Perioperative management decision support system for lung cancer

Abstract

The application provides a lung cancer perioperative management decision support system, comprising a data acquisition terminal which dynamically acquires multi-source diagnosis and treatment data of a lung cancer patient in a whole cycle of a perioperative period; a management decision device which, according to the multi-source diagnosis and treatment data, constructs a dynamic diagnosis and treatment correlation network for the lung cancer patient, calls a multi-scene decision deduction model, simulates and deduces a whole cycle diagnosis and treatment strategy of the dynamic diagnosis and treatment correlation network, obtains a deduction result, generates multiple groups of candidate management decision schemes in combination with the deduction result and dynamic demands of different diagnosis and treatment scenes, iteratively optimizes the multiple groups of candidate management decision schemes, fuses adaptive advantages of each candidate management decision scheme and corrects potential defects of each candidate management decision scheme, and generates a dynamically adapted individualized management decision scheme; and a medical execution terminal which receives and displays the individualized management decision scheme. In this way, the decision scheme is dynamically adjusted, efficient management in the whole cycle can be realized, and the treatment effect and the rehabilitation process are avoided from being affected.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and in particular to a decision support system for perioperative management of lung cancer. Background Technology

[0002] In the field of lung cancer diagnosis and treatment, perioperative management is crucial, but existing technologies have many shortcomings. Currently, most perioperative management methods process multi-source diagnostic and treatment data in a relatively isolated manner, lacking effective integration and dynamic correlation of preoperative, intraoperative, and postoperative data, making it difficult to comprehensively reflect the continuous changes in the patient's treatment status. Furthermore, decision-making in different treatment scenarios lacks systematic deduction and often relies on experience, making it difficult to accurately adapt to individual patient conditions. Moreover, after the management decision plan is generated, there is a lack of effective dynamic adjustment mechanisms, making it impossible to respond promptly to new changes that arise during the treatment process. This limits the accuracy and individualization of decisions, hindering efficient management throughout the entire cycle and impacting patient treatment outcomes and recovery progress. Summary of the Invention

[0003] In view of the aforementioned problems, this disclosure provides a decision support system for perioperative management of lung cancer, which aims to dynamically adjust decision-making schemes, achieve efficient management throughout the entire cycle, and avoid affecting treatment outcomes and recovery progress.

[0004] In conjunction with embodiments of the present invention, a lung cancer perioperative management decision support system is provided, comprising: a data acquisition terminal, a management decision device communicatively connected to the data acquisition terminal, and a medical execution terminal communicatively connected to the management decision device, wherein the management decision device comprises: a network construction engine, a deduction engine communicatively connected to the network construction engine, and an optimization engine communicatively connected to the deduction engine;

[0005] The data acquisition terminal is used to dynamically collect multi-source diagnosis and treatment data of lung cancer patients throughout the entire perioperative period;

[0006] The network construction engine is used to construct a dynamic diagnosis and treatment association network for the lung cancer patient based on the multi-source diagnosis and treatment data. The dynamic diagnosis and treatment association network uses time-series diagnosis and treatment features as nodes and dynamic diagnosis and treatment influence relationships as edges. The node attributes and the association strength of the edges of the dynamic diagnosis and treatment association network are adjusted in real time as the multi-source diagnosis and treatment data is updated.

[0007] The simulation engine is used to call a multi-scenario decision simulation model to simulate and simulate the full-cycle treatment strategy of the dynamic treatment association network, obtain simulation results, and generate multiple sets of candidate management decision schemes by combining the simulation results with the dynamic needs of different treatment scenarios.

[0008] The optimization engine is used to iteratively optimize the multiple sets of candidate management decision schemes, integrate the adaptation advantages of each candidate management decision scheme and correct the potential defects of each candidate management decision scheme, and generate a dynamically adapted personalized management decision scheme.

[0009] The medical execution terminal is used to receive and display the personalized management decision-making scheme.

[0010] Through the above-described technical solution, this disclosure can achieve at least the following effective effects:

[0011] The data acquisition terminal dynamically collects multi-source diagnosis and treatment data throughout the entire treatment cycle. The network construction engine of the management decision-making device constructs a dynamic diagnosis and treatment correlation network, using time-series diagnosis and treatment characteristics as nodes and dynamic diagnosis and treatment influence relationships as edges. It can adjust node attributes and edge correlation strength in real time as data is updated, effectively integrating and dynamically associating preoperative, intraoperative, and postoperative data to comprehensively reflect the continuous changes in the patient's treatment status. The inference engine of the management decision-making device calls multi-scenario decision-making inference models to simulate and infer the entire treatment cycle, generating multiple sets of candidate management decision-making schemes based on dynamic needs. This overcomes the limitations of relying on experience-based decisions and can accurately adapt to individual patient conditions. The optimization engine of the management decision-making device iteratively optimizes candidate schemes, integrating adaptation advantages and correcting potential defects to generate dynamically adapted personalized management decision-making schemes, solving the problem of lacking a dynamic adjustment mechanism. The medical execution terminal receives and displays the schemes, achieving efficient management throughout the entire cycle, effectively improving the accuracy and individualization of decisions, and significantly improving patient treatment outcomes and recovery processes.

[0012] Other features and advantages of this disclosure will be described in detail in the following detailed description section. Attached Figure Description

[0013] The accompanying drawings are provided to further illustrate the present disclosure and form part of the specification. They are used together with the following detailed description to explain the present disclosure, but do not constitute a limitation thereof. In the drawings:

[0014] Figure 1 This is a system block diagram of a lung cancer perioperative management decision support system shown in an embodiment of this disclosure. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] The specific embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit this disclosure.

[0017] In conjunction with embodiments of the present invention, a decision support system for perioperative management of lung cancer is provided, see [link to relevant documentation]. Figure 1 As shown, the lung cancer perioperative management decision support system 100 includes a data acquisition terminal 110, a management decision device 120 communicatively connected to the data acquisition terminal 110, and a medical execution terminal 130 communicatively connected to the management decision device 120. The management decision device 120 includes a network construction engine 1201, a deduction engine 1202 communicatively connected to the network construction engine 1201, and an optimization engine 1203 communicatively connected to the deduction engine 1202.

[0018] The data acquisition terminal 110 can be a device used to collect various diagnosis and treatment-related data generated by lung cancer patients throughout the perioperative period. One or more data acquisition terminals can establish wireless communication connections with the management and decision-making equipment.

[0019] The medical execution terminal is used to receive personalized management decision plans generated by the management decision equipment and display them to medical personnel in an intuitive way so that medical personnel can implement corresponding diagnosis and treatment measures for patients according to the plan. Similarly, one or more medical execution terminals can establish wireless communication connections with the management decision equipment.

[0020] The data acquisition terminal 110 is used to dynamically collect multi-source diagnosis and treatment data of lung cancer patients throughout the entire perioperative period.

[0021] The multi-source diagnostic and treatment data can include preoperative dynamic assessment data stream, intraoperative real-time operation data stream, postoperative continuous monitoring data stream, and cross-cycle diagnostic and treatment feedback data stream. Each data stream is updated in real time with the progress of diagnosis and treatment and has temporal correlation logic.

[0022] In this embodiment, the data acquisition terminal connects to various medical devices and information systems within the hospital through multiple interfaces and methods. For example, it connects to an electronic medical record system to acquire textual data such as the patient's basic information, medical history, and diagnostic results; and it connects to monitors and testing equipment to collect the patient's vital signs and test results in real time. Simultaneously, it can also collect information such as the patient's subjective feelings and recovery status through mobile medical devices or patient self-reporting. The collected data is stored and transmitted according to certain formats and standards to ensure data integrity and accuracy.

[0023] The network construction engine 1201 is used to construct a dynamic diagnosis and treatment association network for the lung cancer patient based on the multi-source diagnosis and treatment data. The dynamic diagnosis and treatment association network uses time-series diagnosis and treatment features as nodes and dynamic diagnosis and treatment influence relationships as edges. The node attributes and the association strength of the edges of the dynamic diagnosis and treatment association network are adjusted in real time as the multi-source diagnosis and treatment data is updated.

[0024] In this embodiment, the collected multi-source diagnostic and treatment data can be preprocessed, including data cleaning, noise reduction, and standardization, to improve data quality. Then, time-series diagnostic and treatment features are extracted from the processed data. These features can be changes in a patient's physiological indicator at different time points, time series of diagnostic and treatment procedures, etc. These features are used as nodes, and the dynamic diagnostic and treatment influence relationships between them are analyzed, such as whether changes in a certain indicator affect the effectiveness of other indicators or diagnostic and treatment procedures. This influence relationship is then treated as edges.

[0025] Furthermore, an initial dynamic diagnosis and treatment association network is constructed based on nodes and edges. As new multi-source diagnosis and treatment data are continuously collected, the network construction engine updates the node attributes of the network in real time, such as adjusting the node values ​​and states; at the same time, it adjusts the association strength of edges according to new influence relationships, so that the dynamic diagnosis and treatment association network can accurately reflect the patient's current diagnosis and treatment status and the relationships between features.

[0026] The simulation engine 1202 is used to call a multi-scenario decision simulation model to simulate and simulate the full-cycle treatment strategy of the dynamic treatment association network, obtain simulation results, and generate multiple sets of candidate management decision schemes by combining the simulation results and the dynamic needs of different treatment scenarios.

[0027] In this embodiment, the inference engine invokes a pre-built multi-scenario decision-making inference model, which includes various rules and logics for different diagnosis and treatment scenarios. A dynamic diagnosis and treatment correlation network is input into the model, and the model simulates and infers the entire cycle of diagnosis and treatment strategies according to a pre-defined simulation process. During the inference process, the model simulates the execution effects of different strategy steps based on the node attributes and edge association strength of the network, predicting the changing trends of various diagnosis and treatment characteristics. For example, in a preoperative risk control scenario, it simulates the impact of different preoperative preparation strategies on the patient's surgical risk; in a postoperative rapid recovery scenario, it simulates the impact of different rehabilitation measures on the patient's recovery speed, etc.

[0028] The simulation results are obtained, including changes in various diagnostic and treatment characteristics and potential risks. Then, combined with the dynamic needs of different diagnostic and treatment scenarios, such as individual differences among patients and different requirements of surgical types, multiple sets of candidate management decision-making schemes are generated from the simulation results. These schemes cover different combinations of strategies and implementation methods.

[0029] The optimization engine 1203 is used to iteratively optimize the multiple sets of candidate management decision schemes, integrate the adaptation advantages of each candidate management decision scheme and correct the potential defects of each candidate management decision scheme, and generate a dynamically adapted personalized management decision scheme.

[0030] In this embodiment, the optimization engine first evaluates multiple candidate management decision-making schemes according to a preset candidate scheme evaluation index system. The evaluation indicators include scenario adaptability (whether the scheme meets the requirements of a specific treatment scenario); temporal coherence (examining the rationality and coherence of each strategy step in the scheme over time); risk control capability (evaluating the scheme's ability to prevent and handle potential risks); rehabilitation promotion efficiency (measuring the scheme's role in promoting patient recovery speed and effectiveness); and resource adaptability (considering whether the scheme's utilization of medical resources is reasonable). By quantifying the performance of each scheme on each evaluation index, a scheme evaluation score matrix is ​​generated.

[0031] Then, based on the scoring matrix, the advantage evaluation indicators of each group of solutions are identified. Combining the temporal evolution logic of the dynamic diagnosis and treatment association network and the scenario collaboration requirements, the advantage strategies of all solutions are integrated to generate a fusion advantage strategy framework. Then, based on the feature interaction relationships in the dynamic diagnosis and treatment association network, the potential adaptation defects in the fusion advantage strategy framework are analyzed, such as conflicts or ineffective collaboration between certain strategy steps.

[0032] Then, these shortcomings are corrected by adjusting the execution content, timing, or intensity of the strategy steps to obtain an optimized solution. The optimized solution is then input into the candidate solution evaluation index system for evaluation. If it does not meet the preset optimal standard, the iterative optimization process of advantage integration and defect correction is repeated until the optimized solution meets the preset optimal standard, ultimately generating a dynamically adapted personalized management decision solution.

[0033] The medical execution terminal 130 is used to receive and display the personalized management decision-making scheme.

