Thoracic surgery postoperative complication early warning system

Abstract

The invention discloses a chest surgery postoperative complication early warning system. The chest surgery postoperative complication early warning system comprises a data acquisition layer, a data transmission layer, a data processing and storage layer, an analysis and early warning layer and a user interface and feedback layer. The data acquisition layer is used for acquiring vital signs of a patient, laboratory inspection, image data and information manually input by medical staff; the data transmission layer transmits the collected data to the data processing and storage layer in real time; the data processing and storage layer preprocesses, stores and manages the collected data; the analysis and early warning layer analyzes the data through a predefined rule and a machine learning model, calculates a risk score in real time, and triggers early warning; and the user interface and feedback layer provides a graphical operation interface, mobile terminal pushing and alarm feedback. According to the invention, the change of the condition of the patient after the thoracic surgery can be accurately monitored in real time, the risk of complication can be early warned in time, decision support is provided for medical personnel, and the postoperative recovery effect of the patient is improved.

Description

Technical Field

[0001] The present invention relates to the field of medical assistance technology, and particularly to a warning system for postoperative complications in thoracic surgery. Background Art

[0002] After thoracic surgery, patients may develop various complications, such as infection, bleeding, respiratory failure, etc. The occurrence of these complications often has a certain degree of concealment and suddenness, making it difficult to detect in the early stage. Once detected, it may have caused serious harm to the patient. Currently, medical staff mainly rely on regular examinations and experience to monitor the postoperative recovery of patients. This method has the following deficiencies: firstly, the monitoring is not real-time enough to capture the subtle changes in the patient's condition in a timely manner; secondly, there is a lack of systematic analysis and warning mechanisms, making it difficult to comprehensively consider the impact of various factors on complications and prone to misjudgment or missed judgment. Therefore, there is an urgent need for a system that can monitor the condition of patients after thoracic surgery in real time and accurately, and timely warn of the risk of complications. Summary of the Invention

[0003] The object of the present invention is to provide a warning system for postoperative complications in thoracic surgery.

[0004] To achieve the above object, the present invention is implemented according to the following technical solution:

[0005] The present invention includes a data acquisition layer, a data transmission layer, a data processing and storage layer, an analysis and warning layer, and a user interface and feedback layer. The data acquisition layer collects the patient's vital signs, laboratory tests, imaging data, and information manually entered by medical staff. The data transmission layer transmits the collected data to the data processing and storage layer in real time. The data processing and storage layer preprocesses, stores, and manages the collected data. The analysis and warning layer analyzes the data through predefined rules and machine learning models, calculates the risk score in real time, and triggers a warning. The user interface and feedback layer provides a graphical operation interface, mobile terminal push, and alarm feedback.

[0006] The vital signs collected by the data acquisition layer include heart rate, blood pressure, respiratory rate, and blood oxygen saturation data collected through the interfaces of the monitor and ventilator devices. The laboratory tests collected include blood routine, inflammatory indicators, and coagulation indicators. The imaging data collected includes X-rays and CTs, and the image recognition system is used to assist in judging the inflammation or bleeding situation. The information manually entered by medical staff includes the patient's main complaints, physical sign changes, and nursing records entered manually.

[0007] The preprocessing of the collected data includes data cleaning, data fusion, and data time series management. The data cleaning processes missing data, outliers, and redundant data, and standardizes the data. The data fusion uses data middleware to integrate data from different devices and systems in a unified format to ensure data consistency. The data time series management assigns timestamps to all data to ensure the accuracy of subsequent real-time monitoring and historical data comparison and analysis. The data processing and storage layer stores and manages the collected data using a database, encrypts data transmission using the SSL / TLS protocol, and implements role-based access control.

[0008] The predefined rules in the analysis and warning layer are to set key indicator thresholds, including abnormal body temperature, blood pressure fluctuations, and elevated white blood cells. An alarm is immediately triggered when the safe range is exceeded. The machine learning model uses historical clinical data to train the model to calculate risk scores. Feature selection includes dynamic changes in vital signs, trends in laboratory indicators, patient age, past medical history, and surgical type. Model fusion combines a rule engine and a statistical model through voting or weighted fusion algorithms to improve the accuracy and robustness of warnings. A data stream engine is used to process high-concurrency real-time data streams to ensure that warning information can be generated and transmitted quickly.

