Vascular surgery intervention intracavity treatment postoperative complication prediction and management system

Abstract

The invention relates to the technical field of vascular surgery, in particular to a vascular surgery interventional intracavity treatment postoperative complication prediction and management system which integrates data acquisition, processing, model training, prediction analysis, comprehensive decision and early warning management and is specially designed for vascular surgery interventional intracavity treatment. The system monitors the postoperative physiological indexes of a patient in real time through multiple sensors, combines clinical data and disease history, carries out physiological sequence prediction and image analysis by utilizing RNN and CNN models, and accurately evaluates the risk of complications. The comprehensive decision-making module generates a personalized management scheme, and the early warning management module monitors in real time and pushes early warning and suggestions in time when abnormity occurs. The system not only improves the prediction accuracy and the management effect, but also optimizes the medical resource configuration, enhances the patient experience, contributes to the development of precise medical treatment and intelligent management, and effectively solves the key problem of postoperative complication prediction and management.

Description

Technical Field

[0001] The present invention relates to the technical field of vascular surgery, in particular to a system for predicting and managing postoperative complications after endovascular treatment in vascular surgery. Background Art

[0002] Endovascular treatment in vascular surgery is an important means of treating vascular diseases today. However, the management of postoperative complications has always been a major challenge in clinical practice. Traditional postoperative management methods mainly rely on regular examinations by medical staff and active reports from patients. This approach has many limitations. First of all, it cannot achieve continuous and real-time monitoring, and may miss key warning opportunities. Secondly, due to the lack of personalized risk assessment, a unified standard is often used for management, making it difficult to meet the special needs of different patients. In addition, traditional methods are inefficient in data integration and analysis, and it is difficult to make full use of multi-source heterogeneous medical data.

[0003] In recent years, with the development of artificial intelligence technology, some medical institutions have begun to try to apply machine learning algorithms to the prediction of postoperative complications. Although these methods have improved the prediction accuracy to a certain extent, there are still many deficiencies. First of all, most existing systems only focus on a single type of data (such as physiological indicators or laboratory test results), ignoring the importance of multi-modal data fusion, resulting in limitations in the comprehensiveness and accuracy of the prediction model. Secondly, many systems use static prediction models and lack the ability of adaptive learning, making it difficult to adapt to the dynamic changes of the patient's condition and the continuous update of medical knowledge. Moreover, existing systems often focus on the prediction of complications and pay insufficient attention to how to effectively transform the prediction results into personalized management strategies.

[0004] The closest prior art usually uses simple machine learning models (such as logistic regression or decision tree) for risk prediction and sets a fixed warning threshold. Although this method is an improvement over traditional manual monitoring, there are still many problems. First of all, simple models are difficult to capture complex non-linear relationships and temporal dependencies, resulting in low prediction accuracy, especially for rare but serious complications. Secondly, the fixed warning threshold cannot adapt to the individual differences of different patients, easily leading to a high false alarm rate or missed alarm rate. In addition, these systems usually lack interpretability of the prediction results, making it difficult for medical staff to understand and trust the decision-making suggestions of the system. Finally, the existing technology lacks consideration in terms of patient participation and multi-center collaboration, limiting the practicality and scalability of the system. Summary of the Invention

[0005] In view of the above problems, the present invention proposes an innovative system for predicting and managing postoperative complications in endovascular treatment of vascular surgery. The system aims to comprehensively improve the prediction accuracy and management effect of postoperative complications through multi-modal data fusion, deep learning algorithms, dynamic early warning mechanisms, and personalized management strategies. The present invention not only focuses on the accurate prediction of complications but also emphasizes converting the prediction results into actionable management suggestions, while considering key factors such as the interpretability of the system, patient participation, and multi-center collaboration.

[0006] The present invention proposes a quality control system for user-generated content based on generative adversarial networks, including:

[0007] A data acquisition module, used for:

[0008] Collecting the basic data, physiological parameters, biochemical indexes, imaging data, clinical data, and disease history data of patients;

[0009] Real-time collecting the postoperative physiological indexes of patients through a variety of sensors and monitoring devices;

[0010] A data processing module, communicatively connected to the data acquisition module, used for:

[0011] Receiving the multi-modal data sent by the data acquisition module;

[0012] Preprocessing the multi-modal data, including data cleaning, feature extraction, and feature selection;

[0013] A model training module, communicatively connected to the data processing module, used for:

[0014] Based on the preprocessed multi-modal data, constructing and training a physiological sequence prediction model and an imaging analysis model;

[0015] The physiological sequence prediction model includes a physiological feature sequence prediction model based on a recurrent neural network (RNN);

[0016] The imaging analysis model includes a medical imaging analysis model based on a convolutional neural network (CNN); A prediction and analysis module, communicatively connected to the model training module, used for:

[0017] Using the physiological sequence prediction model to predict the postoperative physiological indexes of patients; Using the imaging analysis model to analyze the postoperative medical images of patients;

[0018] Based on the prediction and analysis results, evaluating the risk of patients having postoperative complications;

[0019] A comprehensive decision-making module, communicatively connected to the prediction and analysis module, used for:

[0020] Integrate the physiological index prediction results and medical image analysis results;

[0021] Combine the patient's clinical data and medical history to generate a comprehensive risk assessment report;

[0022] Formulate a personalized postoperative management plan based on the risk assessment results;

[0023] An early warning management module, communicatively connected to the comprehensive decision-making module, for:

[0024] Set dynamic early warning thresholds based on the comprehensive risk assessment report;

[0025] Real-time monitor the patient's postoperative indicators and trigger an early warning when abnormalities are detected;

[0026] Push early warning information and personalized management suggestions to medical staff.

[0027] Preferably, the data processing module includes:

[0028] A data cleaning unit, for:

[0029] Detect and process outliers in the collected multi-modal data;

[0030] Fill in missing data and unify the data format;

[0031] A feature extraction unit, communicatively connected to the data cleaning unit, for:

[0032] Extract time series features, statistical features, and frequency domain features from the cleaned data; perform noise reduction and enhancement processing on medical image data;

[0033] A feature selection unit, communicatively connected to the feature extraction unit, for:

[0034] Use filtering methods, wrapper methods, or embedding methods to screen the extracted features;

[0035] Reduce the feature dimension and improve the training efficiency and generalization ability of subsequent models.

