A wearable post-cardiac surgery sign and wound linkage monitoring and care system

Abstract

The application discloses a wearable postoperative sign and wound linkage monitoring and nursing system and relates to the technical field of sign and wound monitoring and nursing.The system comprises a wearable collection terminal, an edge computing unit connected with the wearable collection terminal and a nursing intervention output end in communication connection with the edge computing unit.The wearable collection terminal comprises a main wearable unit and an auxiliary wearable unit, wherein the main wearable unit is pasted on the surface of the median sternotomy incision of a target patient.The application discards a single index determination logic, constructs a false positive exclusion mechanism integrating a clinical knowledge graph, and through sensor abnormal duration determination, activity state correlation, symptom label matching and other logics, can eliminate sensor interference, post-activity physiological fluctuations, sweat interference and other misjudgments, simultaneously realizes accurate differentiation of postoperative absorption heat and wound infection, normal bleeding and active bleeding, and reduces unnecessary medical resource consumption and panic.

Description

Technical Field

[0001] This invention relates to the field of vital signs and wound monitoring and nursing technology, specifically a wearable postoperative cardiac surgery vital signs and wound linkage monitoring and nursing system. Background Technology

[0002] Postoperative cardiac surgery monitoring of vital signs and wound care is a core nursing task during the recovery period after cardiac surgery, especially during the high-risk hospitalization period of 1-2 weeks postoperatively. Its purpose is to promptly detect common postoperative complications such as bleeding, infection, heart failure, and arrhythmias, helping patients recover smoothly. It mainly consists of two parts: vital sign monitoring and wound monitoring. Vital sign monitoring involves dynamic observation of systemic and circulatory functions, continuously monitoring key physiological indicators to detect functional abnormalities early, given the specific high circulatory risk after cardiac surgery. Wound monitoring is specialized care for surgical incisions and drainage. Common cardiac surgery wounds include the main sternal incision, the lower limb incision for harvesting the great saphenous vein in coronary artery bypass graft patients, the wrist incision for harvesting the radial artery, and small chest wall incisions in minimally invasive patients. Most patients also have pericardial / mediastinal / thoracic drainage tubes in place postoperatively. Nursing care for these wounds focuses on wound condition monitoring and drainage tube care.

[0003] The fiber optic sensing-based intelligent monitoring and nursing system for cardiovascular postoperative wounds, patent publication number CN120982991A, significantly improves monitoring accuracy and continuity, and greatly enhances early anomaly detection capabilities. The fiber optic sensor array module employs an anti-electromagnetic interference design, enabling stable operation in environments with equipment such as ECG monitors, avoiding data deviations caused by electromagnetic signals. The combination of distributed and point-based fiber optics allows for monitoring of wound surface temperature and circumferential pressure, as well as capturing microscopic parameters such as subcutaneous interstitial fluid pressure and exudate refractive index, achieving comprehensive multi-dimensional data coverage. The multi-parameter data acquisition module continuously collects data at high frequency, avoiding the omission of anomalies caused by periodic checks; for example, it can detect subtle increases in exudate volume in real time, providing a basis for early intervention in inflammation. The sensor failure self-diagnosis module can quickly locate faults and activate redundancy schemes to ensure uninterrupted monitoring. Compared to the frequent interruptions of existing equipment, data integrity and reliability are significantly improved.

[0004] Because the existing technical solutions for monitoring vital signs and wounds operate completely independently, without utilizing the pathological correlation between the two types of indicators to establish a joint judgment mechanism, the false positive rate of single-dimensional threshold alarms is high. For example, a simple increase in body temperature cannot distinguish between postoperative absorption fever and wound infection, a simple increase in drainage volume cannot distinguish between normal oozing and active bleeding, and it is even more unable to identify occult complications, such as occult pericardial hemorrhage caused by drainage tube blockage and deep mediastinal infection without surface exudate, resulting in delayed early warning and a high rate of missed diagnosis of fatal complications. Summary of the Invention

[0005] The purpose of this invention is to provide a wearable postoperative cardiac surgery vital signs and wound linkage monitoring and nursing system to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a wearable postoperative cardiac surgery vital signs and wound linkage monitoring and nursing system, comprising:

[0007] Wearable data acquisition terminal, edge computing unit connected to the wearable data acquisition terminal, and nursing intervention output terminal communicating with the edge computing unit;

[0008] The wearable data acquisition terminal includes a main wearable unit and a secondary wearable unit. The main wearable unit is attached to the surface of the surgical incision in the center of the sternum of the target patient and integrates a wound monitoring module and a vital sign monitoring module to acquire wound monitoring data and vital sign monitoring data of the target patient, respectively, to obtain wound monitoring dataset and vital sign monitoring dataset.

[0009] The auxiliary wearable unit is attached to the surface of the distal surgical incision of the target patient, including the great saphenous vein incision of the lower extremity and the radial artery incision of the wrist. It is equipped with a miniature wound acquisition module to acquire the distal incision monitoring data of the target patient and obtain the distal incision monitoring dataset.

[0010] The edge computing unit has a built-in clinical data-trained multi-parameter linkage early warning model for vital signs and wounds. By associating the coupling changes of cyclical vital sign parameters and wound parameters at multiple sites, it outputs graded early warning results.

[0011] Based on the output dataset and matching the corresponding complication type labels, the system automatically pushes tiered nursing intervention plans that are suitable for the scenario and the individual characteristics of the patient. At the same time, it automatically generates nursing records and synchronizes them to the hospital's electronic medical record system to obtain intervention push items, forming a complete closed loop of collection-analysis-intervention-archiving.

[0012] Furthermore, the wound monitoring module integrates a microcapacitive exudate sensor, a pH electrochemical sensor, a lattice temperature sensor, a MEMS micro-vibration sensor, and a pluggable ultrasonic drainage tube flow sensor. The method for acquiring the wound monitoring dataset includes:

[0013] Based on the microcapacitive exudation sensor, the incision exudate volume and exudate rate are collected in real time. At the same time, the subcutaneous swelling pressure and swelling amplitude are detected by the change of skin impedance. The corresponding data are output to obtain the first wound data.

[0014] A pH electrochemical sensor collects and outputs the pH value of wound exudate, thus obtaining second wound data;

[0015] The dot matrix temperature sensor is configured with at least three temperature measurement points to collect the local temperature of the wound and the temperature of the surrounding normal skin. After calculating the temperature difference between the two, the data of the third wound is obtained.

[0016] MEMS micro-vibration sensors collect and output data on the amplitude of sternal micro-movements during patient coughing, changes in body position, and upper limb exertion, thus obtaining fourth wound data.

[0017] The pluggable ultrasonic drainage tube flow sensor has a non-contact structure, collects and outputs the drainage fluid flow rate and cumulative drainage volume, and obtains the fifth wound data;

[0018] The wound monitoring dataset is obtained by aligning the timestamps of the first, second, third, fourth, and fifth wound data, associating them with the target patient ID, and then storing them in a structured manner.

