Cerebral stroke sudden sign real-time grading alarm method and system based on edge calculation

Abstract

The invention provides a cerebral apoplexy sudden sign real-time grading alarm method and system based on edge computing, and relates to the technical field of edge computing, and the method comprises the steps: collecting multi-modal physiological sign data in real time through an intelligent monitoring bracelet worn by a patient, the multi-modal physiological sign data comprises heart rate, blood pressure and brain oxygen saturation, and the multi-modal physiological sign data is transmitted to an edge computing gateway deployed in a ward. According to the invention, efficient and accurate monitoring of the stroke high-risk patient is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of edge computing, in particular to a stroke emergency sign real-time grading alarm method and system based on edge computing. BACKGROUND

[0002] As a high incidence, high disability rate and high mortality rate of nervous system diseases, the acute attack of stroke is often closely related to the abnormal changes of the patient's physiological signs, especially the high-risk elderly population with underlying diseases such as hypertension and atrial fibrillation, the risk of suddenness is higher. In the current clinical practice, there are still many pain points in the sign monitoring of stroke patients. The traditional monitoring method mainly depends on the nurses' regular ward rounds or the patients' active calls, which is difficult to capture the real-time sign changes in time and easy to miss the best intervention opportunity. Even if the centralized monitoring system is used, due to the long data transmission link, the delay of algorithm distribution and other problems, the alarm response is lagged, which is difficult to meet the strict requirements of brain stroke emergency time.

[0003] At the same time, the abnormal signs of stroke patients show complex and diverse characteristics. The slight fluctuation of a single index may only be a physiological fluctuation, while the coordinated abnormality of multiple indexes often indicates a serious risk, which puts higher requirements on the grading judgment ability of the sign monitoring system. The existing monitoring system can only realize the simple threshold alarm of a single index, lacks comprehensive analysis of multi-modal sign data, and is easy to produce false alarm and missed alarm. Or because of the complexity of the algorithm, it needs to rely on high-performance computing, which is difficult to realize local real-time processing in the ward and other scenes, limiting its efficient application in the clinical front line. SUMMARY

[0004] The technical problem to be solved by the present application is to provide a stroke emergency sign real-time grading alarm method and system based on edge computing, which realizes efficient and accurate monitoring of high-risk patients with stroke.

[0005] To solve the above technical problems, the technical scheme of the present application is as follows: In a first aspect, a stroke emergency sign real-time grading alarm method based on edge computing, the method comprising: Step 1, real-time acquisition of multi-modal physiological sign data by a smart monitoring bracelet worn by the patient, the multi-modal physiological sign data including heart rate, blood pressure and cerebral oxygen saturation; Step 2, transmitting the multi-modal physiological sign data to an edge computing gateway deployed in the ward; Step 3, the edge computing gateway constructs a multi-dimensional convex hull defining the normal range of the normal sign space based on the normal data in the historical multi-modal physiological sign data, calculates the distance from the real-time multi-modal physiological sign data point to the multi-dimensional convex hull, and initializes the generation of a population containing multiple rule sets; based on the fitness of each rule set in the historical data, selection, crossover and mutation operations are performed to iteratively evolve the rule set, and an optimized rule set is obtained, and the multi-modal physiological sign data is compared and logically judged in real time through the optimized rule set to obtain a judgment result; Step 4, based on the judgment result, if multiple physiological parameters exceed the corresponding first threshold and satisfy the combined logical relationship, it is determined that the first abnormal situation is met and a high-level alarm is generated; if only a single physiological parameter exceeds the corresponding second threshold, it is determined that the second abnormal situation is met and a low-level alarm is generated; Step 5, based on the high-level alarm, an alarm message containing the patient location identifier and real-time sign details is pushed to the designated medical terminal, and a synchronous trigger is performed to cause the on-site alarm device in the ward to emit an audible and visual alarm signal; based on the low-level alarm, notification information prompting re-measurement is sent to the designated nursing terminal.

[0006] Further, the multi-modal physiological sign data is collected in real time by the smart monitoring bracelet worn by the patient, and the multi-modal physiological sign data includes heart rate, blood pressure and cerebral oxygen saturation, including: Step 1.1, the original photoplethysmogram signal is collected by the photoelectric sensor integrated in the bracelet, the radial artery pulsation signal is collected by the pressure sensor, and the brain blood oxygen parameter original signal is collected by the near-infrared spectrum sensor; Step 1.2, the collected original signals are preprocessed, including denoising and filtering processing of the photoplethysmogram signal, baseline correction processing of the radial artery pulsation signal, and motion artifact elimination processing of the near-infrared spectrum signal; Step 1.3, based on the preprocessed signals, the pulse wave feature extraction algorithm is used to calculate the heart rate value, the pulse wave transmission time algorithm is used to calculate the blood pressure value, and the dual-wavelength near-infrared spectrum analysis method is used to calculate the cerebral oxygen saturation value; Step 1.4, the calculated heart rate, blood pressure and cerebral oxygen saturation values are time-aligned and packaged, and the multi-modal physiological sign data in a standardized format is obtained.