[0036] Preferably, the model training module includes:

[0037] An RNN training unit, for:

[0038] Construct a physiological feature sequence prediction model based on a long short-term memory (LSTM) network;

[0039] Use historical patient data to perform offline training on the LSTM model;

[0040] Implement an online fine-tuning function to dynamically update model parameters according to current patient data;

[0041] A CNN training unit for:

[0042] Construct a multi-layer convolutional neural network model for medical image analysis;

[0043] Adopt the transfer learning method to accelerate the training process using a pre-trained model;

[0044] Implement multi-task learning to perform organ segmentation and anomaly detection simultaneously;

[0045] A model evaluation unit, communicatively connected to the RNN training unit and the CNN training unit, for:

[0046] Evaluate the model performance using the cross-validation method;

[0047] Calculate indicators such as the accuracy, sensitivity, and specificity of the model;

[0048] When the model performance does not meet the preset criteria, trigger the model optimization process.

[0049] Preferably, the prediction analysis module includes:

[0050] A physiological index prediction unit for:

[0051] Use the trained LSTM model to predict the key physiological indicators of the patient within the next 24 hours;

[0052] Calculate the deviation between the predicted value and the normal range to quantify the degree of anomaly;

[0053] An image analysis unit for:

[0054] Use the trained CNN model to analyze the patient's postoperative CT, MRI, or X-ray images;

[0055] Detect and locate possible vascular abnormalities such as stenosis, thrombosis, or calcification;

[0056] A risk assessment unit, communicatively connected to the physiological index prediction unit and the image analysis unit, for:

[0057] Calculate the probability of the patient developing different types of complications based on the physiological index prediction results and the image analysis results;

[0058] Use a multi-factor logistic regression model to comprehensively consider the patient's basic characteristics and clinical indicators to generate a risk score.

[0059] Preferably, the comprehensive decision-making module includes:

[0060] A data fusion unit for:

[0061] Integrate the prediction results of physiological indicators, the results of image analysis, and clinical data;

[0062] Use the attention mechanism to perform weighted fusion on data from different sources;

[0063] The decision-making and reasoning unit, which is communicatively connected to the data fusion unit, is used for:

[0064] Based on the fused data, use decision tree or random forest algorithms to classify the complication risks;

[0065] Generate an interpretable risk assessment report, including key influencing factors and their weights;

[0066] The solution generation unit, which is communicatively connected to the decision-making and reasoning unit, is used for:

[0067] According to the risk assessment results, select a suitable basic solution from a preset management solution library;

[0068] Combine the individual characteristics of the patient to make personalized adjustments to the management solution;

[0069] Generate comprehensive management suggestions including drug treatment, rehabilitation training, and lifestyle guidance.

[0070] Preferably, the early warning management module includes:

[0071] The threshold setting unit, which is used for:

[0072] Based on historical data statistics and expert knowledge, set initial early warning thresholds for different types of complications;

[0073] Dynamically adjust the early warning thresholds according to the individual characteristics and real-time status of the patient;

[0074] The monitoring and early warning unit, which is communicatively connected to the threshold setting unit, is used for:

[0075] Compare the physiological indicators of the patient with the preset thresholds in real time;

[0076] When an abnormality is detected, determine the early warning level according to the degree of deviation;

[0077] Trigger the corresponding level of early warning mechanism, including sound and light reminders and message push;

[0078] The intervention management unit, which is communicatively connected to the monitoring and early warning unit, is used for:

[0079] Generate targeted intervention suggestions according to the early warning information;

[0080] Record the intervention measures of medical staff and the response of the patient;

[0081] Evaluate the intervention effect and optimize future early warning and management strategies accordingly.

[0082] Preferably, it further includes an adaptive learning module, which is communicatively connected to the model training module and the prediction analysis module and is used for:

[0083] Collect new data generated during the operation of the system, including prediction results, actual complication occurrences, and intervention effects;

[0084] Regularly update the model parameters using the newly collected data to improve the prediction accuracy of the model;

[0085] Analyze prediction error cases, identify the weaknesses of the model, and perform targeted optimization;

[0086] Dynamically adjust the model structure or feature weights according to the prediction performance of different types of complications.

[0087] Preferably, it further includes a knowledge base management module, which is communicatively connected to the comprehensive decision-making module and is used for:

[0088] Construct and maintain a professional knowledge base related to endovascular treatment in vascular surgery;

[0089] Store typical cases, treatment plans, and the latest research results;

[0090] Provide rule-based auxiliary support for decision-making reasoning;

[0091] Regularly update the content of the knowledge base to ensure that the system decisions are based on the latest medical evidence.

[0092] Preferably, it further includes a patient interaction module, which is communicatively connected to the early warning management module and is used for:

[0093] Interact with patients in real time through a mobile application or smart device;

[0094] Push personalized health education content and rehabilitation guidance to patients;

[0095] Collect patients' subjective symptom reports and quality of life assessment data;

[0096] Provide remote consultation and psychological support services to enhance patient compliance.

[0097] Preferably, it further includes a multi-center collaboration module, which is communicatively connected to the data collection module and the model training module and is used for:

[0098] Establish a secure data sharing mechanism among multiple medical institutions;

[0099] Under the premise of protecting patient privacy, large-scale data sets from different centers are collected;

[0100] Enable federated learning, allowing centers to jointly train and optimize models without directly sharing raw data;

[0101] Compare the prediction performance and management effectiveness of different centers to promote the promotion of best practices.

[0102] The beneficial effects of the present invention are reflected in multiple aspects. At the macro level, the system integrates multi-source heterogeneous data to build a comprehensive and intelligent postoperative management platform, which significantly improves the overall safety and effectiveness of vascular surgical interventional treatment. This not only helps to reduce the incidence of complications, but also optimizes the allocation of medical resources and reduces the workload of medical staff.

[0103] From the perspective of system architecture, the modular design of the present invention achieves high flexibility and scalability. The close collaboration between the various functional modules forms a closed-loop management system. From data collection, preprocessing, model training to prediction analysis, decision support and early warning management, each link has been carefully designed to ensure the efficient operation of the system.

[0104] At the algorithm level, the present invention solves the shortcomings of traditional methods in dealing with complex nonlinear relationships and time series dependencies through deep learning technologies (such as LSTM and CNN). Multimodal data fusion technology effectively integrates information from different sources and provides a more comprehensive and accurate assessment of patient status. The dynamic threshold adjustment mechanism balances the sensitivity and specificity of the warning well, significantly reducing the false alarm rate and missed alarm rate.

[0105] In terms of management strategies, the personalized recommendation system of the present invention fully considers the individual differences of patients and provides tailored management suggestions. This not only improves the pertinence and effectiveness of intervention measures, but also enhances patient compliance. The introduction of the knowledge base management module ensures that the system can absorb the latest medical research results in a timely manner and continuously optimize decision-making strategies.