[0019] Furthermore, the vital signs monitoring module integrates a single-lead ECG sensor, a PPG photoplethysmography sensor, an impedance-type respiratory sensor, a six-axis posture sensor, and an interaction port. The method for acquiring the vital signs monitoring dataset includes:

[0020] A single-lead ECG sensor identifies atrial premature beats, ventricular premature beats, atrial fibrillation, and ventricular tachycardia, outputs heart rate and rhythm classification results, and obtains primary vital sign data.

[0021] The PPG photoplethysmography sensor automatically calibrates blood pressure, simultaneously collects and outputs blood oxygen saturation and pulse rate data, and obtains secondary vital sign data.

[0022] Impedance-type respiratory sensors collect and output respiratory rate and respiratory amplitude data to obtain third vital sign data.

[0023] The six-axis posture sensor identifies the target patient's position, activity level, and whether coughing occurs, and outputs the data to obtain the fourth vital sign.

[0024] The interaction port is a Bluetooth port, which automatically synchronizes the weight and water consumption data of the smart scale and smart water bottle that are bound to it. It also supports users to manually enter data and label symptoms such as chest tightness, chest pain and cough, and outputs the data to obtain the fifth vital sign data.

[0025] The data on the first, second, third, fourth, and fifth vital signs are timestamped, associated with the target patient ID, and then stored in a structured manner to obtain the vital sign monitoring dataset.

[0026] Furthermore, the miniature wound acquisition module includes a miniature temperature sensor and a miniature exudate sensor, and the method for acquiring the distal incision monitoring dataset includes:

[0027] Based on the attachment location of the secondary wearable unit, the incision ID is bound, and the corresponding label is the surface location of the distal surgical incision, thus obtaining the incision binding item;

[0028] A miniature temperature sensor collects and outputs the local temperature of the corresponding incision, and a miniature seepage sensor collects and outputs the seepage volume of the corresponding incision, thus obtaining the incision output data.

[0029] The auxiliary wearable unit synchronizes data with the main wearable unit in real time. Based on the incision binding item, the incision output data output by the auxiliary wearable unit is timestamped, associated with the target patient ID, and then stored in a structured manner to obtain the remote incision monitoring dataset.

[0030] Furthermore, the method for acquiring the vital signs-wound multi-parameter linkage early warning model includes:

[0031] Clinical data patient samples were obtained and categorized into five types: active bleeding / cardiac tamponade, wound / mediastinal infection, poor sternal healing, early heart failure, and normal, resulting in a sample dataset.

[0032] Feature extraction is performed based on the sample dataset, and the features are normalized before being used as model input to obtain the input dataset;

[0033] We trained a hybrid architecture model, built a hard rule engine, set the unmodifiable triggering rules for fatal complications based on the optimal cutoff value calculated from clinical guidelines and large-sample ROC curves, and then built a lightweight pruning and 8-bit quantization XGBoost multi-class model to handle fuzzy scenarios where hard rules were not triggered.

[0034] The input dataset is divided into training, validation, and test sets. Model training and validation are then performed to obtain a multi-parameter linkage early warning model for vital signs and wounds.

[0035] Furthermore, the aforementioned vital signs-wound multi-parameter linkage early warning model incorporates a baseline dynamic update mechanism, including:

[0036] After the device is worn, the sensor calibration process is run, the acquisition threshold is set, and wound monitoring data and vital sign monitoring data in the resting state are continuously acquired. The mean is calculated to generate an initial baseline library.

[0037] Set an update threshold, which is a fixed postoperative time value, to trigger a baseline update. Based on the acquisition threshold, continuously acquire wound monitoring datasets and vital sign monitoring datasets in a resting state, calculate the mean, and replace the original baseline.

[0038] Furthermore, the aforementioned vital signs-wound multi-parameter linkage early warning model incorporates false positive exclusion logic, including:

[0039] An abnormal parameter threshold is set, which is a fixed time value. The duration of abnormal parameters collected by the wearable collection terminal is logically determined based on the collection threshold.

[0040] Furthermore, the method for obtaining the output dataset includes:

[0041] The wound monitoring dataset, vital sign monitoring dataset, and distal incision monitoring dataset are compared with the dynamic baseline library to obtain the variation range of each parameter. The data are then input into the model and the model output results are obtained.

[0042] Based on the results of the graded early warning, at least two judgment rules are set to obtain at least two judgment rule range sets;

[0043] The model output results are judged based on the set of judgment rules to obtain the corresponding level of early warning results, and the output dataset is obtained.

[0044] Furthermore, the method for obtaining the intervention push item includes:

[0045] The nursing intervention output terminal is configured with interactive ports between the inpatient and outpatient terminals. Based on the graded early warning results, corresponding intervention plans are set to obtain intervention level items.

[0046] Identify the scene in which the target patient is located, obtain the real-time scene item, and obtain the corresponding intervention plan based on the output dataset and the combination result of the intervention level item, thereby obtaining the intervention push item.

[0047] Compared with the prior art, the beneficial effects of the present invention are:

[0048] This wearable postoperative cardiac surgery vital signs and wound linkage monitoring and nursing system abandons the logic of single indicator judgment and builds a false positive exclusion mechanism that integrates clinical knowledge graph. Through the logic of judging the duration of abnormal sensors, associating with activity status, and matching symptom tags, it can eliminate misjudgments such as sensor interference, physiological fluctuations after activity, and sweat interference. At the same time, it can accurately distinguish between postoperative absorption heat and wound infection, and normal bleeding and active bleeding, reducing unnecessary medical resource consumption and panic.

[0049] Meanwhile, relying on the multi-dimensional wound monitoring of the main and auxiliary wearable units, including subcutaneous swelling pressure, non-contact flow monitoring of drainage tubes, and other data linked with vital signs, it can promptly detect insidious and fatal complications such as occult pericardial hemorrhage caused by drainage tube blockage and deep mediastinal infection without surface exudate. Compared with the traditional method of manual recording every hour, it can provide early warning and avoid adverse events such as emergency thoracotomy caused by missed diagnoses. Furthermore, based on the hybrid architecture early warning model and combined with the dynamic baseline update mechanism to adapt to the patient's postoperative recovery pattern, it can output three-level graded early warning results and match personalized intervention plans to inpatient / outpatient scenarios, forming a closed loop of collection-analysis-intervention-archiving, which saves valuable time for clinical intervention, and can effectively reduce the risk of complication progression, especially for high-risk patients. Attached Figure Description

[0050] Figure 1This is a schematic diagram of the overall structural distribution of the present invention;

[0051] Figure 2 This is a schematic diagram of the wound monitoring module structure of the present invention;