[0007] Further, the multi-modal physiological sign data is transmitted to the edge computing gateway deployed in the ward, including: Step 2.1, the smart monitoring bracelet establishes a secure communication connection with the edge computing gateway in the ward through the low-power Bluetooth protocol, and encrypts and packages the standardized data frame; Step 2.2, after receiving the encrypted data frame, the edge computing gateway decrypts and verifies the data packet format, checks the data integrity and validity; Step 2.3, adding a timestamp and a patient identifier to the physiological data that passes the verification, and converting it into a standardized data format suitable for edge computing processing; Step 2.4, temporarily storing the processed standardized data in the local cache area of the edge computing gateway.

[0008] Further, the edge computing gateway constructs a multi-dimensional convex hull defining the spatial range of normal signs based on normal data in historical multi-modal physiological data, calculates the distance from the real-time multi-modal physiological data point to the multi-dimensional convex hull, and initializes the generation of a population containing multiple rule sets, including: Step 3.1, extracting historical multi-modal physiological data from the local cache area, and screening data samples representing normal physiological state according to clinically defined normal physiological parameter range values; Step 3.2, based on the data samples representing normal physiological state, using a convex hull generation algorithm to determine the boundary points of normal data in the high-dimensional feature space, and by connecting these boundary points to form the smallest convex polygon enclosing all normal data, the spatial range of normal signs is defined; Step 3.3, calculating the shortest Euclidean distance from the real-time collected multi-modal physiological data point to each surface of the multi-dimensional convex hull as a quantitative indicator of the degree of deviation from the normal state of the data point; Step 3.4, based on the distance index, generating initial rule sets with different parameter thresholds and logical combinations, where each rule contains specific physiological parameter threshold range, logical operation relationship and corresponding weight coefficient; Step 3.5, encoding the generated multiple rule sets as genetic algorithm population individuals to complete the initialization of the population.

[0009] Further, based on the fitness of each rule set in the historical data, selection, crossover and mutation operations are performed to iteratively evolve the rule sets, obtaining optimized rule sets, and through the optimized rule sets, real-time parallel comparison and logical judgment are performed on the multi-modal physiological data to obtain the judgment result, including: Step 3.6, based on historical multi-modal physiological data, calculating the fitness of each rule set in the population, and the fitness is comprehensively evaluated by the warning accuracy and false alarm rate of the rule set on historical data; Step 3.7, according to the fitness evaluation result, using roulette wheel selection method to select high-quality rule set individuals from the current population, retaining rule sets with high fitness, and performing single-point crossover operation on the selected rule set individuals, exchanging part of the rule conditions between different rule sets to generate new rule set combinations; Step 3.8, implementing mutation operation on the newly generated rule sets to randomly adjust the physiological parameter thresholds, logical relationships or weight coefficients in the rules with a preset probability; Step 3.9, repeat steps 3.6 to 3.8 until a preset number of iterations is reached or fitness converges, to obtain an optimized rule set, and perform parallel rule matching on the real-time collected multi-modal physiological sign data through the optimized rule set, and generate abnormal sign judgment results according to the rule condition satisfaction.

[0010] Further, based on the judgment result, if multiple physiological parameters exceed the corresponding first threshold and satisfy the combined logic relationship, it is determined to meet the first abnormal situation and generate a high-level alarm; if only a single physiological parameter exceeds the corresponding second threshold, it is determined to be the second abnormal situation and generate a low-level alarm, including: Step 4.1, receiving the abnormal sign judgment result from the edge computing gateway, the judgment result including the abnormal state identification and corresponding confidence score of each physiological parameter; Step 4.2, analyzing the judgment result, when at least two of heart rate, blood pressure and cerebral oxygen saturation are detected to exceed their respective corresponding first threshold at the same time, and the parameter abnormal duration reaches a preset length, it is determined to meet the first abnormal situation; Step 4.3, when only a single physiological parameter is detected to exceed its corresponding second threshold, and other parameters are within the normal range, it is determined to meet the second abnormal situation; Step 4.4, based on the first abnormal situation judgment result, generating a high-level alarm instruction including patient detailed information, abnormal parameter data and emergency level identification; Step 4.5, based on the second abnormal situation judgment result, generating a low-level alarm instruction including abnormal parameter type and current value.

[0011] Further, based on the high-level alarm, an alarm message containing patient location identification and real-time sign details is pushed to the designated medical terminal, and a synchronous trigger is sent to the ward on-site alarm device to issue an audible and visual alarm signal; based on the low-level alarm, a notification message prompting retest is sent to the designated nursing terminal, including: Step 5.1, based on the high-level alarm instruction, generating an alarm message containing patient bed number, abnormal parameter real-time value and historical trend data; based on the low-level alarm instruction, generating a notification message containing abnormal parameter type and recommended retest time; Step 5.2, querying the corresponding responsible medical staff information based on the patient identity information, determining the target push terminal of the high-level alarm message and the device identification of the ward on-site alarm device; Step 5.3, pushing the high-level alarm message to the designated medical terminal in real time through the hospital intranet communication system, and synchronously sending an audible and visual alarm trigger instruction to the ward on-site alarm device; Step 5.4, distributing the low-level alarm message to the terminal device of the patient's responsible nursing staff through the nursing workstation management system; Step 5.5, record the sending state and receiving confirmation of all alarm messages, and update to the patient monitoring log database.