[0009] The user interface and feedback layer provides a graphical operation interface with a real-time monitoring dashboard, trend charts, and risk curves that intuitively display the patient's status and risk changes, shows detailed data records, alarm reason analysis, and recommended treatment measures for each warning. Medical staff can query historical data, model prediction results, and feedback records. The mobile terminal push and alarm feedback receive alarm information and patient status updates in real time through a mobile terminal application on mobile devices.

[0010] The beneficial effects of the present invention are:

[0011] The present invention is a warning system for postoperative complications in thoracic surgery. Compared with the prior art, the present invention realizes comprehensive collection, efficient processing, accurate analysis, and timely warning of postoperative data of patients through a multi-layer architecture design. Its technical effects are mainly reflected in the following aspects:

[0012] Real-time performance: The data collection layer can obtain information such as patients' vital signs, laboratory test results, and imaging data in real time, and quickly transmit it to the processing and storage layer through the data transmission layer, ensuring the timeliness of the data.

[0013] Accuracy: The analysis and warning layer combines predefined rules and machine learning models to comprehensively analyze the data, can accurately calculate risk scores and trigger warnings, improving the warning accuracy for complications.

[0014] Comprehensiveness: The data collected by the system covers various factors that may affect the occurrence of complications after surgery for patients, including dynamic changes in vital signs, trends in laboratory indicators, imaging findings, etc., enabling a comprehensive assessment of the patient's condition.

[0015] Convenience: The user interface and feedback layer provide a graphical operation interface and mobile terminal push function. Medical staff can view the real-time status and warning information of patients anytime and anywhere, improving work efficiency.

[0016] Security: The data processing and storage layer uses the SSL / TLS protocol to encrypt data transmission and implements role-based access control to ensure the security and privacy of patient data. Description of the Drawings

[0017] Figure 1 It is the system structure schematic diagram of the present invention. Detailed Embodiments

[0018] The present invention will be further described below in conjunction with the drawings and specific embodiments. The schematic embodiments and descriptions of this invention are used to explain the present invention, but not to limit the present invention.

[0019] As Figure 1 shown: The present invention includes a data acquisition layer, a data transmission layer, a data processing and storage layer, an analysis and warning layer, and a user interface and feedback layer. The data acquisition layer acquires the vital signs, laboratory tests, imaging data of patients, and information manually input by medical staff. The data transmission layer transmits the acquired data to the data processing and storage layer in real time. The data processing and storage layer preprocesses, stores, and manages the acquired data. The analysis and warning layer analyzes the data through predefined rules and machine learning models, calculates the risk score in real time, and triggers warnings. The user interface and feedback layer provides a graphical operation interface, mobile terminal push, and alarm feedback.

[0020] The vital signs acquired by the data acquisition layer include heart rate, blood pressure, respiratory rate, and blood oxygen saturation data collected through the interfaces of monitors and ventilator devices. The laboratory tests acquired include blood routine, inflammatory indicators, and coagulation indicators. The imaging data acquired include X-rays and CTs to assist in judging inflammation or bleeding conditions through an image recognition system. The information manually input by medical staff includes the patient's chief complaints, physical sign changes, and nursing records manually entered.

[0021] The preprocessing of the acquired data includes data cleaning, data fusion, and data time series management. The data cleaning processes missing data, outliers, and redundant data, and standardizes the data;

[0022] The data cleaning includes missing value filling, outlier detection, and standardization processing:

[0023] Missing value filling is processed using mean filling or time series interpolation:

[0024]

[0025] Where: At time t, the filled value of the i-th eigenvalue, N(t): the set of non-missing values near time t; |N(t)|: the number of non-missing values near time t;

[0026] Outlier detection is performed using the three-standard-deviation method:

[0027]

[0028] Where: μ: the mean of the feature in the training samples; σ: the standard deviation of the feature in the training samples; x i : the feature value of the i-th sample;

[0029] Standardization is performed using z-score standardization for data normalization:

[0030]

[0031] Where: x' i : the standardized feature value;

[0032] In the data fusion, assume devices D1, D2,..., D m The data streams X1, X2,..., X m , and linear interpolation is used for data alignment:

[0033]

[0034] Where: X' i (t): the interpolated data at time t; X i (t1), X i (t2): the original data values at adjacent time points; t1, t2: adjacent time points;

[0035] The data time series management assigns a timestamp T to each piece of data and stores it in the database:

[0036] D = {(T i , X i )}, i = 1, 2,..., n

[0037] Where: D: the dataset stored in the database; T i : the timestamp of the i-th piece of data, X i : the feature vector of the i-th piece of data.

