A medical multi-scene automatic reporting and full-link closed-loop processing method and system

Abstract

This invention relates to the field of medical data processing technology, specifically to a method and system for automated reporting and closed-loop processing across multiple medical scenarios. The method provided by this invention mainly includes acquiring stage migration data from historical medical records, constructing a stage migration probability model, outputting the migration probability of the current stage based on the risk state vector and stage migration data through the stage migration probability model, setting a migration probability threshold, and determining whether to trigger a stage migration warning based on the migration probability threshold. This method achieves adaptive automated reporting across multiple scenarios, eliminating the need for medical staff to switch systems or manually enter data. Combined with efficient processing capabilities for unstructured data, it significantly reduces manual workload and reporting error rates, thereby improving reporting efficiency.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, and more specifically, to a method and system for automated reporting and closed-loop processing of multiple medical scenarios. Background Technology

[0002] The reporting and handling system in the medical field is a technical system that supports the handling of medical events (including public health emergencies, medical quality and safety incidents, infectious / foodborne diseases, adverse drug and medical device reactions, etc.) from information collection, reporting, transmission, review to analysis and judgment, emergency response, and closed-loop management. It fundamentally addresses issues such as "untimely reporting, inconsistent data, low handling efficiency, and difficulty in tracing the source" in medical events. Simultaneously, it meets the compliance, security, and real-time requirements of the medical industry. Its underlying structure integrates information technology, communication technology, data technology, and intelligent technology, and is combined with medical industry standards to form a proprietary technical architecture.

[0003] However, there are problems such as fragmented reporting scenarios, with reporting of single diseases, adverse medical events, critical values, and infectious diseases requiring different systems. Medical staff need to switch between these systems, resulting in a large workload for manual data entry and low reporting accuracy. Furthermore, the closed-loop handling process is fragmented, with most reporting only completing the "report-review-upload" process, lacking the steps of handling execution, effect verification, and dynamic optimization. Reported items often have the problem of "no feedback or rectification after handling," indicating an incomplete closed-loop process. The existing technology's "closed loop" only covers "report-review-allocation," lacking the steps of handling process traceability, rectification effect verification, and dynamic optimization. The process fragmentation leads to a disconnect between reporting and handling. Secondly, there is also insufficient data collaboration and traceability. Most systems only connect to a single system within the hospital, failing to achieve cross-system and cross-institutional (hospital-CDC-community) collaboration, and lacking full-process data traceability capabilities. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for automated reporting and closed-loop processing of multiple medical scenarios to solve the above-mentioned problems in the prior art.

[0005] This invention is achieved through the following technical solution: A method for automated reporting and closed-loop processing across multiple medical scenarios, characterized by comprising: Acquire multimodal data, construct a disease evolution stage judgment model, and output the current disease evolution stage label and stage confidence based on the multimodal data and the disease evolution stage judgment model; Obtain the risk dimension set for this stage, set sensitivity weights for each risk dimension in the risk dimension set, perform structured mapping on the patient's current condition data based on the risk dimension set, and construct a risk state vector of risk composition under this stage; Acquire stage migration data from historical medical records, construct a stage migration probability model, output the migration probability of the current stage based on the risk state vector and stage migration data, set a migration probability threshold, and determine whether to trigger a stage migration warning based on the migration probability threshold. Based on a predefined closed-loop control strategy library that matches the stage transition probability, corresponding intervention measures are triggered, and the intervention effect is monitored in real time. The strategy is dynamically adjusted according to the changes in the risk state vector.

[0006] Preferably, the acquisition and preprocessing of multimodal data includes: Determine whether the current multimodal data is structured data; If so, use linear interpolation to fill in the missing values ​​and eliminate dimensional differences through Z-score standardization; If not, key clinical features are extracted using an NLP model, transformed into structured feature vectors, and a unified multimodal temporal feature matrix is ​​obtained.

[0007] Preferably, the construction of the disease progression stage judgment model includes: The dynamic features of the preprocessed multimodal time series data are extracted based on the Transformer encoder. The dynamic features are then input into the DBSCAN algorithm model to identify disease stages with similar evolution patterns and obtain clustering results. It has historical clinical and expert experience data, performs semantic annotation on clustering results, defines clinical characteristics and transition thresholds for each stage, and outputs stage labels and stage confidence scores.

[0008] Preferred options also include: When the stage label output by the disease evolution stage judgment model remains consistent for N consecutive time steps, and the stage confidence is greater than or equal to the confidence threshold, the stage judgment result is confirmed. If the confidence level of a stage is lower than the confidence threshold or if the stage switches frequently, a manual review process will be triggered to correct the stage determination.

