Multi-modal sign anesthesia risk real-time evaluation and early warning method
By constructing a multimodal vital sign simulation prediction model, the problem of low accuracy in traditional anesthesia risk assessment has been solved, enabling timely early warning and accurate assessment of anesthesia risks.
Patent Information
- Application Number
- CN202511785518.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-01
- Publication Date
- 2026-03-06
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Traditional anesthesia risk assessment relies on intermittent manual examinations, which makes it difficult to capture rapidly changing physiological fluctuations, resulting in low accuracy of anesthesia risk assessment. Furthermore, abnormalities in a single indicator are easily misinterpreted as interference signals.
By constructing a multimodal vital sign simulation prediction model, and using historical anesthesia record data to classify and statistically analyze the multimodal vital sign data, two multimodal vital sign simulation prediction models, Model 1 and Model 2, are established to handle real-time and stable synchronous changes of data, respectively, thereby improving the accuracy and timeliness of the assessment.
It enhances the systematicness and accuracy of anesthesia risk assessment, improves the timeliness of anesthesia risk warning, and facilitates medical staff in developing timely response plans.
Smart Images
Figure CN121617622A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of anesthesia risk assessment technology, and in particular to a method for real-time assessment and early warning of anesthesia risk based on multimodal physical signs. Background Technology
[0002] Anesthesia risk assessment is an essential medical step for surgical patients before surgery. Real-time assessment and early warning of anesthesia risks before surgery can ensure the safety of anesthetized patients during the operation.
[0003] Traditional anesthesia management relies on intermittent manual checks (such as blood pressure and heart rate), which are insufficient to capture rapidly changing physiological fluctuations. This system integrates multi-dimensional data streams, including ECG, EEG, muscle relaxation, and end-tidal CO2, to construct a continuous vital sign map, enabling the early detection of hidden risks (such as the prodromal phase of malignant hyperthermia). Furthermore, abnormalities in a single indicator may be misinterpreted as interfering signals, easily reducing the accuracy of anesthesia risk index assessment. Summary of the Invention
[0004] To overcome the shortcomings of the prior art, this application provides a method for real-time assessment and early warning of anesthesia risk based on multimodal physical signs.
[0005] This application provides a method for real-time assessment and early warning of anesthesia risk based on multimodal physical signs, the method comprising:
[0006] Step S1: Extract historical real-time change data from historical anesthesia record data to identify changes in the timing of multimodal vital signs that exhibit real-time variations. Extract historical stable data from historical anesthesia record data to identify changes in the timing of multimodal vital signs that exhibit stable synchronous variations. Statistically analyze the time interval characteristics of each adjacent cycle of multimodal vital signs changes in the historical real-time change data, the difference correlation characteristics of multimodal related vital signs items in each adjacent cycle, and the data of each real-time change of multimodal vital signs to obtain preprocessed training set one. Based on preprocessed training set one, establish multimodal vital signs simulation prediction model one.
[0007] Step S2: Statistically analyze the interval characteristics of each change time of the multimodal vital signs data in the historical stable data and the data of each real-time change of the multimodal vital signs to obtain the second preprocessed training set. Based on the second preprocessed training set, establish the second multimodal vital signs simulation prediction model.
[0008] Step S3: Perform dose-based anesthesia on the current target anesthesia patient to obtain real-time anesthesia multimodal vital signs data of the current target anesthesia patient. Input the real-time anesthesia multimodal vital signs data into multimodal vital signs simulation prediction model one or multimodal vital signs simulation prediction model two for testing to obtain pre-prediction multimodal vital signs simulation prediction data. Based on the pre-prediction multimodal vital signs simulation prediction data, obtain the early warning anesthesia risk assessment result.
[0009] Preferably, historical anesthesia record data of historical anesthesia patients are obtained, and a similarity threshold is preset. The similarity between the pre-anesthesia individualized characteristic data of historical anesthesia patients and the current target anesthesia patient is greater than or equal to the similarity threshold.
[0010] Extract the records of changes in multimodal vital signs from the historical anesthesia records, and output the historical real-time change data.
