An artificial intelligence-based in-hospital integrated platform system

Abstract

The application relates to the field of intelligent medical treatment, and discloses an in-hospital integrated platform system based on artificial intelligence, which comprises the following steps: extracting current patient identification information, identifying a surgical patient in an anesthetic state in real time, and automatically associating a preoperative electronic medical record with an intraoperative anesthesia plan; based on an artificial intelligence semantic recognition model, image text extraction and semantic understanding are carried out on parameter content set in real time in a respirator equipment screen; according to the difference between the set parameters and the semantic recognition of the system, a cross-system risk factor atlas is constructed, and a risk linkage score is generated in real time for intraoperative operation behavior; the parameter evolution path under the same type of operation in the past is called, and analogy modeling is carried out in combination with the current patient anesthesia depth, physiological data and the intraoperative stage; when it is confirmed that there is a strategy-level abnormality in setting, the warning level and the display form are dynamically adjusted based on the operation equipment type, the setting personnel role and the current stage load intensity. The application has the advantage of improving the intelligent level of a hospital.

Description

An AI-based in-hospital integrated platform system Technical Field

[0001] This invention relates to the field of smart healthcare, specifically to an in-hospital integrated platform system based on artificial intelligence. Background Technology

[0002] In modern surgery, ventilators are crucial for stabilizing patients' vital signs, especially under general anesthesia, where spontaneous breathing is suppressed, making patients entirely reliant on continuous and precise ventilation. Several ventilator operating parameters, such as tidal volume, respiratory rate, and oxygen concentration, need dynamic adjustment based on intraoperative conditions and the patient's specific vital signs. Oxygen concentration setting is particularly critical; both excessively high and low concentrations can lead to intraoperative complications. Currently, oxygen concentration settings are mostly manually entered by anesthesiologists based on the type of surgery and the patient's condition. However, during rapid patient transitions or shift changes, incomplete or negligent information transmission can easily lead to parameters not being adjusted in a timely manner. This is especially true in emergency situations where medical staff often prioritize life support and medication intervention, neglecting to verify equipment settings. Furthermore, existing medical AI systems largely focus on diagnostic and treatment decision support or image recognition, lacking the ability to perceive and handle specific details related to equipment use during surgery. Currently, there is no systematic solution that can intervene in real-time at the level of detailed equipment parameter settings, combining upstream and downstream system data, using artificial intelligence. Therefore, it is essential to design an AI-based in-hospital integrated platform system to enhance the hospital's intelligence level. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides an integrated hospital platform system based on artificial intelligence, which has the advantage of improving the intelligence level of hospitals and solves the problems mentioned in the background technology.

[0004] To achieve the aforementioned goal of improving the intelligence level of hospitals, this invention provides the following technical solution: an in-hospital integrated platform system based on artificial intelligence, comprising:

[0005] Tag matching module: Extracts current patient identification information, identifies surgical patients under anesthesia in real time, automatically associates preoperative electronic medical records with intraoperative anesthesia plans, determines whether there is an oxygen concentration setting requirement, and if so, proceeds to parameter semantic parsing module;

[0006] Semantic parsing module: Based on an artificial intelligence semantic recognition model, it extracts images and text and performs semantic understanding on the parameters set in real time on the ventilator device screen. Combined with the preoperative contraindications and recommended oxygen concentration range in the electronic medical record, it judges whether there is a semantic deviation in the set value. If so, it enters the instruction risk linkage module.

[0007] Risk linkage module: Based on the set parameters and the semantic differences identified by the system, a cross-system risk factor map is constructed, and risk linkage scoring is performed on intraoperative operation behavior in real time to determine whether the AI ​​intervention conditions are triggered. If so, it enters the intraoperative strategy comparison module.

[0008] Strategy comparison module: retrieves the parameter evolution path under similar surgical procedures in the past, combines the current patient's anesthesia depth, physiological data and intraoperative stage to perform analog modeling, and judges whether there is a strategy deviation in the current set behavior. If so, it enters the early warning trigger feedback module.

[0009] Early warning feedback module: When a policy-level anomaly is confirmed, the early warning level and display format are dynamically adjusted based on the type of operating equipment, the roles of the personnel, and the current load intensity, and graphical intervention suggestions are generated.