[0009] Preferably, the risk state vector constituting the risk at this stage includes:

[0010] In the formula, Let the risk state vector be... From 1 to Each risk dimension in the stage time The risk component below.

[0011] Preferably, the risk components include:

[0012] In the formula, For the first Each risk dimension in the stage time The risk component below, , For the first Each risk dimension in the stage Sensitivity weights under, To map the degree of deviation and trend of the indicator to a function of risk intensity, For the first Each risk dimension at any time The measured value, This indicates the trend of the indicator within a preset time window. To mark the stage of disease evolution The corresponding stage baseline or interval.

[0013] Preferably, the construction phase migration probability model includes: Obtain the set of all disease evolution stages. and risk state vector sequence ,in, for Risk state vector at time t; The transition probability matrix is ​​obtained by training with historical case data. ,in Indicates from stage Towards the stage The probability of migration; The Gaussian mixture model (GMM) was used to fit the data at the stage... The probability of observing the risk state vector.

[0014] Preferably, the step of outputting the migration probability of the current stage based on the risk state vector and stage migration data through a stage migration probability model includes:

[0015] In the formula, To start from the current stage To other stages The migration probability, To be from the stage To other stages The transition probability, For the current stage To any other stage The prior transition probability of migration occurring. In the stage The probability of observing the current risk state vector. In the stage The probability of observing the current risk state vector.

[0016] Preferably, the step of setting a migration probability threshold and determining whether to trigger a phased migration warning based on the migration probability threshold includes: When the migration probability is greater than or equal to the migration probability threshold, a stage migration early warning is triggered, and a signal is issued that attention should be paid to the disease evolution trend.

[0017] Secondly, the present invention also provides a medical multi-scenario automated reporting and end-to-end closed-loop processing system for the aforementioned medical multi-scenario automated reporting and end-to-end closed-loop processing method, comprising: The data processing module is configured to acquire and preprocess multimodal data, construct a disease evolution stage judgment model, and output the current disease evolution stage label and stage confidence based on the preprocessed multimodal data through the disease evolution stage judgment model; obtain the risk dimension set for the stage according to the disease evolution stage label being located in the stage risk dimension mapping library, and set sensitivity weights for each risk dimension in the risk dimension set; and perform structured mapping on the patient's current disease data based on the risk dimension set to construct a risk state vector of risk composition under the stage. The migration probability judgment module is configured to acquire stage migration data from historical medical records, construct a stage migration probability model, output the migration probability of the current stage based on the risk state vector and stage migration data, set a migration probability threshold, determine whether to trigger a stage migration warning based on the migration probability threshold, match the stage migration probability with a predefined closed-loop control strategy library, trigger corresponding intervention measures, monitor the intervention effect in real time, and dynamically adjust the strategy according to changes in the risk state vector.

[0018] The technical solution of the present invention has at least the following advantages and beneficial effects: The method provided by this invention mainly includes acquiring stage migration data from historical medical records, constructing a stage migration probability model, outputting the migration probability of the current stage based on the risk state vector and stage migration data through the stage migration probability model, setting a migration probability threshold, and determining whether to trigger a stage migration warning based on the migration probability threshold. Through this method, multi-scenario adaptive automated reporting is achieved, eliminating the need for medical staff to switch systems or manually fill in data. Combined with efficient processing capabilities for unstructured data, it significantly reduces manual workload and reporting error rates, improving reporting efficiency. Furthermore, it constructs a full-link closed-loop handling mechanism, addressing the shortcomings of existing technical processes. Through multi-dimensional effect verification and dynamic optimization, it ensures that every reported item is "handled, verified, and optimized," significantly improving medical quality control and handling quality. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the process of the present invention; Figure 2 This is a schematic diagram of the system structure of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0022] Please refer to Figures 1-2 This invention provides a method for automated reporting and closed-loop processing across multiple medical scenarios, comprising: S101: Acquire multimodal data and preprocess it, construct a disease evolution stage judgment model, and output the current disease evolution stage label and stage confidence based on the preprocessed multimodal data through the disease evolution stage judgment model; S102: Based on the disease evolution stage label being located in the stage risk dimension mapping library, obtain the risk dimension set for that stage and set a sensitivity weight for each risk dimension in the risk dimension set; S103: Based on the risk dimension set, perform structured mapping on the patient's current condition data to construct a risk state vector of risk composition at this stage; S104: Obtain stage migration data from historical medical records, construct a stage migration probability model, output the migration probability of the current stage based on the risk state vector and stage migration data through the stage migration probability model, set a migration probability threshold, and determine whether to trigger a stage migration warning based on the migration probability threshold. S105: Based on a predefined closed-loop control strategy library for stage transition probability matching, trigger corresponding intervention measures, monitor the intervention effect in real time, and dynamically adjust the strategy according to the changes in the risk state vector.