[0106] In addition, the present invention has also made innovations in patient engagement and multi-center collaboration. The patient interaction module significantly improves patient engagement and satisfaction by providing personalized health education and remote consultation services. The multi-center collaboration module provides a safe and efficient platform for large-scale data sharing and model optimization, which effectively promotes collaboration among medical institutions and the promotion of best practices.

[0107] Finally, the adaptive learning capability of the present invention ensures that the system performance can be continuously improved over time. By continuously learning new data and optimizing models, the system can adapt to the update of medical knowledge and changes in clinical practice, maintaining long-term effectiveness and advancement.

[0108] In summary, through innovative technical solutions and comprehensive system designs, the present invention effectively solves the key problems in the prediction and management of complications after endovascular treatment in vascular surgery. It not only significantly improves the prediction accuracy and management effectiveness, but also shows great potential in enhancing medical quality, optimizing resource allocation, and improving the patient experience, making important contributions to the development of precision medicine and intelligent medical management. BRIEF DESCRIPTION OF THE DRAWINGS

[0109] Figure 1 is a logical block diagram of the system of the present invention;

[0110] Figure 2 is a logical block diagram of the data processing module of the present invention;

[0111] Figure 3 is a logical block diagram of the model training module of the present invention;

[0112] Figure 4 is a logical block diagram of the prediction analysis module of the present invention;

[0113] Figure 5 is a logical block diagram of the comprehensive decision-making module of the present invention;

[0114] Figure 6 is a logical block diagram of the early warning management module of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0115] Please refer to the attached Figure 1-6 , the present invention provides a system for predicting and managing complications after endovascular treatment in vascular surgery, which can effectively predict and manage the possible complications after endovascular treatment in vascular surgery, thereby improving the postoperative rehabilitation effect and quality of life of patients. The following will describe the detailed implementation manners of the present invention.

[0116] Preferably, the system for predicting and managing complications after endovascular treatment in vascular surgery of the present invention includes a data acquisition module 1, a data processing module 2, a model training module 3, a prediction analysis module 4, a comprehensive decision-making module 5, and an early warning management module 6. These modules cooperate closely through communication connections to jointly complete the entire process from data acquisition to early warning management.

[0117] The data acquisition module 1 is used to collect multi-dimensional data of patients. Specifically, this module collects the basic data of patients (such as age, gender, height, weight, etc.), physiological parameters (such as blood pressure, heart rate, body temperature, etc.), biochemical indicators (such as blood glucose, blood lipids, liver and kidney functions, etc.), imaging data (such as CT, MRI, ultrasound, etc.), clinical data (such as diagnosis results, medication conditions, etc.) and disease history data. In addition, the data acquisition module 1 also collects the postoperative physiological indicators of patients in real time through a variety of sensors and monitoring devices. For example, a non-invasive blood pressure monitor can be used to measure blood pressure every 15 minutes, an electrocardiogram monitor can be used to continuously monitor heart rate and rhythm, and a pulse oximeter can be used to monitor blood oxygen saturation in real time, etc. This comprehensive data collection lays a solid foundation for subsequent analysis and prediction.

[0118] The data processing module 2 is communicatively connected to the data acquisition module 1 and is used to receive and process the collected multi-modal data. In one embodiment of the present invention, the data processing module 2 first preprocesses the received data, including data cleaning, feature extraction, and feature selection. During the data cleaning process, the system will remove outliers, such as marking blood pressure values that exceed the normal range by 3 standard deviations as abnormal and processing them. In the feature extraction stage, the system will extract statistical features (such as mean, variance, skewness, etc.) and frequency domain features (such as power spectral density) from time series data. In the feature selection stage, the system uses the recursive feature elimination method (RFE) to select the most relevant feature subset, usually retaining features that can explain 80% of the variance.

[0119] The model training module 3 is communicatively connected to the data processing module 2 and constructs and trains a prediction model based on the preprocessed multi-modal data. The present invention adopts two core models: a physiological feature sequence prediction model based on a recurrent neural network (RNN) and a medical image analysis model based on a convolutional neural network (CNN).

[0120] For the physiological feature sequence prediction model, the present invention preferably uses a long short-term memory network (LSTM). The mathematical expression of the LSTM model is as follows:

[0121] f t =σ(W f ·[h t-1 ,x t +b f ),

[0122] i t =σ(W i ·[h t-1 ,x t +b i ),

[0123]

[0124]

[0125] o t = σ(W o ·[h t-1 , x t +b o ),

[0126] h t = o t *tanh(C t ),

[0127] Among them, f t is the forgetting gate, i t is the input gate, is the candidate memory cell, C t is the current memory cell, o t is the output gate, h t is the hidden state, σ is the sigmoid activation function, and W and b are the weight matrix and bias term respectively.

[0128] The medical image analysis model adopts an improved ResNet50 structure, and the mathematical expression of its core residual block is as follows:

[0129] y = F(x, W i ) + x,

[0130] Among them, x and y are the input and output of the residual block respectively, and F(x, W i ) represents the residual mapping.

[0131] The prediction analysis module 4 is communicatively connected to the model training module 3, and uses the trained model to predict and analyze the postoperative condition of the patient. For example, the LSTM model can predict the trend of blood pressure changes within the next 24 hours of the patient, and the CNN model can detect whether there are signs of vascular stenosis or thrombosis from the postoperative CT images.

[0132] The comprehensive decision-making module 5 integrates the output results of the prediction analysis module 4, combines the clinical data and disease history of the patient, and generates a comprehensive risk assessment report. This module uses a multi-factor logistic regression model to calculate the complication risk, and its mathematical expression is as follows:

[0133]

[0134] Among them, P(Y = 1) represents the probability of occurrence of complications, X i represents each risk factor, and β i is the corresponding regression coefficient.

[0135] Finally, the early warning management module 6 sets dynamic early warning thresholds based on the output of the comprehensive decision-making module 5 and monitors the patient's condition in real time. When an anomaly is detected, the system immediately triggers an alarm and pushes information to medical staff. For example, if it is predicted that the patient's systolic blood pressure may exceed 180 mmHg within the next 2 hours, the system issues a high blood pressure risk warning.

[0136] The data cleaning unit 21 first performs outlier detection and processing on the collected multimodal data. For example, for heart rate data, if a heart rate value at a certain time point is detected to be outside the normal range of 40 - 200 beats per minute, the system marks it as an outlier. For these outliers, the system replaces them with the average of the previous and next values or estimates them using interpolation methods. In addition, the data cleaning unit 21 is also responsible for filling in missing data. For short-term missing data (such as continuous missing data not exceeding 3 time points), the system uses linear interpolation to fill it; for long-term missing data, multiple imputation methods are used, considering the information of other relevant variables to estimate the missing values.