[0038] The data fusion uses a data middleware to integrate data from different devices and systems in a unified format to ensure data consistency. The data time series management stamps timestamps on all data to ensure the accuracy of subsequent real-time monitoring and historical data comparison and analysis. The data processing and storage layer stores and manages the collected data using a database, encrypts data transmission using the SSL / TLS protocol, and implements role-based access control.

[0039] The predefined rules in the analysis and early warning layer are to set key indicator thresholds, including abnormal body temperature, blood pressure fluctuations, and elevated white blood cells. An alarm is immediately triggered when the safe range is exceeded. The machine learning model uses historical clinical data to train the model to calculate the risk score. Feature selection includes dynamic changes in vital signs, trends in laboratory indicators, patient age, past medical history, and surgical type. Model fusion combines a rule engine and a statistical model, and through voting or weighted fusion algorithms, improves the accuracy and robustness of early warning. A data flow engine is used to process high-concurrency real-time data streams to ensure that early warning information can be generated and transmitted quickly.

[0040] The machine learning model includes a logistic regression model, a rule engine, and multi-model fusion;

[0041] The logistic regression model uses logistic regression to calculate the risk score of postoperative complications:

[0042]

[0043] Where:

[0044]

[0045] In the formula: P(y = 1|X): The probability of predicting the occurrence of complications; X = [x1, x2,..., x n : The patient feature vector; x i : The i-th eigenvalue; β i : The weight of the i-th feature; β0: Bias term;

[0046] The loss function is optimized using cross-entropy loss:

[0047]

[0048] The rule engine triggers an early warning based on the set threshold rules:

[0049]

[0050] Including: Body temperature > 38.5°C triggers a high fever early warning; Systolic blood pressure < 90 mmHg triggers a low blood pressure early warning; White blood cell count > 12×10 9 / L triggers an infection early warning;

[0051] Multi - model fusion combines a logistic regression model and a decision tree model, and uses weighted fusion to calculate the final risk score:

[0052] R = γ1R LR + γ2R DT + γ3R NN

[0053] Where: R: The final comprehensive risk score; R LR , R DT , R NN : The predicted risk scores of the logistic regression, decision tree, and neural network; γ i : The model weighting coefficient.

[0054] The user interface and the feedback layer provide a graphical operation interface, offering a real - time monitoring dashboard, trend charts, and risk curves that intuitively display the patient's status and risk changes, showing detailed data records of each warning, analysis of the alarm reasons, and recommended handling measures. Medical staff can query historical data, model prediction results, and feedback records. The mobile terminal push and alarm feedback receive alarm information and patient status updates in real - time through the mobile terminal application on mobile devices.

[0055] In the analysis and early warning layer, the key index threshold uses the exponential smoothing method to calculate the dynamic threshold:

[0056] S t = αx t +(1 - α)S t-1

[0057] Where: S t : The smoothed index value; x t : The original value at the current moment; α: The smoothing coefficient, controlling the sensitivity (0 < α < 1); S t-1 : The smoothed value at the previous moment; When S t > θ, an alarm is triggered.

[0058] The analysis and early warning layer uses a WebSocket - based push mechanism. When an early warning is triggered, it sends a notification to the medical staff's mobile devices:

[0059] M = {(T, ID, R, L)}

[0060] Where: M: The alarm information in the message queue; T: The alarm time; ID: The patient ID; R: The risk score; L: The alarm level.

[0061] Example 1: Early warning of postoperative infection complications

[0062] Patient situation: Patient A, male, 55 years old, underwent thoracic surgery due to lung cancer. After the operation, the patient was admitted to the intensive care unit, and close monitoring of vital signs and complication risks was required.

