the risk feature representation data obtained in step S43.The training process adopts the supervised learning method in machine learning, and uses the labeled data set for model training. The training algorithm can adopt random forest, support vector machine (SVM), neural network and other methods. Through training, the model will learn the relationship between vascular stenosis, blood flow resistance and other hemodynamic characteristics and the risk of deep vein thrombosis, and provide accurate model parameters for risk prediction. In the application scenario, the model can be trained by historical case data to ensure that the model can accurately assess the risk of thrombosis based on the individual's blood vessel and blood flow data. The risk prediction model is used to classify the risk level of the lower limb venous ultrasound image sequence. The risk prediction model obtained in step S44 is applied to the lower limb venous ultrasound image sequence data to classify the risk level of thrombosis. The specific operation includes inputting the vascular stenosis index and blood flow resistance index as input features into the trained risk prediction model. The model calculates a comprehensive risk score based on these data and divides it according to the preset risk level standard. Specifically, when the stenosis index is greater than 0.5 and the resistance index is greater than 0.7, it is judged as high risk; when the stenosis index is between 0.3 and 0.5 or the resistance index is between 0.5 and 0.7, it is judged as medium risk; when the stenosis index is less than 0.3 and the resistance index is less than 0.5, it is judged as low risk. This process can provide the risk level of thrombosis in real time in clinical applications through automated image processing and model prediction methods, and provide decision support for doctors.

[0098] The present invention can calculate the stenosis index of the blood vessel by analyzing the change rate of the cross-sectional area of ​​the vascular lumen, combining the distribution of the vascular wall thickness and the vascular elasticity coefficient. The vascular stenosis index is a key indicator for evaluating whether the blood vessel is stenotic or blocked, and can help doctors identify early signs of vascular disease. The change rate of the lumen cross-sectional area reflects the change of the vascular channel, the wall thickness distribution is related to the degree of vascular sclerosis, and the vascular elasticity coefficient indicates the elasticity and health of the blood vessel. Combining the information of these three, the degree of stenosis and potential risks of the blood vessel can be comprehensively evaluated. This analysis not only helps to monitor the structural changes of the blood vessels, but also provides a basis for further disease diagnosis and intervention. The blood flow resistance index can be extracted through hemodynamic parameters (such as blood flow velocity field distribution, blood flow direction change characteristics and blood flow turbulence index). The blood flow velocity field distribution can show the speed of blood flow, the blood flow direction change characteristics reflect the directional change of blood flow, and the blood flow turbulence index reveals the degree of blood flow disorder. The blood flow resistance index provides a quantitative standard for the difficulty of blood flow. A higher resistance index may indicate that there are lesions (such as stenosis, sclerosis) in the blood vessels, resulting in obstructed blood flow or increased turbulence. The calculation of this index can better understand the relationship between blood flow and vascular health, which helps to evaluate blood flow changes caused by vascular lesions. The vascular stenosis index and blood flow resistance index are mapped into a feature space, and the risk feature representation data of thrombosis is finally obtained through the establishment of a feature weight matrix. The effect of this process is that vascular stenosis and hemodynamic parameters provide information closely related to thrombosis, and the construction of the feature space can combine these data to form a comprehensive risk assessment framework. The feature weight matrix assigns different weights to each feature according to its importance in risk prediction, thereby optimizing the effect of the prediction model. This provides more accurate and comprehensive input data for subsequent risk prediction. Based on the risk feature representation data, the deep vein thrombosis risk assessment model can be trained to obtain a model with predictive ability by learning the relationship between vascular structure and blood flow characteristics and thrombosis. By combining multiple vascular parameters, this model can more accurately identify which factors increase the risk of thrombosis and provide personalized risk prediction. By training this model, it can better provide support for clinical practice, help doctors formulate reasonable intervention measures, and prevent the occurrence of deep vein thrombosis. The trained risk prediction model is used to classify the risk level of lower extremity venous ultrasound image sequences, and the patient's risk of thrombosis can be determined based on different stenosis indexes and resistance indexes. Through clear grading standards (high risk, medium risk, and low risk), doctors can adopt different treatment plans based on the risk level of thrombosis. High-risk patients can undergo more active interventions, such as drug therapy or surgery, while low-risk patients can avoid unnecessary treatment through regular monitoring. Risk level classification provides an important basis for clinical decision-making and ensures the personalization and accuracy of treatment.These steps start from the structural characteristics of the blood vessels and hemodynamic parameters, and comprehensively consider factors such as vascular stenosis and blood flow obstruction to establish a comprehensive deep vein thrombosis risk assessment framework. By establishing a feature space and feature weight matrix and training a deep learning model, this method can provide accurate predictions of thrombosis risk, helping doctors to make timely diagnoses and take appropriate intervention measures. The division of risk levels helps to implement personalized treatment, optimize the allocation of clinical resources, and reduce the risk of complications for patients.