[0010] Preferably, the process for determining whether there is an oxygen concentration setting requirement is as follows:

[0011] Extract the binding information between the intraoperative equipment and the patient, and read the patient's current anesthesia status identifier from the anesthesia information system;

[0012] Determine whether the patient's current surgical stage involves spontaneous breathing depression or gas management requirements;

[0013] Based on the preoperative anesthesia plan and surgical requirements, compare whether the recommended oxygen concentration range has been synchronized with the current ventilator oxygen concentration setting;

[0014] If the current setting is missing or deviates from the recommended range by more than the set ratio threshold, it is determined that there is a need to set an oxygen concentration.

[0015] Preferably, the process of image and text extraction and semantic understanding of the parameters set in real time on the ventilator device screen is as follows:

[0016] Acquire the image stream of the ventilator panel area and perform preprocessing operations on the image area, including edge sharpening, color enhancement and distortion correction;

[0017] Using a trained parameter localization model, identify the oxygen concentration parameter labels and set value blocks in the display screen area;

[0018] Extract the set values ​​and, in conjunction with the device manufacturer's parameter layout dictionary, convert the extracted values ​​into semantic tags.

[0019] Preferably, the process of determining whether the set value has semantic deviation is as follows:

[0020] The extracted oxygen concentration setpoint is matched with the recommended oxygen concentration range in the patient's preoperative electronic medical record to compare whether the current setpoint falls within the recommended range.

[0021] Extract high-risk factor markers from patients and construct a clinical context risk factor set;

[0022] Using a knowledge graph-based semantic bias identification model, the set values ​​are fused with the clinical background to calculate the bias risk score;

[0023] If the deviation risk score is greater than the policy tolerance, it is determined that there is a semantic deviation.

[0024] If the deviation risk score is less than or equal to the policy tolerance, it is determined that there is no semantic deviation.

[0025] Preferably, the process of constructing a cross-system risk factor map is as follows:

[0026] Summarize multi-source data related to patient status from ventilator, anesthesia information system, electronic medical record system and monitoring system, including set parameters, historical oxygen consumption curves, intraoperative monitoring indicators and handover operation records;

[0027] The data is standardized and mapped to a risk factor graph template, and cross-system node connections and dependency paths are constructed through graph embedding technology.

[0028] Mark potential abnormal paths and output a cross-system risk factor map.

[0029] Preferably, the process of real-time risk linkage scoring of intraoperative procedures is as follows:

[0030] Extract the system context state at the moment the current set behavior occurs, including the operator role, the set time point, the physiological parameter status curve, and the equipment load status;

[0031] Based on the correlation paths in the cross-system risk factor graph, identify whether operational behaviors trigger high-weight risk links;

[0032] The timing and intensity of each link are weighted and calculated using a time-series risk scoring model to form a risk linkage score for operational behavior.

[0033] Preferably, the process for determining whether the conditions for AI intervention have been triggered is as follows:

[0034] The risk-linked score of the operational behavior was integrated with the patient's condition stability index for analysis;

[0035] The risk level is dynamically adjusted by combining the emergency level during the anesthesia stage with the current operational intensity of the equipment operators;

[0036] When the combined score of the risk index and the intensity of actual operations exceeds the dynamic intervention threshold, the AI ​​intervention process is triggered.

[0037] When the combined score of the risk index and the intensity of actual operations is less than or equal to the dynamic intervention threshold, the AI ​​intervention process will not be triggered.

[0038] Preferably, the analogy modeling process combining the current patient's anesthesia depth, physiological data, and intraoperative stage is as follows:

[0039] Search the structured database for historical cases with similar anesthesia methods, surgical procedures, and underlying disease characteristics, and select a sample set with a similarity greater than a threshold to the current case;

[0040] Extract the oxygen concentration setting change trajectory, intervention response time, and final postoperative recovery data from these samples to form a strategy evolution template;

[0041] By combining the patient’s current BIS anesthesia depth, hemodynamic status and intraoperative operation labels, an analogy modeling algorithm is used to generate the predicted risk distribution of the current strategy in the historical context.