[0052] Figure 3 This is a schematic diagram of the vital signs monitoring module of the present invention;

[0053] Figure 4 This is a schematic diagram of the multi-parameter linkage early warning model for vital signs and wounds 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] In modern medical care systems, vital sign monitoring and wound monitoring are important means of assessing patient condition and preventing complications. However, in current technical solutions, these two types of monitoring often operate independently, failing to fully utilize the pathological correlation between them. This results in a high false positive rate for single-dimensional threshold alarms, delayed warnings, and even missed diagnoses of fatal complications. The high false positive rate of single-dimensional threshold alarms, relying solely on elevated body temperature as an alarm indicator, cannot distinguish between postoperative absorption fever and wound infection, causing unnecessary panic and additional consumption of medical resources. Similarly, simply increasing drainage volume cannot distinguish between normal oozing and active bleeding, easily delaying the judgment of true bleeding. Existing monitoring methods are unable to identify occult complications such as occult pericardial hemorrhage caused by drainage tube blockage and deep mediastinal infection without surface exudate. These complications often have an insidious onset and progress rapidly. If missed, the consequences can be dire. Due to a lack of comprehensive analysis of physical signs and wound indicators, early warnings often lag behind changes in the condition, failing to buy valuable time for clinical intervention. The technical solution provided in this application, however, abandons the logic of judging based on a single indicator and constructs a false-positive exclusion mechanism that integrates a clinical knowledge graph. Through logic such as judging the duration of abnormal sensor readings, associating activity status, and matching symptom tags, it can eliminate misjudgments caused by sensor interference, post-activity physiological fluctuations, and sweat interference. Simultaneously, it achieves accurate differentiation between postoperative absorption heat and wound infection, and between normal oozing and active bleeding, reducing unnecessary consumption of medical resources and panic. Figure 1 As shown, it includes a wearable data acquisition terminal, an edge computing unit connected to the wearable data acquisition terminal, and a nursing intervention output terminal that is communicatively connected to the edge computing unit.

[0056] The wearable data acquisition terminal includes a main wearable unit and a secondary wearable unit. The main wearable unit is attached to the surface of the surgical incision in the center of the sternum of the target patient and integrates a wound monitoring module and a vital sign monitoring module to acquire wound monitoring data and vital sign monitoring data of the target patient, respectively, to obtain wound monitoring dataset and vital sign monitoring dataset.

[0057] It is important to note that, such as Figure 2 As shown, the wound monitoring module integrates a microcapacitive exudate sensor, a pH electrochemical sensor, a matrix temperature sensor, a MEMS micro-vibration sensor, and a pluggable ultrasonic drainage tube flow sensor. The method for acquiring the wound monitoring dataset includes: real-time acquisition of incision exudate volume and rate based on the microcapacitive exudate sensor, while simultaneously detecting subcutaneous swelling pressure and swelling amplitude through changes in skin impedance, outputting corresponding data to obtain the first wound data; the pH electrochemical sensor acquires and outputs the pH value of the wound exudate to obtain the second wound data; and the matrix temperature sensor is configured with at least three temperature measurement points to collect local wound temperature data. The temperature and the surrounding normal skin temperature are calculated, and the temperature difference between the two is output to obtain the third wound data; the MEMS micro-vibration sensor collects and outputs the amplitude of sternal micro-movement when the patient coughs, changes in body position, and exerts force on the upper limbs, and obtains the fourth wound data; the pluggable ultrasonic drainage tube flow sensor has a non-contact structure, collects and outputs the drainage fluid flow rate and cumulative drainage volume, and obtains the fifth wound data; based on the first, second, third, fourth, and fifth wound data, the timestamps are aligned, the target patient ID is associated, and the data is then structured and saved to obtain the wound monitoring dataset.

[0058] Specifically, the microcapacitive exudate sensor is based on the principle of dielectric constant change of parallel plates. As exudate enters the gap between the plates, the dielectric constant increases, and the corresponding output value is linearly correlated with the amount of exudate. The detection accuracy is set at 0.1 ml, with a sampling interval of 1 min. Simultaneously, it fits subcutaneous swelling pressure through the skin-plate impedance change, with an error ≤10%. The pH electrochemical sensor uses a polytetrafluoroethylene permeable membrane, allowing only hydrogen ions from wound exudate to pass through. The detection range is 5.0-9.0, with an accuracy of 0.1 and a sampling interval of 10 min. The dot matrix temperature sensor is configured with four NTC temperature measurement points, two of which are attached to the incision edge at 0.5°C. Two temperature sensors are placed at a distance of m, 2cm outside the incision on normal skin. The temperature measurement accuracy is 0.1℃, and the sampling interval is 5min. The final temperature difference is automatically calculated by taking the difference between the incision point and the normal skin point on the same side. The MEMS micro-vibration sensor collects the sternal deformation signal based on the piezoelectric effect, with a sampling frequency of 100Hz. It can identify micro-movement amplitudes greater than 0.1mm. The pluggable ultrasonic drainage tube flow sensor is based on the Doppler frequency shift principle. It is non-contactly clamped to the outer wall of the drainage tube without puncturing the tube, avoiding contamination of the drainage system. The detection accuracy is 1ml, and the sampling interval is 1min. It can be directly removed upon discharge without affecting the operation of other sensors.

[0059] In the specific implementation process, a patient returned to the ICU after aortic valve replacement surgery. The main unit was attached to the midline sternal incision, and the drainage tube sensor was clipped to the mediastinal drainage tube. Two hours postoperatively, the sensor measured a cumulative exudate of 48 ml over 30 minutes, an exudate rate of 1.6 ml / min, a 18% decrease in subcutaneous impedance compared to baseline indicating swelling, a pH of 7.2, a temperature difference of 0.3℃ between the incision and the surrounding skin, a sternal micromovement amplitude of 0.8 mm during coughing, and a drainage volume of 42 ml over 30 minutes. All data were aligned with timestamps, associated with the patient ID: XXXXXXXX, and then stored in the wound monitoring dataset.

[0060] It is important to note that, such as Figure 3 As shown, the vital signs monitoring module integrates a single-lead ECG sensor, a PPG photoplethysmography sensor, an impedance-type respiratory sensor, a six-axis posture sensor, and an interactive port. The methods for acquiring the vital signs monitoring dataset include: the single-lead ECG sensor identifies atrial premature beats, ventricular premature beats, atrial fibrillation, and ventricular tachycardia, outputting heart rate and rhythm classification results to obtain the first vital signs data; the PPG photoplethysmography sensor performs automatic blood pressure calibration, simultaneously collects and outputs blood oxygen saturation and pulse rate data to obtain the second vital signs data; the impedance-type respiratory sensor collects and outputs respiratory rate and respiratory amplitude data. The system obtains the third vital sign data; a six-axis posture sensor identifies and outputs the target patient's position, activity intensity, and presence of coughing movements, resulting in the fourth vital sign data; the interaction port is a Bluetooth port, which automatically synchronizes the weight and water intake data of the bound smart scale and smart water bottle, while also supporting user manual data entry and symptom tags such as chest tightness, chest pain, and cough, resulting in the fifth vital sign data; the first, second, third, fourth, and fifth vital sign data are timestamped, associated with the target patient ID, and then structured and saved to obtain the vital sign monitoring dataset.