[0099] Preferably, step S5 comprises the following steps:

[0100] Step S51: establishing a mapping relationship between risk level and warning level based on thrombosis risk assessment data, thereby obtaining warning level determination standard data;

[0101] Step S52: determining the monitoring warning level of the target patient according to the warning level determination standard data, and generating warning information including risk level, warning suggestions and intervention measures, thereby obtaining monitoring warning data;

[0102] Step S53: constructing a three-dimensional blood vessel model, marking abnormal areas, and visualizing blood flow dynamic characteristics based on the blood vessel fluid characteristic data and the thrombosis risk assessment data, thereby obtaining visualized analysis result data;

[0103] Step S54: obtaining standardized monitoring report data according to the monitoring and early warning data and the visual analysis result data;

[0104] Step S55: extracting abnormal features, risk factors and evaluation results in the standardized monitoring report data as structured data, and feeding it back to the vascular ontology knowledge base, thereby obtaining knowledge base update data;

[0105] Step S56: updating data based on the knowledge base to continuously optimize the recognition accuracy of vascular morphology and blood flow characteristics.

[0106] The embodiment of the present invention establishes a mapping relationship between risk level and warning level based on thrombosis risk assessment data. First, according to the thrombosis risk assessment data obtained in step S45, the thrombosis risk prediction value is associated with different warning levels. The specific operation is to define the warning level standard of each level according to different risk levels (high risk, medium risk, low risk). For example, it can be set that high risk corresponds to "red warning", medium risk corresponds to "orange warning", and low risk corresponds to "yellow warning". These standards are adjusted based on clinical experience and historical case data to ensure that the relationship between risk and warning at each level can reflect the actual changes in the condition. For example, for high-risk patients, more stringent monitoring requirements and intervention measures can be set, while the monitoring frequency and intervention intensity of low-risk patients are correspondingly lower. The warning level determination standard data can be generated by an automated algorithm to ensure that it has efficient and real-time application value in different scenarios. The monitoring warning level of the target patient is determined according to the warning level determination standard data. According to the warning level determination standard data obtained in step S51, the patient's monitoring warning level is automatically determined by inputting the thrombosis risk assessment data of the target patient. This process determines the risk level and generates warning information by analyzing the patient's vascular stenosis index and blood flow resistance index, combined with the established warning level mapping relationship. Specifically, when the patient's risk assessment data falls into a certain risk level range, the system will automatically generate warning information for the patient including risk level, warning suggestions and possible intervention measures. For example, for high-risk patients, the system may prompt to strengthen ultrasound monitoring of lower limb veins, regularly check thrombosis, and recommend intervention measures such as anticoagulants. Finally, the generated warning information forms a warning report, which is provided to doctors as monitoring and warning data for decision-making. Based on vascular fluid characteristic data and thrombosis risk assessment data, a three-dimensional vascular model is constructed, abnormal areas are marked, and dynamic characteristics of blood flow are visualized. Combined with the vascular fluid characteristic data and risk assessment data mentioned in step S42 and step S45, a three-dimensional vascular model is constructed. The specific operation includes using medical imaging data (such as CT or MRI images) to perform three-dimensional reconstruction of blood vessels, marking abnormal areas such as vascular stenosis and varicose veins, and combining hemodynamic simulation data to display dynamic characteristics such as velocity field and direction change of blood flow. These data can be processed by 3D visualization software to generate a clear vascular network diagram, helping doctors view the dynamic changes of blood flow on a visualization interface and mark high-risk areas for thrombosis. The visualization analysis result data includes blood flow velocity distribution, turbulence index, changes in vascular wall thickness, etc. This information can provide the relationship between blood flow dynamics and vascular morphology, further assisting clinical diagnosis and treatment decisions. Standardized monitoring report data is obtained based on monitoring and early warning data and visualization analysis result data.Combined with the monitoring and early warning data generated in step S52 and the visual analysis result data obtained in step S53, they are integrated into standardized monitoring report data through data processing and formatting. The specific operation is to output all relevant data in a structured format, including the patient's risk level, vascular morphological characteristics, hemodynamic characteristics, and corresponding early warning suggestions and intervention measures, and form a unified monitoring report template. This template can include basic patient information, test data, risk assessment results, visual images, recommended intervention measures, etc., and the standardized report is automatically generated by the report generation system, so that doctors can quickly obtain complete patient monitoring data. The abnormal features, risk factors and evaluation results in the standardized monitoring report data are extracted as structured data, and fed back to the vascular ontology knowledge base. First, the abnormal features (such as vascular stenosis, abnormal blood flow resistance, etc.), risk factors (such as thrombosis risk assessment value) and evaluation results (such as risk level, early warning level, etc.) in the standardized monitoring report data generated in step S54 are extracted as structured data. The specific operation is to extract key features from the monitoring report through natural language processing (NLP) technology, convert them into structured information, and store them in the vascular ontology knowledge base. These data will be used to update the patient data and clinical feature data in the vascular knowledge base to provide support for subsequent clinical decision-making. Through data extraction and organization, the patient's dynamic monitoring information can be integrated into the knowledge base, which not only improves the availability of data, but also provides a reference for similar cases in the future. Update data based on the knowledge base to continuously optimize the recognition accuracy of vascular morphology and blood flow characteristics. Through the updated data fed back to the knowledge base, the recognition accuracy of vascular morphology and blood flow characteristics is continuously optimized using machine learning and data mining methods. Specific operations include adjusting the vascular morphology recognition algorithm and blood flow dynamic analysis model through the analysis of historical data and new feedback data, so that it can more accurately identify abnormal vascular morphology, unstable blood flow and other characteristics. In the application scenario, the model performance is optimized through the continuous updating of patient data to improve the adaptability to different patient groups and types of vascular abnormalities. Through this process, the recognition of vascular morphology and blood flow characteristics will gradually improve, thereby improving the accuracy of thrombosis risk assessment and providing clinicians with a more reliable basis for diagnosis and treatment.