[0137] The feature extraction unit 22 extracts valuable features from the cleaned data. For time series data, such as continuously monitored blood pressure values, the system extracts statistical features (such as mean, standard deviation, skewness, kurtosis, etc.) and frequency domain features (such as peak frequency and energy of power spectral density). For medical image data, the system uses image processing techniques for noise reduction and enhancement. For example, applying a Gaussian filter to CT images to reduce noise and then using histogram equalization to enhance image contrast.

[0138] The feature selection unit 23 is responsible for selecting the most relevant and most predictive subset from the extracted features. The present invention uses a recursive feature elimination (RFE) algorithm combined with cross-validation for feature selection. The basic idea of the RFE algorithm is to repeatedly construct a model (such as a support vector machine or a random forest), and each time the least important feature is deleted until the preset number of features is reached. In this system, we usually select the smallest feature subset that can explain 80% of the data variance. This can effectively reduce the complexity of the model while retaining key information and improve its generalization ability.

[0139] The RNN training unit 31 is responsible for constructing and training a physiological feature sequence prediction model based on a long short-term memory (LSTM) network. The architecture of the LSTM model includes an input layer, two LSTM hidden layers (each layer contains 128 neurons), and a fully connected output layer. The model is trained using the Adam optimizer, with the initial learning rate set to 0.001, and a learning rate decay strategy is used. To prevent overfitting, we add a Dropout layer between the LSTM layers, with the dropout rate set to 0.5.

[0140] The CNN training unit 32 is used to construct and train a medical image analysis model. We adopted an improved ResNet50 structure and used the transfer learning method to accelerate the training process. Specifically, we used the ResNet50 model pre-trained on the ImageNet dataset as the basis and then fine-tuned it on our medical image dataset. During the fine-tuning process, we froze the parameters of the first 100 layers and only trained the subsequent layers. This can effectively utilize the low-level features captured by the pre-trained model while allowing the model to learn high-level features specific to medical images.

[0141] The model evaluation unit 33 is responsible for evaluating the performance of the trained model. We use the 5-fold cross-validation method to evaluate the generalization ability of the model. For the physiological feature sequence prediction model, we calculate the root mean square error (RMSE) and the mean absolute percentage error (MAPE) to measure the prediction accuracy. For the medical image analysis model, we calculate the accuracy, precision, recall, and F1 score. If the model performance does not meet the preset criteria (e.g., MAPE > 10% or F1 score < 0.8), the system will automatically trigger the model optimization process, such as increasing the training data, adjusting the model structure, or trying different hyperparameters.

[0142] From the above detailed description, we can see that the postoperative complication prediction and management system for endovascular treatment in vascular surgery of the present invention has multiple innovative points and advantages. First, the system adopts a multi-modal data fusion method to comprehensively capture the patient's health status, improving the accuracy and reliability of prediction. Second, the system uses advanced deep learning algorithms, such as LSTM and CNN, which can effectively process time series data and medical image data to achieve accurate prediction. Third, the early warning management module of the system adopts dynamic threshold setting, which can be adjusted according to the patient's individual characteristics and real-time status, reducing false alarms while improving sensitivity. Finally, the system has an adaptive learning ability and can continuously optimize the model according to newly collected data to ensure that the prediction performance continues to improve over time.

[0143] These innovative features enable the system to provide more accurate, timely, and personalized complication prediction and management services for patients after endovascular treatment in vascular surgery, promising to significantly improve the postoperative rehabilitation effect and quality of life of patients, while reducing medical costs and resource consumption.

[0144] The prediction analysis module 4 in the postoperative complication prediction and management system for endovascular treatment in vascular surgery of the present invention includes a physiological index prediction unit 41, an image analysis unit 42, and a risk assessment unit 43. These three units work together to jointly complete the comprehensive analysis and risk prediction of the patient's postoperative condition.

[0145] The physiological index prediction unit 41 uses the trained LSTM model to predict the key physiological indexes of the patient within the next 24 hours. Preferably, the system of the present invention will predict multiple indexes including blood pressure, heart rate, blood oxygen saturation, body temperature, etc. For example, for blood pressure prediction, the system generates predicted values for the next 24 hours every 30 minutes. The prediction results include not only point estimates but also 95% confidence intervals, so that medical staff can more comprehensively evaluate potential risks.

[0146] In one embodiment of the present invention, the physiological index prediction unit 41 will also calculate the deviation between the predicted value and the normal range to quantify the degree of abnormality. For example, for systolic blood pressure, the system regards 90 - 140 mmHg as the normal range. If the predicted value exceeds this range, the system will calculate the standardized deviation score:

[0147]

[0148] where X is the predicted value, μ is the midpoint of the normal range, and σ is one - quarter of the normal range. |Z|>2 is regarded as a significant deviation and requires special attention.

[0149] The imaging analysis unit 42 is responsible for processing the patient's postoperative medical images. Preferably, the system of the present invention can analyze CT, MRI, and X - ray images. Taking CT angiography images as an example, the system uses the trained CNN model for analysis and mainly focuses on the following aspects:

[0150] 1. Vascular stenosis detection: The system can automatically identify the stenotic parts of blood vessels and calculate the degree of stenosis. The degree of stenosis is calculated by the following formula:

[0151]

[0152] Generally, a stenosis degree greater than 50% is regarded as significant stenosis and requires further clinical evaluation.

[0153] 2. Thrombus detection: The system can identify filling defects in blood vessels, which may indicate the presence of thrombus.

[0154] 3. Vascular wall calcification assessment: The system can detect and quantify the degree of calcification of the vascular wall, which is an important indicator of atherosclerosis.

[0155] The risk assessment unit 43 integrates the results of physiological index prediction and imaging analysis to calculate the probabilities of different types of complications occurring in the patient. The present invention uses a multi - factor logistic regression model for risk assessment. The mathematical expression of the model is as follows:

[0156]

[0157] where Y = 1 indicates the occurrence of complications, X irepresent different risk factors, including predicted physiological indicators, imaging analysis results, and the patient's basic characteristics (such as age, previous medical history, etc.). β i is the corresponding regression coefficient, obtained through training with large-scale historical data.