[0063] Data collection:

[0064] Vital sign data: Heart rate, blood pressure, respiratory rate, and blood oxygen saturation data were collected in real-time through the interfaces of the monitor and ventilator equipment: Heart rate: 70 bpm; Blood pressure: 120 / 80 mmHg; Respiratory rate: 18 breaths per minute; Blood oxygen saturation: 98%.

[0065] Laboratory test data: The blood routine examination showed that the white blood cell count was 10×109 / L, and the inflammatory index C-reactive protein (CRP) was 10 mg / L.

[0066] Imaging data: Chest X-ray and CT images on the 3rd day after the operation showed no obvious abnormalities in the lungs.

[0067] Information manually entered by medical staff: The patient complained of pain at the surgical wound after the operation, without symptoms such as fever and cough. The nursing record showed that the patient had a good postoperative recovery.

[0068] Data processing and storage:

[0069] Data cleaning: An outlier (18×109 / L) was detected in the white blood cell count data, and it was judged as an outlier and removed by the three-standard-deviation method. The missing blood oxygen saturation data was filled using the time series interpolation method.

[0070] Data fusion: The data collected by the monitor, ventilator, laboratory test system, and imaging system were integrated through a data middleware, and after unifying the format, they were stored in the database.

[0071] Data time series management: Timestamps were added to all data to ensure the time series of the data, which was convenient for subsequent analysis.

[0072] Analysis and early warning:

[0073] Triggering of predefined rules: The system's predefined rules were set to trigger an infection early warning when the white blood cell count exceeded 12×109 / L. On the 4th day after the operation, the white blood cell count of Patient A rose to 12.5×109 / L, triggering the early warning.

[0074] Calculation of risk score by machine learning model: The logistic regression model calculated the infection risk score as 0.75 based on factors such as the dynamic changes in the vital signs of Patient A, the trends of laboratory indicators, age, and past medical history. The rule engine judged that the body temperature of Patient A was normal according to the set threshold rules, but the white blood cell count increased, triggering an infection early warning. The multi-model fusion algorithm combined the logistic regression model and the decision tree model, and finally calculated the comprehensive risk score as 0.80, indicating that Patient A had a relatively high infection risk.

[0075] User Interface and Feedback:

[0076] Graphical operation interface: The system real-time monitoring dashboard displays the vital signs trend chart, white blood cell count change curve and infection risk score of patient A. It also displays the detailed warning information, including alarm time, patient ID, risk score, alarm level and recommended treatment measures (such as rechecking blood routine, using antibiotics, etc.).

[0077] Mobile terminal push: The system sends alarm information to medical staff through mobile terminal applications, reminding them to check patient A's condition in time and take corresponding measures.

[0078] Results: According to the system warning information, the medical staff promptly re-examined patient A and found that his wound had slight signs of infection. After timely treatment, the infection of patient A was effectively controlled and the condition was prevented from worsening.

[0079] Example 2: Early warning of postoperative bleeding complications

[0080] Patient condition: Patient B, female, 60 years old, underwent thoracic surgery for esophageal cancer. She was admitted to the general ward after surgery and needed to be closely monitored for vital signs and risk of complications.

[0081] Data collection:

[0082] Vital signs data: The monitor collects heart rate, blood pressure, respiratory rate and blood oxygen saturation data in real time. Heart rate: 85bpm; blood pressure: 130 / 85mmHg; respiratory rate: 20 times / minute; blood oxygen saturation: 97%.

[0083] Laboratory test data: Routine blood test showed that the red blood cell count was 4.5×10 12 / L, coagulation function test showed prothrombin time (PT) was 12 seconds.

[0084] Imaging data: Chest CT images on the second day after surgery showed no abnormalities in the lungs, but the drainage volume of the chest drainage tube increased.

[0085] Medical staff manually entered information: the patient complained of chest pain after surgery, the drainage volume of the drainage tube increased, and the nursing record showed that the patient's postoperative recovery was acceptable.

[0086] Data processing and storage:

[0087] Data cleaning: An outlier value (systolic pressure 150 mmHg) was detected in the blood pressure data, which was identified as an outlier using the triple standard deviation method and removed. The missing prothrombin time data was processed using the mean filling method.

[0088] Data fusion: Integrate the data collected by the monitor, laboratory test system, and imaging system through a data middleware, unify the format, and store it in the database.