[0023] The method provided by this invention mainly includes acquiring stage migration data of historical medical records, constructing a stage migration probability model, outputting the migration probability of the current stage based on the risk state vector and stage migration data through the stage migration probability model, setting a migration probability threshold, and determining whether to trigger a stage migration warning based on the migration probability threshold. Through the above methods, multi-scenario adaptive automated reporting is achieved, eliminating the need for medical staff to switch systems or manually fill in data. Combined with efficient unstructured data processing capabilities, it significantly reduces manual workload and reporting error rates, thereby improving reporting efficiency. A full-link closed-loop handling mechanism is constructed to address the shortcomings of existing technical processes. Through AI-powered intelligent allocation, multi-dimensional effect verification, and dynamic optimization, it ensures that every reported item is "handled, verified, and optimized," significantly improving medical quality control and handling quality. A cross-system and cross-institutional collaborative traceability system is established to break down data barriers, enabling real-time data interoperability and full-process immutability and traceability, solving the problems of difficulty in tracing responsibility and disjointed handling, while meeting medical data security and privacy protection requirements. It has differentiated adaptability capabilities, flexibly adapting to the needs of hospitals at different levels. Adopting a lightweight integration model, it does not require the reconstruction of existing systems, reducing implementation costs and facilitating promotion and application in medical institutions at all levels. This invention can effectively improve the automation and standardization of medical reporting and treatment, reduce the workload of medical staff, reduce medical safety risks, and meet the relevant policy requirements of national medical quality control and performance evaluation. It has outstanding substantive features and significant progress, and has broad application prospects.

[0024] In one exemplary embodiment of the present invention, traditional disease assessment often relies on static single-dimensional indicators or empirical staging, which cannot dynamically capture the continuous evolution of a patient's condition over time. Furthermore, information from different modalities of data (such as vital signs, test results, and imaging reports) is not effectively integrated, leading to delayed and inaccurate disease stage determination, making it difficult to support subsequent precise risk assessment and intervention decisions. Therefore, this step aims to: construct a dynamic disease evolution stage determination model by integrating multimodal time-series data, accurately classifying the patient's current disease stage, and providing a foundation for subsequent staged risk assessment and closed-loop control.

[0025] We collect multimodal time-series data of patients throughout their entire lifecycle (including vital signs, laboratory tests, imaging reports, electronic medical record texts, etc.). Through data cleaning, standardization, and feature extraction, we construct a disease evolution stage determination model and output the patient's current disease evolution stage and stage confidence level. This stage division has clear clinical semantics, with different stages corresponding to different pathophysiological characteristics and risk patterns, providing stage constraints for the subsequent construction of risk state vectors.

[0026] Specifically, acquiring and preprocessing multimodal data includes: Determine whether the current multimodal data is structured data, including structured data such as vital signs and test results, and unstructured data such as electronic medical records and image reports; If so, use linear interpolation to fill in the missing values ​​and eliminate dimensional differences through Z-score standardization; If not, key clinical features are extracted using an NLP model, transformed into structured feature vectors, and a unified multimodal temporal feature matrix is ​​obtained.

[0027] The construction of a disease progression assessment model includes: The dynamic features of the preprocessed multimodal time series data are extracted based on the Transformer encoder. The dynamic features are then input into the DBSCAN algorithm model to identify disease stages with similar evolution patterns and obtain clustering results. DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is a classic unsupervised density clustering algorithm that uses the spatial density of sample points to divide clusters. It can automatically discover clusters of arbitrary shapes and identify noise points without specifying the number of clusters in advance.

[0028] It has historical clinical and expert experience data, performs semantic annotation on clustering results, defines clinical characteristics and transition thresholds for each stage, and outputs stage labels and stage confidence scores, with confidence scores ranging from 0 to 1.

[0029] It also includes phase stability verification: When the stage label output by the disease evolution stage judgment model remains consistent for N consecutive time steps, and the stage confidence is greater than or equal to the confidence threshold, the stage judgment result is confirmed. N can be three. If the confidence level of a stage is lower than the confidence threshold or if the stage switches frequently, a manual review process will be triggered to correct the stage determination. The confidence threshold can be 0.8.