[0158] Preferably, the system of the present invention establishes independent risk assessment models for different types of complications (such as bleeding, thrombosis, restenosis, etc.). The risk scores output by the system range from 0 to 100, where 0 - 20 is considered low risk, 21 - 50 is medium risk, 51 - 80 is high risk, and 81 - 100 is extremely high risk. This grading method helps medical staff quickly identify high-risk patients and take corresponding preventive measures.

[0159] The comprehensive decision-making module 5 of the present invention includes a data fusion unit 51, a decision-making reasoning unit 52, and a solution generation unit 53. These three units work together to convert various data and analysis results into actionable management suggestions.

[0160] The data fusion unit 51 is responsible for integrating data from different sources, including physiological index prediction results, imaging analysis results, and clinical data. In a preferred embodiment of the present invention, the system uses an attention mechanism to perform weighted fusion on data from different sources. The calculation formula for the attention weight is as follows:

[0161]

[0162] where, e i is the attention score calculated through a small neural network, and α i is the final attention weight. This method allows the system to dynamically adjust the importance of different data sources, improving the accuracy and interpretability of decision-making.

[0163] The decision-making reasoning unit 52 classifies the complication risks based on the fused data. The present invention preferably uses the random forest algorithm for decision-making reasoning because the random forest has good generalization ability and interpretability. The random forest consists of multiple decision trees, and the prediction results of each tree are voted or averaged to obtain the final prediction. The Gini index is used as the criterion for judging node splitting during the construction of the decision tree:

[0164]

[0165] where, D is the data set, K is the number of categories, and p k is the proportion of samples in the k-th category.

[0166] Based on the risk assessment results, the solution generation unit 53 selects a suitable basic solution from a preset management solution library and makes personalized adjustments. For example, for patients evaluated as having a high bleeding risk, the system may recommend adjusting the anticoagulant drug dosage, increasing the frequency of coagulation function monitoring, and formulating a detailed care plan. The system of the present invention also takes into account the individual characteristics of the patient (such as age, comorbidities, etc.) to fine-tune the management solution to ensure the feasibility and effectiveness of the solution.

[0167] The early warning management module 6 of the present invention includes a threshold setting unit 61, a monitoring and warning unit 62, and an intervention management unit 63. These units together constitute a dynamic and intelligent early warning system that can timely detect potential risks and provide intervention suggestions.

[0168] The threshold setting unit 61 is responsible for setting early warning thresholds for different types of complications. An innovation of the present invention is the adoption of a dynamic threshold strategy. The initial threshold is set based on large-scale historical data statistics and expert knowledge, and then the system adjusts it according to the individual characteristics and real-time status of the patient. For example, for systolic blood pressure, the general early warning threshold may be set at 160 mmHg, but for patients with a history of hypertension, the system may adjust the threshold to 150 mmHg. The mathematical model for threshold adjustment is as follows:

[0169]

[0170] where T adjusted is the adjusted threshold, T base is the basic threshold, f i is the influencing factor (such as age, previous medical history, etc.), and w i is the corresponding weight.

[0171] The monitoring and warning unit 62 compares the patient's physiological indicators with the preset thresholds in real time. When an abnormality is detected, the system determines the warning level according to the degree of deviation. The present invention adopts a three-level warning mechanism: yellow (mild abnormality), orange (moderate abnormality), and red (severe abnormality). The determination of the warning level not only considers the degree of deviation of a single indicator but also comprehensively evaluates the change trends of multiple related indicators. For example, for suspected bleeding complications, the system will simultaneously monitor the changes in blood pressure, heart rate, and hemoglobin level.

[0172] The intervention management unit 63 generates targeted intervention suggestions based on the warning information. The system of the present invention includes a rule-based expert system that can automatically generate a preliminary intervention plan according to the warning type and severity. For example, for a mild increase in blood pressure, the system may recommend increasing the frequency of blood pressure monitoring and reviewing the medication situation; while for a severe blood pressure abnormality, the system will recommend immediately notifying the doctor and preparing emergency medications.

[0173] In addition, the intervention management unit 63 is also responsible for recording the actual intervention measures of medical staff and the responses of patients. This information is used to evaluate the intervention effect and continuously optimize future early warning and management strategies through machine learning algorithms. The present invention adopts a reinforcement learning method, regarding each intervention as an action and the degree of improvement of the patient's state as a reward, and improving the decision-making ability of the system through continuous trial and error and learning.

[0174] The system of the present invention further includes an adaptive learning module 7. This module is communicatively connected to the model training module 3 and the prediction analysis module 4 and is responsible for the continuous optimization and performance improvement of the system.

[0175] The adaptive learning module 7 first collects new data generated during the operation of the system, including prediction results, actual complication occurrences, and intervention effects. This data is used to regularly update the model parameters and improve the prediction accuracy. The present invention adopts an online learning algorithm, allowing the model to adapt to new data without completely retraining. For the LSTM model, we use the Stochastic Gradient Descent (SGD) method for incremental learning, and the update formula is as follows:

[0176]

[0177] where θ t is the current model parameter, η is the learning rate, and L(θ t ) is the loss function.

[0178] The adaptive learning module 7 is also responsible for analyzing prediction error cases, identifying the weaknesses of the model, and performing targeted optimization. For example, if the system finds that the prediction accuracy for a certain type of complication is relatively low, it may increase the weight of relevant features or conduct focused training on the data of this type of complication.

[0179] In addition, the system of the present invention dynamically adjusts the model structure or feature weights according to the prediction performance of different types of complications. For example, if it is found that certain physiological indicators are particularly important for predicting specific complications, the system may increase the weights of these indicators in the model or construct a separate sub-model for these indicators.

[0180] Through the above detailed description, we can see that the postoperative complication prediction and management system for endovascular treatment in vascular surgery of the present invention has the characteristics of comprehensiveness, intelligence, and adaptability. The system can not only accurately predict potential risks, but also provide personalized management suggestions and continuously improve its own performance through continuous learning. This innovative method is expected to significantly improve the postoperative management quality of vascular surgery intervention, reduce the complication incidence rate, and improve the patient prognosis.

[0181] The postoperative complication prediction and management system for vascular surgical interventional endovascular therapy of the present invention further includes a knowledge base management module 8. This module is communicatively connected to the comprehensive decision-making module 5, providing professional knowledge support and auxiliary decision-making functions for the entire system.

[0182] The knowledge base management module 8 is first responsible for constructing and maintaining a comprehensive knowledge base related to vascular surgical interventional endovascular therapy. This knowledge base contains multiple sub-modules, covering aspects such as surgical techniques, complication types, treatment plans, and the latest research results. Preferably, the knowledge base of the present invention organizes information using an ontology structure, which can effectively represent complex medical concepts and the relationships between them.