[0042] Output the degree of match between the current behavior setting and the risk setting path in the analog template.

[0043] Preferably, the process of determining whether there is a strategy deviation in the current set behavior is as follows:

[0044] By comparing the current oxygen concentration value with the optimal trajectory range in the analog model, we can determine the magnitude and time dimension of its falling into the offset interval.

[0045] By combining the real-time changes in the patient's vital signs at the time the set behavior occurs, the subsequent physiological chain reactions are assessed;

[0046] Based on the strategy deviation scoring function, the degree of deviation of the current set behavior from the historical analogy model is quantified, and corrections are made by combining the historical operating habits of the set personnel with the error model;

[0047] When the policy bias score exceeds the warning value predicted by the system's self-learning model, it indicates that there is a policy bias in the current behavior setting.

[0048] Preferably, the process of generating graphical intervention suggestions is as follows:

[0049] The system automatically selects an early warning template based on the current deviation type, surgical stage, and equipment type, and distinguishes between color levels, shape symbols, and information display levels.

[0050] By integrating specific indicators of strategy deviation with system judgment logic, a graphical suggestion box is generated on the interface, which includes parameter value comparison, risk prediction trend chart and recommended adjustment range, thus generating graphical intervention suggestions.

[0051] Compared with existing technologies, this invention provides an artificial intelligence-based in-hospital integrated platform system, which has the following beneficial effects:

[0052] This invention utilizes a tag matching module to automatically associate patient identity with key preoperative information, accurately identifying whether specific actions require attention. A semantic parsing module extracts image and text from device screen settings and performs clinical semantic analysis, significantly improving the real-time performance and accuracy of parameter recognition. A risk linkage module integrates cross-system data and knowledge graph technology to score operational behaviors and potential risk paths, providing dynamic response capabilities. A strategy comparison module uses analogical modeling to analyze current settings against historical optimal strategy trajectories, enhancing the ability to identify strategy-level deviations. An early warning feedback module intelligently adjusts intervention levels and suggestion presentation methods based on operator role, equipment load, and intraoperative stage, improving clinical acceptability and user-friendliness. Overall, the system enhances the intelligence level of anesthesia management, reduces intraoperative risks caused by setting errors, and possesses good clinical applicability and prospects for widespread adoption. Attached Figure Description

[0053] Figure 1 is a schematic diagram of the system of the present invention. Detailed Implementation

[0054] 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.

[0055] Example 1: As shown in Figure 1, an artificial intelligence-based in-hospital integrated platform system according to an embodiment of the present invention includes:

[0056] Tag matching module: Extracts current patient identification information, identifies surgical patients under anesthesia in real time, automatically associates preoperative electronic medical records with intraoperative anesthesia plans, determines whether there is a need for oxygen concentration setting, and if so, proceeds to parameter semantic parsing module.

[0057] The process for determining whether an oxygen concentration setting requirement exists in the label matching module is as follows:

[0058] The system extracts the binding information between the intraoperative equipment and the patient, and reads the patient's current anesthesia status identifier from the anesthesia information system. After the patient enters the operating room, the system automatically identifies the patient through surgical scheduling information, wristband identification code, electronic medical record number, etc., and uniquely binds the patient to key intraoperative equipment such as anesthesia machine or ventilator. The equipment uploads its equipment number, equipment type and network address information through its built-in communication interface. The system establishes a mapping relationship between the equipment information and the patient's unique identity and records the binding time. The system accesses the anesthesia information system through the interface protocol to extract the current anesthesia stage information of the patient. The anesthesia stage usually includes the anesthesia preparation period, induction period, maintenance period, awakening period and recovery period. Each stage is manually marked by the anesthesiologist or automatically identified by the system based on events such as drug injection time and respiratory status, and the current status identifier code is updated.

[0059] Determine whether the patient's current surgical stage involves spontaneous breathing depression or requires gas management; based on the patient's anesthesia stage, surgical procedure, ventilation method, and medication use, determine whether the current stage is a critical phase requiring precise management of inhaled oxygen concentration. The assessment process includes the following dimensions:

[0060] Anesthesia stage assessment: If the current stage is the induction or maintenance phase, and muscle relaxants or sedatives are detected, the system will initially determine that the patient is in a state of suppressed spontaneous breathing.