[0011] Extract the data from historical anesthesia records that show stable and synchronous changes in the timing of changes in various multimodal vital signs, and output historical stable data.
[0012] Preferably, based on the historical real-time change data, if the data of each multimodal vital sign in the historical real-time change data are not synchronized in terms of change time sequence, then the change trend feature statistics are performed on the time interval between each two adjacent cycles of change of each multimodal vital sign data to obtain the time change interval trend feature one.
[0013] Based on the trend characteristic of the time variation period, extract the data of the real-time changes of multimodal vital signs from the historical real-time change data to obtain the data of the vital signs to be evaluated.
[0014] The time-varying interval trend features and the vital sign data to be evaluated are combined to form the first training feature set.
[0015] Preferably, if the number of multimodal related features changes in the historical real-time change data, then the difference feature correlation trend characteristics are statistically analyzed for the multimodal related features that change between each two adjacent cycles to obtain the feature change trend characteristics.
[0016] Based on the changing trend characteristics of the vital signs, the data of various real-time changes of multimodal vital signs are extracted from the historical real-time change data to obtain the second set of vital signs data to be evaluated.
[0017] The variation trend characteristics of the vital signs and the data of the vital signs to be evaluated are combined to form the second training feature set;
[0018] The first training feature set and the second training feature set are combined into a preprocessing training set 1. Based on the preprocessing training set 1, a multimodal physiological characteristic simulation prediction model 1 is established.
[0019] Preferably, based on the historical stable data, if the historical stable data contains data changes in which the priority order of the multimodal vital signs remains unchanged, and the data changes of the multimodal vital signs are time-synchronous and the time interval between each two adjacent periodic changes is the same, then the data of each multimodal vital sign that changes in real time are extracted from the historical stable data to obtain the third vital sign data to be evaluated.
[0020] If the historical stable data contains data changes where the priority of the multimodal vital signs remains unchanged, and the multimodal vital signs are time-synchronous and the time interval between each two adjacent cycles is different, then extract the data of each multimodal vital sign that changes in real time from the historical stable data to obtain the fourth vital sign data to be evaluated.
[0021] Furthermore, the variation trend characteristics of each data point in the historical stable data were statistically analyzed to obtain the second variation trend characteristic of the time interval.
[0022] The three sets of vital signs to be evaluated, the four sets of vital signs to be evaluated, and the trend characteristics of time variation intervals are collected into a preprocessing training set two. Based on the preprocessing training set two, a multimodal vital sign simulation prediction model two is established.
[0023] Preferably, an anesthesia risk assessment model is established based on historical multimodal vital signs data after anesthesia treatment in historical anesthesia records and the corresponding historical anesthesia risk assessment results.
[0024] The target anesthesia patient is given dose-based anesthesia treatment. Based on the dose of anesthesia treatment, an anesthesia assessment interval is preset, and a pre-treatment pre-assessment interval is extracted from the preset anesthesia assessment interval.
[0025] Based on the pre-treatment assessment interval, real-time anesthesia multimodal vital signs data of the current target anesthesia patient are obtained.
[0026] Preferably, the real-time anesthesia multimodal sign data is input into the anesthesia risk assessment model for testing to obtain an initial anesthesia risk assessment result. If the initial anesthesia risk assessment result is determined to be safe, then based on the remaining pre-pre-assessment intervals after removing the pre-treatment pre-assessment intervals in the anesthesia assessment interval, the real-time anesthesia multimodal sign data is input into either multimodal sign simulation prediction model one or multimodal sign simulation prediction model two for testing to obtain pre-pre-multimodal sign simulation prediction data.
[0027] The pre-anesthesia multimodal sign simulation prediction data is input into the anesthesia risk assessment model for testing to obtain the early warning anesthesia risk assessment results.