[0107] The present invention establishes a mapping relationship between risk level and warning level according to thrombosis risk assessment data, and can realize quantitative assessment and dynamic warning of thrombosis risk. By defining the standard data for judging the warning level, it can help doctors quickly identify high-risk patients and take corresponding monitoring and intervention measures according to different risk levels. This mapping relationship ensures that there is a clear definition between different risk levels and warning levels, and optimizes the efficiency of treatment and monitoring. According to the warning level judgment standard, warning information including risk level, warning suggestions and intervention measures is generated, which can provide personalized medical plans for patients in a timely and accurate manner. Doctors can take corresponding intervention measures such as drug treatment, regular examination or emergency treatment based on these warning information. Warning information not only helps to improve the accuracy of treatment, but also avoids delaying the treatment time and improves the cure rate and survival rate of patients. By constructing a three-dimensional vascular model based on vascular fluid characteristic data and thrombosis risk assessment data, the structure and dynamic characteristics of blood flow of the blood vessel can be intuitively displayed. This visual analysis enables doctors to more clearly understand the abnormal areas inside the blood vessel, such as the location of stenosis or thrombosis, hemodynamic changes, etc., thereby providing a basis for treatment decisions. Marking abnormal areas and visualizing blood flow dynamic characteristics help doctors quickly identify high-risk areas and provide guidance for taking targeted treatment measures. Generate standardized monitoring reports based on monitoring and early warning data and visual analysis result data. This report has a unified format, which is convenient for patient follow-up, doctor diagnosis and treatment, and medical record preservation. Standardized reports display risk assessment results, vascular status, blood flow changes, and early warning information in a systematic and structured manner, which helps to ensure the traceability and sustainability of the treatment process. This also provides clear and systematic support for subsequent medical evaluations and adjustments to treatment plans. By extracting abnormal features, risk factors, and assessment results in standardized reports as structured data and feeding them back to the vascular ontology knowledge base, the content of the database can be continuously updated and optimized. This data feedback mechanism can help the system update knowledge about vascular health, fluid characteristics, and thrombosis risks in real time, thereby improving the accuracy of subsequent evaluations and predictions. The update of the knowledge base helps to strengthen the learning ability of machine learning models, improve the accuracy of recognition and prediction, and provide more reliable decision support for clinicians. Based on the updated data of the vascular ontology knowledge base, by continuously optimizing the recognition accuracy of vascular morphology and blood flow characteristics, it can ensure that the system continues to improve its diagnostic and predictive capabilities after long-term operation. With the introduction of new data, the learning and recognition capabilities of the model are enhanced, making vascular health assessment, thrombosis risk prediction, and intervention measures more accurate. The effect of this step is that the system can gradually evolve as more clinical data accumulates, thereby providing higher quality medical services. These steps together build a complete thrombosis risk assessment and intervention system, from risk assessment, determination of warning levels to personalized patient intervention, generation of standardized monitoring reports, and then to updating and continuous optimization of the knowledge base.The effect of the entire process is that it can provide accurate and personalized risk assessment, monitor changes in patients' vascular health in real time, and support decision-making through visual analysis. Feedback mechanisms and updates to the knowledge base further improve the system's learning ability, enabling it to continuously optimize and adapt to new clinical data. This closed-loop system provides doctors with a powerful support tool to ensure the timeliness and effectiveness of treatment, thereby improving patients' health and reducing the risk of thrombosis-related complications.