[0012] In a second aspect, a stroke burst sign real-time grading alarm system based on edge computing includes: An acquisition module is configured to collect multi-modal physiological sign data in real time through a smart monitoring bracelet worn by a patient, the multi-modal physiological sign data including heart rate, blood pressure, and cerebral oxygen saturation, and transmit the multi-modal physiological sign data to an edge computing gateway deployed in a ward. A computing module is configured to construct a multi-dimensional convex hull defining a normal sign space range based on normal data in historical multi-modal physiological sign data by the edge computing gateway, calculate a distance from a real-time multi-modal physiological sign data point to the multi-dimensional convex hull to initialize a population including a plurality of rule sets, select, cross, and mutate each rule set in the historical data to iteratively evolve the rule sets to obtain optimized rule sets, and perform real-time parallel comparison and logical judgment on the multi-modal physiological sign data through the optimized rule sets to obtain a judgment result. An alarm module is configured to determine that a first abnormal situation is met and generate a high-level alarm if a plurality of physiological parameters exceed corresponding first threshold values and satisfy a combined logical relationship based on the judgment result, and determine a second abnormal situation and generate a low-level alarm if only a single physiological parameter exceeds a corresponding second threshold value. A processing module is configured to push an alarm message including a patient location identifier and real-time sign details to a designated medical care terminal based on the high-level alarm, and synchronously trigger a sound and light alarm signal of a ward on-site alarm device, and send notification information prompting re-measurement to a designated nursing terminal based on the low-level alarm.

[0013] In a third aspect, a computing device includes: One or more processors; A storage device configured to store one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the method.

[0014] In a fourth aspect, a computer readable storage medium has a program stored therein, when the program is executed by a processor, the method is implemented.

[0015] The above-mentioned scheme of the present application at least includes the following beneficial effects: The multi-modal physiological sign data is collected in real time through the intelligent monitoring bracelet and transmitted to the ward edge computing gateway, the normal sign range is defined by using the multi-dimensional convex hull, the local real-time analysis and judgment of the data are realized by combining the genetic algorithm optimization rule set, and the hierarchical alarm is generated according to the abnormal situation and accurately pushed to the corresponding terminal. Not only the response speed is improved through edge computing to meet the timeliness demand of cerebral apoplexy first aid, but also the accuracy of abnormal judgment is improved by means of multi-parameter comprehensive analysis and optimization rule set, and the false alarm and missed alarm are reduced. Meanwhile, the accurate scheduling of medical care resources is realized through the hierarchical alarm mechanism, the monitoring quality and clinical intervention efficiency of the cerebral apoplexy high-risk patients are improved, and technical support is provided for the timely disposal of cerebral apoplexy emergency. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 is a flowchart of a cerebral apoplexy emergency sign real-time hierarchical alarm method based on edge computing provided by an embodiment of the present application.

[0017] Figure 2 is a schematic diagram of a cerebral apoplexy emergency sign real-time hierarchical alarm system based on edge computing provided by an embodiment of the present application. DETAILED DESCRIPTION

[0018] Exemplary embodiments of the present disclosure will be described in greater detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be accurately conveyed to those skilled in the art.

[0019] As shown in Figure 1 An embodiment of the present application proposes a cerebral apoplexy emergency sign real-time hierarchical alarm method based on edge computing, which comprises the following steps: Step 1, real-time collection of multi-modal physiological sign data through the intelligent monitoring bracelet worn by the patient, wherein the multi-modal physiological sign data includes heart rate, blood pressure and cerebral oxygen saturation; Step 2, transmission of the multi-modal physiological sign data to the edge computing gateway deployed in the ward; Step 3, the edge computing gateway constructs a multi-dimensional convex hull defining the normal sign space range based on the normal data in the historical multi-modal physiological sign data, calculates the distance from the real-time multi-modal physiological sign data point to the multi-dimensional convex hull, and initializes a population containing multiple rule sets; the rule sets are evolved iteratively by selection, crossover and mutation operations based on the fitness of each rule set in the historical data, and the optimized rule sets are obtained, and the multi-modal physiological sign data is compared and logically judged in real time by using the optimized rule sets to obtain the judgment result; Step 4, based on the judgment result, if multiple physiological parameters exceed the corresponding first threshold and meet the combined logic relationship, it is determined that the first abnormal situation is met and a high-level alarm is generated; if only a single physiological parameter exceeds the corresponding second threshold, it is determined that the second abnormal situation is met and a low-level alarm is generated; Step 5, based on the high-level alarm, an alarm message containing the patient location identifier and real-time vital sign details is pushed to the designated medical terminal, and a synchronous trigger is performed to cause the on-site alarm device in the ward to emit a sound and light alarm signal; based on the low-level alarm, notification information prompting re-measurement is sent to the designated nursing terminal.