[0028] Compared with the prior art, the present invention has the following characteristics and beneficial effects:
[0029] By training a multimodal vital sign simulation model based on the characteristics of historical anesthesia records of patients under anesthesia, this approach avoids the pitfalls of traditional techniques that analyze multiple vital sign data using single, non-fusion data or simply analyze the visual changes in multiple vital signs to assess anesthesia risk. This approach fails to further analyze hidden data, such as the temporal characteristics of changes in multiple vital sign data and the correlation trends of changes in the number of vital sign items showing data changes. This can lead to insufficient anesthesia risk detection data and reduced accuracy of the assessment results. Therefore, before training the multimodal vital sign simulation model, historical anesthesia record data is processed... The data is divided into two categories: the first category is historical real-time change data where the changes in various multimodal vital signs occur in real time; the second category is historical stable data where the changes in various multimodal vital signs occur synchronously in real time. Because the hidden data change characteristics of these two categories of data are different, a differential data feature statistical processing method is subsequently used to reduce the interference between diverse data under different conditions, enhance the systematic nature of the entire anesthesia risk assessment work, and finally, by using the real-time anesthesia multimodal vital sign data of the current target anesthesia patient, a pre-anesthesia multimodal vital sign simulation prediction data is performed to improve the timeliness of anesthesia risk warning and facilitate medical staff to formulate timely response plans. Attached Figure Description
[0030] Figure 1 This embodiment is a flowchart illustrating the steps of a real-time assessment and early warning method for anesthesia risk based on multimodal physical signs. Detailed Implementation
[0031] The present invention will be further described in detail below with reference to the following embodiments.
[0032] Reference Figure 1 A real-time assessment and early warning method for anesthesia risk using multimodal physical signs, comprising the following steps:
[0033] Step S1: Extract historical real-time change data from historical anesthesia record data, where the occurrence time sequence of changes in multimodal vital signs varies in real time. Extract historical stable data from historical anesthesia record data, where the occurrence time sequence of changes in multimodal vital signs varies synchronously. Statistically analyze the time interval characteristics of each adjacent cycle change of multimodal vital signs data in the historical real-time change data, the difference correlation characteristics of multimodal related vital signs items in each adjacent cycle change, and the data of each real-time change of multimodal vital signs to obtain preprocessed training set one. Based on preprocessed training set one, establish multimodal vital sign simulation prediction model one.
[0034] Step S2: Statistically analyze the time interval characteristics of each change in the multimodal vital signs data in the historical stable data and the real-time changes of various data of the multimodal vital signs to obtain the second preprocessed training set. Based on the second preprocessed training set, establish the second multimodal vital signs simulation prediction model.
[0035] Step S3: Perform dose-based anesthesia on the current target anesthesia patient to obtain real-time anesthesia multimodal sign data of the current target anesthesia patient. Input the real-time anesthesia multimodal sign data into multimodal sign simulation prediction model one or multimodal sign simulation prediction model two for testing to obtain pre-pregnancy multimodal sign simulation prediction data. Based on the pre-pregnancy multimodal sign simulation prediction data, obtain the early warning anesthesia risk assessment result.
[0036] Specifically, a multimodal vital sign simulation model is trained based on the characteristics of historical anesthesia records of patients undergoing anesthesia. However, the variations in multimodal vital sign data are diverse. To avoid the pitfalls of traditional techniques that analyze multiple vital sign data using single, non-fusion data or only analyze the visual data of changes in multiple vital signs to assess anesthesia risk, hidden data such as the temporal characteristics of changes in multiple vital sign data and the correlation trends of changes in the number of vital sign items showing data changes are not further analyzed. This can lead to insufficient anesthesia risk detection data and reduced accuracy of the assessment results. Therefore, before training the multimodal vital sign simulation model, historical anesthesia record data is used to... The data is divided into two categories: the first category is historical real-time change data where the changes in the multimodal vital signs occur in real time; the second category is historical stable data where the changes in the multimodal vital signs occur synchronously in real time. Because the hidden data change characteristics of these two categories of data are different, a differential data feature statistical processing method is subsequently used to reduce the interference between diverse data under different conditions and enhance the systematic nature of the entire anesthesia risk assessment work. Finally, by using the real-time anesthesia multimodal vital sign data of the current target anesthesia patient, a pre-anesthesia multimodal vital sign simulation prediction data is performed to improve the timeliness of anesthesia risk warning and facilitate medical staff to formulate timely response plans.