[0108] The present invention also provides a monitoring system for dynamically monitoring the blood flow status of the lower extremity veins, which is used to execute the monitoring method for dynamically monitoring the blood flow status of the lower extremity veins as described above. The monitoring system for dynamically monitoring the blood flow status of the lower extremity veins comprises:

[0109] The vascular knowledge construction module is used to obtain the target patient's lower limb venous ultrasound image sequence and establish a vascular space probability distribution map based on anatomical prior knowledge; the vascular space probability distribution map is encoded into conditional prior information to construct a vascular ontology knowledge base;

[0110] A multi-level diffusion modeling module is used to construct a multi-level diffusion probability model based on a vascular ontology knowledge base, wherein the multi-level diffusion probability model specifically includes a first-level diffusion model for spatially locating the main vascular orientation in an ultrasound image, a second-level diffusion model for identifying vascular branch structures and interweaving relationships, and a third-level diffusion model for analyzing the morphological characteristics of a single vascular segment;

[0111] The vascular reconstruction and feature extraction module is used to reconstruct the vascular structure and extract the blood flow features of the lower limb venous ultrasound image sequence using a multi-level diffusion probability model, thereby obtaining vascular fluid feature data;

[0112] A risk assessment model building module is used to establish a deep vein thrombosis risk assessment model based on vascular structure and blood flow characteristics according to vascular fluid characteristic data, and to perform deep vein thrombosis risk assessment on lower extremity venous ultrasound image sequences, thereby obtaining thrombosis risk assessment data;

[0113] The dynamic monitoring module is used to determine the monitoring warning level based on the thrombosis risk assessment data, generate a monitoring report containing vascular structure reconstruction images, blood flow dynamic characteristics and risk assessment analysis, and feed the monitoring report back to the vascular ontology knowledge base to continuously optimize the accuracy of vascular identification and risk assessment after orthopedic surgery.

[0114] The present invention also provides a storage medium, comprising:

[0115] Memory for storing computer programs;

[0116] The processor is used to implement the above-mentioned method for dynamically monitoring the lower extremity venous blood flow status when executing the computer program.