[0061] Specifically, the single-lead ECG sensor uses a silver chloride dry electrode with a sampling frequency of 250Hz. It incorporates the Pan-Tompkins algorithm to automatically identify four types of arrhythmias: atrial premature beats, ventricular premature beats, atrial fibrillation, and ventricular tachycardia. It outputs heart rate and rhythm classification results every minute. The PPG photoplethysmography sensor is based on the principle of hemoglobin absorbance changes and automatically calibrates blood pressure every 5 minutes using pulse wave transit time, with a measurement error ≤5mmHg. It simultaneously outputs blood oxygen saturation and pulse rate data. The impedance-type respiratory sensor measures the impedance changes caused by chest respiratory movements. It calculates respiratory parameters with a sampling frequency of 50Hz, outputting respiratory rate and amplitude. The six-axis posture sensor fits human posture based on accelerometer and gyroscope data, with a sampling frequency of 50Hz. It automatically distinguishes four body positions (supine, lateral, prone, and standing) and three activity intensities (resting, mild activity, and moderate activity). It also identifies coughing actions through chest vibration characteristics. The interaction port uses Bluetooth 5.0 protocol and can automatically pair with paired smart scales and smart water bottles. Users can manually enter food intake, urine output, and bowel output, as well as three symptom tags (chest tightness, chest pain, and cough) through the associated mini-program.

[0062] In the specific implementation process, a patient got out of bed and moved around on the afternoon of the 3rd day after surgery. The six-axis sensor identified it as moderate activity intensity. The single-lead ECG measured a heart rate of 102 beats / min and sinus rhythm. The PPG measured a blood pressure of 105 / 65 mmHg and blood oxygen 98%. The impedance sensor measured a respiratory rate of 22 breaths / min. The smart water cup synchronized the cumulative water intake of 800ml that day. The user entered no discomfort symptoms. All data were stored in the vital signs monitoring dataset after being aligned with the timestamp.

[0063] The auxiliary wearable unit is attached to the surface of the distal surgical incision of the target patient, including the great saphenous vein incision of the lower extremity and the radial artery incision of the wrist. It is equipped with a miniature wound acquisition module to acquire the distal incision monitoring data of the target patient and obtain the distal incision monitoring dataset.

[0064] It is important to note that the micro-wound acquisition module includes a micro-temperature sensor and a micro-exudate sensor. The method for acquiring the remote incision monitoring dataset includes: binding the incision ID based on the attachment position of the sub-wearing unit, which is labeled as the surface location of the remote surgical incision, to obtain the incision binding item; the micro-temperature sensor collects and outputs the local temperature of the corresponding incision, and the micro-exudate sensor collects and outputs the exudate volume of the corresponding incision, to obtain the incision output data; the sub-wearing unit synchronizes data with the main wearing unit in real time, and the incision output data output by the sub-wearing unit is timestamped based on the incision binding item, associated with the target patient ID, and then stored in a structured manner to obtain the remote incision monitoring dataset.

[0065] Specifically, the secondary wearable unit uses a transparent, low-sensitivity pressure-sensitive adhesive patch with a diameter of 2cm and a thickness of 1mm, weighing ≤3g in total. It does not interfere with medical staff's visual observation of the incision status. The built-in button battery allows for continuous use for 14 days. The miniature temperature sensor has a detection accuracy of 0.1℃ and a sampling interval of 5min, while the miniature exudate sensor has a detection accuracy of 0.1ml and a sampling interval of 5min. It communicates with the main wearable unit via Bluetooth Low Energy, with a communication distance ≤5m and a reconnection time ≤1s. There is no data disconnection when the patient gets out of bed. Each secondary wearable unit must be bound to a unique incision ID during application. The corresponding incision locations are labeled. For example, if a patient undergoes coronary artery bypass grafting, the great saphenous vein of the left leg and the radial artery of the right leg are used as graft vessels during the operation. After the operation, two auxiliary wearable units are attached, which are respectively bound to the ID: left great saphenous vein incision and ID: right radial artery incision. On the first day after the operation, the miniature sensor of the left leg incision measures a local temperature of 37.6℃, which is 0.4℃ higher than the surrounding skin, and an exudate volume of 0.2ml. The temperature of the right radial artery incision is measured to be 37.1℃ and the exudate volume is 0ml. After the data is synchronized to the main unit, it is aligned according to the incision ID and timestamp, associated with the patient ID, and stored in the remote incision monitoring dataset.

[0066] The edge computing unit has a built-in clinical data-trained multi-parameter linkage early warning model for vital signs and wounds. By associating the coupling changes of cyclical vital sign parameters and wound parameters at multiple sites, it outputs graded early warning results.

[0067] It is important to note that, such as Figure 4 As shown, the acquisition method of the vital signs-wound multi-parameter linkage early warning model includes: acquiring clinical data patient samples and classifying them into five categories: active bleeding / cardiac tamponade, wound / mediastinal infection, poor sternal healing, early heart failure, and normal, to obtain a sample dataset; extracting features from the sample dataset and normalizing the features before using them as model input to obtain an input dataset; training a hybrid architecture model, building a hard rule engine, setting an unmodifiable trigger rule for fatal complications based on the optimal cutoff value calculated from clinical guidelines and large-sample ROC curves, and then building a lightweight pruned, 8-bit quantized XGBoost multi-classification model to handle ambiguous scenarios where hard rules are not triggered; dividing the input dataset into training, validation, and test sets, and then training and validating the model to obtain the vital signs-wound multi-parameter linkage early warning model.