[0037] The specific step S1 includes the following sub-steps:
[0038] Obtain historical anesthesia records of patients who have undergone anesthesia, and set a similarity threshold. The similarity between the pre-anesthesia individualized characteristic data of the historical anesthesia patient and the current target anesthesia patient is greater than or equal to the similarity threshold.
[0039] Extract the real-time data of changes in various multimodal vital signs from historical anesthesia records, and output the historical real-time change data.
[0040] Extract the data from historical anesthesia records that show stable and synchronous changes in the timing of changes in various multimodal vital signs, and output historical stable data.
[0041] Based on historical real-time change data, if the data of each multimodal vital sign in the historical real-time change data are not synchronized in terms of change time sequence, then the change trend characteristics of each data of each multimodal vital sign are statistically analyzed for the time interval between two adjacent cycles of change, and the first trend characteristic of time change interval is obtained.
[0042] Based on the trend characteristics of time variation, various data of real-time changes in multimodal vital signs are extracted from historical real-time change data to obtain the first set of vital sign data to be evaluated.
[0043] The time-varying trend characteristics and the data of the vital signs to be evaluated are combined to form the first training feature set.
[0044] If the number of multimodal related features changes in the historical real-time change data, then the trend characteristics of the difference in the multimodal related features changes between each two adjacent cycles are statistically analyzed to obtain the trend characteristics of the feature changes.
[0045] Based on the trend characteristics of the changes in vital signs, the data of various changes in multimodal vital signs in real time are extracted from the historical real-time change data to obtain the second set of vital signs data to be evaluated.
[0046] The trend characteristics of changes in vital signs and the data of vital signs to be evaluated are combined to form the second training feature set.
[0047] The first training feature set and the second training feature set are combined into a preprocessing training set 1. Based on the preprocessing training set 1, a multimodal physiological characteristic simulation prediction model 1 is established.
[0048] Specifically, this includes pre-anesthesia individualized characteristic data (including age, gender, health status, body type, and pre-anesthesia vital signs), historical anesthesia record data (including anesthetic dosage, vital signs with varying data, amplitude of data changes for each vital sign, time interval between two changes for the same vital sign, temporal sequence of data changes for each vital sign, and anesthesia risk assessment results), and a preset similarity threshold (e.g., preset at 90%, obtained by statistically analyzing the similarity of individualized characteristics of patients with the same anesthesia risk assessment data changes corresponding to multiple modal vital signs in historical anesthesia record data; this threshold can also be determined independently, but it can only be higher, not lower. Multimodal vital signs include vital signs: electrocardiogram, blood pressure, and blood oxygen saturation). Blood pressure, body temperature, and respiratory mechanics parameters (tidal volume or minute ventilation, peak airway pressure, plateau pressure, etc.) are monitored. Historical real-time variation data (e.g., times t1, t2, t3, t4, t5, in seconds) are analyzed. Within time t1, blood pressure, oxygen saturation, and body temperature fluctuate. If the fluctuation sequence is blood pressure first, then oxygen saturation, then body temperature (if all fluctuations are within the normal range, the difference lies in the fluctuations occurring based on the pre-anesthesia data of the anesthetized patient; therefore, the changes in each multimodal vital sign are the amplitudes of these fluctuations based on the pre-anesthesia data of the anesthetized patient), and within time t2, the fluctuating vital signs are oxygen saturation, body temperature, and blood pressure, and so on. Comparing these data with time t... 1. If a time-series change occurs, then the recorded data where the changes in the multimodal vital signs data occur in real time is called historical real-time change data. Historically