[0034] In this embodiment, the medical execution terminal establishes a communication connection with the management decision-making device to receive personalized management decision-making plans sent by the management decision-making device. The terminal employs an intuitive interface design, displaying the plans to medical personnel in the form of charts, text descriptions, etc. For example, it displays the treatment strategy steps for different time periods in a timeline format, detailing the specific content, timing, and precautions for each step. Medical personnel can implement corresponding treatment operations for patients based on the displayed plans and the patient's actual situation, ensuring that patients receive perioperative treatment and care according to a scientifically sound plan. Simultaneously, the medical execution terminal can also record the execution status of medical personnel and patient feedback information, and feed this information back to the management decision-making device.

[0035] The aforementioned technical solution dynamically collects multi-source diagnosis and treatment data throughout the entire treatment cycle via a data acquisition terminal. The network construction engine of the management decision-making device constructs a dynamic diagnosis and treatment correlation network, using time-series diagnosis and treatment characteristics as nodes and dynamic diagnosis and treatment influence relationships as edges. It can adjust node attributes and edge correlation strength in real time as data is updated, effectively integrating and dynamically associating preoperative, intraoperative, and postoperative data to comprehensively reflect the continuous changes in the patient's treatment status. The management decision-making device's deduction engine calls multi-scenario decision deduction models to simulate and deduce the entire treatment cycle, generating multiple sets of candidate management decision-making schemes based on dynamic needs. This overcomes the limitations of relying on experience-based decisions and can accurately adapt to individual patient conditions. The management decision-making device's optimization engine iteratively optimizes candidate schemes, integrating adaptation advantages and correcting potential defects to generate dynamically adapted personalized management decision-making schemes, solving the problem of lacking a dynamic adjustment mechanism. The medical execution terminal receives and displays the schemes, achieving efficient management throughout the entire cycle, effectively improving the accuracy and individualization of decisions, and significantly improving patient treatment outcomes and recovery progress.

[0036] In a preferred embodiment, the network building engine is specifically used for:

[0037] The multi-source diagnostic and treatment data are analyzed for temporal feature evolution and fused with cross-dimensional information through a pre-trained cross-stage feature interaction model to generate a temporal diagnostic and treatment feature set.

[0038] In this embodiment, the pre-trained cross-stage feature interaction model contains complex neural network structures, such as recurrent neural networks (RNNs) and their variants, long short-term memory networks (LSTMs) or gated recurrent units (GRUs), which are capable of processing time-series data. The model first preprocesses the input multi-source diagnostic data, transforming data of different formats into a unified numerical vector form that the model can process. Then, utilizing the model's time-series processing capabilities, it analyzes the changing trends of the data at different time points and uncovers the evolutionary patterns of time-series features.

[0039] Meanwhile, the model fuses diagnostic and treatment information from different dimensions through nonlinear transformations of multi-layer neural networks, such as comprehensively analyzing patients' clinical symptoms, examination results, and treatment measures. After a series of calculations and processing, the model extracts a set of temporal diagnostic and treatment features from multi-source diagnostic and treatment data that has temporal characteristics and integrates multi-dimensional information. These features can more comprehensively and accurately reflect the changes in the patient's diagnostic and treatment status over time.

[0040] Based on the set of time-series diagnostic and treatment features, extract the core attributes of each time-series diagnostic and treatment feature;

[0041] The core attributes may include the treatment stage corresponding to the time-series diagnosis and treatment features, the types of diagnosis and treatment indicators corresponding to the time-series diagnosis and treatment features, the dynamic change parameters corresponding to the time-series diagnosis and treatment features, and the clinical significance descriptions corresponding to the time-series diagnosis and treatment features.

[0042] In this embodiment, feature analysis algorithms are used to extract core attributes. For example, for numerical features, statistical analysis methods are used to calculate their mean, variance, maximum, minimum, and other statistical measures. These statistical measures can reflect the basic distribution and range of variation of the feature and serve as part of the core attributes. For textual features, such as patient symptom descriptions, natural language processing techniques are used for word frequency statistics, keyword extraction, and other operations to identify key information that represents the symptom features as core attributes. For categorical features, the frequency and distribution of each category are statistically analyzed to determine its main categories and related features as core attributes. In this way, key core attributes can be extracted from each temporal diagnostic feature.

[0043] Each of the time-series diagnostic features is used as a node in the dynamic diagnostic association network. A unique feature identifier is assigned to each node in the dynamic diagnostic association network, and the corresponding core attribute is embedded in the node.

[0044] In this embodiment of the disclosure, when constructing the dynamic diagnosis and treatment association network, each time-series diagnosis and treatment feature is regarded as a node in the network. To accurately distinguish different nodes, a unique feature identifier is assigned to each node. This identifier can be in the form of numerical encoding, a string, etc., ensuring uniqueness throughout the entire network. Then, the extracted corresponding core attributes are embedded into the nodes using a specific data structure, such as key-value pairs, where the name of the core attribute is used as the key and the attribute value is stored internally within the node. In this way, each node not only has a unique identifier but also contains rich core attribute information, enabling a more accurate description of the characteristics and state of the time-series diagnosis and treatment feature.

[0045] The driving influence of the sequential diagnostic and treatment features that occur earlier on the sequential diagnostic and treatment features that occur later on the sequential diagnostic and treatment features that have the same time sequence on the sequential diagnostic and treatment features that occur earlier on the sequential diagnostic and treatment features is extracted to construct the dynamic diagnostic and treatment influence relationship.

[0046] Specifically, the nodes of the perioperative dynamic diagnosis and treatment association network in the dynamic node set are sorted chronologically according to the timestamp identifier, and divided into nodes with earlier chronological order, synchronous nodes, and nodes with later chronological order. The nodes with earlier chronological order correspond to the chronological diagnosis and treatment characteristics of the early diagnosis and treatment stage, the synchronous nodes correspond to the chronological diagnosis and treatment characteristics within the same time window, and the nodes with later chronological order correspond to the chronological diagnosis and treatment characteristics of the later diagnosis and treatment stage.

[0047] In this embodiment of the disclosure, a causal analysis method is used to investigate the influence of chronologically preceding diagnostic and treatment features on chronologically subsequent features. By analyzing the changes in chronologically subsequent features when chronologically preceding features change in historical data, it is determined whether a driving relationship exists. For example, if a patient's preoperative physiological indicator (chronologically preceding feature) changes significantly, and the postoperative recovery speed (chronologically subsequent feature) also changes accordingly, and statistical analysis shows that this change is significant, then a driving effect is considered to exist.

[0048] To investigate the synergistic effects among time-series diagnostic and treatment features with similar timelines, correlation analysis and cluster analysis were employed. Correlation coefficients were calculated between different features; high correlation coefficients indicated a potential synergistic relationship. Simultaneously, clustering algorithms were used to group features with similar trends together to further analyze their synergistic effects.

[0049] To analyze the feedback influence of later-time sequential diagnostic and treatment features on earlier-time features, a feedback model is established. For example, adjusting the preoperative treatment plan based on the postoperative patient's recovery (later-time feature) (earlier-time feature), and comparing simulation and actual data, the effect and degree of this feedback adjustment are analyzed, thereby constructing a comprehensive dynamic diagnostic and treatment influence relationship.

[0050] Based on the dynamic diagnosis and treatment influence relationship, the edges of the dynamic diagnosis and treatment association network are constructed, and the dynamic node set is associated and combined with the edge structure of the dynamic diagnosis and treatment association network to generate an initial diagnosis and treatment association network.

[0051] Specifically, the edge structure of the perioperative dynamic diagnosis and treatment association network is constructed based on the identified dynamic diagnosis and treatment influence relationships. Each edge of the perioperative dynamic diagnosis and treatment association network corresponds to a set of diagnosis and treatment influence relationships between nodes of the perioperative dynamic diagnosis and treatment association network. The attributes of the edges of the perioperative dynamic diagnosis and treatment association network include influence type, influence degree and influence timeliness parameters. The influence degree parameter is quantified and characterized by feature interaction intensity.

[0052] In this embodiment of the disclosure, based on the constructed dynamic diagnosis and treatment influence relationships, the corresponding edge type and attributes are defined for each influence relationship. For example, for a driving influence relationship, the edge attributes may include driving strength, driving direction, etc.; for a cooperative influence relationship, the edge attributes may include cooperative degree, cooperative mode, etc.; for a feedback influence relationship, the edge attributes may include feedback strength, feedback delay, etc.

[0053] Then, based on the type and attributes of these edges, an edge structure is established among the dynamic node set, connecting nodes with corresponding influence relationships. During the connection process, it is ensured that the direction of the edge is consistent with the direction of the influence relationship; for example, the edge driving the influence relationship points from the node with a earlier time sequence to the node with a later time sequence. By associating and combining the dynamic node set with the edge structure of the dynamic diagnosis-treatment association network, a complete initial diagnosis-treatment association network is formed. This network can initially reflect the complex relationships between the perioperative diagnosis-treatment characteristics of lung cancer patients.

[0054] The system acquires the update status of multi-source diagnosis and treatment data in real time, identifies newly added diagnosis and treatment features, and adjusts the number of nodes, the core attributes of the nodes, and the association strength of the edges in the initial diagnosis and treatment association network based on the newly added diagnosis and treatment features, thereby forming a dynamic treatment course association network that evolves dynamically with the diagnosis and treatment process.

[0055] By establishing a real-time communication connection with the data acquisition system, the network construction engine can obtain the update status of multi-source diagnostic and treatment data in real time. When new diagnostic and treatment data is generated, the same feature extraction method as before is used to determine the new diagnostic and treatment features. For new diagnostic and treatment features, it is first determined whether a new node needs to be added to the initial diagnostic and treatment association network. If the new feature is a feature with independent significance that has not appeared before, a new node is added to the network, and a unique feature identifier is assigned to it. At the same time, its core attributes are extracted and embedded into the node.

[0056] For existing nodes, their core attributes are updated based on new data to ensure that node information is consistent with the latest treatment status. Simultaneously, the strength of edges in the dynamic treatment association network is adjusted based on the impact of new features on other features. For example, if a new postoperative complication feature has a new feedback impact on a preoperative risk assessment feature, the feedback strength attribute of the edge connecting these two nodes is adjusted. By continuously adjusting the initial treatment association network based on new treatment features, it can dynamically evolve with the treatment process, always accurately reflecting the relationship between the patient's current treatment status and features.

[0057] In this embodiment of the disclosure, the association strength of the edges of the perioperative dynamic diagnosis and treatment association network is adjusted according to the newly added or updated diagnosis and treatment feature interaction relationship. The edge weights of the perioperative dynamic diagnosis and treatment association network corresponding to the significant influence relationship are enhanced, the edge weights of the perioperative dynamic diagnosis and treatment association network corresponding to the slight influence relationship are weakened, and redundant edges of the perioperative dynamic diagnosis and treatment association network that have no actual impact are eliminated.

[0058] The adjusted perioperative dynamic diagnosis and treatment network is hierarchically optimized according to the preoperative, intraoperative, and postoperative time stages. Within each level, the nodes of the perioperative dynamic diagnosis and treatment network are connected by immediate influence edges, and between levels, the nodes of the perioperative dynamic diagnosis and treatment network are connected by cross-stage influence edges, forming a multi-dimensional network structure that dynamically evolves with the diagnosis and treatment process.

[0059] The aforementioned technical solution utilizes a pre-trained cross-stage feature interaction model to deeply analyze multi-source diagnostic and treatment data, achieving temporal feature evolution and cross-dimensional information fusion. This accurately generates a temporal set of diagnostic and treatment features, laying a solid foundation for network construction. Core attributes are extracted and embedded into nodes, each assigned a unique identifier to ensure complete and distinguishable node information. The constructed dynamic diagnostic and treatment influence relationships comprehensively cover the driving, synergistic, and feedback effects between different temporal diagnostic and treatment features, making the network relationships more aligned with actual diagnostic and treatment logic. Edges constructed based on these relationships are associated and combined with the dynamic node set to form an initial diagnostic and treatment association network with a reasonable structure. Real-time acquisition of data update status identifies newly added diagnostic and treatment features, dynamically adjusting the number of nodes, core attributes, and edge association strength to form a dynamic diagnostic and treatment association network that evolves dynamically with the diagnostic and treatment process. This effectively solves the problems of isolated data and lack of dynamic association in existing technologies.