[0068] Specifically, clinical data samples were obtained, including wound monitoring parameters, vital sign monitoring parameters, and distal incision parameters. Wound monitoring parameters included incision exudate volume, exudate rate, subcutaneous swelling pressure, swelling amplitude, wound exudate pH, local wound temperature, surrounding normal skin temperature, patient cough, changes in body position, sternal micromotion amplitude during upper limb exertion, drainage flow rate, and cumulative drainage volume. Vital sign monitoring parameters included atrial premature beats, ventricular premature beats, atrial fibrillation, ventricular tachycardia type, blood oxygen saturation, pulse rate, respiratory rate, respiratory amplitude, body position, activity intensity, presence of coughing, chest tightness, chest pain, and cough symptom labels. Distal incision parameters included lower limb incision temperature and exudate volume. All data were collected at fixed sampling intervals, such as 1 minute, starting from the end of surgery, forming a time-series matrix. Where M is the time step and N is the number of parameter channels, the data is labeled and classified, and each segment... A one-hot encoded label vector is assigned This corresponds to five complication states: active bleeding / cardiac tamponade, wound / mediastinal infection, poor sternal healing, early heart failure, and normal status, resulting in a labeled sample dataset:

[0069] ;

[0070] in For the entire labeled sample dataset, containing multiple patients right, This represents the total number of samples (time series segments) in the dataset. For each sample's time series data, it is represented by a time window. Slide the slider, set the time window to 30 minutes, and extract multidimensional feature vectors. , The total number of feature dimensions enables the model to capture dynamic changes and cross-modal coupling relationships. Statistical features include descriptions of the numerical distribution within the window, including:

[0071] Mean: ;

[0072] Standard deviation: ;

[0073] Coefficient of variation: ;

[0074] in This represents the value of a certain monitoring parameter at time t. Indicates the number of sampling points within the window

[0075] Trend characteristics reflect whether parameters are continuously rising or falling, including:

[0076] Linear regression slope: ;

[0077] The final value deviates from the initial value by a percentage:

[0078] ;

[0079] in This represents the time mean of the parameter x within the window, used for trend and cross-correlation. For time variables, The average value of the time variable. Indicating the start and end times of the time window, multi-parameter coupling features are used to quantify the temporal dependence between two physiological parameters and are crucial for the model to identify linkage anomalies, including:

[0080] Cross-correlation delays, such as the lag relationship between heart rate increase and increased drainage volume:

[0081] ;

[0082] Pick and the maximum cross-correlation number at this time As a feature, among This is a time delay variable used to calculate the cross-correlation between two signals. To maximize the allowable time delay search range, the two signals are at the optimal time delay. The cross-correlation coefficient at each location reflects the strength of the linkage. This represents the values ​​of two different parameters, such as heart rate and drainage volume, at time t.

[0083] The multidimensional deviation index represents the gradient between the local temperature at the incision site and the whole body temperature, defining the whole-body reference temperature:

[0084] ;

[0085] The gradient of the sternal incision: ;

[0086] Its mean and rate of change are the characteristics, among which Temperature at the radial artery incision site in the wrist. Temperature at the incision site of the great saphenous vein in the lower extremity. The average temperature at multiple points on the surface of the sternal incision;

[0087] Shock index derivative characteristics are Among them, the estimated systolic blood pressure Extracted from pulse wave conduction time Mean, minimum, etc. within the window;

[0088] Perform a short-time Fourier transform on the MEMS vibration signal to calculate the proportion of low-frequency energy:

[0089] ;

[0090] The calculation result is That is, the proportion of low-frequency energy. The short-time Fourier transform amplitude spectrum of the vibration signal. The sampling frequency of the original signal. For frequency variables;

[0091] Heart rate variability frequency domain index is The Lomb-Scargle periodogram is used to calculate the RR interval sequence from ECG, and a feature vector is generated for each sliding window. All samples are combined into a feature matrix F, corresponding to the label Y;

[0092] To eliminate dimensional differences, each feature is standardized using Z-score:

[0093] ;

[0094] in , These are statistics for the training set only, applied to the validation and test sets. The normalized input dataset is denoted as... , For the j-th original feature value, This is the mean of the feature on the training set. The standard deviation of this feature on the training set. These are the standardized eigenvalues, with a mean of 0 and a standard deviation of 1. The standardized feature matrix for all samples;

[0095] Following this, a hybrid architecture model was trained, consisting of a "hard rule engine" and a "lightweight XGBoost multi-classifier." This model balances zero-latency capture of emergency situations with fine-grained discrimination of ambiguous states. For potentially fatal complications, based on clinical guidelines and large-sample ROC analysis of retrospective data, immutable, immediate triggering rules were set to determine the optimal cutoff. ,by For example, positive samples with a clear gold standard (secondary thoracotomy for hemostasis or imaging confirmation of cardiac tamponade) and negative samples (without such complications) with continuous drainage data were selected. The true positive rate (TPR) and false positive rate (FPR) at each candidate threshold were calculated, and ROC curves were plotted. The Youden index was selected.

[0096] Maximize the cutoff value Balancing sensitivity and specificity:

[0097] ;

[0098] Similarly, it can be calculated , which is the emergency threshold of heart rate. To estimate the emergency low threshold of systolic blood pressure, and based on clinical safety, the Youden index can be fine-tuned near the optimal value to absolutely guarantee 100% sensitivity, allowing for a certain number of false positives. Finally, the hard rules are solidified in the edge computing unit in the form of a decision table.

[0099] In the lightweight XGBoost multi-class model, for fuzzy scenes where hard rules are not triggered, a gradient boosting tree model is used for five-class prediction, given training data. ,in Let be the normalized feature vector of the i-th sample. Given the true label vector of the i-th sample, XGBoost iteratively adds a regression tree. To minimize the regularization loss:

[0100] ;

[0101] in for The predicted total score for each tree. The total number of trees is used to boost the gradient. For the first A tree of return, For the front The predicted total score vector for each tree has a length of 5. For sample i in class c, The predicted raw score of sample i in class c. For loss functions, cross-entropy is used for multi-class classification. To incorporate regularization into the overall objective function, multi-class classification employs softmax loss:

[0102] ;

[0103] Regular terms , The number of leaves. Leaf weights For hyperparameters, a histogram approximation algorithm is used during training to accelerate the search for split points, where... For the first The regularization term for the trees, To control model complexity, let's define the number of leaf nodes in a tree. This is the weight vector of the leaf nodes. The minimum gain threshold for splitting, a regularization parameter, controls the pruning intensity. The L2 regularization coefficient is used to prevent excessively large leaf weights. Maximum tree depth and subsample ratio are also set.

[0104] ;

[0105] Column sampling:

[0106] To prevent overfitting, post-pruning is performed based on feature importance, sorting by the sum of gains at the time of segmentation, and removing branches corresponding to features with less than 0.1% of total importance. This simplifies the tree structure, and 8-bit quantization is used for deployment on the edge-side MCU. The leaf weights and feature thresholds in the tree model are quantized from 32-bit floating-point to 8-bit integers. Symmetric linear quantization is employed.