stable data (e.g., if the vital signs with data fluctuations at time t1 are blood pressure, blood oxygen saturation, and body temperature, and the same vital signs with data fluctuations at times t2, t3, t4, and t5 occur in the same time series, then the recorded data where the changes in the multimodal vital signs data occur in a stable and synchronous time series is called historically stable data). The time-series trend characteristic is also important (e.g., if the time point where blood pressure data fluctuates within time t1 is j1, then at time t2...). If the time point of data fluctuation within the interval is j2, then the time interval between the two adjacent periodic changes t1 and t2 is j2-j1. If it is I1, then I2, I3, and In are similarly calculated. A trend graph is constructed for I1, I2, I3, and In. The same processing is performed on the other changing vital signs, thus obtaining the first trend feature of the time interval. The first set of vital signs to be evaluated (i.e., the changing data of each vital sign in the historical real-time change data located at t1, t2, t3, t4, and t5) is statistically analyzed. The trend feature of the vital sign changes (e.g., the vital signs are interrelated; if the data changes in time t1 are blood pressure, blood oxygen saturation, and body temperature, and the data changes in time t2 are electrocardiogram, blood pressure, blood oxygen saturation, and body temperature).Within time period t3, the data that changed were electrocardiogram (ECG), blood pressure, oxygen saturation, body temperature, minute ventilation, and central venous oxygen saturation. Therefore, the associated vital signs appearing in the two periods t1 and t2 changed to newly added vital signs: ECG. Similarly, the associated vital signs appearing in the two periods t3 and t2 changed to newly added vital signs: minute ventilation and central venous oxygen saturation, and so on. The collected newly added vital signs were then statistically analyzed for the changing paths of associated vital signs (e.g., constructing a knowledge graph of all multimodal vital signs appearing in historical anesthesia records to obtain a knowledge graph library of vital sign associations). The spectral library contains the trend of changes in related vital signs (i.e., the trend features of vital sign changes), the second set of vital sign data to be evaluated (based on the first set of vital sign data to be evaluated, and so on), and the first multimodal vital sign simulation prediction model (first training feature set: such as constructing an information mapping pairing table between the first time-varying segment trend feature and the first set of vital sign data to be evaluated, to form a matching information library corresponding to the first time-varying segment trend feature and the first set of vital sign data to be evaluated, if it is Ku1. Second training feature set: vital sign change trend features and the second set of vital sign data to be evaluated, and so on, to obtain Ku2, Ku1 and Ku2 are the first multimodal vital sign simulation prediction model).
[0049] The specific step S2 includes the following sub-steps:
[0050] Based on historical stable data, if the priority order of the multimodal vital signs to which the data changes remains unchanged, and the data changes of the multimodal vital signs are time-synchronous and the time interval between each two adjacent periodic changes is the same, then the data of each multimodal vital sign that changes in real time are extracted from the historical stable data to obtain the third vital sign data to be evaluated.
[0051] If the historical stable data contains data changes in which the priority of the multimodal vital signs remains unchanged, and the multimodal vital signs are time-synchronized and the time interval between each two adjacent cycles is different, then the data of each multimodal vital sign that changes in real time is extracted from the historical stable data to obtain the fourth set of vital signs to be evaluated.
[0052] Furthermore, the variation trend characteristics of each data point in the historical stable data were statistically analyzed to obtain the second variation trend characteristic of the time interval.
[0053] The three sets of vital signs to be evaluated, the four sets of vital signs to be evaluated, and the trend characteristics of time variation intervals are compiled into preprocessing training set two. Based on preprocessing training set two, a multimodal vital sign simulation prediction model two is established.