[0061] Auxiliary diagnostic indicators: If the patient has endotracheal intubation, laryngeal mask airway insertion, or is in fully controlled ventilation mode, further confirmation is needed that the patient is unable to autonomously adjust the composition of the inhaled gas.

[0062] Surgical procedure requirement identification: If the surgical procedure is thoracoscopic surgery, one-lung ventilation, prone spinal surgery, or other procedures involving lung isolation and high ventilation requirements, the system assumes that the procedure requires fine control of oxygen concentration. The surgical procedure information is provided by the ProcedureType field in the preoperative plan, and the system determines whether the field contains the GasControlRequired flag.

[0063] In summary, if any judgment condition is met, the system will consider that the current stage requires attention to the inhaled oxygen concentration setting status.

[0064] Based on the preoperative anesthesia plan and surgical requirements, compare the recommended oxygen concentration range with the current ventilator oxygen concentration setting to see if they are synchronized; read the preoperative anesthesia plan form, which usually includes recommended inhaled oxygen concentration ranges set for the patient's underlying diseases, surgical complexity, and intraoperative ventilation goals. For example, for patients without underlying respiratory diseases, the recommended range is 30%–40%; if there is chronic obstructive pulmonary disease or low tidal volume ventilation is planned during the operation, the recommended range may be adjusted to 35%–50% or higher.

[0065] If the current setting is missing or deviates from the recommended range by more than the set ratio threshold, it is determined that there is a need to set an oxygen concentration.

[0066] Semantic parsing module: Based on an artificial intelligence semantic recognition model, it extracts images and text from the parameters set in real time on the ventilator device screen and performs semantic understanding. Combined with the preoperative contraindications and recommended oxygen concentration range in the electronic medical record, it judges whether there is a semantic deviation in the set value. If so, it enters the instruction risk linkage module.

[0067] The semantic parsing module performs image and text extraction and semantic understanding of the parameters set in real time on the ventilator device screen as follows:

[0068] Acquire the image stream of the ventilator panel area and perform preprocessing operations on the image area, including edge sharpening, color enhancement and distortion correction;

[0069] Edge sharpening: Employs Sobel or Laplacian operators to enhance the edges of characters on the display, improving the clarity of the recognition boundaries of numbers and units;

[0070] Color enhancement: Apply histogram equalization or LAB color space adjustment algorithms to improve screen brightness and contrast, and compensate for the effects of screen reflection and low light.

[0071] Distortion correction: If the image has perspective distortion caused by viewpoint shift or wide-angle lens, use affine transformation or OpenCV's perspective transformation function to correct it;

[0072] Using a trained parameter localization model, the oxygen concentration parameter label and the set value block in the display area are identified; a lightweight object detection network is used for parameter region localization. The parameter localization model accepts a pre-processed screen image as input and outputs the coordinates of the labeled bounding box. The model training samples cover ventilator interface images from multiple manufacturers, and the labeled content includes: oxygen concentration label text and set value display area.

[0073] Extract the set values ​​and, in conjunction with the device manufacturer's parameter layout dictionary, convert the extracted values ​​into semantic tags.

[0074] The process of determining whether a set value has a semantic deviation in the semantic parsing module is as follows:

[0075] The extracted oxygen concentration setpoint is matched with the recommended oxygen concentration range in the patient's preoperative electronic medical record (EMR) to compare whether the current setpoint falls within the recommended range. The current set oxygen concentration value is obtained in real time from the clinical monitoring system or anesthesia machine interface and expressed in numerical form, such as 40%, 60%, etc. The patient's preoperative electronic medical record (EMR) is parsed to extract the recommended oxygen concentration range, which is given by the anesthesiologist or attending physician in the preoperative assessment based on risk factors such as the patient's underlying diseases (e.g., COPD, asthma, heart failure), for example, a recommended oxygen concentration of [30%–50%]. A matching algorithm is used to determine the range between the setpoint and the recommended range. If the setpoint falls within the range, it is initially determined to be "no deviation". If it falls outside the range, the next step is to determine whether there is a semantic deviation (i.e., whether the current setpoint and the recommended value have a risk difference due to clinical semantic factors).