[0107] ;

[0108] For weights and thresholds, zero point Scaling factor:

[0109] ;

[0110] Quantized value:

[0111] ;

[0112] in This is the original 32-bit floating-point value. The quantization scaling factor linearly maps the floating-point range to [-127, 127]. The maximum and minimum values ​​in a set of parameters, To quantize the zero point, we set it to 0 here. The quantized 8-bit integer values ​​are used directly in inference with integer operations, and the results are then dequantized back to probabilities. Quantization reduces the model size by approximately 75% and supports efficient inference on edge devices. The hard rule and XGBoost fusion logic works as follows: for each inference sample, it first passes through the hard rule engine; if any rule is matched, the corresponding complication label and highest warning level are directly output; if no rule is matched, features are extracted, normalized, and fed into the quantized XGBoost to output various probabilities. ;

[0113] Pick As a prediction category, and mapped to a graded early warning result based on the interval where the maximum probability lies, Let be the probability vectors for each category in the final output of the model;

[0114] Next, data partitioning, training, and model validation were performed. To avoid data leakage caused by different segments from the same patient, a stratified partitioning was adopted based on patient ID: training set: 70% of patients, validation set: 15% of patients, test set: 15% of patients, with the proportion of the five labels remaining relatively consistent across sets. XGBoost was trained on the training set, and early stopping was performed using the validation set. The process stopped when the multi-class log loss on the validation set did not decrease for 20 consecutive rounds. A 5-fold cross-validation grid search was used to optimize the following parameters:

[0115] Learning rates: 0.01, 0.05, 0.1;

[0116] Maximum tree depth: 3, 4, 5;

[0117] Subsamples: 0.7, 0.8;

[0118] L2 regularization λ: 1, 5, 10

[0119] The hyperparameter combination with the highest macro-average F1 score on the validation set is selected as the final model.

[0120] Report precision, recall, F1 score, and macro-average F1 for each category on the test set:

[0121] ;

[0122] Among them, The macro-average F1 score, the arithmetic mean of the F1 scores for each class, measures the overall performance of the classifier. Let be the recall rate for class c, representing the proportion of true individuals belonging to that class that were correctly predicted. The accuracy rate for class cc is the proportion of correctly predicted cases belonging to that class. The early warning time is the median number of minutes before the clinical diagnosis time for true positive samples. The final "sign-wound multi-parameter linkage early warning model" consists of hard rule logic blocks, lightweight XGBoost model files, normalized parameters and tree structure, feature extraction configuration, and normalized parameters. It can be directly embedded and deployed on edge computing units to achieve real-time, high-precision, and low-latency multi-parameter linkage early warning.

[0123] It is important to note that the multi-parameter linkage early warning model for vital signs and wounds has a built-in dynamic baseline update mechanism, which includes: after the device is worn, running the sensor calibration process, setting the acquisition threshold, continuously acquiring wound monitoring datasets and vital sign monitoring datasets in the resting state, calculating the mean to generate an initial baseline library; setting the update threshold, which is a fixed postoperative time value, triggering the baseline update, and based on the acquisition threshold, continuously acquiring wound monitoring datasets and vital sign monitoring datasets in the resting state, calculating the mean and replacing the original baseline.

[0124] Specifically, the principle of baseline updating is to adapt to the normal recovery pattern of patients after cardiac surgery. In the early postoperative period, heart rate and body temperature will gradually decrease as the inflammatory response subsides. Fixing the baseline can easily misjudge fluctuations in normal recovery parameters as abnormalities. After the device is put on, a 10-minute sensor calibration process is first run to confirm that all sensor data are within the normal physiological range. Then, the patient's resting state is automatically identified. The six-axis sensor determines that the patient is lying flat, without activity, and without coughing. The collection threshold is set to 1 hour, and all vital signs and wound parameters are continuously collected for 1 hour. The mean is calculated to generate an initial baseline library. The update threshold is set to automatically trigger baseline updating every 72 hours after surgery. The mean of 1 hour of resting state parameters is calculated to replace the original baseline. If the patient has abnormal conditions such as fever or arrhythmia during the collection period, the period is automatically skipped and the collection is postponed to the next resting period to avoid abnormal values ​​contaminating the baseline. On the day of the patient's discharge, 1 hour of resting state parameters are automatically collected, the mean is calculated, and an independent outpatient rehabilitation baseline library is generated and stored separately from the inpatient baseline.

[0125] In the specific implementation process, a patient's resting heart rate was 88 beats / min and the local wound temperature was 37.2℃ on the day of surgery, which was used as the initial baseline. On the third day after surgery, the patient's resting heart rate dropped to 76 beats / min and the local wound temperature was 36.8℃. The system automatically updated the baseline. If the patient developed a lung infection and had a body temperature of 39.1℃ on the second day after surgery, the parameters during that period were automatically marked as abnormal and were not included in the baseline update.

[0126] It should be noted that the vital signs-wound multi-parameter linkage early warning model has built-in false positive exclusion logic, including: setting abnormal parameter thresholds, which are fixed time values, and making logical judgments on the duration of abnormal parameters collected by the wearable acquisition terminal based on the acquisition thresholds.

[0127] Specifically, the false positive exclusion logic is based on a clinical knowledge graph. An abnormal parameter threshold is set at 30 seconds. If the abnormal parameter collected by the sensor lasts for less than 30 seconds, it is determined to be an abnormal value caused by sensor displacement, electromagnetic interference, or touch, and is directly excluded from the warning judgment. If the heart rate increases by ≥10 beats / min from the baseline and the systolic blood pressure decreases by ≥10 mmHg from the baseline, and the six-axis posture sensor identifies the patient as being at a moderate to high activity level, it is determined to be a physiological fluctuation after activity, excluding alarms for active bleeding and heart failure. If the core body temperature is ≥38.5℃, and the user enters respiratory symptom tags such as cough, sputum, and sore throat, it is determined to be a respiratory infection, excluding alarms for wound / mediastinal infection. If the exudate pH is ≥7.5, and the microcapacitive exudate sensor detects an exudate volume of <0.1ml, it is determined to be interference from sweat on the skin surface, excluding alarms for wound infection.

[0128] During the actual implementation, a nurse accidentally touched the main wearable unit during rounds. The ECG sensor briefly showed a heart rate of 152 beats / min for only 12 seconds. The system determined this to be interference and eliminated it directly. When the patient got out of bed and went to the bathroom, the heart rate rose to 105 beats / min and the systolic blood pressure dropped to 92 / 60 mmHg. The six-axis sensor identified this as moderate activity intensity, and the system automatically excluded the active bleeding alarm to avoid false alarms.

[0129] Based on the output dataset and matching the corresponding complication type labels, the system automatically pushes tiered nursing intervention plans that are suitable for the scenario and the individual characteristics of the patient. At the same time, it automatically generates nursing records and synchronizes them to the hospital's electronic medical record system to obtain intervention push items, forming a complete closed loop of collection-analysis-intervention-archiving.

[0130] It should be noted that the method for obtaining the output dataset includes: comparing the wound monitoring dataset, vital sign monitoring dataset, and distal incision monitoring dataset with the dynamic baseline library to obtain the variation range of each parameter, inputting them into the model and obtaining the model output results; based on the graded early warning results, setting at least two judgment rules to obtain at least two judgment rule range sets; judging the model output results based on the judgment rule range sets to obtain the corresponding level of early warning results and obtain the output dataset.