[0089] Data time series management: Add timestamps to all data to ensure the time series of the data, which is convenient for subsequent analysis.

[0090] Analysis and early warning:

[0091] Trigger by predefined rules: The system's predefined rules are set to trigger a bleeding warning when the drainage volume of the thoracic drainage tube exceeds 200 ml / hour. On the 3rd day after the operation, the drainage volume of the thoracic drainage tube of Patient B reached 220 ml / hour, triggering the warning.

[0092] Calculation of risk score by machine learning model: The logistic regression model calculates the bleeding risk score of 0.85 based on factors such as the dynamic changes in the vital signs of Patient B, the trend of coagulation indicators, age, and past medical history. The rule engine determines that the drainage volume of the thoracic drainage tube of Patient B is abnormal according to the set threshold rules, triggering a bleeding warning. The multi-model fusion algorithm combines the logistic regression model and the decision tree model, and finally calculates the comprehensive risk score of 0.90, indicating that Patient B has a high bleeding risk.

[0093] User interface and feedback:

[0094] Graphical operation interface: The system's real-time monitoring dashboard displays the trend chart of the vital signs of Patient B, the change curve of the drainage volume of the thoracic drainage tube, and the bleeding risk score. It shows the detailed warning information, including the alarm time, patient ID, risk score, alarm level, and recommended treatment measures (such as rechecking coagulation function, hemostatic treatment, etc.).

[0095] Push to mobile terminal: The system sends alarm information to medical staff through the mobile terminal application, reminding medical staff to check the condition of Patient B in time and take corresponding measures.

[0096] Results:

[0097] Based on the system's warning information, medical staff rechecked Patient B in time and found that the increase in the drainage volume of his thoracic drainage tube was due to postoperative intrathoracic bleeding. After timely treatment, the bleeding of Patient B was effectively controlled, avoiding further complications.

[0098] Summary: Through the above two embodiments, it can be seen that the postoperative complication warning system of the present invention can effectively collect the postoperative data of patients in real time, perform accurate analysis through predefined rules and machine learning models, timely warn of the risk of complications, and provide intuitive monitoring and alarm information for medical staff through the user interface and feedback layer. This helps medical staff to detect and handle the postoperative complications of patients in time, improving the postoperative recovery effect and safety of patients.

[0099] The technical solution of the present invention is not limited to the limitations of the above specific embodiments. Any technical deformation made according to the technical solution of the present invention falls within the protection scope of the present invention.

Claims

1. A postoperative complication warning system for thoracic surgery, characterized in that: It includes a data collection layer, a data transmission layer, a data processing and storage layer, an analysis and warning layer, and a user interface and feedback layer. The data collection layer collects patients' vital signs, laboratory tests, imaging data, and information manually entered by medical staff. The data transmission layer transmits the collected data to the data processing and storage layer in real time. The data processing and storage layer preprocesses, stores, and manages the collected data. The analysis and warning layer analyzes the data through predefined rules and machine learning models, calculates the risk score in real time, and triggers an alarm. The user interface and feedback layer provides a graphical operation interface, mobile terminal push, and alarm feedback.

2. The postoperative complication warning system for thoracic surgery according to claim 1, wherein: The vital signs collected by the data collection layer include heart rate, blood pressure, respiratory rate, and blood oxygen saturation data collected through the interfaces of monitors and ventilator devices. The collected laboratory tests include blood routine, inflammatory indicators, and coagulation indicators. The collected imaging data includes X-rays and CTs, and the image recognition system is used to assist in judging inflammation or bleeding conditions. The information manually entered by medical staff includes the patient's chief complaint, physical sign changes, and nursing records entered manually.

3. The postoperative complication warning system for thoracic surgery according to claim 2, wherein: The preprocessing of the collected data includes data cleaning, data fusion, and data time series management. The data cleaning processes missing data, outliers, and redundant data, and standardizes the data. The data fusion uses data middleware to integrate data from different devices and systems in a unified format to ensure data consistency. The data time series management stamps all data with timestamps to ensure the accuracy of subsequent real-time monitoring and historical data comparison and analysis. The data processing and storage layer stores and manages the collected data using a database, encrypts the data transmission using the SSL / TLS protocol, and implements role-based access control.