[0020] In the embodiment of the application, multi-modal data such as heart rate, blood pressure, and cerebral oxygen saturation are collected in real time by the smart bracelet and transmitted to the ward edge computing gateway, the normal vital sign range is defined by means of multi-dimensional convex hull, and the local real-time analysis is realized by combining the rule set optimization, then the hierarchical alarm is generated according to the abnormal situation and accurately pushed, the response speed is improved through the local processing of edge computing, the timeliness requirement of cerebral apoplexy first aid is met, the accuracy of abnormal identification is improved and the rational scheduling of medical resources is realized through the multi-parameter comprehensive judgment and hierarchical alarm logic, and the timeliness and accuracy of vital sign monitoring of cerebral apoplexy high-risk patients are enhanced.

[0021] In a preferred embodiment of the application, the above step 1 can include: Step 1.1, collecting the original photoplethysmogram signal through the photoelectric sensor integrated in the bracelet, collecting the radial artery pulsation signal through the pressure sensor, and collecting the original brain blood oxygen parameter signal through the near-infrared spectrum sensor, specifically including: for heart rate monitoring, a high-precision photoelectric sensor is embedded in the wrist inner side of the bracelet and adheres to the skin, green light with a wavelength of 520-570nm or red light with a wavelength of 600-700nm is used as the light source, the original photoplethysmogram signal PPG signal is collected in real time by continuously emitting a light beam and receiving the light signal reflected by the skin tissue and blood, the PPG signal can reflect the periodic change of the blood volume in the blood vessel with the heart beat, and the sampling frequency is set to 100-200Hz to ensure the signal details; for blood pressure monitoring, a micro pressure sensor is built-in the bracelet corresponding to the radial artery position, a pre-set flexible film structure slightly adheres to the skin, senses the pulsation pressure change of the radial artery vessel wall with the heart contraction and diastole, collects the original radial artery pulsation signal, and the sampling frequency is synchronized with the PPG signal to ensure the time correlation with the heart rate signal; For cerebral oxygen saturation monitoring, the contact end of the wristband front end extending to the forehead or temple integrates a near-infrared spectroscopy sensor, which emits two specific wavelengths of near-infrared light, usually 760 nm and 850 nm, and receives reflected light signals after penetrating the superficial tissues of the skull, collecting raw signals of cerebral blood oxygen parameters. The signal can reflect the concentration ratio of oxygenated hemoglobin and deoxygenated hemoglobin in brain tissue, and the sampling frequency is 50-100Hz to balance accuracy and power consumption.

[0022] Step 1.2, preprocessing the collected raw signals, including denoising and filtering processing for photoplethysmography wave signal, baseline correction processing for radial artery pulsation signal, and motion artifact elimination processing for near-infrared spectroscopy signal, specifically including: for the photoplethysmography wave signal, PPG signal, which is susceptible to power frequency interference, motion noise and electronic noise, the preprocessing adopts two-step filtering, first through a capacitance coupling circuit or a digital notch filter to suppress power frequency interference, and then a low-pass filter below 20Hz, such as a Butterworth low-pass filter, to filter out high-frequency motion noise, while smoothing the signal through a sliding window mean method to eliminate baseline drift, and finally obtaining a smooth PPG waveform; for the radial artery pulsation signal, its main interference is baseline drift caused by changes in sensor and skin contact pressure, respiratory motion and low-frequency noise, preprocessing adopts baseline correction algorithm, by extracting the low-frequency trend component in the signal, such as using the mean value of a 10-second sliding window as the baseline estimate, subtracting the trend component from the original signal, making the baseline of the pulsation signal regress to near zero value, highlighting the pressure fluctuation characteristics of arterial contraction or diastole; for the near-infrared spectroscopy signal, the main interference is motion artifact, which is caused by the relative displacement of the sensor and the skin due to patient head movement, causing light intensity mutation, preprocessing adopts motion artifact elimination strategy, first through signal mutation detection algorithm, such as threshold-based variance analysis, to identify motion artifact segments, and then use the waveform similarity of adjacent normal signals to repair the segments, while combining adaptive filtering algorithm to separate and remove residual motion interference components, and retain the effective signals reflecting blood oxygen changes.

[0023] Step 1.3, based on the pre-processed signal, the pulse wave feature extraction algorithm is used to calculate the heart rate value, the pulse wave transmission time algorithm is used to calculate the blood pressure value, and the dual-wavelength near-infrared spectral analysis method is used to calculate the cerebral oxygen saturation value, which specifically includes: for the pre-processed PPG signal, the pulse wave feature extraction algorithm is used, first, the peak value is detected, such as the joint determination based on the first derivative and the second derivative to identify the peak position of each pulse wave, and the time interval between adjacent two peaks, i.e. the pulse period, is recorded; then, the average pulse period is obtained by averaging 10-20 consecutive pulse periods; the number of heartbeats per minute is calculated, and abnormal periods, such as periods deviating from the average value by more than 20%, are removed to avoid errors, and finally the stable heart rate value is output; the pulse wave transmission time algorithm is used, first, the peak time is extracted from the pre-processed PPG signal, and the corresponding peak time is extracted from the radial artery pulse signal, and the time difference between the two is calculated as PTT, i.e. the time for the pulse wave to transmit from the heart to the radial artery; based on the clinically verified negative correlation between PTT and blood pressure, combined with the patient's basic blood pressure data at admission, such as systolic pressure and diastolic pressure, the mapping relationship between PTT and blood pressure is established, and the real-time calculated PTT is converted into the corresponding systolic and diastolic pressure values; the dual-wavelength near-infrared spectral analysis method is used, based on the pre-processed near-infrared light signals of two wavelengths 760nm and 850nm, the Lambert-Beer law is used, and the attenuation coefficients of the light signals absorbed by the brain tissue under the two wavelengths are calculated; due to the difference in absorption coefficients of oxygenated hemoglobin and deoxygenated hemoglobin at these two wavelengths, the proportion of the attenuation coefficients of the two is compared and analyzed, and the percentage of oxygenated hemoglobin in total hemoglobin in the brain tissue, i.e. the cerebral oxygen saturation value, is calculated, and the result is rounded to one decimal place to ensure accuracy.