[0060] In a preferred embodiment, the network building engine is specifically used for:

[0061] Based on the causal logic obtained from the analysis that changes in the sequential diagnostic features in the preceding time series cause changes in the sequential diagnostic features in the following time series, the direction and intensity of the driving influence are determined, and based on the direction and intensity of the driving influence, the driving influence of the sequential diagnostic features in the preceding time series on the sequential diagnostic features in the following time series is determined.

[0062] Among them, time-series diagnostic and treatment features are a series of features that reflect the changes in a patient's diagnostic and treatment status over time, after multi-source diagnostic and treatment data have undergone time-series feature evolution analysis and cross-dimensional information fusion processing and are arranged in chronological order.

[0063] Among them, the driving effect is that when the temporal characteristics of the preceding time sequence change, the temporal characteristics of the following time sequence will also change accordingly. This influence of the preceding characteristics on the following characteristics is the direction and degree of influence.

[0064] In this embodiment, causal analysis algorithms, such as Granger causality tests, are used to analyze time-series diagnostic feature data. By comparing the statistical relationships between different time series data, it is determined whether changes in earlier features can significantly explain changes in later features. If a significant correlation exists, the direction of the driving effect is further determined, i.e., whether the later feature increases or decreases when the earlier feature increases or decreases.

[0065] To determine the strength of the driving effect, quantitative analysis methods are employed, such as calculating correlation coefficients and regression coefficients. The closer the correlation coefficient is to 1 or -1, the stronger the linear relationship between the two features, and the greater the driving effect. The regression coefficient directly reflects the change in the subsequent feature for every unit change in the preceding feature. By combining causal logic judgments and quantitative analysis results, the driving effect of the preceding time-series diagnostic and treatment features on the subsequent time-series features can be accurately determined.

[0066] Based on the interaction between different time-series diagnostic and treatment features within the same time window obtained from the analysis, the synergistic promotion or mutual constraint relationship between the time-series diagnostic and treatment features is determined, and based on the synergistic promotion or mutual constraint relationship, the synergistic influence between the time-series diagnostic and treatment features with the same time sequence is determined;

[0067] Among them, synergistic influence refers to the interaction between different temporalized diagnosis and treatment characteristics within the same time window, which presents a synergistic promotion or mutual constraint relationship.

[0068] In this embodiment of the disclosure, correlation analysis and cluster analysis are performed on different time-series diagnostic and treatment features within the same time window. Correlation analysis calculates the correlation coefficient between features to determine the degree of linear correlation. If the correlation coefficient is positive and significant, it indicates a synergistic relationship between the features; if the correlation coefficient is negative and significant, it may indicate a mutually restrictive relationship. Cluster analysis divides features into different categories based on their similarity. Features within the same category may have similar trends, further verifying the synergistic relationship between them.

[0069] Simultaneously, considering the nonlinear interactions between features, machine learning algorithms, such as decision trees and neural networks, are used to uncover complex synergistic or constraint patterns among features. By synthesizing the results of multiple analytical methods, the synergistic promotion or mutual constraint relationships between time-series diagnostic and treatment features with the same time sequence are accurately determined, thereby identifying their synergistic effects.

[0070] Based on the evaluation feedback effect of the later-order sequential diagnosis and treatment features on the earlier-order sequential diagnosis and treatment features obtained from the analysis, the basis for verifying or correcting the accuracy of the evaluation of the earlier-order sequential diagnosis and treatment features can be determined by the diagnosis and treatment results corresponding to the later-order sequential diagnosis and treatment features.

[0071] In this embodiment, actual treatment result data corresponding to later-ordered time-series diagnostic features and initial assessment data corresponding to earlier-ordered time-series diagnostic features are collected. The actual treatment results are compared and analyzed with the initial assessment. If the actual results match the assessment results, it indicates that the earlier-ordered feature assessment has higher accuracy; if there is a deviation, the cause of the deviation is further analyzed.

[0072] By establishing a feedback model, the impact of various factors on assessment accuracy is considered, such as individual patient differences and unexpected situations during treatment. Statistical analysis methods, such as analysis of variance and regression analysis, are used to determine the degree and direction of the influence of later-time treatment outcomes on the accuracy of earlier-time feature assessments.

[0073] If the actual results are better than the assessment results, it may indicate that the assessment of the earlier-time features was too conservative and needs to be revised upwards; conversely, it may need to be revised downwards. In this way, the basis for verifying or correcting the accuracy of the assessment of the earlier-time features is determined by the treatment results corresponding to the later-time sequential diagnostic and treatment features.

[0074] Based on the validity results corresponding to the verification and the correction direction corresponding to the correction basis, determine the feedback influence of the later-ordered sequential diagnosis and treatment features on the earlier-ordered sequential diagnosis and treatment features;

[0075] Among them, feedback impact refers to the impact generated by the verification or correction of the accuracy of the assessment of the sequential diagnosis and treatment characteristics corresponding to the later time sequence diagnosis and treatment characteristics, which verify or correct the accuracy of the assessment of the earlier time sequence diagnosis and treatment characteristics.

[0076] In this embodiment of the disclosure, the validity results corresponding to the verification are quantitatively evaluated, for example, by calculating the degree of improvement or decrease in evaluation accuracy. Based on the correction direction corresponding to the correction criteria, the magnitude and direction of the feedback impact are determined. If the correction direction is positive, meaning that the evaluation accuracy of features with earlier time sequences needs to be improved, and the verification results show that there is room for improvement in evaluation accuracy, then the feedback impact is positive, and its magnitude depends on the degree and probability of improvement. Conversely, if the correction direction is negative, the feedback impact is negative.

[0077] Meanwhile, considering the timeliness of feedback effects, the feedback effect of treatment outcomes at different stages on earlier feature assessments may vary. For example, short-term postoperative complications may provide a more direct and strong feedback on preoperative risk assessments, while long-term recovery effects may provide a relatively weaker and somewhat delayed feedback. Taking into account effectiveness results, correction directions, and timeliness, mathematical models, such as exponential and linear models, can be used to accurately determine the feedback impact of later-time sequential treatment features on earlier-time sequential treatment features.

[0078] Based on the timeliness and intensity of the effects of the driving effects, the synergistic effects, and the feedback effects, the values ​​of the corresponding influence factors are adjusted, and the dynamic diagnosis and treatment influence relationship is constructed based on the values ​​of the influence factors.

[0079] Among them, the impact factor is a parameter used to quantify the magnitude and timing of driving impact, synergistic impact and feedback impact. By adjusting its value, a dynamic diagnosis and treatment impact relationship can be constructed.

[0080] In this embodiment, the driving effect, synergistic effect, and feedback effect are analyzed for their respective duration and intensity. The duration of the effect can be determined through time series analysis, such as calculating the duration of the effect and the delay in its occurrence. The intensity of the effect is evaluated based on quantitative indicators such as correlation coefficients, regression coefficients, and feedback magnitude determined in the preceding steps. The corresponding influence factor values ​​are adjusted according to the analysis results of the duration and intensity of the effect. For example, a larger influence factor value is assigned to an effect with a longer duration and greater intensity; a smaller influence factor value is assigned to an effect with a shorter duration and weaker intensity.

[0081] Simultaneously, considering the interactions and cumulative effects among different influences, system dynamics methods or complex network models are employed to substitute the adjusted values ​​of influence factors into the model, constructing a dynamic relationship between diagnosis and treatment influences. This relationship can continuously update and adjust the values ​​of influence factors based on new data as the diagnosis and treatment process progresses, thereby reflecting the complex dynamic relationships between time-series diagnosis and treatment characteristics in real time.

[0082] The aforementioned technical solutions precisely analyze the complex relationships between time-series diagnostic and treatment features. By analyzing the causal logic of feature changes before and after the time sequence, the direction and intensity of driving influences are clarified, thereby accurately identifying the driving influences and providing a basis for understanding the causal context of the diagnostic and treatment process. In-depth exploration of the interactions between different features within the same time window identifies synergistic or mutually restrictive relationships, thus clarifying synergistic effects and contributing to a comprehensive understanding of the synergistic mechanisms among diagnostic and treatment features.

[0083] Simultaneously, analyzing the feedback effect of later-stage features on earlier assessments helps determine the basis for verification or correction, and thus the impact of feedback can effectively correct biases in early diagnostic and treatment decisions. Furthermore, by adjusting the values ​​of influencing factors based on the timeliness and intensity of each influence, a dynamic diagnostic and treatment influence relationship is constructed, making the relationship more closely aligned with the dynamic changes in actual diagnosis and treatment. This overcomes the limitation of existing technologies that lack systematic extrapolation in decision-making across different diagnostic and treatment scenarios, laying the foundation for constructing a scientifically sound and rational dynamic diagnostic and treatment correlation network, significantly improving the accuracy and individualization of decision-making, and enhancing patient treatment outcomes and recovery processes.

[0084] In a preferred embodiment, the network building engine is specifically used for:

[0085] The preoperative dynamic assessment data stream, intraoperative real-time operation data stream, postoperative continuous monitoring data stream, and cross-cycle diagnosis and treatment feedback data stream in the multi-source diagnosis and treatment data are each time-series sliced ​​to obtain multiple diagnosis and treatment data segments with timestamps.

[0086] Time-series slicing divides a continuous data stream into multiple segments with specific time intervals, based on the time dimension. Each segment is marked with a timestamp to indicate the data's position on the timeline.

[0087] In this embodiment of the disclosure, the time range and sampling frequency of diagnostic and treatment data streams from different sources are clearly defined. For preoperative dynamic assessment data streams, appropriate slice intervals are set according to the assessment time nodes and frequency, such as slices every half hour or hour. For intraoperative real-time operation data streams, due to their real-time nature, slices can be taken according to key operation nodes or fixed time intervals (such as every 5 minutes). For postoperative continuous monitoring data streams, slices are taken, such as every 1 hour or every 2 hours, depending on the importance and frequency of change of the monitored indicators. For cross-cycle diagnostic and treatment feedback data streams, slices are taken according to the feedback time period.

[0088] Then, using data processing algorithms, data segments are sequentially extracted from the start time of the data stream according to the set slicing intervals. During the extraction process, a precise timestamp is added to each segment, accurate to the second or even millisecond, to ensure accurate recording of the data segment's position on the timeline. In this way, the continuous multi-source medical data stream is transformed into multiple discrete medical data segments with timestamps.

[0089] The diagnostic and treatment data segments are input into the cross-stage feature interaction model to extract the temporal evolution features of each diagnostic and treatment data segment. The temporal evolution features include the changing trend of the diagnostic and treatment indicators, the changing rate of the diagnostic and treatment indicators, and the description of the correlation between the changes of the diagnostic and treatment indicators.

[0090] In this embodiment, the diagnostic and treatment indicator data in each segment are first preprocessed, including data cleaning (removing outliers and noisy data) and data standardization (converting data of different dimensions into a unified standard). Then, for the extraction of the trend, the model uses trend analysis algorithms, such as moving average and exponential smoothing, to fit the changes of the diagnostic and treatment indicators over time, thereby determining whether they show an upward, downward, or fluctuating trend.

[0091] To calculate the rate of change, the model calculates the ratio of the difference between diagnostic and treatment indicators between adjacent time points to the time interval, thus obtaining the amount of change per unit time and measuring the rate of change. When analyzing the correlation of changes, the model employs correlation analysis methods, such as calculating the Pearson correlation coefficient and Spearman's rank correlation coefficient, to determine whether linear or nonlinear relationships exist between different diagnostic and treatment indicators, and the strength of these relationships. In this way, the model can accurately extract the temporal evolution characteristics of the diagnostic and treatment indicators, including their changing trends, rates of change, and descriptions of their correlations, from each segment of diagnostic and treatment data.

[0092] Based on the information overlap between the time-series evolution features aligned on the time axis, cross-dimensional correlation and fusion are performed on the non-redundant time-series evolution features to construct the interaction mapping relationship between the non-redundant time-series evolution features and generate a preliminary feature interaction set.

[0093] Information overlap measures the similarity of information contained between different temporal evolution features. Calculated using a specific algorithm, its value typically ranges from 0 to 1, with higher values ​​indicating greater overlap. Cross-dimensional correlation fusion integrates non-redundant temporal evolution features from different dimensions (such as different treatment stages or indicators) to establish connections between them, providing a more comprehensive reflection of the treatment process. Interaction mapping describes the relationships of interaction and influence between different temporal evolution features, revealing how changes in one feature trigger changes in others.