[0131] Specifically, three judgment rules are set: Level 1 warning, Level 2 warning, and Level 3 warning. Level 1 warning, which meets any one of the hard rules, corresponds to a risk of fatal complications, labeled as active bleeding / cardiac tamponade. Hard rule 1: Drainage flow rate ≥50ml / 30min + heart rate increase from baseline ≥10 bpm + systolic blood pressure decrease from baseline ≥10 mmHg; Hard rule 2: No drainage flow ≥10min + heart rate ≥110 bpm + systolic blood pressure ≤90 / 60 mmHg + subcutaneous swelling pressure increase from baseline ≥20%; Level 2 warning occurs when Level 1 warning is not triggered, and X... The GBoost model outputs a complication risk probability ≥90%, corresponding to a complication risk requiring medical intervention, labeled as wound / mediastinal infection / poor sternal healing / early heart failure. Level 3 warnings occur when Level 1 and Level 2 warnings are not triggered, and the XGBoost model outputs a risk probability ≥70% and <90%, corresponding to a minor abnormal risk requiring adjustment of rehabilitation behavior, labeled as abnormal volume overload / poor lower extremity venous return / drainage tube blockage / sternal compression. A risk probability <70% is considered normal and no warning is generated. All warning results are associated with timestamps and complication labels and then stored in the output dataset.

[0132] In the specific implementation process, 1.5 hours after the operation, the drainage flow sensor measured 52 ml of drainage in 30 minutes, the heart rate increased from 85 beats / min to 97 beats / min, and the systolic blood pressure decreased from 112 mmHg to 98 mmHg, triggering a level 1 warning, labeled as active bleeding, and stored in the output dataset. On the 4th day after the operation, the temperature difference between the wound and the surrounding skin was 0.9℃, the pH of the exudate was 7.4, the body temperature was 38.2℃, and the XGBoost output infection probability was 82%, which was judged as a level 3 warning, labeled as suspected infection, and stored in the output dataset.

[0133] It should be noted that the method for obtaining intervention push items includes: configuring the inpatient and outpatient interaction ports on the nursing intervention output end, setting corresponding intervention plans based on the graded early warning results to obtain intervention level items; identifying the scenario in which the target patient is located to obtain real-time scenario items; and obtaining the corresponding intervention plan based on the output dataset and the combination results of intervention level items, thereby obtaining intervention push items.

[0134] Specifically, since the nursing intervention output end is configured with inpatient and outpatient interaction ports, it first identifies the target patient's scenario, determining whether it is inpatient or outpatient. Simultaneously, since there are three levels of tiered warnings: Level 1, Level 2, and Level 3, there are also three corresponding intervention plans. The Level 1 warning intervention plan is as follows: in inpatient scenarios, triggering the nurse station's audible and visual alarm, and simultaneously pushing the patient's medical record, current monitoring data, and intervention guidelines, including supine immobilization, establishing intravenous access, blood preparation, and preparing a thoracotomy kit, to the mobile terminals of the attending physician and responsible nurse; in outpatient scenarios, pushing emergency guidelines to the mobile terminals of the patient, family members, and follow-up physicians. The Level 2 warning intervention plan is as follows: in inpatient scenarios, personalized procedures for corresponding complications are pushed to the mobile terminals of the patient, family members, and follow-up physicians. The guidelines are pushed to the responsible nurse's terminal. Infection risk guidelines include sampling and culture, aseptic dressing, and retesting of infection indicators. Poor sternal healing guidelines include checking for sternal fixation and pressing the sternum when coughing. Heart failure risk guidelines include adjusting diuretic dosage and limiting daily fluid intake. In out-of-hospital scenarios, home care guidelines are pushed to the patient / family's mini-program. Infection risk guidelines include uploading wound photos. Poor sternal healing guidelines and heart failure risk guidelines include a follow-up visit to the hospital the next day. The three-level early warning corresponding intervention plan is as follows: In inpatient scenarios, rehabilitation reminders are pushed to the patient's bedside screen and the responsible nurse's terminal. Abnormal volume overload reminders include reducing fluid intake and increasing ambulation. Poor lower limb return reminders are pushed to the patient's mini-program. In out-of-hospital scenarios, rehabilitation reminders are pushed to the patient's mini-program. Sternal compression reminders include adjusting to a supine position. Prolonged sitting or standing reminders include moving the lower limbs.

[0135] In the specific implementation process, the existing patient, Mr. Zhang, male, 68 years old, had a 12-year history of type 2 diabetes with poor blood sugar control. His preoperative glycated hemoglobin was 7.8%, and he had an 8-year history of hypertension. Preoperative diagnosis was coronary atherosclerotic heart disease with severe aortic stenosis. He underwent median sternotomy for coronary artery bypass grafting, using the left great saphenous vein and right radial artery as graft vessels, plus aortic valve bioprosthetic valve replacement. The intraoperative cardiopulmonary bypass lasted 87 minutes, with a blood loss of 320 ml. He belonged to a high-risk group for postoperative infection, bleeding, and poor sternal healing. The patient returned postoperatively. Immediately after ICU admission, the patient was fitted with this monitoring system. The main wearable unit was attached to the surface of the midline sternal surgical incision. The removable ultrasonic drainage tube flow sensor was clipped to the outer wall of the mediastinal drainage tube. Two auxiliary wearable units were attached to the surfaces of the left great saphenous vein incision and the right radial artery incision, respectively, and were bound to the incision IDs: left lower extremity graft incision and right radial artery incision, respectively. After 10 minutes of sensor calibration, resting parameters were continuously collected for 1 hour to generate the initial baseline: resting heart rate 82 bpm, systolic blood pressure 128 mmHg, and local wound temperature 37.1℃. The sternal micromovement amplitude was ≤0.5mm, and the drainage flow rate was ≤10ml / 30min. At 2 hours and 15 minutes post-surgery, the system monitored that the drainage tube flow sensor showed no drainage fluid output for 12 consecutive minutes. The nurse had previously manually squeezed the drainage tube without finding any blockage, mistakenly believing that bleeding had decreased. The vital signs module monitored a heart rate increase to 97 beats / min, an increase of 15 beats / min from baseline, and a systolic blood pressure decrease to 102 mmHg, a decrease of 26 mmHg from baseline. The wound module's microcapacitive exudate sensor detected a 28% increase in subcutaneous swelling pressure from baseline, indicating subcutaneous hematoma. The system triggered a Level 1 emergency alarm, indicating a risk of occult cardiac tamponade. Simultaneously, the ICU nursing station triggered an audible and visual alarm, and the attending physician received push notifications of patient data and intervention guidelines on their mobile phones: immediately lay the patient flat, establish a second intravenous access, squeeze the drainage tube, and prepare for bedside ultrasound to screen for pericardial effusion. The physician immediately performed bedside ultrasound, confirming that there was 120ml of blood in the pericardium and that the tip of the drainage tube was blocked by a blood clot. After the nurse squeezed the drainage tube, 80ml of dark red blood clots were expelled. The patient's heart rate and blood pressure gradually returned to baseline levels, avoiding the need for emergency open-chest surgery to prevent fatal cardiac tamponade.