[0183] In an embodiment of the present invention, the core of the knowledge base is a medical knowledge graph based on a graph database. This graph uses nodes to represent medical concepts (such as diseases, symptoms, treatment methods, etc.) and edges to represent the relationships between concepts (such as causing, treating, etc.). For example, the thrombus node may be connected to the anticoagulant therapy node through a treatment edge. This structure allows the system to perform complex semantic queries and inferences.

[0184] The knowledge base management module 8 is also responsible for storing and managing typical cases. Each case is annotated with detailed metadata, including patient basic information, diagnosis results, treatment processes, complication situations, etc. These cases provide valuable reference materials for the system, especially when dealing with rare or complex situations. The present invention uses a content-based recommendation algorithm that can quickly retrieve similar historical cases based on the current patient's situation to provide support for medical decision-making.

[0185] In addition, the knowledge base management module 8 also contains a treatment plan library. This plan library includes not only standard treatment guidelines but also various personalized treatment plans and their implementation effects. The system uses a reinforcement learning algorithm to continuously optimize these plans and adjusts the recommendation weights of the plans according to the actual application effects.

[0186] To ensure the timeliness and accuracy of the knowledge base, the system of the present invention designs an automatic update mechanism. This mechanism uses natural language processing (NLP) technology to regularly extract the latest information from medical literature databases, clinical guidelines, and expert consensuses and integrate it into the existing knowledge base. During the integration process, the system checks the consistency of the new and old information. If conflicts are found, they will be marked for expert review.

[0187] The knowledge base management module 8 provides rule-based auxiliary support for decision-making reasoning. For example, when the system detects that a patient has a certain complication risk, it will automatically query the relevant preventive measures and treatment methods in the knowledge base and provide this information to the comprehensive decision-making module 5 to assist in generating more comprehensive and accurate management suggestions.

[0188] The system of the present invention further includes a patient interaction module 9. This module is communicatively connected to the early warning management module 6, aiming to strengthen the interaction between the system and the patient and improve patient participation and compliance.

[0189] The core of the patient interaction module 9 is a mobile application that patients can access via a smartphone or tablet. The application provides an intuitive and user-friendly interface, enabling patients to conveniently view their health data, receive medical advice, and perform self-management.

[0190] In a preferred embodiment of the present invention, the patient interaction module 9 has the following main functions:

[0191] First, this module can push personalized health education content to patients. The system screens relevant information from the knowledge base based on the patient's specific situation (such as type of surgery, complication risk, etc.) and uses natural language generation (NLG) technology to transform professional medical knowledge into a form that is easy for patients to understand. For example, for a patient who has just undergone vascular stent implantation, the system will push content such as post-stent precautions and guidance on the use of anticoagulant drugs.

[0192] Second, the patient interaction module 9 provides rehabilitation guidance functions. The system generates personalized rehabilitation plans according to the patient's recovery progress, including exercise suggestions, diet guidance, etc. These suggestions take into account the patient's physical condition and complication risk to ensure safety and effectiveness. For example, for patients at risk of bleeding, the system will recommend avoiding strenuous exercise and provide suitable mild exercise programs.

[0193] In addition, this module is also responsible for collecting subjective symptom reports and quality of life assessment data from patients. Patients can regularly fill out standardized questionnaires through the application, such as pain rating scales, quality of life scales, etc. The system uses natural language processing technology to analyze the patient's free text input and extract key information. Combining these subjective data with objective physiological indicators provides a more comprehensive assessment of the patient's condition for the medical team.

[0194] The patient interaction module 9 also provides remote consultation and psychological support services. Patients can communicate with medical staff in writing or via video through the application to promptly solve problems encountered during the postoperative recovery process. For patients detected by the system to have mental health risks, mental health resources and online consultation services will be actively pushed.

[0195] Finally, this module also includes an intelligent reminder system to enhance the patient's treatment compliance. The system automatically generates medication reminders, follow-up appointments, etc. according to medical orders and promptly reminds patients through push notifications. The frequency and method of reminders will be adjusted according to the patient's personal preferences and feedback to achieve the best effect.

[0196] The system of the present invention further includes a multi - center collaboration module 10. This module is communicatively connected to the data acquisition module 1 and the model training module 3, aiming to achieve data sharing and model collaborative optimization among multiple medical institutions.

[0197] The multi - center collaboration module 10 first establishes a secure data sharing mechanism. The present invention adopts a blockchain - based decentralized data sharing framework to ensure data security and privacy protection. Each participating medical institution joins the network as a node, and the data access rights and usage rules are specified through smart contracts.

[0198] During the data sharing process, the system of the present invention adopts homomorphic encryption technology. This technology allows calculations to be performed on encrypted data without decrypting the original data. Specifically, assume there are two medical institutions A and B, which respectively have patient data $x_A$ and $x_B$. Through homomorphic encryption, $f(Ex_A, Ex_B)$ can be calculated, where $E()$ represents the encryption function and $f()$ represents a certain statistical or machine learning operation. In this way, each institution can contribute its data for joint analysis without revealing the original data.

[0199] Another important function of the multi - center collaboration module 10 is to implement federated learning. In this mode, each medical institution does not directly share the original data, but trains the model locally and then only shares the model parameters. The present invention adopts an improved federated averaging (FedAvg) algorithm, and its update formula is as follows:

[0200]

[0201] where, w t +1 is the global model parameter, w t +1 k is the local model parameter of the k - th institution, n k is the number of samples of this institution, and n is the total number of samples. This method allows the system to utilize large - scale datasets from multiple institutions for training while protecting patient privacy.

[0202] In addition, the multi - center collaboration module 10 is also responsible for comparing the prediction performance and management effects of different centers. The system uses standardized evaluation metrics (such as the AUC - ROC curve, calibration plot, etc.) to evaluate the model performance of each center. Through comparative analysis, the system can identify the best practices and promote the popularization of these practices among centers.

[0203] In a preferred embodiment of the present invention, the multi - center collaboration module 10 further includes a Knowledge Distillation mechanism. This mechanism allows the transfer of knowledge from large and complex models trained by multiple institutions to a smaller and more efficient model. Specifically, the system first trains respective teacher models at each center, and then uses the soft labels of these teacher models to train a unified student model. This process can be expressed as:

[0204] L KD =(1 - α)L CE (y,σ(z s / T))+αT 2 L KL (σ(z t / T),σ(z s / T)),

[0205] where L CE is the cross - entropy loss, L KL is the KL divergence, z t and z s are the logits of the teacher model and the student model respectively, T is the temperature parameter, and α is the balancing factor. This method not only improves the generalization ability of the model but also reduces the computational complexity, making the model more suitable for deployment on mobile devices.