[0094] In this embodiment, the extracted temporal evolution features are aligned along a timeline to ensure temporal comparability of features from different stages and with different indicators. Then, information overlap calculation algorithms, such as those based on cosine similarity or Jaccard similarity coefficients, are used to calculate the information overlap between each pair of temporal evolution features. A suitable threshold is set; when the information overlap is below this threshold, the two features are considered non-redundant. For non-redundant temporal evolution features, association rule mining algorithms, such as the Apriori algorithm and FP-growth algorithm, are used to analyze their potential associations.

[0095] By mining frequent itemsets and association rules, significant interactions and influences between features are identified. Based on the mining results, a non-redundant interaction mapping relationship between temporally evolving features is constructed. This mapping relationship can be presented in the form of charts, matrices, or rules, clearly showing the interaction patterns between different features. Finally, all interaction mapping relationships are integrated to generate a preliminary feature interaction set.

[0096] The following steps are executed in a loop: based on the temporal evolution characteristics corresponding to the newly received multi-source diagnosis and treatment data streams, the preliminary feature interaction set is incrementally updated, new interaction mapping relationships are added, and the temporalized diagnosis and treatment feature set is generated.

[0097] In this embodiment of the disclosure, upon receiving a new data stream, temporal slicing and feature extraction are performed according to the preceding steps to obtain newly added temporal evolution features. Then, these newly added temporal evolution features are compared and analyzed with the features in the preliminary feature interaction set.

[0098] First, the information overlap between the newly added feature and existing features in the set is checked. If the information overlap between the newly added feature and existing features is below a set threshold, it is considered non-redundant. For non-redundant new features, the association rule mining algorithm mentioned earlier is used to analyze its potential interaction relationships with other features in the set. Based on the analysis results, the preliminary feature interaction set is incrementally updated, adding new interaction mapping relationships to the set.

[0099] Simultaneously, existing interaction mapping relationships that may change due to new data are re-evaluated and adjusted to ensure their accuracy and timeliness. By continuously repeating this step, the initial feature interaction set is constantly supplemented and improved as new data continues to flow in, ultimately generating a time-seriesd set of diagnostic and treatment features that comprehensively and accurately reflects the changes in the diagnostic and treatment process over time.

[0100] The aforementioned technical solution first performs time-series slicing on different data streams from multi-source diagnostic and treatment data to obtain time-stamped data fragments, making the data more organized and analyzable. These data fragments are then input into a cross-stage feature interaction model to extract time-series evolution features, covering the trends, rates, and correlations of changes in diagnostic and treatment indicators, comprehensively and accurately capturing key information from the diagnostic and treatment process. Based on the information overlap of features after time-axis alignment, non-redundant features are cross-dimensionally correlated and fused to construct an interaction mapping relationship, generating a preliminary feature interaction set. This effectively integrates scattered data features and avoids information redundancy.

[0101] By iteratively executing steps and based on the temporal evolution characteristics corresponding to the newly added multi-source diagnostic and treatment data streams in real time, the initial feature interaction set is incrementally updated, and a dynamic update mechanism for generating a temporal diagnostic and treatment feature set by supplementing new interaction mapping relationships is established. This ensures that the system can reflect the continuous changes in the patient's diagnostic and treatment status in a timely manner, providing accurate, comprehensive and real-time data support for the subsequent construction of a dynamic diagnostic and treatment association network, and greatly improving the scientific nature and accuracy of decision-making.

[0102] In a preferred embodiment, the network building engine is specifically used for:

[0103] Based on the dimensions of diagnostic and treatment indicators, diagnostic and treatment behavior, and diagnostic and treatment results, the non-redundant temporal evolution features are classified into multiple dimensional feature subsets.

[0104] The diagnostic and treatment indicator dimension includes quantitative indicators used to measure the patient's health status and treatment effectiveness, such as physiological indicators like blood pressure, blood sugar, and heart rate, as well as pathological indicators like tumor size and inflammation severity. The diagnostic and treatment behavior dimension records the various procedures and behaviors performed on the patient by medical staff during the medical process, such as surgical procedures, medication regimens, and examinations. The diagnostic and treatment outcome dimension reflects the patient's final state after treatment, including clinical outcomes such as cure, improvement, no cure, and death, as well as recovery time and the degree of improvement in quality of life.

[0105] Among them, non-redundant temporal evolution features are key features that reflect changes in diagnostic and treatment indicators over time after screening and removing duplicate or highly similar information. These features have unique information value at different points in time.

[0106] In this embodiment, the specific definitions and scope of the three dimensions—treatment indicators, treatment behaviors, and treatment outcomes—are clearly defined. For non-redundant temporal evolution features, their respective categories are analyzed one by one. For example, if a feature reflects the change in a patient's blood pressure over time, since blood pressure is a treatment indicator, this feature is categorized under the treatment indicator dimension. If a feature describes the change in the frequency of a certain procedure during surgery over time, since surgical procedures are treatment behaviors, this feature is categorized under the treatment behavior dimension. If a feature reflects the trend of a patient's recovery time, and recovery time is related to treatment outcomes, this feature belongs to the treatment outcome dimension. In this way, all non-redundant temporal evolution features can be accurately assigned to their corresponding dimensions, forming multiple feature subsets with clearly defined dimensional attributes.

[0107] Extract the dynamic change pattern of the temporal evolution feature representation within each of the dimensional feature subsets, and generate a core feature vector based on the dynamic change pattern;

[0108] Among them, the core feature vector is a set of numerical vectors used to highly summarize and represent the dynamic change law of the dimensional feature subset.

[0109] In this embodiment, time series analysis methods, such as Autoregressive Moving Average (ARMA) and Autoregressive Integrated Moving Average (ARIMA), are used to analyze the temporal evolution characteristics of each dimensional feature subset. These methods can uncover the data's trends, seasonality, and periodicity over time. For example, for the blood pressure feature subset of the diagnostic indicator dimension, analysis can reveal the daily fluctuation patterns and long-term trends of blood pressure. After extracting the dynamic patterns, feature extraction algorithms, such as Principal Component Analysis (PCA) and Independent Component Analysis (ICA), are used to extract key information from the complex dynamic patterns and generate core feature vectors. These core feature vectors can summarize the main dynamic characteristics of the dimensional feature subset in a concise numerical form, reducing the data dimensionality.

[0110] Based on the core feature vector, analyze the correlation between each pair of dimensional feature subsets, and based on the correlation, determine the driving logic of the change of the temporal evolution feature corresponding to one of the two dimensional feature subsets on the temporal evolution feature corresponding to the other, and determine the first interactive mapping relationship between the two dimensional feature subsets.

[0111] In this embodiment, correlation analysis methods, such as calculating the Pearson correlation coefficient and Spearman rank correlation coefficient, are used to analyze the core feature vectors of different dimensional feature subsets to measure the degree of linear or nonlinear correlation between them. If the core feature vectors of two dimensional feature subsets are highly correlated, causal analysis methods, such as the Granger causality test, are further applied to determine whether a causal relationship exists between them. For example, analyzing the core feature vectors of the diagnostic behavior dimension (such as medication regimen) and the diagnostic indicator dimension (such as blood pressure), if a causal relationship is found between the adjustment of the medication regimen and the change in blood pressure, further investigation is conducted into how changes in factors such as which drugs, dosages, or durations of medication in the regimen drive the corresponding changes in blood pressure. Through this detailed analysis, the first interaction mapping relationship between the two dimensional feature subsets is determined, clarifying the interaction mechanism between different dimensional features.

[0112] In this embodiment of the disclosure, the relationship between the feature subset of the diagnostic and treatment indicator dimension and the feature subset of the diagnostic and treatment behavior dimension is analyzed, the driving logic of the temporal evolution feature changes of the feature subset of the diagnostic and treatment indicator dimension on the temporal evolution feature of the feature subset of the diagnostic and treatment behavior dimension is identified, and the interactive mapping relationship between the feature subset of the diagnostic and treatment indicator dimension and the feature subset of the diagnostic and treatment behavior dimension is constructed.

[0113] This study analyzes the relationship between the feature subsets of the diagnosis and treatment behavior dimension and the feature subsets of the diagnosis and treatment outcome dimension, identifies the influence path of the temporal evolution characteristics of the feature subsets of the diagnosis and treatment behavior dimension on the temporal evolution characteristics of the feature subsets of the diagnosis and treatment outcome dimension, and constructs the interactive mapping relationship between the feature subsets of the diagnosis and treatment behavior dimension and the feature subsets of the diagnosis and treatment outcome dimension.

[0114] This study analyzes the correlation between the feature subsets of the diagnosis and treatment outcome dimension and the feature subsets of the diagnosis and treatment indicator dimension, identifies the correction logic of the temporal evolution feature feedback of the feature subset of the diagnosis and treatment outcome dimension on the assessment of the temporal evolution feature subset of the diagnosis and treatment indicator dimension, and constructs the interactive mapping relationship between the feature subsets of the diagnosis and treatment outcome dimension and the feature subsets of the diagnosis and treatment indicator dimension.

[0115] Based on the core feature vector, analyze the collaborative correlation between different temporal evolution features within the same dimensional feature subset, and determine the second interactive mapping relationship between different temporal evolution features within the same dimensional feature subset;

[0116] In this embodiment of the disclosure, for different temporal evolution features within the same feature subset of the same dimension, correlation analysis is first performed to identify feature pairs that are related. Then, clustering analysis methods, such as K-means clustering and hierarchical clustering, are used to group features with similar dynamic change patterns into one class. By analyzing the interrelationships between features within the same class and the differences between features in different classes, the synergistic relationships between them are determined. For example, in the dimension of diagnostic indicators, if the temporal evolution features of systolic blood pressure and diastolic blood pressure are clustered into one class, it indicates that they have a synergistic relationship during the change process. Further analysis can reveal the consistency or complementarity of their change magnitude, change time, etc., thereby determining the second interactive mapping relationship between different temporal evolution features within the same feature subset of the same dimension, revealing the interaction mode of features within the same dimension.

[0117] Based on the first interaction mapping relationship and the second interaction mapping relationship, a feature interaction matrix is ​​constructed, and based on the feature interaction matrix, the temporal evolution feature combination corresponding to the core interaction relationship is extracted to generate a preliminary feature interaction set.

[0118] In this embodiment, the temporal evolution features within all dimensional feature subsets are used as rows and columns, and the values ​​of matrix elements are determined based on the first and second interaction mapping relationships. For example, if a first or second interaction mapping relationship exists between two features, the corresponding matrix element is assigned a larger value, indicating a strong interaction between them; if no interaction relationship exists, a smaller value or zero is assigned. This method constructs a feature interaction matrix, intuitively displaying the interaction strength and relationship type between different features. Then, feature selection algorithms, such as those based on information gain or the Gini index, are used to extract the temporal evolution feature combinations corresponding to the core interaction relationships from the feature interaction matrix. These core interaction relationships are those that have a significant impact on the outcome and are closely intertwined throughout the diagnosis and treatment process. Combining their corresponding features generates a preliminary feature interaction set.

[0119] The aforementioned technical solution first classifies non-redundant time-series evolution features according to diagnostic indicators, behaviors, and outcomes, resulting in multiple dimensional feature subsets. This is akin to precisely categorizing complex data, laying a clear framework for subsequent analysis. Next, it extracts the dynamic change patterns of features within each dimensional feature subset and generates core feature vectors, deeply exploring the inherent patterns of the data and highlighting key information. By analyzing the correlations between feature subsets of different dimensions using core feature vectors, the driving logic is determined, deriving the first interaction mapping relationship, which clearly presents the causal connections between different dimensions. Simultaneously, it analyzes the synergistic correlations of different features within the same dimension, determining the second interaction mapping relationship, accurately grasping the interaction between features within the same dimension.

[0120] Furthermore, a feature interaction matrix is ​​constructed based on the first and second interaction mapping relationships, and the temporal evolution feature combinations corresponding to the core interaction relationships are extracted to generate a preliminary feature interaction set, comprehensively integrating feature interaction information from different dimensions and within the same dimension. This effectively solves the problems of isolated data and lack of correlation analysis in existing technologies, providing accurate basis for constructing dynamic diagnosis and treatment correlation networks, and greatly improving the accuracy and individualization of decision-making.