[0024] Step 1.4, the calculated heart rate, blood pressure and cerebral oxygen saturation values are time-aligned and packaged, and the standardized format of multi-modal physiological sign data is obtained, which specifically includes: due to the slight differences in the calculation frequencies of heart rate, blood pressure and cerebral oxygen saturation, the time stamps of the high-precision clock built-in the bracelet are used as the reference to match the values of the three types of parameters in time. For the data points that do not coincide in time, linear interpolation method is used to supplement the intermediate values, such as only heart rate and cerebral oxygen data at a certain time, the blood pressure value at the previous and next time is used to interpolate to obtain the estimated value of blood pressure at that time, and finally a three-dimensional parameter sequence with completely aligned time dimension is formed; according to the pre-set standardized data format, the parameter data after time alignment is integrated into a structured data frame, and through time alignment and standardized packaging, the multi-modal physiological sign data has a unified format and time reference.

[0025] In the embodiment of the present application, by integrating various special sensors to collect raw signals, combining noise reduction filtering, baseline correction, motion artifact elimination and other preprocessing methods to improve signal quality, and then using professional algorithms suitable for each parameter to accurately calculate heart rate, blood pressure and cerebral oxygen saturation values, finally through time alignment and standardized packaging to ensure data consistency, the comprehensiveness and accuracy of multi-modal sign data acquisition are guaranteed, and the accuracy of stroke sign abnormality judgment is supported.

[0026] In a preferred embodiment of the present application, the above step 2 can include: Step 2.1, the intelligent monitoring bracelet establishes a secure communication connection with the edge computing gateway in the ward through the low-power Bluetooth protocol, encrypts and packages the standardized data frame, specifically including: after the bracelet is turned on, it automatically scans the broadcast specific service UUID in the surrounding environment, which is used to identify the edge computing gateway of the stroke sign monitoring service. After discovering the target gateway, the bracelet initiates a connection request, and communicates and negotiates through the low-power Bluetooth protocol, including handshaking confirmation, transmission rate adaptation and connection interval setting; after the connection is established, both parties complete identity verification through the preset security authentication mechanism: the bracelet sends an authentication request containing the device unique identifier to the gateway, the gateway verifies whether the identifier is in the pre-authorized device list, and returns an encrypted session key after verification; the standardized data frame generated by the bracelet, including heart rate, blood pressure, cerebral oxygen saturation and basic metadata, is encrypted and packaged, the session key is used to symmetrically encrypt the payload part of the data frame, and the encryption algorithm identifier, such as AES-128-CBC, the initial vector and the data packet length information, is added to the data packet header, and the message authentication code, HMAC-SHA256, is added to the tail, and finally the encrypted data frame conforming to the BLE protocol specification is formed.

[0027] Step 2.2, after receiving the encrypted data frame, the edge computing gateway performs decryption verification and parses the data packet format, checks the data integrity and validity, which specifically includes: the gateway receives the encrypted data frame sent by the bracelet through BLE continuous listening and receiving, first extracts the encryption algorithm identifier and initial vector in the data packet header, and calls the corresponding decryption algorithm. For example, AES-128-CBC, the payload part is decrypted using the previously negotiated session key to obtain the original data frame; after decryption, the gateway extracts the message authentication code in the tail of the data packet, and compares it with the locally recalculated HMAC-SHA256 value based on the original data frame. If they are consistent, it is confirmed that the data has not been tampered with; at the same time, check whether the data packet length is consistent with the length declared in the header, and eliminate invalid data with abnormal length; parse the data frame that passes the integrity check according to the preset standardized protocol, extract the heart rate, blood pressure, cerebral oxygen saturation physiological parameter values, and bracelet device identifier, original collection timestamp and other metadata; during the parsing process, the parameter values are range- checked, such as heart rate range limited to 30-200 beats / min, and abnormal values outside the reasonable range are eliminated to ensure data validity.

[0028] Step 2.3, add timestamps and patient identity identifiers to the physiological sign data that passes the verification, and convert it to a standardized data format suitable for edge computing processing, which specifically includes: based on the NTP synchronous clock built-in the gateway, calibrate with the hospital intranet time server, error ≤1ms, record the precise time of data reception completion, replace the local timestamp of the bracelet to eliminate the clock deviation between devices; the gateway queries the pre-stored device and patient mapping table according to the unique identifier of the bracelet device, which is pre-configured by the nurse through the management terminal and contains the binding relationship between device ID and patient ID, associates the corresponding patient unique identifier to the current data to ensure accurate correspondence between data and patient identity; integrate the scattered parameter values, timestamps, patient identifiers and other information into structured data that can be directly parsed by the edge computing engine, such as using ProtocolBuffers format to reduce data volume, wherein the physiological parameter part is organized in the form of key-value pairs of parameter name, value, unit and confidence, and the metadata part is separately encapsulated.