[0206] Through the above - detailed description, we can see that the postoperative complication prediction and management system for endovascular treatment in vascular surgery of the present invention has innovative designs in knowledge management, patient interaction, and multi - center collaboration. The addition of these modules enables the system not only to provide accurate predictions and personalized management suggestions but also to continuously learn and optimize, while fully considering the patient experience and the utilization of multi - center data. This comprehensive and innovative approach is expected to significantly improve the postoperative management level of vascular surgery endovascular treatment and provide higher - quality and more personalized medical services for patients.

[0207] To verify the superiority of the postoperative complication prediction and management system for endovascular treatment in vascular surgery of the present invention, we designed a simulation study to compare the performance of the embodiments of the present invention with two comparative examples in actual application scenarios.

[0208] Simulation conditions:

[0209] This study simulated the postoperative management process of the vascular surgery department of a Class - III Grade - A hospital and selected 200 patients who received endovascular treatment as the research subjects. The research period was 30 days after the operation, and three common complications, namely bleeding, thrombosis, and restenosis, were mainly concerned.

[0210] Example 1: The complete system of the present invention is used for postoperative management.

[0211] Comparative Example 1: Traditional manual monitoring and fixed-threshold warning methods are used for postoperative management.

[0212] Comparative Example 2: A simple machine learning model (such as logistic regression) is used for risk prediction, but it does not include real-time monitoring and dynamic adjustment functions.

[0213] Detection indicators and methods:

[0214] 1. Accuracy of complication prediction: The area under the ROC curve (AUC) is used to evaluate the prediction performance of the model.

[0215] 2. Incidence of complications: The number of actually occurring complications within 30 days is recorded and divided by the total number of patients.

[0216] 3. False alarm rate: The number of false warnings is divided by the total number of warnings.

[0217] 4. Missed alarm rate: The number of complication events that failed to be warned is divided by the total number of actually occurring complications.

[0218] 5. Average response time: The average time interval from the system issuing a warning to the medical staff taking intervention measures.

[0219] 6. Patient satisfaction: A Likert 5-point scale is used for questionnaire surveys to evaluate the patient's satisfaction with the management process.

[0220] 7. Work efficiency of medical staff: Questionnaire surveys are used to evaluate the time investment and workload of medical staff in postoperative management.

[0221] Detection results:

[0222] Index Example 1 Comparative Example 1 Comparative Example 2 Accuracy rate of complication prediction (AUC) 0.92 0.71 0.83 Incidence of complications 5.5％ 12％ 8.5％ False alarm rate 8％ 25％ 15％ Missed alarm rate 3％ 18％ 10％ Average response time (minutes) 7.5 25 15 Patient satisfaction (1-5 points) 4.6 3.2 3.8 Improvement of medical staff work efficiency 35％ - 15％

[0223] Analysis and interpretation are as follows:

[0224] 1. Accuracy of complication prediction: The system of the present invention (Example 1) demonstrates significantly superior prediction performance, with an AUC reaching 0.92, far higher than 0.71 of the traditional method (Comparative Example 1) and 0.83 of the simple machine learning model (Comparative Example 2). This indicates that the multi-modal data fusion and deep learning algorithm of the present invention can more accurately capture complex risk factors and their interactions.

[0225] 2. Incidence of complications: The complication incidence rate of Example 1 (5.5%) is significantly lower than that of Comparative Example 1 (12%) and Comparative Example 2 (8.5%). This result proves that the system of the present invention can not only accurately predict risks but also effectively reduce the actual incidence of complications through timely intervention measures.

[0226] 3. False positive rate and false negative rate: The system of the present invention performs excellently in these two indicators. The false positive rate (8%) and false negative rate (3%) are both significantly lower than those of the other two methods. This reflects the advantages of the dynamic threshold adjustment and multi-factor comprehensive analysis of this system, which can maintain a high specificity while improving the sensitivity.

[0227] 4. Average response time: The average response time of Example 1 (7.5 minutes) is much lower than that of the other two methods. This benefits from the real-time monitoring and intelligent early warning functions of the system of the present invention, enabling medical staff to identify and respond to potential risks more quickly.

[0228] 5. Patient satisfaction: The system of the present invention has obtained a high patient satisfaction score (4.6 points), which may be related to the personalized health education, rehabilitation guidance, and convenient remote consultation functions provided by the system.

[0229] 6. Work efficiency of medical staff: Compared with the traditional method, the system of the present invention helps medical staff improve their work efficiency by 35%. This improvement stems from the intelligent and automated characteristics of the system, reducing the time investment of medical staff in daily monitoring and data analysis.

[0230] Based on the above results, Example 1 is considered the best embodiment. It performs excellently in all key indicators, especially in terms of prediction accuracy, reduction of complication incidence, and improvement of medical staff work efficiency.

[0231] These test results fully demonstrate the superiority of the system of the present invention in the prediction and management of postoperative complications in endovascular interventional treatment in vascular surgery. The system not only improves the accuracy of prediction but also effectively reduces the complication incidence through timely and accurate intervention measures. At the same time, the intelligent and personalized characteristics of the system significantly improve patient satisfaction and the work efficiency of medical staff.

[0232] It is particularly noteworthy that the system of the present invention performs excellently in reducing the false positive rate and false negative rate, which is of great significance for clinical practice. A low false positive rate means reducing unnecessary waste of medical resources and patient anxiety, while a low false negative rate ensures that high-risk situations can be detected and handled in a timely manner. The achievement of this balance benefits from the dynamic threshold adjustment and multi-modal data fusion technology of the system.

[0233] In addition, the significantly shortened average response time of the system indicates that it can effectively support the medical team for rapid decision-making and intervention. In the management of postoperative complications in vascular surgery, time is often a key factor, and rapid response may be the key to avoiding serious consequences.

[0234] Generally speaking, these results indicate that the system of the present invention not only achieves innovation at the technical level, but more importantly, it demonstrates significant value in actual clinical applications and has the potential to become a powerful tool for improving the quality of postoperative management in vascular surgery and enhancing patient prognosis.

[0235] It should be noted that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the principles of the present invention shall be included within the protection scope of the present invention.