[0076] High-risk factor markers were extracted from patients to construct a clinical context risk factor set. High-risk factor information was extracted from EMR, preoperative assessment forms, ICD diagnoses, and laboratory test results, including but not limited to: underlying respiratory diseases (such as COPD, pulmonary fibrosis); cardiovascular diseases (such as coronary heart disease, heart failure); advanced age (such as >75 years old) or low weight (BMI <18); surgical type characteristics (such as open-chest surgery, neurosurgery). The extracted risk factors were structured and converted into labels or vectors that can be used as model input.

[0077] Using a knowledge graph-based semantic bias identification model, a bias risk score is derived by fusing setpoints with clinical context. A medical knowledge graph (including relationships such as disease-treatment-risk-physiological indicators) is introduced, with nodes covering disease names, recommended treatment parameters, safety ranges, and complication risks. Setpoints, recommended intervals, and risk factor sets are mapped to corresponding nodes in the knowledge graph, establishing semantic context links. For example: oxygen concentration = 60% → risk of hyperoxia inhalation; COPD + hyperoxia concentration → potential Increased risk of retention; using graph neural networks or graph attention mechanisms to aggregate and calculate the current set value with the patient's atlas context to generate a bias risk score, which is usually a continuous value between 0 and 1. The higher the score, the greater the risk of semantic bias.

[0078] If the deviation risk score is greater than the policy tolerance, it is determined that there is a semantic deviation.

[0079] If the deviation risk score is less than or equal to the policy tolerance, it is determined that there is no semantic deviation.

[0080] Risk linkage module: Based on the set parameters and the semantic differences identified by the system, a cross-system risk factor map is constructed, and risk linkage scoring is performed on intraoperative operation behavior in real time to determine whether the conditions for AI intervention are triggered. If so, the intraoperative strategy comparison module is entered.

[0081] The process of constructing a cross-system risk factor map in the risk linkage module is as follows:

[0082] Summarize multi-source data related to patient status from ventilator, anesthesia information system, electronic medical record system and monitoring system, including set parameters, historical oxygen consumption curves, intraoperative monitoring indicators and handover operation records;

[0083] The data is standardized and mapped to a risk factor graph template, and cross-system node connections and dependency paths are constructed through graph embedding technology.

[0084] Mark potential abnormal pathways, such as conflicts between preoperative oxygenation contraindications and intraoperative settings, and inconsistencies between equipment settings and patient vital sign trends, and output a cross-system risk factor map.

[0085] The real-time risk linkage scoring process for intraoperative procedures in the risk linkage module is as follows:

[0086] Extract the system context state at the moment the current set behavior occurs, including the operator role, the set time point, the physiological parameter status curve, and the equipment load status;

[0087] Based on the correlation paths in the cross-system risk factor graph, it identifies whether an operational behavior triggers a high-weight risk link; it maps the currently set behavior to the corresponding node in the risk factor graph, such as setting... =80% → Node: High oxygen setting; Use graph indexing mechanism to quickly retrieve the upstream and downstream relationships of this node in the graph; Traverse risk paths directly or indirectly connected to the node of this operation to identify whether they form a closed chain (e.g., COPD → high oxygen). → (Decrease); Each link has a preset weight (from clinical knowledge graph, data statistics, or expert scoring), such as 0.85 for the high oxygen chain and 0.75 for the low body weight intraoperative anesthesia fluctuation chain; For paths triggered by operational behaviors, it is determined whether the complete triggering conditions are met (as shown in the figure, all key condition nodes on the knowledge graph chain are activated), otherwise it is not counted as a risk linkage; Example: Only when "COPD" + "high" "+" The simultaneous occurrence of the three factors of "decline" activates the hyperoxia risk chain; the same operational behavior may trigger multiple chains, and the system detects all relevant risk chains in parallel and prepares to input them into the time series risk scoring model;

[0088] The timing and intensity of each link are weighted and calculated using a time-series risk scoring model to form a risk linkage score for operational behavior.