[0029] Step 2.4, the processed standardized data is temporarily stored in the local cache area of the edge computing gateway, specifically including: the first-level cache is a memory buffer area, adopting a ring queue structure to store high-frequency vital sign data in the last 24 hours in real time, supporting millisecond-level read and write, and meeting the requirement of fast data retrieval in real-time analysis; the second-level cache is a local solid-state storage, and historical data is stored in a directory structure according to patient ID and date, data in the memory exceeding 24 hours is automatically archived to the solid-state storage in the early morning every day, ensuring long-term traceability of data; the vital sign data of each patient is indexed by time, recording the sampling time of different parameters, supporting fast query according to time range and parameter type, and a data expiration policy is set to automatically clean up historical data exceeding 30 days, avoiding buffer overflow; double copies of key data are stored, respectively saved in the memory and the solid-state storage, preventing data loss caused by sudden power failure of the gateway.

[0030] In the embodiment of the application, the low-power Bluetooth protocol is used to establish a secure communication connection and encrypt and encapsulate the data frame, thereby ensuring the security and low-power characteristics of the multi-modal physiological vital sign data transmission; the decryption verification, format analysis and integrity check of the edge computing gateway ensure the reliability and effectiveness of the received data; the time stamp and patient identifier are added and converted into a standardized format suitable for edge computing; temporary storage in the local cache area facilitates historical data calling and fast access, and the overall process not only meets the safe and efficient data transmission requirement in the ward scene, but also lays a foundation for high-quality data support through real-time processing of the edge node, thereby improving the stability and response speed of the stroke sign monitoring.

[0031] In a preferred embodiment of the application, step 3 can include: Step 3.1, extracting historical multi-modal physiological data from local cache area, screening data samples representing normal physiological state according to clinically defined normal physiological parameter range values, specifically including: first, the gateway extracts the multi-modal physiological history data of the current patient in the past 7-30 days from the local secondary cache, including heart rate, blood pressure, and cerebral oxygen saturation records every 5 minutes, while associating the corresponding data collection timestamps and patient basic information such as age and basic medical history, ensuring that the data covers different time periods such as morning, night, and after exercise to reflect the diversity of physiological state, second, according to the clinically recognized normal physiological parameter range, the screening standard is set, the heart rate is limited to 60-100 times / min, the systolic pressure is 90-140 mmHg, the diastolic pressure is 60-90 mmHg, and the cerebral oxygen saturation is 90%-100% for non-hypoxemia patients; for patients with underlying diseases, such as hypertension history, the individual normal range is adjusted according to clinical orders, such as the upper limit of systolic pressure can be widened to 150 mmHg; finally, the extracted historical data is screened point by point, and abnormal data points outside the normal range are removed, while excluding records with clear interference factors at the time of data collection, such as intense activity, emotional excitement, and temporary device failure; the screened data is checked for time distribution uniformity to ensure that there are enough samples in each period, and finally a multi-modal data sample set representing the normal physiological state of the patient is formed.

[0032] Step 3.2, based on the data samples representing the normal physiological state, the convex hull generation algorithm is used to determine the boundary points of the normal data in the high-dimensional feature space, and by connecting these boundary points, a minimum convex polygon is formed to enclose all normal data, defining the spatial range of normal signs, specifically including: first, the screened normal data samples are mapped to a high-dimensional feature space, with heart rate, systolic pressure, and cerebral oxygen saturation forming a three-dimensional feature space, each normal data point corresponding to a coordinate point in space, such as heart rate value, systolic pressure value, and cerebral oxygen saturation value; second, a convex hull generation algorithm suitable for high-dimensional data is used, such as the QuickHull algorithm, to determine the boundary points, the algorithm first randomly selects a data point as the initial vertex, and by calculating the distance from all other points to the vertex, the farthest point is selected as the first boundary point; based on this, other boundary points that can form a convex hull surface are iteratively found, i.e. those points that can make all normal data points on one side; finally, all boundary points are connected to form a closed three-dimensional convex polyhedron, and the connection relationship between the boundary points is determined through geometric calculation to ensure that the convex hull can completely enclose all normal data points and there is no redundant boundary, i.e. removing any boundary point will cause some normal data points to be exposed outside the convex hull; the interior and surface of the convex hull are defined as the normal sign range of the patient, and any data point outside the convex hull is initially determined to be deviated from the normal state.