[0117] The present invention can improve the accuracy of vascular detection by acquiring the lower limb venous ultrasound image sequence of the target patient and establishing a vascular space probability distribution map in combination with anatomical prior knowledge. Anatomical prior knowledge enables the spatial position, direction, size and other characteristics of the vascular structure to be more effectively encoded and expressed, providing a solid foundation for subsequent vascular identification and feature extraction. Encoding the vascular space probability distribution map as conditional prior information and constructing a vascular ontology knowledge base can help the system to deeply understand and analyze the vascular structure. The construction of the knowledge base can not only integrate the vascular data of different cases, but also provide rich reference materials for subsequent image analysis, modeling, risk assessment, etc., thereby improving the system's ability to recognize and understand complex vascular morphology. The role of this module is to construct a multi-level diffusion probability model based on the vascular ontology knowledge base, so as to realize the hierarchical analysis of the vascular structure in the lower limb venous ultrasound image. The multi-level model can accurately model the different scales and complex structures of the blood vessels, which helps to improve the recognition ability of characteristics such as vascular direction, vascular branching structure and blood flow dynamics. The first-level diffusion model locates the main blood vessel direction in the ultrasound image, which can help the system accurately determine the basic direction and distribution of the blood vessels, laying the foundation for the subsequent analysis of the branch structure and interweaving relationship of the blood vessels. The second-level model can identify the branch structure of the blood vessels and their interweaving relationship, improve the recognition ability of the complex structure of the blood vessels, especially in the vascular network involving multiple branches, and can effectively distinguish the relative relationship between the blood vessels. The third-level model analyzes the morphological characteristics of a single blood vessel segment to provide more detailed structural feature data for subsequent vascular health assessment, and help identify abnormalities such as changes in the vascular wall, vascular stenosis, and bending. By using a multi-level diffusion probability model to reconstruct the vascular structure and extract blood flow features from the ultrasound image sequence of the lower limb veins, accurate vascular fluid feature data can be obtained. The core advantage of this module is that it can comprehensively model the morphology and function of the blood vessels, and provide important basic data for subsequent risk assessment and diagnosis by extracting the dynamic characteristics of the blood flow. This step can reflect the changing trend of blood flow in the blood vessels in real time, reveal the abnormality of hemodynamics, and help assess the risk of thrombosis. According to the vascular fluid characteristic data, a deep vein thrombosis risk assessment model based on vascular structure and blood flow characteristics is established. Through the comprehensive analysis of hemodynamic characteristics, vascular morphological characteristics and other data, the risk of thrombosis in patients can be quantitatively assessed. Predicting thrombosis risk through the model can provide doctors with quantitative risk assessment, help doctors make more accurate diagnosis and treatment decisions, and provide early warning and intervention measures. Risk assessment data provides a basis for subsequent treatment and monitoring, can help identify high-risk patients, and provide support for the formulation of personalized treatment plans. Through risk assessment, doctors can better plan patient monitoring plans and treatment strategies.This module determines the monitoring and warning level based on the thrombosis risk assessment data, and generates a monitoring report containing vascular structure reconstruction images, blood flow dynamic characteristics and risk assessment analysis. These reports not only help doctors to monitor patients regularly, but also issue early warning information in time when the patient's thrombosis risk changes to ensure the health of the patient. The report integrates the vascular health status, hemodynamic characteristics and potential risks, providing doctors with a detailed and intuitive decision support tool. Feeding back the monitoring report to the vascular ontology knowledge base can provide data support for the system to continuously optimize the recognition accuracy of vascular morphology and blood flow characteristics. With the continuous accumulation of monitoring data, the knowledge base can self-update and optimize to improve the system's recognition and prediction capabilities. This feedback mechanism enables the system to evolve continuously, thereby improving the diagnosis and treatment support capabilities for vascular diseases. These modules combine vascular anatomical knowledge, image analysis, fluid dynamics modeling, risk assessment and dynamic monitoring. Through the synergistic effect of each module, a comprehensive thrombosis risk assessment and intervention system is formed. The system can achieve efficient and accurate vascular health assessment, thrombosis risk prediction and monitoring and early warning, and comprehensively improve the decision support capabilities of clinicians. At the same time, the system can continuously optimize its recognition capabilities through feedback mechanisms and knowledge base updates, and gradually achieve accurate management of vascular health. This process not only helps reduce the occurrence of thrombosis-related diseases, but also improves the treatment effect and survival rate of patients.

[0118] In the above embodiments, the method for dynamically monitoring the lower extremity venous blood flow status can be applied to the risk assessment of deep vein thrombosis after orthopedic surgery and the analysis of the influence of body position on the lower extremity venous hemodynamics.

[0119] The goal of this method is to dynamically monitor the blood flow status of the lower extremity veins and evaluate the risk of deep vein thrombosis by combining vascular structure and blood flow characteristics. Orthopedic patients, especially those who have undergone lower extremity fracture surgery, are prone to deep vein thrombosis (DVT) due to long-term bed rest, limited activity, and blood flow changes during the postoperative recovery period. Therefore, this method is highly relevant to the orthopedic patient population, especially for those who need to continuously monitor blood circulation and prevent thrombosis. The dynamic blood flow monitoring and thrombosis risk assessment in the method just meet clinical needs. Through ultrasound image sequences, vascular structure reconstruction and blood flow feature extraction, combined with deep vein thrombosis risk assessment, it can provide accurate dynamic monitoring for orthopedic patients and meet the needs of orthopedic clinics for postoperative thrombosis prevention and blood flow status assessment. The multi-level diffusion probability model can accurately monitor and evaluate the health status of lower extremity blood vessels in orthopedic patients by hierarchical analysis of vascular direction, branching structure, interweaving relationship and morphological characteristics. For orthopedic patients who have undergone lower extremity fracture surgery or have been bedridden for a long time, the health of the blood vessels and blood flow status directly affect the rehabilitation process. Through the precise analysis of multi-level models, it is not only possible to identify the risk of thrombosis, but also to guide clinicians to develop personalized rehabilitation treatment plans for patients. This method fully meets the clinical needs of orthopedics, especially the monitoring and assessment of thrombosis risks in patients after lower limb fracture surgery, long-term bed rest or limited activity, through technologies such as dynamic monitoring of lower limb venous blood flow status, analysis of vascular structure and blood flow characteristics, and assessment of deep vein thrombosis risk. Although the original method did not particularly emphasize the orthopedic background at the technical level, its process and application scenarios are highly consistent with the actual needs of orthopedic patients, and can provide orthopedic patients with solutions for thrombosis monitoring, assessment and risk warning. Therefore, the original method has strong orthopedic application potential, and can provide real-time and accurate thrombosis monitoring services for patients in the postoperative rehabilitation period, which meets the urgent needs of the orthopedic field for dynamic assessment of blood flow status and thrombosis prevention.