[0030] This step outputs the label for the patient's current stage of disease progression. The stage confidence level provides core input for step two, “Configure the set of staged risk dimensions”, ensuring that subsequent risk assessments are accurately matched with stage characteristics; at the same time, the multimodal time series feature matrix will provide raw data support for step three, “Construction of risk state vectors”.

[0031] If this step is not performed, subsequent risk assessments will lack clear stage constraints, leading to semantic confusion of risks at different stages and making it impossible to achieve accurate dynamic risk management.

[0032] In one exemplary embodiment of the present invention, existing risk assessment systems often employ a fixed set of risk dimensions, failing to consider the differences in disease evolution stages. This results in the homogenization of the risk weights and clinical significance of the same indicator at different stages, failing to accurately reflect stage-specific risks. Therefore, this step aims to: dynamically configure the corresponding set of risk dimensions and stage-sensitive weights for the disease evolution stage output in Step 1, ensuring the stage-appropriateness and clinical interpretability of subsequent risk assessments.

[0033] Based on the disease evolution stage labels obtained from S101, a predefined stage-risk dimension mapping library is invoked to extract the set of core risk dimensions for that stage, and stage sensitivity weights are assigned to each dimension. This mapping library was constructed by clinical experts in conjunction with the pathophysiological characteristics of each stage, with different risk dimensions and weight configurations corresponding to different stages, ensuring a high degree of matching between risk assessment and stage characteristics.

[0034] Specifically, this includes a phase-risk dimension mapping library built by a multidisciplinary team of clinical experts (including emergency, critical care, and internal medicine specialists) based on clinical guidelines and real-world data. A multidisciplinary team of clinical experts (including emergency, critical care, and internal medicine specialists) has constructed a phase-risk dimension mapping library based on clinical guidelines and real-world data. Each stage of disease evolution This corresponds to a set of risk dimensions:

[0035] in, For the stage The set of risk dimensions For the stage The Several core risk dimensions, such as "lactic acid level" during the shock phase and "procalcitonin" during the infection phase; Assign stage sensitivity weights to each risk dimension The sensitivity weights are determined using the Analytic Hierarchy Process (AHP), and their sum is 1, reflecting the relative importance of the dimension at the current stage.

[0036] The system is based on stage labels Automatically retrieve the corresponding risk dimension set from the mapping library. With sensitivity weight : (1) If a phase changes, update the risk dimension set and weight configuration in real time; (2) Support clinical experts to adjust the contents of the mapping library through a visual interface to achieve dynamic iterative optimization.

[0037] It also includes dimension validity verification, which verifies the validity of the configured set of risk dimensions: (1) Use feature importance analysis (such as random forest feature importance) to verify the discriminative power of each dimension at the current stage; (2) If the importance score of a certain dimension is lower than the preset threshold (e.g., <0.05), an early warning will be triggered and experts will be prompted to reassess the clinical value of that dimension.

[0038] Risk dimensions set for this step With sensitivity weight This step provides core dimensional constraints and weight inputs for subsequent risk state vector construction, ensuring the stage-specific adaptability and clinical interpretability of the risk state vectors. Simultaneously, the dynamic iteration mechanism of the mapping library supports subsequent system optimization. Without this step, subsequent risk state vectors will use fixed dimensions and weights, failing to reflect stage-specific risks and significantly reducing the accuracy and clinical value of risk assessment.

[0039] In one exemplary embodiment of the present invention, existing risk assessment methods often compress multi-dimensional indicators into a single risk score, which has significant drawbacks: the contributions of different risk sources easily cancel each other out, masking key risk signals; they cannot distinguish the essential difference between "a change in risk structure" and "an overall increase in risk"; and they are difficult to output interpretable structured information to support subsequent risk migration modeling and closed-loop control decisions. Therefore, this step aims to: under the premise of clearly defining the stage of disease evolution, reconstruct the patient's current risk from a "single score" into a "risk state vector" with clear structural meaning, so that the risk expression can reflect the relative contribution of different risk sources at the current stage, providing stateful input for subsequent risk migration modeling and closed-loop control.

[0040] Based on the disease progression stage assessment, a structured mapping is performed on the patient's current condition data for a predefined set of risk dimensions at that stage, constructing a risk state vector to characterize the patient's risk composition at that stage. This risk state vector consists of multiple risk components, each corresponding to a clinically significant source of risk, clearly describing the internal structure of the patient's current risk status.