[0033] Step 3.3, calculate the shortest Euclidean distance from the real-time collected multi-modal physiological data point to each surface of the multi-dimensional convex hull as a quantitative indicator of the degree of deviation from the normal state, specifically including: first, obtain the real-time collected and pre-processed multi-modal physiological data, convert it to a coordinate point in a three-dimensional feature space, such as real-time heart rate, real-time systolic blood pressure, and real-time cerebral oxygen saturation, to ensure that the coordinate dimensions match the convex hull space completely; second, determine the spatial relationship between the real-time data point and the multi-dimensional convex hull, if the data point is inside or on the surface of the convex hull, it is determined to be within the normal range, and the distance value is recorded as 0; if it is outside the convex hull, the shortest distance from each surface of the convex hull needs to be calculated; in the specific calculation, each triangular face of the convex hull is traversed, and the perpendicular distance from the real-time data point to each triangular face, i.e. the shortest Euclidean distance, is calculated by spatial geometric method, and the smallest distance value is selected as the quantitative indicator of the deviation of the data point from the normal physiological space; the greater the distance value, the more significant the degree of deviation of the real-time physiological data from the patient's normal state.

[0034] Step 3.4, based on the distance index, generate an initial rule set with different parameter threshold values and logical combinations, where each rule contains specific physiological parameter threshold ranges, logical operation relationships and corresponding weight coefficients, specifically including: first, according to the distance index distribution characteristics, such as normal data point distance 0, slightly deviated data point distance 0-5 units, significantly deviated 5-10 units, divide multiple distance intervals, each interval corresponds to different abnormal severity; second, design rules for each distance interval containing specific parameter thresholds, logical combinations and weight coefficients, parameter thresholds, set specific thresholds for heart rate, blood pressure and cerebral oxygen saturation based on clinical experience, such as distance 5-10 unit interval corresponding to heart rate > 100 times / min, systolic blood pressure > 150 mmHg, cerebral oxygen saturation < 85%; logical operation relationship, according to the clinical significance of multi-parameter cooperative abnormality, set "and" "or" logic, such as heart rate > 100 and systolic blood pressure > 150 as combined abnormality, cerebral oxygen saturation < 85 as single abnormality; weight coefficient, according to the clinical relevance of parameters to stroke onset, such as cerebral oxygen saturation abnormality weight set to 0.4, blood pressure abnormality set to 0.3, heart rate abnormality set to 0.3; finally, generate multiple differentiated initial rule sets, adjust the threshold range, logical relationship, weight coefficient, form 50-100 different rule sets, ensure that the rules cover a variety of abnormal situations, provide rich initial samples for subsequent optimization.

[0035] Step 3.5 encodes the generated rule sets into individuals for the genetic algorithm population, completing the population initialization. This includes: First, determining the encoding method: using real-number encoding to convert each rule set into a chromosome that the genetic algorithm can process. The chromosome length is consistent with the number of parameters contained in the rule set. For example, if a rule set contains 3 parameters (threshold heart rate, systolic blood pressure, cerebral oxygen saturation), 2 logical relations, and 3 weight coefficients, then the chromosome consists of 3+2+3=8 real-number genes. Each gene corresponds to a specific parameter in the rule set, such as the first gene corresponding to the heart rate threshold, with a value range of 100-120. Second, setting the population size: based on the computing power of the edge computing gateway, the population size is controlled at 50-80 individuals to ensure population diversity. Finally, completing the population initialization: randomly selecting individuals from the generated rule sets to ensure that the gene values ​​of each individual are within a preset reasonable range, such as the threshold not exceeding the clinically possible range, and the sum of the weight coefficients is 1. Simultaneously, randomly perturbing a small number of genes, such as fine-tuning the threshold by 10%, increases population diversity, ultimately forming the initial population for the genetic algorithm.

[0036] In this embodiment of the invention, historical data is extracted from the local cache and normal samples are screened according to clinical standards; a high-dimensional normal vital sign space boundary is constructed using a convex hull generation algorithm, which accurately defines the range of normal physiological state and avoids the limitations of single threshold judgment; the degree of deviation of vital signs from normal state is quantified by calculating the Euclidean distance from real-time data points to the convex hull, providing an intuitive basis for abnormal judgment; an initial rule set containing diverse parameter thresholds, logical combinations, and weight coefficients is generated based on the distance index and encoded into a genetic algorithm population, preserving the collaborative analysis capability of multimodal data. The overall process improves the accuracy of abnormal judgment of stroke vital signs by combining clinical standards and algorithms.

[0037] In a preferred embodiment of the present invention, step 3 above may include: Step 3.6: Based on historical multimodal physiological sign data, calculate the fitness of each rule set in the population. Fitness is comprehensively evaluated by the early warning accuracy and false alarm rate of the rule set on historical data. Specifically, this includes: First, extracting historical multimodal physiological sign data containing known abnormal and normal states from the local cache of the edge computing gateway as validation samples. Selecting the signs records of stroke patients in the ward within the past 3 months, where known abnormal samples are signs data within 30 minutes before the onset of a clinically diagnosed stroke, such as a combination of sudden increase in blood pressure, rapid heart rate, and decreased cerebral oxygen, a total of 50-100 sets are collected; normal state samples are signs data of patients when their condition is stable during the same period, conforming to the clinical normal range, a total of 200-300 sets are collected to ensure that the samples cover different time periods and individual patient differences; Second, each rule set is used to... The rule set is applied to validate samples. For known anomalous samples, it is determined whether the rule set can trigger an alert, i.e., whether the rule conditions are met. For normal samples, it is determined whether the rule set falsely triggers an alert. Then, two key indicators are calculated: the alert accuracy rate is the number of anomalous samples correctly identified by the rule set divided by the total number of anomalous samples, reflecting the rule's ability to capture real anomalies; the false alarm rate is the number of normal samples falsely alerted by the rule set divided by the total number of normal samples, reflecting the rule's anti-interference ability. Finally, the fitness is determined by combining the two indicators using a weighted summation method, where the alert accuracy rate is weighted at 0.7 and the false alarm rate is weighted at 0.3. The fitness is calculated using the logic: fitness = accuracy × 0.7 + (1 - false alarm rate) × 0.3. A higher value indicates better rule set performance. A specific fitness value is assigned to each rule set.