[0089] The process for determining whether an AI intervention condition has been triggered in the risk linkage module is as follows:

[0090] The risk-linked score of the operational behavior was integrated with the patient's condition stability index for analysis;

[0091] The risk level is dynamically adjusted by combining the emergency level during the anesthesia stage with the current operational intensity of the equipment operators;

[0092] When the combined score of the risk index and the intensity of actual operations exceeds the dynamic intervention threshold, the AI ​​intervention process is triggered.

[0093] When the combined score of the risk index and the intensity of actual operations is less than or equal to the dynamic intervention threshold, the AI ​​intervention process will not be triggered.

[0094] Strategy comparison module: retrieves the parameter evolution path under similar surgical procedures in the past, combines the current patient's anesthesia depth, physiological data and intraoperative stage to perform analog modeling, and judges whether there is a strategy deviation in the current set behavior. If so, it enters the early warning trigger feedback module.

[0095] The strategy comparison module performs an analogy modeling by combining the current patient's anesthesia depth, physiological data, and intraoperative stage as follows:

[0096] Search the structured database for historical cases with similar anesthesia methods, surgical procedures, and underlying disease characteristics, and select a sample set with a similarity greater than a threshold to the current case;

[0097] Extracting oxygen concentration change trajectories, intervention response times, and final postoperative recovery data from these samples to form a strategy evolution template; the oxygen concentration in the matched samples ( The set values ​​are reconstructed over time to form a "set trajectory map"; the specific time, magnitude, and direction (increase or decrease) of each set adjustment are captured; and the depth of anesthesia (BIS value) is also monitored. , Alignment of synchronous physiological indicator curves; marking of each adjusted target indicator (e.g.) The time required for a significant response (such as an increase or stabilization) to occur; for example: After increasing from 50% to 70%, The time required to increase from 89% to 95% was 180 seconds; tag-adjustment-response pairing; extraction of postoperative PACU recovery score, whether transferred to ICU, and postoperative complications (such as...). Information such as pulmonary edema (e.g., atelectasis); the strategy trajectory is labeled according to the results as "low risk," "caution required," or "high risk"; the trajectories and results of multiple similar samples are classified and clustered to form a strategy evolution template.

[0098] Template A: Steady-state oxygenation strategy → Improved stability → Good postoperative recovery;

[0099] Template B: Sudden increase in oxygen concentration + high BIS fluctuation → Slow recovery → Moderately delayed recovery;

[0100] Template C: Frequent adjustment of oxygen concentration → Drastic fluctuations lead to an increased ICU transfer rate;

[0101] Combining the patient's current BIS anesthesia depth, hemodynamic status, and intraoperative operation tags, an analogical modeling algorithm is used to generate the predicted risk distribution of the currently set strategy in the historical context; the current intraoperative set behavior and its corresponding time point are captured; and the BIS values ​​before and after that time point (e.g., the trend of change from 68 to 58), HR / MAP curves, etc., are extracted simultaneously. Changes, etc.; synchronously acquire set behavior tags (such as " (Increase by 20%, operator permission level, occurrence stage, etc.); Use a combination of K-nearest neighbors + probability model or deep analogy network (such as Siamese Network); match the current policy features with the trajectory in the policy template; calculate the risk level distribution probability corresponding to the policy in historical samples; output the risk probability distribution; output the degree of matching between the current set behavior and the risk set path in the analogy template.

[0102] The process by which the strategy comparison module determines whether there is a strategy deviation in the currently set behavior is as follows:

[0103] By comparing the current oxygen concentration value with the optimal trajectory range in the analog model, the magnitude and time dimension of its falling into the offset interval are determined; the current... The set values ​​and their time points (e.g., 35 minutes into the procedure) (Adjust from 60% to 80%); Create a set behavior node with a timestamp and set amplitude label; Call the optimal trajectory template previously generated by aggregating similar historical cases, which contains the ideal range of variation for oxygen concentration settings; Compare whether the current setting value falls within the optimal trajectory tolerance range for that time point; Calculate the offset amplitude: Dm = Calculate the offset time window; if it is not set within the recommended time period, mark the time misalignment.