[0041] Specifically, the risk state vector constituting the risks at this stage includes:

[0042] In the formula, Let the risk state vector be... From 1 to Each risk dimension in the stage time The risk component below, The number of risk dimensions.

[0043] Secondly, the risk components include:

[0044] In the formula, For the first Each risk dimension in the stage time The risk component below, , For the first Each risk dimension in the stage Sensitivity weights under, To map the degree of deviation and trend of the indicator to a function of risk intensity, For the first Each risk dimension at any time The measured value, This indicates the trend of the indicator within a preset time window. To mark the stage of disease evolution The corresponding stage baseline or interval.

[0045] When constructing risk state vectors, the following should be followed: (1) Within the same stage of disease progression, the set of risk dimensions remains consistent; (2) The dimensional order and semantics of the risk state vector are fixed within the stage to ensure the comparability between subsequent risk states; (3) When the disease progression stage changes, the risk dimension set and weight configuration can be switched.

[0046] This mechanism avoids unreasonable comparisons of risk states at different stages, ensuring the effectiveness of risk evolution analysis from a structural perspective.

[0047] Unlike the traditional method of obtaining a single risk score through a weighted sum of multiple indicators, this step preserves the independent representation of each risk source in the risk state vector, enabling the identification of the following key scenarios: (1) The overall risk level has not changed much, but the composition of risk has changed significantly; (2) One key risk dimension continues to deteriorate, while other dimensions remain stable; (3) The importance of different risk dimensions shifts at different stages.

[0048] This step outputs a risk state vector with clear structural meaning and stage semantics. This vector will be used as the state input for calculating the risk migration probability in step four, to determine whether the patient's condition has a tendency to jump to other disease evolution stages.

[0049] Existing technologies that use only a single risk score as input cannot characterize changes in risk structure. Risk migration modeling will degenerate into a simple judgment based on a threshold, which cannot support the fine-grained triggering of subsequent closed-loop control strategies.

[0050] In one exemplary embodiment of the present invention, traditional disease monitoring only focuses on the risk level of the current stage and cannot predict the migration trend of the disease to other stages, resulting in delayed intervention decisions and missed opportunities for optimal treatment. Therefore, this step aims to: construct a disease evolution stage migration probability model based on the risk state vector output in step three, quantify the probability of a patient transitioning from the current stage to other stages, and provide a forward-looking basis for triggering subsequent closed-loop control strategies.

[0051] A stage transition probability model is constructed using a Hidden Markov Model (HMM) combined with risk state vectors. Specifically, the construction of the stage transition probability model includes: Obtain the set of all disease evolution stages. and risk state vector sequence ,in, for Risk state vector at time t; The transition probability matrix is ​​obtained by training with historical case data. ,in Indicates from stage Towards the stage The probability of migration; The Gaussian mixture model (GMM) was used to fit the data at the stage... The probability of observing the risk state vector.

[0052] Secondly, based on the risk state vector and stage transition data, the transition probability of the current stage is output through the stage transition probability model, including:

[0053] In the formula, To start from the current stage To other stages The migration probability, To be from the stage To other stages The transition probability, For the current stage To any other stage (including stages) The prior transition probability of migration occurring. In the stage The probability of observing the current risk state vector. In the stage The probability of observing the current risk state vector.

[0054] And set a migration probability threshold, and determine whether to trigger a phased migration warning based on the migration probability threshold, including: When the migration probability is greater than or equal to the migration probability threshold, a stage migration early warning is triggered, and a signal is issued indicating the need to monitor the disease's evolution trend. The threshold can be dynamically adjusted according to clinical needs to balance the sensitivity and specificity of the early warning.

[0055] This step outputs the disease progression stage migration probability matrix and stage migration early warning signal, providing a forward-looking decision-making basis for step five, "Triggering and Execution of Closed-Loop Control Strategy," to ensure the timeliness and accuracy of intervention measures.

[0056] Closed-loop control strategy triggering and execution based on stage transition probability: Current clinical interventions largely rely on physicians' experience and judgment, lacking a closed-loop control mechanism based on dynamic risk evolution. This leads to inaccurate timing and imprecise measures, impacting patient prognosis. Therefore, this step aims to: construct a closed-loop control strategy library based on the stage transition probabilities output from step four, enabling automatic triggering and dynamic adjustment of intervention measures to ensure timeliness, accuracy, and effectiveness of interventions.