4. The thoracic surgery postoperative complication warning system according to claim 3, characterized in that: The predefined rules of the analysis and warning layer are to set key indicator thresholds, including abnormal body temperature, blood pressure fluctuations, and elevated white blood cells. An alarm is immediately triggered when the safety range is exceeded. The machine learning model uses historical clinical data to train the model to calculate the risk score. The feature selection includes the dynamic changes of vital signs, the trends of laboratory indicators, the patient's age, past medical history, and surgical type. The model fusion combines the rule engine and the statistical model, and uses voting or weighted fusion algorithms to improve the accuracy and robustness of the warning. The data stream engine is used to process high-concurrency real-time data streams to ensure that warning information can be generated and transmitted quickly.

5. The postoperative complication warning system for thoracic surgery according to claim 4, characterized in that: The user interface and feedback layer provides a graphical operation interface with a real-time monitoring dashboard, trend chart, and risk curve that intuitively display the patient's status and risk changes, and shows the detailed data records, alarm reason analysis, and recommended treatment measures for each alarm. Medical staff can query historical data, model prediction results, and feedback records. The mobile terminal push and alarm feedback receive alarm information and patient status updates on mobile devices in real time through the mobile terminal application.

6. The postoperative complication warning system for thoracic surgery according to claim 5, wherein: The data cleaning includes missing value filling, outlier detection, and standardization processing: Missing value filling is processed using mean filling or time series interpolation: Wherein: The filled value of the i-th eigenvalue at time t, N(t): the set of non-missing values near time t; |N(t)|: the number of non-missing values near time t; Outlier detection uses the three-standard-deviation method for outlier detection: Where: μ: Mean of the feature in the training samples; σ: Standard deviation of the feature in the training samples; x i : Feature value of the i-th sample; The standardization process uses z-score standardization for data normalization: where: x' i : the standardized eigenvalue; In the data fusion, assume devices D1, D2, ..., D m collect data streams X1, X2, ..., X m , and use linear interpolation for data alignment: Where: X' i (t): Interpolation data at time t; X i (t1), X i (t2): Original data values at adjacent time points; t1, t2: Adjacent time points; The data time series management assigns a timestamp T to each piece of data and stores it in the database: D = {(T i , X i )}, i = 1, 2,..., n Where: D: the data set stored in the database; T i : the timestamp of the i-th data, X i : the feature vector of the i-th data.

7. The postoperative complication warning system for thoracic surgery according to claim 6, wherein: The machine learning model includes a logistic regression model, a rule engine, and multi-model fusion; The logistic regression model uses logistic regression to calculate the postoperative complication risk score: Where: Where: P(y = 1|X): the probability of predicting the occurrence of complications; X = [x1, x2,..., x n : the patient feature vector; x i : the i-th eigenvalue; β i : the weight of the i-th feature; β0: the bias term; The loss function is optimized using cross-entropy loss: The rule engine triggers an alarm based on the set threshold rules: Including: body temperature > 38.5°C triggers a high fever warning; systolic blood pressure < 90 mmHg triggers a low blood pressure warning; white blood cell count > 12×10 9 / L triggers an infection warning; Multi-model fusion combines the logistic regression and decision tree models and uses weighted fusion to calculate the final risk score: R = γ1R LR + γ2R DT + γ3R NN Where: R: Final comprehensive risk score; R LR , R DT , R NN : Predicted risk scores of logistic regression, decision tree, and neural network; γ i : Model weighting coefficient.

8. The postoperative complication warning system for thoracic surgery according to claim 7, characterized in that: In the analysis and warning layer, the key index threshold uses the exponential smoothing method to calculate the dynamic threshold: S t = αx t + (1 - α)S t-1 Where: S t : The smoothed index value; x t : The raw value at the current moment; α: The smoothing coefficient, controlling the sensitivity (0 < α < 1); S t-1 : The smoothed value at the previous moment; When S t > θ, an alarm is triggered.

9. The postoperative complication warning system for thoracic surgery according to claim 8, wherein: The analysis and warning layer uses a WebSocket-based push mechanism to send notifications to the mobile devices of medical staff when an alarm is triggered: M = {(T, ID, R, L)} Where: M: Alarm information in the message queue; T: Alarm time; ID: Patient ID; R: Risk score; L: Alarm level.

Images