[0120] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0121] The above description is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A method for dynamically monitoring the blood flow status of lower extremity veins, characterized in that: The following steps are involved: Step S1: Obtain a sequence of lower limb venous ultrasound images of a target patient, and establish a vascular space probability distribution map based on anatomical prior knowledge; encode the vascular space probability distribution map into conditional prior information to construct a vascular ontology knowledge base; Step S2: constructing a multi-level diffusion probability model according to the vascular ontology knowledge base, wherein the multi-level diffusion probability model specifically includes a first-level diffusion model for spatially locating the main vascular directions in the ultrasound image, a second-level diffusion model for identifying the vascular branching structure and interweaving relationship, and a third-level diffusion model for analyzing the morphological characteristics of a single vascular segment; Step S3: using a multi-level diffusion probability model to reconstruct the vascular structure and extract blood flow characteristics from the lower limb venous ultrasound image sequence, thereby obtaining vascular fluid characteristic data; Step S4: establishing a deep vein thrombosis risk assessment model based on vascular structure and blood flow characteristics according to the vascular fluid characteristic data, and performing deep vein thrombosis risk assessment on the lower extremity venous ultrasound image sequence, thereby obtaining thrombosis risk assessment data; Step S5: Determine the monitoring warning level based on the thrombosis risk assessment data, generate a monitoring report containing vascular structure reconstruction images, blood flow dynamic characteristics and risk assessment analysis, and feed the monitoring report back to the vascular ontology knowledge base to continuously optimize the accuracy of postoperative vascular identification and risk assessment in orthopedics.

2. The method for dynamically monitoring the blood flow status of the lower extremity veins according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: Scanning the lower extremity veins of the target patient using a high-resolution ultrasound device to collect continuous cross-sectional images and longitudinal images of the venous system including the femoral vein, popliteal vein, and small saphenous vein, thereby obtaining a lower extremity venous ultrasound image sequence; Step S12: acquiring standard anatomical structure data based on human lower limb veins; constructing a standardized knowledge base of vein direction, position and morphological characteristics according to the standard anatomical structure data; Step S13: registering the standardized knowledge base with the lower extremity venous ultrasound image sequence, and correcting the angle deviation or size scaling error in the scanning process in real time, thereby obtaining a venous anatomy priori knowledge base, wherein the venous anatomy priori knowledge base includes the spatial distribution and anatomical hierarchical relationship of each major vein; Step S14: Calculate the probability value of each pixel belonging to the main blood vessel area according to the venous anatomy prior knowledge base and the lower extremity venous ultrasound image sequence, thereby generating a vascular space probability distribution map; Step S15: hierarchically representing the vascular space probability distribution map into spatial position features, structural features, and association features, and performing standardization on each layer of features to obtain conditional prior information encoding data; Step S16: Fusing the conditional prior information encoding data with the venous anatomy prior knowledge base to construct a vascular ontology knowledge base.