[0054] Specifically, for the third vital sign data to be evaluated (if the historical stable data contains data changes in the multimodal vital sign items with unchanged priority order: the priority order of vital sign items that change in time periods t1, t2, t3, t4, and t5 is blood pressure, blood oxygen saturation, and body temperature; the multimodal vital sign data changes are time-synchronous: the time points at which each vital sign item changes in time periods t1, t2, t3, t4, and t5 are the same, and the time interval between each two adjacent cycles is the same: if the time point at which blood pressure data fluctuates in time period t1 is j1, and the time point at which blood pressure data fluctuates in time period t2 is j...), then... 2. The time interval between two adjacent cycles, t1 and t2, is j2 - j1 = I1. If the time point when blood pressure fluctuates within time t3 is j3, and the time point when blood pressure fluctuates within time t2 is j2, then the time interval between two adjacent cycles, t3 and t2, is j3 - j2 = I1. And so on. Therefore, statistical analysis is performed on the fluctuation data at t1, t2, t3, t4, and t5 to obtain the third set of data for the vital signs to be evaluated. The fourth set of data for the vital signs to be evaluated (data fluctuations where the time interval between two adjacent cycles is not the same: if the time point when blood pressure fluctuates within time t1 is j1, and the time point when blood pressure fluctuates within time t2 is j2, then the time interval between two adjacent cycles, t3 and t2, is j3 - j2 = I1.) If the time point where the data fluctuation occurs within the interval is j2, then the time interval between the two adjacent cycles of t1 and t2 is j2-j1=I1. If the time point where the blood pressure data fluctuation occurs within time t3 is j3, and the time point where the data fluctuation occurs within time t2 is j2, then the time interval between the two adjacent cycles of t3 and t2 is j3-j2=I2, I1≠I2, and so on. Therefore, at this time, statistical analysis of the fluctuation data at t1, t2, t3, t4, and t5 is performed to obtain the fourth characteristic data to be evaluated. The second characteristic of the time variation interval (based on the first characteristic of the time variation interval, and so on, the difference being that the...) belongs to... The statistical conditions are based on historical stable data for feature statistics. Multimodal vital sign simulation prediction model two (vital sign data three to be evaluated: constructing an information mapping pairing table between t1, t2, t3, t4, t5 and the vital sign data three to be evaluated, forming a one-to-one matching information database between t1, t2, t3, t4, t5 and the vital sign data three to be evaluated, if it is Ku3. Vital sign data four to be evaluated: constructing an information mapping pairing table between the time-varying interval trend feature two and the vital sign data four to be evaluated, forming a one-to-one matching information database between the time-varying interval trend feature two and the vital sign data four to be evaluated, if it is Ku4).
[0055] The specific step S3 includes the following sub-steps:
[0056] An anesthesia risk assessment model was established based on historical multimodal vital signs data after anesthesia treatment in historical anesthesia records and the corresponding historical anesthesia risk assessment results.
[0057] The target anesthesia patient is given dose-based anesthesia treatment. Based on the dose of anesthesia treatment, anesthesia assessment interval is preset, and pre-treatment pre-assessment intervals are extracted from the preset anesthesia assessment intervals.
[0058] Based on the pre-treatment assessment interval, real-time multimodal anesthesia signs data of the current target anesthesia patient are obtained.
[0059] Real-time anesthesia multimodal vital signs data are input into the anesthesia risk assessment model for testing to obtain the initial anesthesia risk assessment result. If the initial anesthesia risk assessment result is determined to be safe, then based on the remaining pre-pre-assessment intervals after removing the pre-treatment pre-assessment intervals in the anesthesia assessment interval, the real-time anesthesia multimodal vital signs data are input into either Multimodal Vital Sign Simulation Prediction Model 1 or Multimodal Vital Sign Simulation Prediction Model 2 for testing to obtain pre-pregnancy multimodal vital signs simulation prediction data.
[0060] The pre-anesthesia risk assessment model is tested by inputting the pre-anesthesia multimodal signs simulation prediction data into the model to obtain the early warning anesthesia risk assessment results.