[0121] In a preferred embodiment, the inference engine is specifically used for:

[0122] The dynamic diagnosis and treatment association network is input into the multi-scenario decision inference model, and the dynamic node attributes, edge association strength, and time sequence hierarchy of the nodes in the dynamic diagnosis and treatment association network are extracted to generate a network dynamic feature descriptor.

[0123] Among them, the network dynamic feature descriptor is used to summarize and describe the feature set of dynamic characteristics of nodes and edges in the dynamic diagnosis and treatment association network, including the dynamic attributes of nodes, the association strength of edges, and the temporal hierarchical structure of the network.

[0124] In this embodiment of the disclosure, the nodes in the network are first traversed. For each node, time series analysis methods, such as moving average and exponential smoothing, are used to analyze the changes in its attribute values ​​at different time points, thereby extracting dynamic node attributes. These attributes may include dynamic information over time, such as the patient's vital signs and the degree of disease progression.

[0125] Next, for the edges in the network, the correlation strength of the edges is quantified by calculating indicators such as correlation coefficients and mutual information between nodes, reflecting the closeness of the association between different diagnostic and treatment elements. For the temporal hierarchical structure, a hierarchical clustering algorithm or a time window-based hierarchical analysis method is used to divide the network into different levels according to the characteristics and relationships of nodes and edges in different time periods, clarifying the inclusion and association relationships between each level. Combining the extracted dynamic node attributes, edge correlation strength, and temporal hierarchical structure, a dynamic feature descriptor for the network is generated.

[0126] The network dynamic feature descriptor is compared with the scenario adaptation conditions in the multi-scenario strategy library to identify the core diagnosis and treatment scenarios for the lung cancer patient. The multi-scenario strategy library includes preoperative risk prevention and control scenarios, intraoperative precision operation scenarios, postoperative rapid recovery scenarios, and emergency treatment scenarios for complications.

[0127] In this embodiment, adaptation conditions for each core diagnosis and treatment scenario are obtained from a multi-scenario strategy library. These conditions are typically a set of thresholds or rules regarding node attributes, edge association strength, and temporal hierarchical structure. The generated network dynamic feature descriptor is then compared with these adaptation conditions one by one. For example, in a preoperative risk control scenario, adaptation conditions might include certain physiological indicators of the patient (such as blood pressure and blood sugar) needing to be within a certain range, and the association strength of specific edges related to the disease needing to be lower than a certain threshold.

[0128] By comparing the data in the network dynamic feature descriptor with these conditions, it is determined which core treatment scenario the current lung cancer patient's condition falls under. If the conditions for a preoperative risk control scenario are met, the patient is identified as being in a preoperative risk control scenario.

[0129] An initial strategy is generated by combining the basic management strategy corresponding to the core diagnosis and treatment scenario, the dynamic node attributes and edge association strength corresponding to the dynamic diagnosis and treatment association network;

[0130] In this embodiment, the basic management strategy corresponding to the scenario is retrieved from a multi-scenario strategy library. The basic management strategy provides a fundamental framework and principles for the diagnosis and treatment process. For example, in a rapid postoperative recovery scenario, the basic management strategy might include guiding principles for early mobilization and nutritional support. Then, the dynamic node attributes and edge association strength information in the dynamic diagnosis and treatment association network are integrated into the basic management strategy. For instance, based on the patient's current vital signs (dynamic node attributes) and the association strength between different diagnosis and treatment stages (edge ​​association strength), the activity time and intensity, and the dosage of nutritional support in the basic management strategy are individually adjusted. In this way, an initial strategy that meets the basic requirements of the core diagnosis and treatment scenario and adapts to the patient's current dynamic condition can be generated.

[0131] Extract the changing patterns of node attributes over time, the logic for adjusting the association strength of edges, and the interaction evolution paths between levels from the dynamic diagnosis and treatment association network to generate temporal evolution logic;

[0132] Among them, the temporal evolution logic is used to describe the logical system of the changes in node attributes over time, the adjustment of edge association strength, and the paths and rules of interaction evolution between different levels in the dynamic diagnosis and treatment association network.

[0133] In this embodiment, trend analysis methods from time series analysis, such as linear trend analysis and nonlinear trend analysis, are used to fit the data on the changes in node attributes over time in the dynamic diagnosis-treatment association network, extracting the changing patterns of node attributes, such as whether a patient's indicator shows an upward trend, a downward trend, or fluctuates. Regarding the edge association strength adjustment logic, by analyzing the changes in edge association strength at different time points, combined with the changes in node attributes, causal analysis methods, such as Granger causality tests, are used to determine which factors lead to the adjustment of edge association strength, thereby clarifying the edge association strength adjustment logic.

[0134] For the interaction and evolution paths between levels, network dynamics analysis is used to study how the interactions between nodes and edges at different levels drive the evolution of the network structure over time, thus determining the interaction and evolution paths between levels. Combining these three aspects, a temporal evolution logic is generated, which can comprehensively describe the dynamic changes and patterns of the dynamic diagnosis-treatment association network in the time dimension.

[0135] Based on the aforementioned temporal evolution logic, the initial strategy is simulated throughout its entire execution process. The dynamic changes of various diagnostic and treatment features and possible interactive feedbacks during the execution of the initial strategy are predicted, and the simulation results are obtained.

[0136] In this embodiment, a simulation model is constructed based on temporal evolution logic, and an initial strategy is input into the model. The simulation model gradually simulates the execution process of the initial strategy throughout the entire cycle, following the node attribute change patterns, edge association strength adjustment logic, and inter-level interaction evolution paths described by the temporal evolution logic. During the simulation, the dynamic changes of various diagnostic and treatment characteristics, such as the patient's vital signs and treatment effect indicators, are tracked in real time.

[0137] Simultaneously, the simulation considers the interactive feedback between different diagnostic and treatment features. For example, the implementation of a certain treatment measure may affect the patient's physiological indicators, thereby affecting the effectiveness of other treatment measures. Through this comprehensive simulation, the dynamic changes of each diagnostic and treatment feature and possible interactive feedback situations during the execution of the initial strategy are predicted, ultimately yielding the simulation results. The simulation results can provide decision-making references for medical staff, helping them to assess the feasibility and effectiveness of the initial strategy and adjust and optimize the diagnostic and treatment strategy in a timely manner.

[0138] The aforementioned technical solution first inputs the dynamic diagnosis-treatment association network into a multi-scenario decision-making model, accurately extracting node attributes, edge association strength, and time-series hierarchical structure to generate a dynamic feature descriptor that clearly presents the network's key features. Comparing this dynamic feature descriptor with a multi-scenario strategy library quickly identifies core diagnosis-treatment scenarios for lung cancer patients, covering key aspects such as preoperative, intraoperative, postoperative, and complication management, ensuring comprehensive and targeted decision-making. Finally, combining the basic management strategies for the core diagnosis-treatment scenarios with the network's dynamic features, an initial strategy is generated.

[0139] Furthermore, the system extracts the patterns of node attribute changes, edge association strength adjustment logic, and hierarchical interaction evolution paths from the network to generate temporal evolution logic, providing in-depth insights into the dynamic trends of the diagnosis and treatment process. Based on this, the initial strategy is simulated throughout its entire lifecycle, accurately predicting the dynamic changes and interactive feedback of various diagnosis and treatment characteristics, and obtaining the deduction results. This effectively solves the problems of existing technology decision-making lacking systematic deduction and difficulty in accurately adapting to individual circumstances, providing strong support for generating scientific and reasonable personalized management decision-making solutions, and significantly improving the perioperative management effect of lung cancer.

[0140] In a preferred embodiment, the inference engine is specifically used for:

[0141] The initial strategy is decomposed into multiple sequential execution steps, each sequential execution step corresponding to a time window in the perioperative period, and each sequential execution step includes execution content and execution conditions;

[0142] The sequential execution steps are specific execution units that break down the initial strategy according to the time sequence of the perioperative period. Each step corresponds to a specific time window, specifying the execution content and conditions within that window.

[0143] In this embodiment of the disclosure, multiple time windows are divided according to the perioperative timeline, each window corresponding to a specific stage, such as the preoperative preparation stage, the intraoperative procedure stage, and the postoperative recovery stage. For each time window, specific content to be performed in that stage is extracted from the initial strategy, such as preoperative examination items, intraoperative anesthesia methods and surgical procedure steps, and postoperative rehabilitation training and medication plans.

[0144] At the same time, the execution conditions for each task are clearly defined. For example, some examinations need to be performed when the patient's physical condition is stable, and some surgical procedures need to be carried out under specific equipment conditions. Through this breakdown, the initial strategy is transformed into a series of sequential execution steps with clear time and condition requirements.

[0145] Using the current state of the dynamic diagnosis and treatment association network as the initial simulation state, the execution process of the initial strategy is simulated step by step according to the order of the temporal execution steps. After each round of the temporal execution steps, the dynamic node attributes and edge association strength in the dynamic diagnosis and treatment association network are updated based on the temporal evolution logic.

[0146] In this embodiment, the current state of the dynamic diagnosis-treatment association network is used as the starting point for the simulation. This state reflects the patient's various diagnosis-treatment characteristics at the current moment and the relationships between them. The execution of each step is simulated sequentially according to the order of the time-sequential execution steps.

[0147] During each step, the dynamic node attributes in the dynamic diagnosis and treatment association network are updated based on the content and conditions of that step, combined with the temporal evolution logic. For example, if a step involves administering a certain drug to a patient, the values ​​of the patient's relevant physiological indicators (node ​​attributes) are adjusted according to the drug's mechanism of action and the rules governing the drug's impact on node attributes in the temporal evolution logic.

[0148] Simultaneously, based on the changes in node attributes after each step, and according to the adjustment rules for edge association strength in the temporal evolution logic, the association strength between nodes is updated to reflect the changes in the relationships between diagnostic and treatment features. Through this gradual simulation and updating, the dynamic diagnostic and treatment association network can reflect the changes in patient status and diagnostic and treatment relationships in real time during the execution of the initial strategy.

[0149] During the simulation of each of the time-sequential execution steps, the impact of the time-sequential execution steps on the diagnostic and treatment features corresponding to the nodes in the dynamic diagnostic and treatment association network is monitored, the changing trend, magnitude and rate of change of the diagnostic and treatment features are predicted, and feature change prediction results are generated.

[0150] In this embodiment of the disclosure, data monitoring and analysis techniques are used to collect real-time data on changes in node attributes. Based on this data, time series analysis methods, such as the Autoregressive Moving Average (ARMA) model and the Autoregressive Integrated Moving Average (ARIMA) model, are employed to predict the changing trends of diagnostic and treatment characteristics, determining whether they are increasing, decreasing, or fluctuating.

[0151] Simultaneously, by calculating the magnitude of data changes, the magnitude of changes in diagnostic and therapeutic characteristics, such as the range of blood pressure values, is determined. Using derivative or difference methods, the rate of change of these characteristics, i.e., the degree of change per unit time, is calculated. Based on the above analysis, a characteristic change prediction result is generated, which can describe in detail the specific impact of each time-sequential execution step on the diagnostic and therapeutic characteristics.

[0152] The interaction between different diagnostic and treatment features in the predicted feature change results is analyzed to identify new feature interaction relationships caused by the execution of the initial strategy, and to predict possible synergistic or constraint feedback. The new feature interaction relationship is an interaction relationship that did not exist before the execution of the initial strategy but was formed by changes in diagnostic and treatment features caused by the execution of the initial strategy.

[0153] Among them, synergistic feedback refers to the interaction between different diagnostic and treatment features, which produces mutually reinforcing and enhancing effects, thus benefiting patient recovery and the achievement of treatment goals. Constraint-restriction feedback, on the other hand, refers to the mutual constraints and limitations between different diagnostic and treatment features, which may hinder patient recovery or affect the achievement of treatment goals.

[0154] In this embodiment, the prediction results of feature changes are analyzed in depth to study the interactions between different diagnostic and treatment features. Correlation analysis and causal analysis are used to determine whether there are mutual influence relationships between the various diagnostic and treatment features. By comparing the changes in diagnostic and treatment features before and after the initial strategy is implemented, feature interactions that did not exist before the initial strategy was implemented but newly emerged after the strategy implementation caused changes in the diagnostic and treatment features.