[0057] Construction of closed-loop control strategy library: A closed-loop control strategy library was constructed by a team of clinical experts based on the pathophysiological characteristics and migration risks at each stage. (1) Each strategy corresponds to a stage migration scenario (such as migration from "stable stage" to "deterioration stage"), which includes intervention measures (such as drug adjustment, life support upgrade), trigger threshold (such as migration probability ≥ 0.3) and effect evaluation indicators (such as risk state vector change); (2) The strategy library supports dynamic iteration and can be continuously optimized based on clinical practice data and expert feedback.

[0058] Strategy Triggering and Execution: When the stage transition probability reaches the trigger threshold, the system automatically matches and executes the corresponding intervention strategy: (1) Automatically generate intervention instructions (such as medical orders and nursing measures) and push them to the clinical work terminal; (2) Collect patient data after intervention in real time, update risk state vector, and evaluate intervention effect.

[0059] Dynamic strategy adjustment: Based on the changes in the risk state vector after intervention, the intervention strategy is dynamically adjusted: (1) Several prognostic risk state vectors evolve in a benign direction (e.g., risk components decrease, migration probability decreases), maintaining the current strategy; (2) Several pre-treatment effects are not good (such as the risk component continues to rise and the migration probability does not decrease), triggering strategy upgrades (such as upgrading from drug intervention to life support) or switching to backup strategies.

[0060] This step outputs closed-loop control intervention records and intervention effect evaluation reports, providing core feedback data for step six, "System Dynamic Optimization," while directly improving patient prognosis and enhancing clinical treatment efficiency.

[0061] Dynamic system optimization based on end-to-end data: The dynamic changes in clinical practice and patient conditions require systems to have the ability to continuously iterate and optimize. Existing systems are mostly statically designed and cannot adapt to the evolving clinical needs, leading to a decline in long-term effectiveness. Therefore, this step aims to: build a dynamic optimization mechanism for the system based on full-process data (including stage determination, risk assessment, intervention implementation, and prognostic results), continuously improve model accuracy and intervention effectiveness, and ensure the long-term clinical value of the system.

[0062] Collect data from steps S101 to S105 throughout the entire process, and construct a system optimization database, including: (1) Core indicators such as stage judgment accuracy, risk state vector interpretability, migration probability prediction accuracy, and intervention effectiveness; (2) Use machine learning algorithms (such as XGBoost) to analyze the correlation between indicators and system parameters and identify optimization points (such as changes in the feature importance of the stage judgment model and unreasonable trigger thresholds of intervention strategies).

[0063] Dynamically optimize execution: Based on the data analysis results, targeted optimizations were performed on each module of the system: (1) Model optimization: Retrain the disease evolution stage judgment model and the stage transition probability model, and update the feature weights and transition probability matrix; (2) Rule optimization: Adjust the trigger thresholds and intervention measures of the risk dimension set, risk state vector construction logic, and closed-loop control strategy library; (3) Iterative verification: The leave-one-out method is used to cross-validate the optimization effect and ensure that the core indicators (such as the accuracy of stage judgment and the effectiveness of intervention) are improved by ≥5%.

[0064] Optimization feedback and iteration: The optimization results are fed back to the clinical expert team for clinical validation and evaluation. Based on the feedback, the optimization strategy is further adjusted to form a continuous optimization closed loop of "data collection - analysis and optimization - validation and feedback - iterative upgrade".

[0065] This step outputs a system optimization report and iteratively upgraded system parameters, providing continuous performance assurance for subsequent clinical applications, ensuring that the system always adapts to the dynamic changes in clinical needs, and maintaining long-term clinical value.

[0066] A multi-scenario automated reporting and end-to-end closed-loop processing system for medical applications, used to execute the aforementioned multi-scenario automated reporting and end-to-end closed-loop processing method for medical applications, including: The data processing module is configured to acquire and preprocess multimodal data, construct a disease evolution stage judgment model, and output the current disease evolution stage label and stage confidence based on the preprocessed multimodal data through the disease evolution stage judgment model; obtain the risk dimension set for the stage according to the disease evolution stage label being located in the stage risk dimension mapping library, and set sensitivity weights for each risk dimension in the risk dimension set; and perform structured mapping on the patient's current disease data based on the risk dimension set to construct a risk state vector of risk composition under the stage. The migration probability judgment module is configured to acquire stage migration data from historical medical records, construct a stage migration probability model, output the migration probability of the current stage based on the risk state vector and stage migration data, set a migration probability threshold, determine whether to trigger a stage migration warning based on the migration probability threshold, match the stage migration probability with a predefined closed-loop control strategy library, trigger corresponding intervention measures, monitor the intervention effect in real time, and dynamically adjust the strategy according to changes in the risk state vector.