[0061] Specifically, an anesthesia risk assessment model (such as historical multimodal vital signs data (e.g., heart rate, blood pressure, blood oxygen saturation, etc.) and historical anesthesia risk assessment results (low / medium / high anesthesia risk level or continuous numerical score): Y=ax+bz+cr+W, where Y refers to the historical anesthesia risk assessment result (score value), a is the heart rate variation data, b is the blood pressure variation data, c is the blood oxygen saturation variation data, x is the unknown weighting factor between heart rate and risk, z is the unknown weighting factor between blood pressure and risk, r is the unknown weighting factor between blood oxygen saturation and risk, and W is the error compensation value. This model combines known historical multimodal vital signs data and historical anesthesia risk assessment results. Substitute the values to obtain x, z, and r), preset the anesthesia assessment interval (based on the statistical analysis of the effective anesthesia assessment intervals from historical anesthesia records of patients, if it is T), preprocess the pre-assessment interval (if it is Ti, Ti is less than T, and Ti is greater than or equal to T / 3, because Ti cannot be too short: if it is too short, the statistical data characteristics will not be significant enough, and misjudgment is likely to occur), and the initial anesthesia risk assessment result (i.e., substitute the real-time anesthesia multimodal sign data into the trained Y=ax+bz+cr+W for calculation to assess the early warning anesthesia risk assessment result, based on historical anesthesia risk assessment results (anesthesia risk level low / medium / high or continuous numerical score): If the score is Z1-Z2, a score less than or equal to Z1 indicates a low anesthesia risk level; a score greater than Z1 and less than Z2 indicates a moderate anesthesia risk level; and a score greater than or equal to Z2 indicates a high anesthesia risk level. If the score of the anesthesia risk assessment result is less than Z1, it indicates a low anesthesia risk level within the Ti time period. To ensure the timeliness of anesthesia risk warnings for current anesthetized patients, it is also necessary to predict the changes in anesthesia multimodal signs during the remaining pre-assessment interval of T-Ti to assess whether the anesthesia risk is continuously increasing. The anesthesia risk assessment result (firstly, the characteristics of the real-time anesthesia multimodal sign data, whether it is real-time changing data or stable data, are determined; if it is real-time changing data or stable data, it is determined that the real-time changing data is the real-time changing data or stable data). If the data changes over time, the system will then determine the characteristics of any asynchronous changes in the timing of the changes in the multimodal vital signs or changes in the number of related multimodal vital signs. If the former is the case, the real-time anesthesia multimodal vital sign data will be matched with the information in Ku1 to obtain pre-anesthesia multimodal vital sign simulation prediction data (i.e., the changes in multimodal vital signs after T-Ti in the real-time anesthesia multimodal vital sign data). Finally, the pre-anesthesia multimodal vital sign simulation prediction data will be substituted into the trained Y=ax+bz+cr+W for calculation to evaluate the anesthesia risk assessment results. Based on the initial anesthesia risk assessment results, this process will be repeated to promptly notify medical staff to adopt pre-prepared plans.
[0062] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.
Claims
1. A multi-modal body sign anesthetic risk real-time assessment pre-warning method, characterized in that, The method comprises the following steps: Step S1, extracting historical real-time change type data of each data change of the multi-modal signs from the historical anesthesia record data, extracting historical stable type data of each data change of the multi-modal signs from the historical anesthesia record data, and obtaining a pre-processing training set one by statistically analyzing the time interval characteristics of each adjacent two cycle changes of the multi-modal signs, the difference correlation characteristics of each adjacent two cycle changes of the multi-modal signs, and the data of the real-time changes of the multi-modal signs in the historical real-time change type data, and establishing a multi-modal sign simulation prediction model one according to the pre-processing training set one; Step S2, obtaining a pre-processing training set two by statistically analyzing the time interval characteristics of each change of the multi-modal signs and the data of the real-time changes of the multi-modal signs in the historical stable type data, and establishing a multi-modal sign simulation prediction model two according to the pre-processing training set two; Step S3, obtaining real-time anesthesia multi-modal sign data of a current target anesthesia patient by performing dose anesthesia treatment on the current target anesthesia patient, inputting the real-time anesthesia multi-modal sign data into the multi-modal sign simulation prediction model one or the multi-modal sign simulation prediction model two for testing, obtaining pre-anesthesia multi-modal sign simulation prediction data, and obtaining an early warning anesthesia risk assessment result according to the pre-anesthesia multi-modal sign simulation prediction data.