3. The method for dynamically monitoring the blood flow status of the lower extremity veins according to claim 2, characterized in that: Step S2 includes the following steps: Step S21: extracting standard morphological features and spatial distribution features of blood vessels according to the blood vessel ontology knowledge base, thereby constructing a feature priori matrix; Step S22: embedding the feature prior matrix into the conditional control module of the preset diffusion model, thereby obtaining basic diffusion model framework data; Step S23: establishing a noise prediction network of the first-level diffusion model based on the basic diffusion model framework data, and using a progressive noise reduction strategy to process the ultrasound image sequence in the vascular ontology knowledge base, thereby obtaining the main vascular contour positioning data; Step S24: Based on the main blood vessel contour positioning data, a region of interest is set around the main blood vessel contour, and a local feature enhancement network of the second-level diffusion model is constructed based on the region of interest, so as to obtain blood vessel branch structure recognition data, wherein the range of the region of interest is 3 times the diameter of the main blood vessel; Step S25: establishing a morphological feature extraction network of the third-level diffusion model according to the vascular branch structure recognition data, calculating the vascular cross-sectional features and wall thickness variation features, thereby obtaining vascular morphological feature data; Step S26: multi-level feature fusion is performed on the main blood vessel contour positioning data, the blood vessel branch structure recognition data and the blood vessel morphology feature data, and a feature association mechanism between levels is constructed, thereby obtaining a multi-level diffusion probability model.

4. The method for dynamically monitoring the blood flow status of the lower extremity veins according to claim 3, characterized in that: Step S23 includes the following steps: Step S231: constructing a noise prediction network of a U-Net structure based on the basic diffusion model framework data, wherein the U-Net includes an encoder and a decoder, the encoder is used for feature extraction, and the decoder is used for blood vessel contour reconstruction; Step S232: Initializing the training parameters of the noise prediction network to obtain network training parameters, wherein the network training parameters specifically include an initial noise ratio of 0.5, a noise reduction step of 0.1, and a feature retention threshold of 0.8; Step S233: performing initial sampling of the ultrasound image sequence in the vascular ontology knowledge base according to the network training parameters, and performing a first round of noise prediction, thereby obtaining initial vascular contour prediction data; Step S234: iteratively processing the initial blood vessel contour prediction data based on a progressive noise reduction strategy to obtain optimized blood vessel contour data, wherein the iterative processing specifically includes gradually reducing the noise ratio in each iteration and retaining blood vessel contour points with a confidence level greater than a preset feature retention threshold; Step S235: When the noise ratio drops to 0.1, the optimized blood vessel contour data is subjected to isolated point removal, contour line smoothing, and contour continuity optimization, thereby obtaining the main blood vessel contour positioning data.

5. The method for dynamically monitoring the blood flow status of the lower extremity veins according to claim 4, characterized in that: Step S24 includes the following steps: Step S241: determining the range of the region of interest according to the main blood vessel contour positioning data, setting the radius of the region of interest to three times the diameter of the main blood vessel, and smoothing the region boundary to obtain the mask data of the region of interest; Step S242: constructing a local feature enhancement network based on the mask data of the region of interest, thereby obtaining a feature enhancement network model, wherein the local feature enhancement network adopts a residual connection structure, including a feature extraction module, an attention module and a feature fusion module; Step S243: performing local contrast enhancement and detail enhancement processing on the image in the region of interest, thereby obtaining enhanced image data; Step S244: using the feature enhancement network model to perform vascular branch structure recognition on the enhanced image data, thereby obtaining initial branch recognition data, wherein the vascular branch structure recognition specifically sets the minimum diameter threshold of branch recognition to 1 / 5 of the main blood vessel diameter, and recognizes branch structures of different scales; Step S245: performing post-processing optimization on the initial branch identification data and calculating the spatial interleaving degree between branches, thereby obtaining blood vessel branch structure identification data, wherein the post-processing optimization includes removing false branches, repairing branch breakpoints, and optimizing branch connection relationships.

6. The method for dynamically monitoring the blood flow status of the lower extremity veins according to claim 5, characterized in that: Step S25 includes the following steps: Step S251: constructing a morphological feature extraction network according to the blood vessel branch structure recognition data, thereby obtaining a feature extraction network model, wherein the morphological feature extraction network adopts a multi-branch parallel structure, including a cross-sectional feature analysis branch and a wall thickness feature analysis branch; Step S252: using the cross-sectional feature analysis branch in the feature extraction network model to calculate the morphological features of the blood vessel cross section, thereby obtaining cross-sectional feature data, wherein the morphological features of the blood vessel cross section include the lumen area change rate, the cross-sectional roundness index, and the cross-sectional eccentricity; Step S253: extracting the vascular wall thickness variation characteristics based on the wall thickness feature analysis branch in the feature extraction network model, thereby obtaining wall thickness feature data; Step S254: Fusing the cross-sectional feature data and the wall thickness feature data to obtain vascular morphology feature data.