[0104] By combining the patient's real-time vital sign changes at the time the set behavior occurred, the subsequent physiological chain reactions were assessed; key parameters were extracted 3–5 minutes before and after the set behavior occurred. , MAP, HR, BIS; transforming data into trend lines (increase / decrease / volatility) and derived indicators (such as rate of change, coefficient of variation); identifying anomalous reaction chains after a set behavior, for example: Significant increase → Decline → Unstable HR; Adjustments were made to detect significant fluctuations in BIS; the results were compared with known physiological response pathways in the atlas to determine if any abnormal response labels were triggered; an impact factor and weight were assigned to each pathway to generate a cumulative score.

[0105] Based on the strategy deviation scoring function, the degree of deviation of the current set behavior from the historical analogy model is quantified, and corrections are made by combining the historical operating habits of the set personnel with the error model;

[0106] When the policy bias score exceeds the warning value predicted by the system's self-learning model, it indicates that there is a policy bias in the current behavior setting.

[0107] Early warning feedback module: When a policy-level anomaly is confirmed, the early warning level and display format are dynamically adjusted based on the type of operating equipment, the roles of the personnel, and the current load intensity, and graphical intervention suggestions are generated.

[0108] The process of generating graphical intervention suggestions in the early warning feedback module is as follows:

[0109] The system automatically selects an early warning template based on the current deviation type, surgical stage, and equipment type, and distinguishes between color levels, shape symbols, and information display levels.

[0110] By integrating specific indicators of strategy deviation with system judgment logic, a graphical suggestion box is generated on the interface, which includes parameter value comparison, risk prediction trend chart and recommended adjustment range, thus generating graphical intervention suggestions.

[0111] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0112] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An integrated platform system for hospitals based on artificial intelligence, characterized in that, include: The tag matching module extracts current patient identification information, identifies surgical patients under anesthesia in real time, automatically links preoperative electronic medical records and intraoperative anesthesia plans, and determines whether there is a need for oxygen concentration setting. If so, it proceeds to the parameter semantic analysis module. The semantic analysis module, based on an artificial intelligence semantic recognition model, extracts image text and performs semantic understanding on the parameters set in real time on the ventilator screen. Combining this with preoperative contraindications and recommended oxygen concentration ranges in the electronic medical record, it determines whether there is semantic deviation in the set values. If so, it proceeds to the instruction risk linkage module. The risk linkage module, based on the semantic differences between the set parameters and those identified by the system, constructs a cross-system risk factor map, performs real-time risk linkage scoring on intraoperative operations, and determines whether AI intervention conditions are triggered. If so, it proceeds to the intraoperative strategy comparison module. The real-time risk linkage scoring process for intraoperative operations involves extracting the system context state at the time the current set behavior occurs, including the operator's role, the set time point, physiological parameter status curves, and equipment load. Based on the correlation paths in the cross-system risk factor map, it is identified whether the operation behavior triggers a high-weight risk link; the triggering time and intensity of each link are weighted and calculated using a time-series risk scoring model to form a risk linkage score for the operation behavior; the process of determining whether the AI ​​intervention condition is triggered is as follows: the risk linkage score of the operation behavior is fused and analyzed with the patient's state stability index; the risk level is dynamically corrected by combining the emergency level of the anesthesia stage and the current operation intensity of the equipment operators. When the combined score of the risk index and the intensity of actual operations exceeds the dynamic intervention threshold, the AI ​​intervention process is triggered. When the combined score of the risk index and the intensity of actual operation is less than or equal to the dynamic intervention threshold, the AI ​​intervention process is not triggered; Strategy comparison module: retrieves the parameter evolution path under similar surgical procedures in the past, combines the current patient's anesthesia depth, physiological data and intraoperative stage to perform analog modeling, and judges whether there is a strategy deviation in the current set behavior. If so, it enters the early warning trigger feedback module. Early warning feedback module: When a policy-level anomaly is confirmed, the early warning level and display format are dynamically adjusted based on the type of operating equipment, the roles of the personnel, and the current load intensity, and graphical intervention suggestions are generated.