2. The method of claim 1, wherein the method further comprises: Step S1 comprises: obtaining historical anesthesia record data of a historical anesthesia patient, presetting a similarity threshold, and determining that the similarity between the historical anesthesia patient and the individualized characteristic data of the current target anesthesia patient before anesthesia is greater than or equal to the similarity threshold; extracting record data of each data change of the multi-modal signs from the historical anesthesia record data, and outputting historical real-time change type data; extracting record data of each data change of the multi-modal signs from the historical anesthesia record data, and outputting historical stable type data.
3. The method of claim 2, wherein the method further comprises: Step S1 further comprises: if the time sequences of the changes of the multi-modal signs in the historical real-time change type data are not synchronized, performing change trend characteristic statistical analysis on the time intervals of each adjacent two cycle changes of the multi-modal signs, and obtaining time change interval trend characteristics one according to the historical real-time change type data; extracting the data of the real-time changes of the multi-modal signs from the historical real-time change type data according to the time change interval trend characteristics one, and obtaining evaluation sign data one; the time change interval trend characteristics one and the evaluation sign data one are combined to form a first training feature set.
4. The method of claim 3, wherein the method further comprises: Step S1 further comprises: if the number of the multi-modal associated sign items of the data changes in the historical real-time change type data changes, performing difference sign item correlation change trend characteristic statistical analysis on each adjacent two cycle changes of the multi-modal associated sign items, and obtaining sign item change trend characteristics; extracting the data of the real-time changes of the multi-modal signs from the historical real-time change type data according to the sign item change trend characteristics, and obtaining evaluation sign data two; the sign item change trend characteristics and the evaluation sign data two are combined to form a second training feature set; The first training feature set and the second training feature set are integrated into a pre-processing feature set one, and a multi-modal sign simulation prediction model one is established according to the pre-processing feature set one.
5. The method of claim 4, wherein the method further comprises: Step S2 includes: According to the historical stable data, if the multi-modal sign item priority order to which the data variation belongs in the historical stable data is unchanged, and the multi-modal sign item data variation is time sequence synchronization and each adjacent two time interval of cycle variation is different, each item of data of the real-time change of the multi-modal sign is extracted from the historical stable data, and the to-be-evaluated sign data three is obtained; If the multi-modal sign item priority order to which the data variation belongs in the historical stable data is unchanged, and the multi-modal sign item is time sequence synchronization and each adjacent two time interval of cycle variation is different, each item of data of the real-time change of the multi-modal sign is extracted from the historical stable data, and the to-be-evaluated sign data four is obtained; And the variation trend characteristics of the interval of each variation of the multi-modal sign data in the historical stable data are counted, and the time variation interval trend characteristics two are obtained. The to-be-evaluated sign data three, the to-be-evaluated sign data four and the time variation interval trend characteristics two are integrated into a pre-processing feature set two, and a multi-modal sign simulation prediction model two is established according to the pre-processing feature set two.
6. The method of claim 5, wherein the method further comprises: Step S3 includes: According to the historical multi-modal sign data after anesthesia treatment in the historical anesthesia record data and the corresponding evaluated historical anesthesia risk evaluation result, an anesthesia risk evaluation model is established; A target anesthesia patient is subjected to a dose of anesthesia treatment, and according to the dose of the anesthesia treatment, a pre-processing pre-estimation interval is extracted from a pre-set anesthesia evaluation interval. According to the pre-processing pre-estimation interval, real-time anesthesia multi-modal sign data of the target anesthesia patient is obtained.
7. The method of claim 6, wherein the method further comprises: Step S3 also includes: The real-time anesthesia multi-modal sign data is input into the anesthesia risk evaluation model for testing to obtain an initial anesthesia risk evaluation result, if the initial anesthesia risk evaluation result is determined to be safe, the real-time anesthesia multi-modal sign data is input into the multi-modal sign simulation prediction model one or the multi-modal sign simulation prediction model two for testing to obtain pre-estimation multi-modal sign simulation prediction data according to the remaining pre-estimation interval of the anesthesia evaluation interval except the pre-processing pre-estimation interval; The pre-estimation multi-modal sign simulation prediction data is input into the anesthesia risk evaluation model for testing to obtain a pre-warning anesthesia risk evaluation result.