[0155] For newly identified feature interactions, their potential impact is further analyzed. If different diagnostic and treatment features mutually promote each other, accelerating the patient's recovery process or improving treatment outcomes, it is predicted as a synergistic feedback; conversely, if different diagnostic and treatment features mutually restrict each other, leading to adverse reactions in patients or affecting treatment outcomes, it is predicted as a constraint feedback. This helps to comprehensively understand the complex impact of initial strategy execution on the diagnostic and treatment process.

[0156] For potential collaborative or restrictive feedback, analyze its impact on subsequent sequential execution steps, generate feedback impact assessment results, and determine the inference results based on the characteristic change prediction results and feedback impact assessment results of each sequential execution step.

[0157] In this embodiment, synergistic feedback may enable patients to better tolerate certain treatment measures in subsequent steps, thereby allowing adjustments to the intensity or timing of those steps; while constraint feedback may prevent certain subsequent steps from being performed as planned, necessitating changes to the treatment plan or additional interventions. A feedback impact assessment result is generated by comprehensively evaluating the impact of these feedbacks on subsequent steps.

[0158] Finally, the predicted results of characteristic changes and the feedback impact assessment results for each sequential execution step are integrated to comprehensively consider the changes in diagnostic and treatment characteristics during the initial strategy execution process and their impact on subsequent steps, thereby determining the final projection result. The projection result can provide healthcare professionals with a comprehensive prediction and assessment of the effectiveness of the initial strategy execution.

[0159] The aforementioned technical solution decomposes the initial strategy into multiple sequential execution steps corresponding to different time windows. Each step clearly defines the execution content and conditions, making the strategy execution more systematic and operable. Using the current state of the dynamic diagnosis-treatment network as the initial simulation state, the execution process is simulated step-by-step in sequence, and node attributes and edge association strength are updated according to the temporal evolution logic, accurately simulating the dynamic changes of the network during actual diagnosis and treatment.

[0160] Furthermore, during the simulation, the impact of each step on the diagnostic and treatment characteristics is monitored, and the trends, magnitudes, and rates of change are predicted, generating characteristic change prediction results. This allows for early insight into the development trajectory of the diagnostic and treatment characteristics. Analyzing the interactions between different diagnostic and treatment characteristics identifies new characteristic interactions triggered by the execution of the initial strategy, predicting synergistic or restrictive feedback, and effectively capturing the potential impact of strategy execution. The impact of feedback analysis on subsequent steps is analyzed, generating evaluation results, and combining these with the characteristic change prediction results to determine the overall projection outcome. This achieves accurate projection of the initial strategy throughout its entire lifecycle and in all aspects, providing solid support for formulating scientific and reasonable personalized management decisions, and significantly improving the quality of perioperative management and patient prognosis in lung cancer patients.

[0161] In a preferred embodiment, the inference engine is specifically used for:

[0162] Based on the direction of change, magnitude of change, and corresponding time-sequential execution steps of all changed diagnostic and treatment features extracted from the feature change prediction results, the direct or indirect correlation between the changed diagnostic and treatment features is identified to obtain the feature change correlation.

[0163] Among them, feature change correlation refers to the direct or indirect relationship between changing diagnostic and treatment features, reflecting the mutual influence of changes in diagnostic and treatment features during the initial strategy execution process.

[0164] In this embodiment of the disclosure, all diagnostic and treatment features that have changed are accurately extracted from the feature change prediction results, the direction of change (increase or decrease) and the magnitude of change (specific value or range) of each feature are clarified, and the time-sequential execution steps corresponding to these changes are recorded.

[0165] Then, association analysis algorithms, such as the Apriori algorithm, are used to perform pairwise analysis on these changing diagnostic features. For each pair of diagnostic features, their changes in each sequential execution step are examined. If two diagnostic features change simultaneously in the same sequential execution step, and the direction of change follows a certain pattern (such as simultaneous increase or simultaneous decrease), then a direct correlation is considered to exist between them.

[0166] If changes in two diagnostic features are not directly reflected in the same sequential execution step, but a connection can be established through changes in other intermediate diagnostic features—for example, a change in diagnostic feature A leads to a change in diagnostic feature B, and a change in diagnostic feature B in turn triggers a change in diagnostic feature C—then there is an indirect relationship between diagnostic features A and C. Through this comprehensive analysis, the direct or indirect relationships between all changing diagnostic features can be identified, thus revealing the feature change relationships.

[0167] Based on the feature change association and the original interaction relationship of nodes in the dynamic diagnosis and treatment association network, it is distinguished whether the association between the changed diagnosis and treatment features belongs to a new feature interaction relationship.

[0168] In this embodiment, a dynamic diagnostic-therapeutic association network is used as a reference framework, where the original interaction relationships of nodes reflect the stable connections between diagnostic and therapeutic features before the initial strategy is executed. The previously obtained feature change associations are compared and analyzed with the original interaction relationships of nodes in the dynamic diagnostic-therapeutic association network. For each pair of changed diagnostic and therapeutic features, it is examined whether there are corresponding edges in the dynamic diagnostic-therapeutic association network to represent their original interaction relationships.

[0169] If there already exists an edge in the dynamic diagnosis and treatment association network that represents the association between these two diagnosis and treatment features, and the nature of the association (such as a promoting or inhibiting relationship) is consistent with the association reflected by the feature change, then it means that this association is not a new feature interaction relationship, but a continuation of the original relationship.

[0170] Conversely, if there is no edge representing the relationship between these two diagnostic features in the dynamic diagnostic-treatment association network, or if the relationship properties of the original edge are different from those of the feature change relationship, then it can be determined that this relationship belongs to a new feature interaction relationship, which is newly generated after the diagnostic-treatment feature change is triggered by the execution of the initial strategy.

[0171] If the new feature interaction relationship is such that a positive change in one of the diagnostic features promotes a positive change in another diagnostic feature, or inhibits a negative change in another diagnostic feature, then the two diagnostic features have the synergistic promotion feedback.

[0172] In this embodiment of the disclosure, for each pair of diagnostic and therapeutic features with a new feature interaction relationship, their changes are closely monitored. If one diagnostic and therapeutic feature shows a positive change (such as an increase in indicators, a reduction in symptoms, etc.), and the other diagnostic and therapeutic feature also shows a positive change, and data analysis (such as correlation analysis, regression analysis, etc.) can prove that the former promotes the positive change of the latter, for example, by establishing a mathematical model to find that if the former changes by one unit, the latter will correspondingly change positively by a certain proportion, then a positive promoting relationship can be considered to exist.

[0173] In addition, if one diagnostic feature changes positively while another diagnostic feature originally changes negatively (such as a decrease in indicators or worsening of symptoms), but the former inhibits the negative change of the latter, thus slowing down or stopping the negative change of the latter, this also indicates that there is a synergistic feedback between the two diagnostic features.

[0174] If a new feature interaction relationship manifests as a change in one of the diagnostic features restricting a positive change in another diagnostic feature, or exacerbating a negative change in another diagnostic feature, then the two diagnostic features exhibit the constraint feedback.

[0175] In this embodiment of the disclosure, if one of the diagnostic features changes (whether positive or negative), while the other diagnostic feature was originally changing positively, but the change in the former restricts the positive change in the latter, for example, slowing down the rate of positive change, reducing the magnitude of positive change, or even stopping the positive change, after verifying this restrictive relationship with data, it can be determined that there is a constraint feedback between the two diagnostic features.

[0176] Furthermore, if a change in one diagnostic feature exacerbates a negative change in another diagnostic feature, such as causing the original indicator to decrease faster or the symptoms to worsen more severely, confirming this exacerbating effect by establishing a corresponding analytical model also indicates that there is a constraint feedback between the two diagnostic features.

[0177] The above technical solution accurately identifies direct or indirect relationships between features based on the direction and magnitude of changes in diagnostic and therapeutic features and the corresponding time-sequential execution steps in the feature change prediction results, thus obtaining feature change relationships. This is akin to clarifying the complex changes in diagnostic and therapeutic features. Therefore, based on the existing interaction relationships of nodes in the feature change relationships and dynamic diagnostic and therapeutic relationship network, it can accurately distinguish whether a new feature interaction relationship exists, effectively avoiding information confusion.

[0178] Furthermore, for new feature interactions, their type is further determined. If they exhibit positive promotion or inhibition of negative changes, they are identified as synergistic promotion feedback, which helps to uncover positive interactions during diagnosis and treatment, providing direction for enhancing treatment effects. If they exhibit limiting positive changes or exacerbating negative changes, they are identified as constraint feedback, which can promptly identify potential risks and formulate countermeasures in advance. Through precise identification and classification, the inference engine can comprehensively assess the impact of initial strategy implementation, providing a scientific basis for adjusting and optimizing decisions, and significantly improving the accuracy and safety of perioperative management for lung cancer.

[0179] In a preferred embodiment, the optimization engine is specifically used for:

[0180] According to the candidate scheme evaluation index system, multiple groups of candidate management decision schemes are evaluated one by one, and the performance of each group of candidate management decision schemes on each evaluation index is quantified to generate a scheme evaluation score matrix. The evaluation indexes in the candidate scheme evaluation index system include scenario adaptability, temporal continuity, risk prevention and control capability, rehabilitation promotion efficiency, and resource adaptability.

[0181] Among these, scenario adaptability measures the degree of matching between candidate management decision-making solutions and specific medical scenarios (such as different stages of the perioperative period, different patient types, etc.), reflecting the applicability and effectiveness of the solution in a specific scenario. Temporal coherence refers to the execution sequence and logical coherence of candidate management decision-making solutions over time, ensuring that operations and decisions at different time points are interconnected and progressed reasonably. Risk control capability assesses the ability of candidate management decision-making solutions to prevent and control potential medical risks (such as complications, adverse reactions, etc.), reflecting the safety and reliability of the solution.

[0182] Among them, rehabilitation promotion efficiency is used to measure the effect of candidate management decision-making schemes on promoting the patient's rehabilitation process, and is reflected by relevant indicators (such as rehabilitation time, rehabilitation degree, etc.). Resource suitability is used to examine the degree of rational utilization of medical resources (such as human, material, and financial resources) by candidate management decision-making schemes, to ensure the rationality and efficiency of resource allocation.

[0183] In this embodiment of the disclosure, for scenario adaptability, the matching elements between the solution and different medical scenarios, such as patient condition and surgical type, are analyzed, and corresponding scoring criteria are set for scoring. For temporal coherence, the reasonableness of the solution is evaluated and quantified based on the time sequence and logical relationship of each step in the solution.

[0184] In this embodiment of the disclosure, the risk prevention and control capability can be assessed by collecting data from similar past cases, analyzing the effectiveness of the proposed solutions in preventing and responding to various risks, and assigning corresponding scores.

[0185] In this embodiment of the disclosure, the efficiency of rehabilitation promotion can be assessed by comparing changes in patients' rehabilitation indicators before and after the implementation of the program. Resource suitability can be scored based on the program's demand for and utilization of various medical resources.

[0186] Then, following a quantitative approach, each of the multiple candidate management decision-making schemes was evaluated, and the score of each scheme on each evaluation indicator was recorded. Finally, these scores were organized into a matrix form, namely the scheme evaluation score matrix, so as to intuitively compare the performance of each scheme on different indicators.

[0187] Based on the scheme evaluation score matrix, the advantage evaluation indicators of each group of candidate management decision schemes are identified, and according to the temporal evolution logic and scenario collaboration requirements of the dynamic diagnosis and treatment association network, all the advantage strategy content is integrated to generate an integrated advantage strategy framework.

[0188] Among them, the scheme evaluation score matrix presents the quantitative scores of each group of candidate management decision schemes on each evaluation indicator in matrix form, intuitively showing the performance of the schemes in various aspects.

[0189] Among them, the advantage evaluation index is the indicator that performs well in the evaluation index system for each group of candidate management decision-making schemes, representing the scheme's advantage in this aspect.