[0136] 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 embodiments and their equivalents.

Claims

1. A wearable postoperative cardiac surgery vital signs and wound linkage monitoring and nursing system, comprising: Wearable data acquisition terminal, edge computing unit connected to the wearable data acquisition terminal, and nursing intervention output terminal communicating with the edge computing unit; Its features are: The wearable data acquisition terminal includes a main wearable unit and a secondary wearable unit. The main wearable unit is attached to the surface of the surgical incision in the center of the sternum of the target patient and integrates a wound monitoring module and a vital sign monitoring module to acquire wound monitoring data and vital sign monitoring data of the target patient, respectively, to obtain wound monitoring dataset and vital sign monitoring dataset. The auxiliary wearable unit is attached to the surface of the distal surgical incision of the target patient, including the great saphenous vein incision of the lower extremity and the radial artery incision of the wrist. It is equipped with a miniature wound acquisition module to acquire the distal incision monitoring data of the target patient and obtain the distal incision monitoring dataset. The edge computing unit has a built-in clinical data-trained multi-parameter linkage early warning model for vital signs and wounds. By associating the coupling changes of cyclical vital sign parameters and wound parameters at multiple sites, it outputs graded early warning results. Based on the output dataset and matching the corresponding complication type labels, the system automatically pushes tiered nursing intervention plans that are suitable for the scenario and the individual characteristics of the patient. At the same time, it automatically generates nursing records and synchronizes them to the hospital's electronic medical record system to obtain intervention push items, forming a complete closed loop of collection-analysis-intervention-archiving.

2. The wearable postoperative cardiac surgery vital signs and wound linkage monitoring and nursing system according to claim 1, characterized in that: The methods for obtaining the wound monitoring dataset include: The wound monitoring module integrates a microcapacitive exudate sensor, a pH electrochemical sensor, a dot matrix temperature sensor, a MEMS micro-vibration sensor, and a pluggable ultrasonic drainage tube flow sensor. Based on the wound monitoring module, wound data of the target patient is acquired, associated with the target patient ID, and then structured and saved to obtain a wound monitoring dataset.

3. The wearable postoperative cardiac surgery vital signs and wound linkage monitoring and nursing system according to claim 1, characterized in that: The methods for obtaining the vital signs monitoring dataset include: The vital signs monitoring module integrates a single-lead ECG sensor, a PPG photoplethysmography sensor, an impedance-type respiratory sensor, a six-axis posture sensor, and an interaction port. Based on the vital signs monitoring module, the target patient's vital signs data are acquired, associated with the target patient ID, and then structured and saved to obtain the vital signs monitoring dataset.

4. The wearable postoperative cardiac surgery vital signs and wound linkage monitoring and nursing system according to claim 1, characterized in that: The method for obtaining the remote incision monitoring dataset includes: The miniature wound acquisition module integrates a miniature temperature sensor and a miniature exudate sensor. It binds the incision ID based on the attachment position of the secondary wearable unit, and marks the corresponding position on the surface of the distal surgical incision to obtain the incision binding item. The auxiliary wearable unit synchronizes data with the main wearable unit in real time. Based on the incision binding item, the incision output data output by the auxiliary wearable unit is associated with the target patient ID and then stored in a structured manner to obtain the remote incision monitoring dataset.

5. The wearable postoperative cardiac surgery vital signs and wound linkage monitoring and nursing system according to claim 1, characterized in that: The method for obtaining the vital signs-wound multi-parameter linkage early warning model includes: Clinical data patient samples were obtained and categorized into five types: active bleeding / cardiac tamponade, wound / mediastinal infection, poor sternal healing, early heart failure, and normal, resulting in a sample dataset. Feature extraction is performed based on the sample dataset, and the features are normalized before being used as model input to obtain the input dataset; We trained a hybrid architecture model, built a hard rule engine, set the unmodifiable triggering rules for fatal complications based on the optimal cutoff value calculated from clinical guidelines and large-sample ROC curves, and then built a lightweight pruning model to handle fuzzy scenarios where hard rules were not triggered. The input dataset is divided into training, validation, and test sets. Model training and validation are then performed to obtain a multi-parameter linkage early warning model for vital signs and wounds.

6. The wearable postoperative cardiac surgery vital signs and wound linkage monitoring and nursing system according to claim 1, characterized in that: The vital signs-wound multi-parameter linkage early warning model has a built-in baseline dynamic update mechanism, including: After the device is worn, the sensor calibration process is run, the acquisition threshold is set, and wound monitoring data and vital sign monitoring data in the resting state are continuously acquired. The mean is calculated to generate an initial baseline library. Set an update threshold, which is a fixed postoperative time value, to trigger a baseline update. Based on the acquisition threshold, continuously acquire wound monitoring datasets and vital sign monitoring datasets in a resting state, calculate the mean, and replace the original baseline.

7. The wearable postoperative cardiac surgery vital signs and wound linkage monitoring and nursing system according to claim 1, characterized in that: The vital signs-wound multi-parameter linkage early warning model has a built-in false positive exclusion logic, including: An abnormal parameter threshold is set, which is a fixed time value. The duration of abnormal parameters collected by the wearable data collection terminal is logically judged based on the collection threshold. If the duration of abnormal parameters does not reach the abnormal parameter threshold, it is judged as abnormal interference and is directly removed without entering the warning judgment.

8. The wearable postoperative cardiac surgery vital signs and wound linkage monitoring and nursing system according to claim 1, characterized in that: The methods for obtaining the output dataset include: The wound monitoring dataset, vital sign monitoring dataset, and distal incision monitoring dataset are compared with the dynamic baseline library to obtain the variation range of each parameter. The data are then input into the model and the model output results are obtained. Based on the results of the graded early warning, at least two judgment rules are set to obtain at least two judgment rule range sets; The model output results are judged based on the set of judgment rules to obtain the corresponding level of early warning results, and the output dataset is obtained.

9. A wearable postoperative cardiac surgery vital signs and wound linkage monitoring and nursing system according to claim 1, characterized in that: The method for obtaining the intervention push item includes: The nursing intervention output terminal is configured with interactive ports between the inpatient and outpatient terminals. Based on the graded early warning results, corresponding intervention plans are set to obtain intervention level items. Identify the scene in which the target patient is located, obtain the real-time scene item, and obtain the corresponding intervention plan based on the output dataset and the combination result of the intervention level item, thereby obtaining the intervention push item.

Images