7. The method for dynamically monitoring the blood flow status of the lower extremity veins according to claim 6, characterized in that: Step S4 includes the following steps: Step S41: analyzing the change rate of the vascular lumen cross-sectional area according to the vascular fluid characteristic data, thereby obtaining vascular stenosis index data, wherein the vascular structural parameters include the change rate of the vascular lumen cross-sectional area, the distribution of vascular wall thickness, and the vascular elasticity coefficient; Step S42: extracting hemodynamic parameters according to the vascular fluid characteristic data, thereby obtaining blood flow resistance index data, wherein the hemodynamic parameters include blood flow velocity field distribution, blood flow direction change characteristics, and blood flow turbulence index; Step S43: constructing a feature space of a deep vein thrombosis risk assessment model, mapping the vascular stenosis index data and the blood flow resistance index data to the feature space, and establishing a feature weight matrix, thereby obtaining risk feature representation data; Step S44: training a deep vein thrombosis risk assessment model based on the risk feature representation data, thereby obtaining a risk prediction model; Step S45: Use the risk prediction model to divide the risk level of the lower limb venous ultrasound image sequence to obtain thrombosis risk assessment data, where the risk level division specifically includes when the stenosis index is greater than 0.5 and the resistance index is greater than 0.7, it is judged as high risk; when the stenosis index is between 0.3 and 0.5 or the resistance index is between 0.5 and 0.7, it is judged as medium risk; when the stenosis index is less than 0.3 and the resistance index is less than 0.5, it is judged as low risk.

8. The method for dynamically monitoring the blood flow status of the lower extremity veins according to claim 7, characterized in that: Step S5 includes the following steps: Step S51: establishing a mapping relationship between risk level and warning level based on thrombosis risk assessment data, thereby obtaining warning level determination standard data; Step S52: determining the monitoring warning level of the target patient according to the warning level determination standard data, and generating warning information including risk level, warning suggestions and intervention measures, thereby obtaining monitoring warning data; Step S53: constructing a three-dimensional blood vessel model, marking abnormal areas, and visualizing blood flow dynamic characteristics based on the blood vessel fluid characteristic data and the thrombosis risk assessment data, thereby obtaining visualized analysis result data; Step S54: obtaining standardized monitoring report data according to the monitoring and early warning data and the visual analysis result data; Step S55: extracting abnormal features, risk factors and evaluation results in the standardized monitoring report data as structured data, and feeding it back to the vascular ontology knowledge base, thereby obtaining knowledge base update data; Step S56: updating data based on the knowledge base to continuously optimize the recognition accuracy of vascular morphology and blood flow characteristics.

9. A monitoring system for dynamically monitoring the blood flow status of lower extremity veins, characterized in that: The monitoring method for dynamically monitoring the blood flow status of the lower extremity veins according to claim 1, wherein the monitoring system for dynamically monitoring the blood flow status of the lower extremity veins comprises: The vascular knowledge construction module is used to obtain the target patient's lower limb venous ultrasound image sequence and establish a vascular space probability distribution map based on anatomical prior knowledge; the vascular space probability distribution map is encoded into conditional prior information to construct a vascular ontology knowledge base; A multi-level diffusion modeling module is used to construct a multi-level diffusion probability model based on a vascular ontology knowledge base, wherein the multi-level diffusion probability model specifically includes a first-level diffusion model for spatially locating the main vascular orientation in an ultrasound image, a second-level diffusion model for identifying vascular branch structures and interweaving relationships, and a third-level diffusion model for analyzing the morphological characteristics of a single vascular segment; The vascular reconstruction and feature extraction module is used to reconstruct the vascular structure and extract the blood flow features of the lower limb venous ultrasound image sequence using a multi-level diffusion probability model, thereby obtaining vascular fluid feature data; A risk assessment model building module is used to establish a deep vein thrombosis risk assessment model based on vascular structure and blood flow characteristics according to vascular fluid characteristic data, and to perform deep vein thrombosis risk assessment on lower extremity venous ultrasound image sequences, thereby obtaining thrombosis risk assessment data; The dynamic monitoring module is used to determine the monitoring warning level based on the thrombosis risk assessment data, generate a monitoring report containing vascular structure reconstruction images, blood flow dynamic characteristics and risk assessment analysis, and feed the monitoring report back to the vascular ontology knowledge base to continuously optimize the accuracy of vascular identification and risk assessment after orthopedic surgery.

10. A storage medium, characterized in that: include: Memory for storing computer programs; A processor is used to implement the monitoring method for dynamically monitoring the lower extremity venous blood flow status as described in any one of claims 1 to 8 when executing the computer program.

Images