[0190] In this embodiment of the disclosure, starting from the scheme evaluation score matrix, for each group of candidate management decision schemes, the evaluation indicators with higher scores are identified, and these indicators are the advantage evaluation indicators of the scheme. Next, the temporal evolution logic of the dynamic diagnosis and treatment association network is analyzed in depth to understand the temporal sequence and mutual influence of different diagnosis and treatment links. At the same time, considering the scenario collaboration requirements, the collaborative requirements of various diagnosis and treatment elements under different scenarios are clarified.

[0191] Then, the advantageous strategies corresponding to the advantage evaluation indicators in all candidate management decision-making schemes are extracted and integrated. During the integration process, the advantageous strategies are reasonably arranged and combined according to the temporal sequence and scenario requirements of the dynamic diagnosis and treatment association network, ensuring that the integrated strategy framework is coherent in time and synergistic in scenario, forming a fusion advantage strategy framework that includes multiple advantages.

[0192] By combining the feature interaction relationships in the dynamic diagnosis and treatment association network, the adaptation defects in the fusion advantage strategy framework are corrected to obtain an optimized solution, which is then input into the candidate solution evaluation index system for evaluation.

[0193] In this embodiment, the fusion advantage strategy framework and the dynamic diagnosis-treatment association network are compared and analyzed to identify parts of the framework that do not match or conflict with the feature interaction relationships; these parts are the adaptation defects. For example, if the order of two diagnosis-treatment steps in the framework is reversed compared to the order required by the feature interaction relationships, it may lead to poor treatment results.

[0194] To address the adaptation deficiencies, corrections are made based on the patterns of feature interaction relationships, adjusting relevant content within the strategy framework. After correction, an optimized solution is obtained. This optimized solution is then input back into the candidate solution evaluation index system, and its performance on each evaluation index is re-evaluated using the previous quantitative methods to verify the optimization effect.

[0195] If the optimization scheme fails to meet the preset optimal standard, the advantage fusion and defect correction are repeatedly executed for iterative optimization until the optimization scheme meets the preset optimal standard. Then, the iteration stops, and the corresponding candidate management decision scheme is used as a dynamically adapted personalized management decision scheme.

[0196] In this embodiment of the disclosure, a clear preset optimal standard can be set. This standard comprehensively considers multiple aspects such as scenario adaptability, temporal continuity, risk prevention and control capabilities, rehabilitation promotion efficiency, and resource adaptability. It is a comprehensive measurement indicator.

[0197] When the optimized solution fails to meet the preset optimal standard after evaluation, it indicates that there is still room for improvement. At this point, the advantage fusion step is repeated to further explore the advantageous strategies of each candidate management decision-making solution and conduct more in-depth integration and optimization; at the same time, the adaptation defects in the optimized solution are corrected by combining the feature interaction relationships in the dynamic diagnosis and treatment association network again.

[0198] By repeatedly performing these two steps, the optimization plan is iteratively improved. After each iteration, the plan's performance is re-evaluated until the optimized plan reaches the preset optimal standard. At this point, iteration stops, and the candidate management decision plan corresponding to the optimized plan that finally meets the standard is used as a dynamically adapted personalized management decision plan. This plan can be flexibly adjusted according to the individual patient's condition and dynamic changes in the medical scenario to achieve personalized management.

[0199] The aforementioned technical solution, based on an evaluation system encompassing indicators such as scenario adaptability, temporal coherence, risk control capabilities, rehabilitation promotion efficiency, and resource adaptability, evaluates multiple candidate management decision-making schemes one by one, generating a scheme evaluation score matrix to comprehensively and accurately quantify the performance of each scheme. Based on the score matrix, it identifies advantageous evaluation indicators and integrates these advantages by combining the temporal evolution logic of the dynamic diagnosis-treatment association network and the scenario collaboration requirements, generating a fusion advantage strategy framework. This framework effectively integrates the strengths of each scheme, improving the scientific rigor and comprehensiveness of the strategy. Furthermore, it corrects adaptation defects in the framework by incorporating feature interaction relationships, resulting in an optimized scheme that is then evaluated again, forming a closed-loop optimization process.

[0200] Furthermore, if the optimized plan fails to meet the preset optimal standard, iterative optimization is carried out through repeated synergy of advantages and correction of defects until the standard is met, ultimately generating a dynamically adapted personalized management decision-making plan. This achieves precise customization of decision-making plans, fully considers individual patient differences and dynamic changes in the diagnosis and treatment process, significantly improves the accuracy and effectiveness of perioperative management for lung cancer, and provides patients with higher quality and more personalized medical services.

[0201] In a preferred embodiment, the optimization engine is further configured to:

[0202] After each round of iterative optimization, key diagnostic and treatment data corresponding to the newly added diagnostic and treatment data in the multi-source diagnostic and treatment data are extracted, and the impact of the key diagnostic and treatment data on the perioperative diagnostic and treatment status of the lung cancer patients is analyzed.

[0203] Among them, key diagnostic and treatment data are core data in the newly added diagnostic and treatment data that have a significant impact on the perioperative diagnosis and treatment status of lung cancer patients and can directly or indirectly determine the direction and effect of diagnosis and treatment decisions, such as key examination results and information on sudden changes in the disease.

[0204] In this embodiment of the disclosure, after each round of iterative optimization, the optimization engine first accurately locates newly added diagnostic and treatment data from the vast information database of multi-source diagnostic and treatment data. Utilizing data mining and feature extraction techniques, and based on pre-set key indicators and rules, it filters out data that plays a crucial role in the perioperative diagnostic and treatment status of lung cancer patients, i.e., key diagnostic and treatment data. This key data may involve newly appearing symptoms in patients, abnormal changes in important examination results, etc.

[0205] Next, using medical expertise and data analysis models, we delved into the intrinsic relationship between key diagnostic and treatment data and the patient's current treatment status. By comparing changes in physiological indicators and disease severity before and after receiving the new diagnostic and treatment data, and combining clinical experience with relevant research findings, we comprehensively assessed the specific impact of key diagnostic and treatment data on the patient's perioperative disease progression, treatment outcomes, and the occurrence of complications.

[0206] Based on the impact of the diagnosis and treatment status, determine the adaptability of each strategy step in the optimization scheme obtained after this iteration with the newly added diagnosis and treatment data, and based on the adaptability, determine the target strategy steps from the strategy steps whose execution conditions do not match the current diagnosis and treatment status or whose execution effect cannot meet the current diagnosis and treatment needs.

[0207] Among them, adaptability is the degree of matching between each strategy step in the optimized scheme after this round of iteration and the newly added diagnosis and treatment data in terms of execution conditions and expected effects, reflecting whether the strategy steps can effectively utilize the new data and adapt to the current diagnosis and treatment situation.

[0208] In this embodiment of the disclosure, for each strategy step, from the perspective of execution conditions, it is checked whether the required diagnosis and treatment data conditions are consistent with the newly added diagnosis and treatment data. For example, if a strategy step requires a patient to reach a certain indicator within a specific range, but the newly added data shows that the indicator exceeds this range, then the execution conditions do not match.

[0209] Regarding implementation effectiveness, based on changes in the treatment status reflected in key diagnostic and treatment data, it is determined whether the strategy steps can achieve the expected results. If the patient's condition does not improve as expected or even worsens, it indicates poor implementation effectiveness. Through comprehensive evaluation, target strategy steps that do not match the implementation conditions with the current treatment status or whose implementation effects cannot meet current needs are identified.

[0210] For the target strategy steps, based on the diagnostic and treatment characteristics and correlations of the newly added diagnostic and treatment data, the execution content, timing, or intensity of the target strategy steps are adjusted to generate adjustment strategy steps;

[0211] In this embodiment of the disclosure, after defining the target strategy steps, the optimization engine conducts an in-depth analysis of the diagnostic and treatment characteristics of the newly added diagnostic and treatment data, including the data type, numerical range, and trend of change, as well as the correlation between these data and other diagnostic and treatment data, such as the mutual influence of changes in a certain indicator with other indicators. Based on these characteristics and correlations, the target strategy steps are adjusted accordingly.

[0212] If the execution content does not match, modify the specific operations of the strategy steps, such as adjusting the medication dosage or changing the treatment method; if the timing of execution is inappropriate, re-determine the implementation time of the strategy steps based on changes in the patient's condition and new data prompts.

[0213] If the implementation intensity is insufficient or excessive, the intensity of the strategy steps should be adjusted based on the severity of the condition and treatment effects reflected in the data. Through these adjustments, new adjusted strategy steps are generated to better suit the current diagnosis and treatment status and needs of the patient.

[0214] The adjustment strategy steps are incorporated into the optimization scheme after this iteration. The timing structure and collaborative logic of the optimization scheme are adjusted to obtain a new optimization scheme, and it is determined whether the new optimization scheme meets the preset optimal standard.

[0215] Among them, temporal structure refers to the sequential order and arrangement of the various strategy steps in the optimization plan over time. A reasonable temporal structure ensures that the diagnosis and treatment process proceeds in an orderly manner. Collaborative logic refers to the relationships and rules governing the cooperation and interaction among the various strategy steps in the optimization plan. Good collaborative logic can maximize the overall diagnosis and treatment effect.

[0216] In this embodiment, the generated adjustment strategy steps are integrated into the optimization scheme after this iteration. This process requires comprehensive consideration of the overall temporal structure and collaborative logic of the optimization scheme. Regarding the temporal structure, the order of each strategy step is rearranged to ensure that the adjusted strategy steps are logically connected with other steps in time, avoiding time conflicts or logical confusion.

[0217] In terms of collaborative logic, we analyze and adjust the interaction between the strategy steps and other steps, optimize their coordination, and enable all steps of the whole plan to work together to improve the overall diagnosis and treatment effect.

[0218] After adjustments, a new optimized plan is obtained. Finally, based on the preset optimal standard, the new plan is comprehensively evaluated from multiple dimensions such as disease control effect, patient recovery speed, and resource utilization efficiency to determine whether it meets the preset optimal standard. If it does not meet the standard, the next round of iteration and optimization will continue.

[0219] After each round of iteration and optimization, the above technical solution accurately extracts key information from the newly added medical data, deeply analyzes its impact on the patient's treatment status, and can promptly capture subtle changes. Based on the impact of the treatment status, it determines the adaptability of each strategy step in the optimization plan to the new data, accurately identifying target strategy steps with mismatched execution conditions or unsatisfactory results, thus avoiding strategy failure due to data changes. For target strategy steps, it flexibly adjusts the execution content, timing, or intensity based on the characteristics and relationships of the newly added medical data, generating adjusted strategy steps to make the strategy more flexible and targeted.

[0220] Furthermore, the adjustment strategy steps are integrated into the optimization plan, and its temporal structure and collaborative logic are adjusted to obtain a new optimized plan, which is then re-evaluated to see if it meets the preset optimal standard. This closed-loop optimization mechanism ensures that the management decision-making plan can be continuously optimized with the dynamic changes in medical data, always closely aligned with the actual medical needs of patients. This provides more scientific, personalized, and effective management decision support for lung cancer perioperative patients, powerfully promoting the improvement of treatment outcomes.

[0221] The preferred embodiments of the present disclosure have been described in detail above with reference to the accompanying drawings. However, the present disclosure is not limited to the specific details of the above embodiments. Within the scope of the technical concept of the present disclosure, various changes, modifications, substitutions and variations can be made to these embodiments, and all such changes, modifications, substitutions and variations fall within the protection scope of the present disclosure.