Claims

1. A system for predicting and managing complications after endovascular interventional treatment of vascular surgery, characterized in that: include: Data acquisition module for: Collect patients' basic data, physiological parameters, biochemical indicators, imaging data, clinical data and disease history data; The patient's postoperative physiological indicators are collected in real time through a variety of sensors and monitoring devices; A data processing module is connected to the data acquisition module for: Receiving multimodal data sent by the data acquisition module; Preprocessing the multimodal data, including data cleaning, feature extraction and feature selection; A model training module is connected to the data processing module for: Based on the preprocessed multimodal data, construct and train a physiological sequence prediction model and an image analysis model; The physiological sequence prediction model includes a physiological feature sequence prediction model based on a recurrent neural network (RNN); The image analysis model includes a medical image analysis model based on a convolutional neural network (CNN); The prediction analysis module is connected to the model training module for: Predicting the patient's postoperative physiological indicators using the physiological sequence prediction model; Analyzing the postoperative medical images of the patient using the image analysis model; Based on the prediction and analysis results, assess the patient's risk of postoperative complications; The comprehensive decision-making module is in communication with the prediction and analysis module and is used to: Integrating the physiological index prediction results and medical image analysis results; Combine the patient's clinical data and medical history to generate a comprehensive risk assessment report; Develop a personalized postoperative management plan based on the risk assessment results; The early warning management module is in communication with the comprehensive decision-making module and is used to: Based on the comprehensive risk assessment report, set dynamic warning thresholds; Monitor patients’ postoperative indicators in real time and trigger warnings when abnormalities are detected; Push early warning information and personalized management suggestions to medical staff.

2. The system according to claim 1, characterized in that The data processing module includes: a data cleaning unit, which is used to: Perform outlier detection and processing on the collected multimodal data; Fill in missing data and unify data formats; A feature extraction unit is connected in communication with the data cleaning unit and is used to: Extract time series features, statistical features, and frequency domain features from the cleaned data; De-noise and enhance medical imaging data; A feature selection unit is communicatively connected to the feature extraction unit and is used to: Use filtering, packaging or embedding methods to filter the extracted features; Reduce feature dimensions and improve the training efficiency and generalization ability of subsequent models.

3. The system according to claim 1, characterized in that The model training module includes: an RNN training unit, which is used to: Construct a physiological feature sequence prediction model based on long short-term memory (LSTM) network; Offline training of the LSTM model using historical patient data; Implement online fine-tuning function to dynamically update model parameters based on current patient data; CNN training unit, used for: Build a multi-layer convolutional neural network model for medical image analysis; Use transfer learning methods and use pre-trained models to speed up the training process; Implement multi-task learning to perform organ segmentation and anomaly detection simultaneously; A model evaluation unit, which is in communication with the RNN training unit and the CNN training unit, and is used to: evaluate model performance using a cross-validation method; Calculate the model's accuracy, sensitivity, and specificity; When the model performance does not meet the preset standards, the model optimization process is triggered.

4. The system according to claim 1, characterized in that The prediction analysis module includes: Physiological indicator prediction unit, used for: Use the trained LSTM model to predict the patient’s key physiological indicators within the next 24 hours; Calculate the deviation between the predicted value and the normal range and quantify the degree of abnormality; Image analysis unit for: Use the trained CNN model to analyze the patient’s postoperative CT, MRI, or X-ray images; Detect and locate possible vascular abnormalities such as stenosis, thrombus, or calcification; The risk assessment unit is in communication with the physiological index prediction unit and the image analysis unit, and is used to: Calculate the probability of different types of complications in patients based on the prediction results of physiological indicators and image analysis results; A multivariate logistic regression model was used to generate a risk score by comprehensively considering the patients' basic characteristics and clinical indicators.

5. The system according to claim 1, characterized in that The comprehensive decision-making module includes: Data fusion unit for: Integrate physiological index prediction results, imaging analysis results and clinical data; Use the attention mechanism to perform weighted fusion of data from different sources; A decision reasoning unit is communicatively connected with the data fusion unit and is used for: Based on the fused data, the complication risk was classified using decision tree or random forest algorithm; Generate interpretable risk assessment reports, including key influencing factors and their weights; A solution generation unit is connected in communication with the decision reasoning unit and is used to: According to the risk assessment results, select a suitable basic plan from the preset management plan library; Personalize the management plan based on the individual characteristics of the patient; Generate comprehensive management recommendations including medication, rehabilitation training, and lifestyle guidance.

6. The system according to claim 1, characterized in that The early warning management module includes: Threshold setting unit, used to: Set initial warning thresholds for different types of complications based on historical data statistics and expert knowledge; Dynamically adjust the warning threshold according to the individual characteristics and real-time status of the patient; A monitoring and early warning unit is connected in communication with the threshold setting unit and is used to: Compare the patient's physiological indicators with preset thresholds in real time; When an anomaly is detected, the warning level is determined based on the degree of deviation; Trigger the corresponding level of early warning mechanism, including sound and light reminders and message push; The intervention management unit is connected in communication with the monitoring and early warning unit and is used to: Generate targeted intervention suggestions based on early warning information; Recording of health care provider interventions and patient responses; Evaluate the effectiveness of interventions and optimize future early warning and management strategies accordingly.

7. The system according to claim 1, characterized in that It also includes an adaptive learning module, which is in communication with the model training module and the prediction analysis module and is used to: Collect new data generated during the operation of the system, including predicted outcomes, actual complication occurrences, and intervention effects; Regularly update model parameters with newly collected data to improve the model’s predictive accuracy; Analyze forecast error cases, identify model weaknesses and perform targeted optimization; Dynamically adjust the model structure or feature weights according to the prediction performance of different types of complications.

8. The system according to claim 1, characterized in that It also includes a knowledge base management module, which is in communication with the comprehensive decision-making module and is used to: Build and maintain a professional knowledge base related to vascular surgery interventional endovascular treatment; Store typical cases, treatment plans and the latest research results; Provide rule-based auxiliary support for decision reasoning; The knowledge base content is updated regularly to ensure that system decisions are based on the latest medical evidence.

9. The system according to claim 1, characterized in that It also includes a patient interaction module, which is communicatively connected with the early warning management module and is used for: Interact with patients in real time via mobile apps or smart devices; Push personalized health education content and rehabilitation guidance to patients; Data on subjective symptom reports and quality of life assessments were collected from patients; Provide remote consultation and psychological support services to enhance patient compliance.

10. The system according to claim 1, characterized in that It also includes a multi-center collaboration module, which is communicatively connected with the data acquisition module and the model training module and is used for: Establish a secure data sharing mechanism among multiple medical institutions; Under the premise of protecting patient privacy, large-scale data sets from different centers are collected; Enable federated learning, allowing centers to jointly train and optimize models without directly sharing raw data; Compare the prediction performance and management effectiveness of different centers to promote the promotion of best practices.

Images