[0067] The solution using the present invention has at least the following effects: This multi-scenario adaptive automated reporting mechanism differs from the "single-scenario" design limitations of existing technologies. It constructs a "general reporting engine + scenario adaptation module," which can simultaneously adapt to four core medical reporting scenarios: single disease, adverse medical events, critical values, and infectious diseases. It also supports users to customize and add new scenarios (such as hospital-acquired infections, abnormal consumables, etc.) through a visual interface without refactoring the system. Through a "scenario tag library + dynamic rule configuration," it achieves multi-scenario adaptation. Combined with NLP and a self-developed medical big data model, it can efficiently process unstructured data such as handwritten medical records and voice reports, with a conversion accuracy of ≥98%. It solves the problems of weak unstructured data processing capabilities and poor scenario adaptability of existing technologies, and creatively realizes integrated automated reporting across multiple scenarios.

[0068] The "reporting-allocation-disposal-verification-optimization" closed-loop disposal mechanism overcomes the shortcomings of existing technologies in terms of "closed-loop process gaps." By combining AI intelligent allocation algorithms with blockchain traceability technology, it achieves a seamless closed loop throughout the entire process. An AI allocation model is constructed using the Analytic Hierarchy Process (AHP) and K-means clustering algorithms, which can automatically allocate responsible parties based on scenario type, disposal priority, and department / personnel qualifications, with a response time of ≤30 seconds. A multi-dimensional verification module for disposal effectiveness is introduced, which automatically determines the rectification effect by comparing data before and after disposal using AI; if the standard is not met, a second rectification is triggered. A new dynamic optimization module is added, which automatically optimizes reporting rules and allocation algorithms based on data from the entire reporting and disposal process, forming a virtuous cycle of "reporting-disposal-optimization." This creatively solves the problems of existing technologies lacking verification, optimization, and accountability in disposal.

[0069] This cross-level, cross-system collaborative linkage and full-dimensional data traceability system differs from the limitations of existing technologies that only allow collaboration within a single hospital system. It adopts a standardized HL7 FHIR interface to achieve seamless integration with internal hospital systems such as HIS, LIS, EMR, and RIS, as well as national / local medical quality control platforms, disease control platforms, and community health service platforms, enabling real-time data exchange. Based on a blockchain consortium chain architecture and data fingerprint technology, it assigns a unique traceability identifier to each reported data and treatment record, supporting multi-level drill-down queries from "hospital-department-doctor-patient". This achieves tamper-proof traceability throughout the entire process of data collection, reporting, treatment, and verification, while also meeting medical data privacy protection requirements. It creatively solves the pain points of existing technologies, such as poor cross-system and cross-institutional collaboration and the inability to trace data in all dimensions.

[0070] The differentiated adaptation and low-cost implementation design, unlike the shortcomings of existing technologies that "cannot adapt to hospitals of different levels", provides a personalized rule configuration interface. Tertiary hospitals can add detailed reporting indicators and strengthen quality control requirements, while community hospitals can simplify the reporting process and remove unnecessary indicators. Adopting the "interface adaptation + lightweight integration" model, it can directly connect to the existing systems of medical institutions at all levels without reconstruction, reducing implementation costs and creatively achieving differentiated adaptation for hospitals of different levels, thereby improving the system's practicality and promotion value.