[0222] It should also be noted that the various specific technical features described in the above embodiments can be combined in any suitable manner without contradiction, and such combinations should also be considered as part of this disclosure. To avoid unnecessary repetition, this disclosure will not further describe the various possible combinations. The technical scope of this application is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A decision support system for perioperative management of lung cancer, characterized in that, include: The system includes a data acquisition terminal, a management decision-making device communicatively connected to the data acquisition terminal, and a medical execution terminal communicatively connected to the management decision-making device. The management decision-making device includes a network construction engine, a deduction engine communicatively connected to the network construction engine, and an optimization engine communicatively connected to the deduction engine. The data acquisition terminal is used to dynamically collect multi-source diagnosis and treatment data of lung cancer patients throughout the entire perioperative period; The network construction engine is used to construct a dynamic diagnosis and treatment association network for the lung cancer patient based on the multi-source diagnosis and treatment data. The dynamic diagnosis and treatment association network uses time-series diagnosis and treatment features as nodes and dynamic diagnosis and treatment influence relationships as edges. The node attributes and the association strength of the edges of the dynamic diagnosis and treatment association network are adjusted in real time as the multi-source diagnosis and treatment data is updated. The simulation engine is used to call a multi-scenario decision simulation model to simulate and simulate the full-cycle treatment strategy of the dynamic treatment association network, obtain simulation results, and generate multiple sets of candidate management decision schemes by combining the simulation results with the dynamic needs of different treatment scenarios. The optimization engine is used to iteratively optimize the multiple sets of candidate management decision schemes, integrate the adaptation advantages of each candidate management decision scheme and correct the potential defects of each candidate management decision scheme, and generate a dynamically adapted personalized management decision scheme. The medical execution terminal is used to receive and display the personalized management decision-making scheme; Specifically, the network building engine is used for: The multi-source diagnostic and treatment data are analyzed for temporal feature evolution and fused with cross-dimensional information through a pre-trained cross-stage feature interaction model to generate a temporal diagnostic and treatment feature set. Based on the set of time-series diagnostic and treatment features, extract the core attributes of each time-series diagnostic and treatment feature; Each of the time-series diagnostic features is used as a node in the dynamic diagnostic association network. A unique feature identifier is assigned to each node in the dynamic diagnostic association network, and the corresponding core attribute is embedded in the node. The driving influence of the sequential diagnostic and treatment features that occur earlier on the sequential diagnostic and treatment features that occur later on the sequential diagnostic and treatment features that have the same time sequence on the sequential diagnostic and treatment features that occur earlier on the sequential diagnostic and treatment features is extracted to construct the dynamic diagnostic and treatment influence relationship. Based on the dynamic diagnosis and treatment influence relationship, the edges of the dynamic diagnosis and treatment association network are constructed, and the dynamic node set is associated and combined with the edge structure of the dynamic diagnosis and treatment association network to generate an initial diagnosis and treatment association network. The system acquires the update status of multi-source diagnosis and treatment data in real time, identifies newly added diagnosis and treatment features, and adjusts the number of nodes, the core attributes of the nodes, and the association strength of the edges in the initial diagnosis and treatment association network based on the newly added diagnosis and treatment features, thereby forming a dynamic treatment course association network that evolves dynamically with the diagnosis and treatment process.

2. The lung cancer perioperative management decision support system according to claim 1, characterized in that, The network building engine is specifically used for: Based on the causal logic obtained from the analysis that changes in the sequential diagnostic features in the preceding time series cause changes in the sequential diagnostic features in the following time series, the direction and intensity of the driving influence are determined, and based on the direction and intensity of the driving influence, the driving influence of the sequential diagnostic features in the preceding time series on the sequential diagnostic features in the following time series is determined. Based on the interaction between different time-series diagnostic and treatment features within the same time window obtained from the analysis, the synergistic promotion or mutual constraint relationship between the time-series diagnostic and treatment features is determined, and based on the synergistic promotion or mutual constraint relationship, the synergistic influence between the time-series diagnostic and treatment features with the same time sequence is determined; Based on the evaluation feedback effect of the later-order sequential diagnosis and treatment features on the earlier-order sequential diagnosis and treatment features obtained from the analysis, the basis for verifying or correcting the accuracy of the evaluation of the earlier-order sequential diagnosis and treatment features can be determined by the diagnosis and treatment results corresponding to the later-order sequential diagnosis and treatment features. Based on the validity results corresponding to the verification and the correction direction corresponding to the correction basis, determine the feedback influence of the later-ordered sequential diagnosis and treatment features on the earlier-ordered sequential diagnosis and treatment features; Based on the timeliness and intensity of the effects of the driving effects, the synergistic effects, and the feedback effects, the values ​​of the corresponding influence factors are adjusted, and the dynamic diagnosis and treatment influence relationship is constructed based on the values ​​of the influence factors.

3. The lung cancer perioperative management decision support system according to claim 1, characterized in that, The network building engine is specifically used for: The preoperative dynamic assessment data stream, intraoperative real-time operation data stream, postoperative continuous monitoring data stream, and cross-cycle diagnosis and treatment feedback data stream in the multi-source diagnosis and treatment data are each time-series sliced ​​to obtain multiple diagnosis and treatment data segments with timestamps. The diagnostic and treatment data segments are input into the cross-stage feature interaction model to extract the temporal evolution features of each diagnostic and treatment data segment. The temporal evolution features include the changing trend of the diagnostic and treatment indicators, the changing rate of the diagnostic and treatment indicators, and the description of the correlation between the changes of the diagnostic and treatment indicators. Based on the information overlap between the time-series evolution features aligned on the time axis, cross-dimensional correlation and fusion are performed on the non-redundant time-series evolution features to construct the interaction mapping relationship between the non-redundant time-series evolution features and generate a preliminary feature interaction set. The following steps are executed in a loop: based on the temporal evolution characteristics corresponding to the newly received multi-source diagnosis and treatment data streams, the preliminary feature interaction set is incrementally updated, new interaction mapping relationships are added, and the temporalized diagnosis and treatment feature set is generated.

4. The lung cancer perioperative management decision support system according to claim 3, characterized in that, The network building engine is specifically used for: Based on the dimensions of diagnostic and treatment indicators, diagnostic and treatment behavior, and diagnostic and treatment results, the non-redundant temporal evolution features are classified into multiple dimensional feature subsets. Extract the dynamic change pattern of the temporal evolution feature representation within each of the dimensional feature subsets, and generate a core feature vector based on the dynamic change pattern; Based on the core feature vector, analyze the correlation between each pair of dimensional feature subsets, and based on the correlation, determine the driving logic of the change of the temporal evolution feature corresponding to one of the two dimensional feature subsets on the temporal evolution feature corresponding to the other, and determine the first interactive mapping relationship between the two dimensional feature subsets. Based on the core feature vector, analyze the collaborative correlation between different temporal evolution features within the same dimensional feature subset, and determine the second interactive mapping relationship between different temporal evolution features within the same dimensional feature subset; Based on the first interaction mapping relationship and the second interaction mapping relationship, a feature interaction matrix is ​​constructed, and based on the feature interaction matrix, the temporal evolution feature combination corresponding to the core interaction relationship is extracted to generate a preliminary feature interaction set.

5. The lung cancer perioperative management decision support system according to claim 1, characterized in that, The inference engine is specifically used for: The dynamic diagnosis and treatment association network is input into the multi-scenario decision inference model, and the dynamic node attributes, edge association strength, and time sequence hierarchy of the nodes in the dynamic diagnosis and treatment association network are extracted to generate a network dynamic feature descriptor. The network dynamic feature descriptor is compared with the scenario adaptation conditions in the multi-scenario strategy library to identify the core diagnosis and treatment scenarios for the lung cancer patient. The multi-scenario strategy library includes preoperative risk prevention and control scenarios, intraoperative precision operation scenarios, postoperative rapid recovery scenarios, and emergency treatment scenarios for complications. An initial strategy is generated by combining the basic management strategy corresponding to the core diagnosis and treatment scenario, the dynamic node attributes and edge association strength corresponding to the dynamic diagnosis and treatment association network; Extract the changing patterns of node attributes over time, the logic for adjusting the association strength of edges, and the interaction evolution paths between levels from the dynamic diagnosis and treatment association network to generate temporal evolution logic; Based on the aforementioned temporal evolution logic, the initial strategy is simulated throughout its entire execution process. The dynamic changes of various diagnostic and treatment features and possible interactive feedbacks during the execution of the initial strategy are predicted, and the simulation results are obtained.

6. The lung cancer perioperative management decision support system according to claim 5, characterized in that, The inference engine is specifically used for: The initial strategy is decomposed into multiple sequential execution steps, each sequential execution step corresponding to a time window in the perioperative period, and each sequential execution step includes execution content and execution conditions; Using the current state of the dynamic diagnosis and treatment association network as the initial simulation state, the execution process of the initial strategy is simulated step by step according to the order of the temporal execution steps. After each round of the temporal execution steps, the dynamic node attributes and edge association strength in the dynamic diagnosis and treatment association network are updated based on the temporal evolution logic. During the simulation of each of the time-sequential execution steps, the impact of the time-sequential execution steps on the diagnostic and treatment features corresponding to the nodes in the dynamic diagnostic and treatment association network is monitored, the changing trend, magnitude and rate of change of the diagnostic and treatment features are predicted, and feature change prediction results are generated. The interaction between different diagnostic and treatment features in the predicted feature change results is analyzed to identify new feature interaction relationships caused by the execution of the initial strategy, and to predict possible synergistic or constraint feedback. The new feature interaction relationship is an interaction relationship that did not exist before the execution of the initial strategy but was formed by changes in diagnostic and treatment features caused by the execution of the initial strategy. For potential collaborative or restrictive feedback, analyze its impact on subsequent sequential execution steps, generate feedback impact assessment results, and determine the inference results based on the characteristic change prediction results and feedback impact assessment results of each sequential execution step.

7. The lung cancer perioperative management decision support system according to claim 6, characterized in that, The inference engine is specifically used for: Based on the direction of change, magnitude of change, and corresponding time-sequential execution steps of all changed diagnostic and treatment features extracted from the feature change prediction results, the direct or indirect correlation between the changed diagnostic and treatment features is identified to obtain the feature change correlation. Based on the feature change association and the original interaction relationship of nodes in the dynamic diagnosis and treatment association network, it is distinguished whether the association between the changed diagnosis and treatment features belongs to a new feature interaction relationship. If the new feature interaction relationship is such that a positive change in one of the diagnostic features promotes a positive change in another diagnostic feature, or inhibits a negative change in another diagnostic feature, then the two diagnostic features have the synergistic promotion feedback. If the new feature interaction relationship manifests as a change in one of the diagnostic features restricting a positive change in another diagnostic feature, or exacerbating a negative change in another diagnostic feature, then the two diagnostic features exhibit the constraint feedback.

8. The lung cancer perioperative management decision support system according to any one of claims 1-7, characterized in that, The optimization engine is specifically used for: According to the candidate scheme evaluation index system, multiple groups of candidate management decision schemes are evaluated one by one, and the performance of each group of candidate management decision schemes on each evaluation index is quantified to generate a scheme evaluation score matrix. The evaluation indexes in the candidate scheme evaluation index system include scenario adaptability, temporal continuity, risk prevention and control capability, rehabilitation promotion efficiency, and resource adaptability. Based on the scheme evaluation score matrix, the advantage evaluation indicators of each group of candidate management decision schemes are identified, and according to the temporal evolution logic and scenario collaboration requirements of the dynamic diagnosis and treatment association network, all the advantage strategy content is integrated to generate an integrated advantage strategy framework. By combining the feature interaction relationships in the dynamic diagnosis and treatment association network, the adaptation defects in the fusion advantage strategy framework are corrected to obtain an optimized solution, which is then input into the candidate solution evaluation index system for evaluation. If the optimization scheme fails to meet the preset optimal standard, the advantage fusion and defect correction are repeatedly executed for iterative optimization until the optimization scheme meets the preset optimal standard. Then, the iteration stops, and the corresponding candidate management decision scheme is used as a dynamically adapted personalized management decision scheme.

9. The lung cancer perioperative management decision support system according to claim 8, characterized in that, The optimization engine is also used for: After each round of iterative optimization, key diagnostic and treatment data corresponding to the newly added diagnostic and treatment data in the multi-source diagnostic and treatment data are extracted, and the impact of the key diagnostic and treatment data on the perioperative diagnostic and treatment status of the lung cancer patients is analyzed. Based on the impact of the diagnosis and treatment status, determine the adaptability of each strategy step in the optimization scheme obtained after this iteration with the newly added diagnosis and treatment data, and based on the adaptability, determine the target strategy steps from the strategy steps whose execution conditions do not match the current diagnosis and treatment status or whose execution effect cannot meet the current diagnosis and treatment needs. For the target strategy steps, based on the diagnostic and treatment characteristics and correlations of the newly added diagnostic and treatment data, the execution content, timing, or intensity of the target strategy steps are adjusted to generate adjustment strategy steps; The adjustment strategy steps are incorporated into the optimization scheme after this iteration. The timing structure and collaborative logic of the optimization scheme are adjusted to obtain a new optimization scheme, and it is determined whether the new optimization scheme meets the preset optimal standard.

Images