2. The hospital-wide integrated platform system based on artificial intelligence according to claim 1, characterized in that, The process for determining whether an oxygen concentration setting requirement exists is as follows: extract the binding information between the intraoperative equipment and the patient, and read the patient's current anesthesia status identifier from the anesthesia information system; determine whether the patient's current surgical stage involves spontaneous breathing depression or gas management requirements; combine the preoperative anesthesia plan and surgical procedure requirements, and compare whether the recommended oxygen concentration range and the current ventilator oxygen concentration setting value are synchronized; if the current setting value is missing or deviates from the recommended range by more than the set ratio threshold, it is determined that an oxygen concentration setting requirement exists.

3. The hospital-wide integrated platform system based on artificial intelligence according to claim 2, characterized in that, The process of extracting text and semantic understanding of the parameters set in real time on the ventilator device screen is as follows: acquire the image stream of the ventilator panel area, perform preprocessing operations on the image area, including edge sharpening, color enhancement and distortion correction; and use the trained parameter localization model to identify the oxygen concentration parameter label and the set value block in the display area. Extract the set values ​​and, in conjunction with the device manufacturer's parameter layout dictionary, convert the extracted values ​​into semantic tags.

4. The hospital-wide integrated platform system based on artificial intelligence according to claim 3, characterized in that, The process of determining whether there is semantic bias in the set value is as follows: the extracted oxygen concentration set value is matched with the recommended oxygen concentration range in the patient's preoperative electronic medical record, and it is compared whether the current set value falls within the recommended range; the patient's high-risk factor markers are extracted to construct a set of clinical context risk factors; using a knowledge graph-based semantic bias identification model, the set value and clinical background are fused and calculated to obtain a bias risk score. If the deviation risk score is greater than the policy tolerance, it is determined that there is semantic deviation; if the deviation risk score is less than or equal to the policy tolerance, it is determined that there is no semantic deviation.

5. The hospital-wide integrated platform system based on artificial intelligence according to claim 4, characterized in that, The process of constructing a cross-system risk factor map is as follows: Aggregate multi-source data related to patient status from ventilator, anesthesia information system, electronic medical record system, and monitoring system, including set parameters, historical oxygen consumption curves, intraoperative monitoring indicators, and shift handover operation records; standardize the data and map it to a risk factor map template; and construct cross-system node connections and dependency paths through graph embedding technology. Mark potential abnormal paths and output a cross-system risk factor map.

6. The hospital-wide integrated platform system based on artificial intelligence according to claim 1, characterized in that, The analogy modeling process, which combines the current patient's anesthesia depth, physiological data, and intraoperative stage, is as follows: Historical cases with similar anesthesia methods, surgical procedures, and underlying disease characteristics are retrieved from a structured database; a sample set with similarity greater than a threshold to the current case is selected; oxygen concentration setting change trajectory, intervention response time, and final postoperative recovery data are extracted from these samples to form a strategy evolution template; and the current patient's BIS anesthesia depth, hemodynamic status, and intraoperative operation labels are combined with an analogy modeling algorithm to generate the predicted risk distribution of the current strategy in the historical context. Output the degree of match between the current behavior setting and the risk setting path in the analog template.

7. The hospital-wide integrated platform system based on artificial intelligence according to claim 1, characterized in that, The process for determining whether the current set behavior has a strategy deviation is as follows: compare the current set oxygen concentration value with the optimal trajectory range in the analog model to determine the magnitude and time dimension of its falling into the deviation range; combine the real-time vital sign changes of the patient when the set behavior occurs to assess the subsequent physiological chain reactions; based on the strategy deviation scoring function, quantify the degree of deviation of the current set behavior in the analog historical model, and make corrections based on the historical operating habits and error model of the setter; when the strategy deviation score exceeds the warning value predicted by the system's self-learning model, it indicates that the current set behavior has a strategy deviation.

8. The hospital-wide integrated platform system based on artificial intelligence according to claim 1, characterized in that, The process of generating graphical intervention suggestions is as follows: based on the current deviation type, surgical stage, and equipment type, an early warning presentation template is automatically selected, and color levels, shape symbols, and information display levels are distinguished; the specific indicators of strategy deviation and system judgment logic are integrated to generate a graphical suggestion box on the interface that includes parameter value comparison, risk prediction trend chart, and recommended adjustment range, thus generating graphical intervention suggestions.

Images