[0071] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0072] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. This computer software product, stored in a storage medium, includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0073] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A medical multi-scene automatic reporting and full-link closed-loop treatment method, characterized in that, include: Acquire multimodal data, construct a disease evolution stage judgment model, and output the current disease evolution stage label and stage confidence based on the multimodal data and the disease evolution stage judgment model; Obtain the risk dimension set for this stage, set sensitivity weights for each risk dimension in the risk dimension set, perform structured mapping on the patient's current condition data based on the risk dimension set, and construct a risk state vector of risk composition under this stage; Acquire stage migration data from historical medical records, construct a stage migration probability model, output the migration probability of the current stage based on the risk state vector and stage migration data, set a migration probability threshold, and determine whether to trigger a stage migration warning based on the migration probability threshold. Based on a predefined closed-loop control strategy library that matches the stage transition probability, corresponding intervention measures are triggered, and the intervention effect is monitored in real time. The strategy is dynamically adjusted according to the changes in the risk state vector. 2.The medical multi-scene automatic reporting and full-link closed-loop processing method according to claim 1, wherein, The acquisition and preprocessing of multimodal data includes: Determine whether the current multimodal data is structured data; If so, use linear interpolation to fill in the missing values ​​and eliminate dimensional differences through Z-score standardization; If not, key clinical features are extracted using an NLP model, transformed into structured feature vectors, and a unified multimodal temporal feature matrix is ​​obtained. 3.The medical multi-scene automatic reporting and full-link closed-loop processing method according to claim 2, characterized in that, The constructed disease progression stage judgment model includes: The dynamic features of the preprocessed multimodal time series data are extracted based on the Transformer encoder. The dynamic features are then input into the DBSCAN algorithm model to identify disease stages with similar evolution patterns and obtain clustering results. It has historical clinical and expert experience data, performs semantic annotation on clustering results, defines clinical characteristics and transition thresholds for each stage, and outputs stage labels and stage confidence scores. 4.The medical multi-scene automatic reporting and full-link closed-loop processing method of claim 3, wherein, Also includes: When the stage label output by the disease evolution stage judgment model remains consistent for N consecutive time steps, and the stage confidence is greater than or equal to the confidence threshold, the stage judgment result is confirmed. If the confidence level of a stage is lower than the confidence threshold or if the stage switches frequently, a manual review process will be triggered to correct the stage determination.

5. The method for automated reporting and closed-loop processing of multiple medical scenarios according to claim 4, characterized in that, The risk state vector constituting the risk at this stage includes: In the formula, Let be the risk state vector. From 1 to Each risk dimension in the stage time The risk component below.

6. The method for automated reporting and closed-loop processing of multiple medical scenarios according to claim 5, characterized in that, The risk components include: In the formula, For the first Each risk dimension in the stage time The risk component below, , For the first Each risk dimension in the stage Sensitivity weights under, To map the degree of deviation and trend of the indicator to a function of risk intensity, For the first Each risk dimension at any time The measured value, This indicates the trend of the indicator within a preset time window. To mark the stage of disease evolution The corresponding stage baseline or interval.

7. The method for automated reporting and closed-loop processing of multiple medical scenarios according to claim 6, characterized in that, The migration probability model during the construction phase includes: Obtain the set of all disease evolution stages. and risk state vector sequence ,in, for Risk state vector at time t; The transition probability matrix is ​​obtained by training with historical case data. ,in Indicates from stage Towards the stage The probability of migration; The Gaussian mixture model (GMM) was used to fit the data at the stage... The probability of observing the risk state vector.

8. A method for automated reporting and closed-loop processing of multiple medical scenarios according to claim 7, characterized in that, The step of outputting the migration probability of the current stage based on the risk state vector and stage migration data through the stage migration probability model includes: In the formula, To start from the current stage To other stages The migration probability, To be from the stage To other stages The transition probability, To start from the current stage To any other stage The prior transition probability of migration occurring In the stage The probability of observing the current risk state vector. In the stage The probability of observing the current risk state vector.

9. A method for automated reporting and closed-loop processing of multiple medical scenarios according to claim 8, characterized in that, The process of setting a migration probability threshold and determining whether to trigger a phased migration warning based on the migration probability threshold includes: When the migration probability is greater than or equal to the migration probability threshold, a stage migration early warning is triggered, and a signal is issued that attention should be paid to the disease evolution trend.

10. A multi-scenario automated reporting and end-to-end closed-loop processing system for medical care, characterized in that, The method for automated reporting and closed-loop processing of multiple medical scenarios as described in any one of claims 1-9 includes: The data processing module is configured to acquire and preprocess multimodal data, construct a disease evolution stage judgment model, and output the current disease evolution stage label and stage confidence based on the preprocessed multimodal data through the disease evolution stage judgment model; obtain the risk dimension set for the stage according to the disease evolution stage label being located in the stage risk dimension mapping library, and set sensitivity weights for each risk dimension in the risk dimension set; and perform structured mapping on the patient's current disease data based on the risk dimension set to construct a risk state vector of risk composition under the stage. The migration probability judgment module is configured to acquire stage migration data from historical medical records, construct a stage migration probability model, output the migration probability of the current stage based on the risk state vector and stage migration data, set a migration probability threshold, determine whether to trigger a stage migration warning based on the migration probability threshold, match the stage migration probability with a predefined closed-loop control strategy library, trigger corresponding intervention measures, monitor the intervention effect in real time, and dynamically adjust the strategy according to changes in the risk state vector.

Images