[0038] Step 3.7: Based on the fitness assessment results, a roulette wheel selection method is used to select high-fitness rule set individuals from the current population, retaining rule sets with high fitness. A single-point crossover operation is performed on the selected rule set individuals, exchanging some rule conditions between different rule sets to generate new rule set combinations. Specifically, this includes: roulette wheel selection; assigning selection probabilities to each rule set in the population based on the calculated fitness values; summing the fitness of all rule sets to obtain the total fitness; and assigning the selection probability of each rule set to its own fitness divided by the total fitness. Rule sets with higher fitness have a greater probability of being selected; for example, a rule set with a fitness percentage of 30% has a 30% probability of being selected. Individuals are randomly selected from the current population, and a total number of rule sets equal to the population size are selected. Repeated selection is allowed to ensure that high-fitness individuals are retained multiple times, forming the parent population for crossover.

[0039] The single-point crossover operation pairs the rule sets in the parent group, generating two new rule sets for each pair. Specifically, the crossover point is determined by treating the parameter sequence of the rule set as a linear sequence and randomly selecting a position in the sequence, such as between the 3rd and 4th parameters, as the crossover point. Then, the rule conditions after the crossover point of the two rule sets are swapped. For example, the part after the crossover point of parent A is heart rate > 110, logical 'AND', and weight 0.3, while the corresponding part of parent B is heart rate > 105, logical 'OR', and weight 0.2. After the crossover, child A inherits the part before the crossover point of parent A plus the part after the crossover point of parent B, while child B inherits the opposite. By crossover and fusing the characteristics of high-quality rules, a new rule set with both diversity and effectiveness is generated.

[0040] Step 3.8 involves performing mutation operations on the newly generated rule set. This involves randomly adjusting physiological parameter thresholds, logical relationships, or weighting coefficients within the rules with a preset probability. Specifically, this includes: setting a preset mutation probability of 1%-5% based on the edge gateway's computing power, typically 3%, meaning each newly generated rule set has a 3% probability of triggering mutation; if mutation is triggered, randomly selecting 1-2 parameters from the rule set for adjustment. Physiological parameter thresholds are fine-tuned within clinically reasonable ranges, such as adjusting the systolic blood pressure threshold from 180 mmHg to 178-182 mmHg, and the heart rate threshold from 110 beats / minute to 108-112 beats / minute, ensuring the range of change is controlled. The deviation should not exceed 5% of the original threshold to avoid the rule deviating from clinical practice; for logical relationships, if the rule contains logical operators such as "AND" or "OR", one of them should be randomly switched, such as changing heart rate > 110 and blood pressure > 180 to heart rate > 110 or blood pressure > 180; for weighting coefficients, the weights of each parameter should be finely adjusted by ±0.05, such as adjusting the weight of cerebral oxygen saturation from 0.4 to 0.35-0.45. After adjustment, it needs to be renormalized to ensure that the sum of the weights of all parameters is 1; if the adjusted threshold exceeds the clinically accepted range or is logically contradictory, the variation should be abandoned and parameters should be randomly selected again for adjustment to ensure that the varied rule still has clinical reference value.

[0041] Step 3.9: Repeat steps 3.6 to 3.8 until the preset number of iterations or fitness convergence is reached, resulting in an optimized rule set. Parallel rule matching is then performed on the real-time collected multimodal physiological characteristic data using this optimized rule set. Abnormal characteristic judgment results are generated based on the fulfillment of rule conditions. Specifically, this includes: setting a maximum number of iterations, typically 50-100 times based on the edge gateway's computing power (usually 60 iterations are sufficient to balance effect and efficiency); or determining fitness convergence and stopping iteration when the average fitness change of the population is less than 0.01 over 5 consecutive iterations; selecting the top 3-5 rule sets with the highest fitness from the final iteration population to form a rule pool. These rule sets are then validated using historical data. The system exhibits an early warning accuracy rate >90% and a false alarm rate <5%, covering different types of anomaly combinations. When new multimodal physiological data is transmitted to the edge gateway, all rule sets in the rule pool simultaneously perform parallel judgments on the data. Each rule set independently checks whether the real-time data meets its own parameter thresholds and logical relationships. Based on the judgment results of all rules, if at least two rule sets determine anomalies, the comprehensive anomaly confidence score is calculated by combining the weight coefficients in the rules. For example, if a rule with a high weight determines an anomaly, the confidence score is increased by 0.6; if a rule with a low weight determines an anomaly, the confidence score is increased by 0.4. When the confidence score is ≥0.8, it is determined to be a significant anomaly; when the confidence score is between 0.5 and 0.8, it is determined to be a suspected anomaly; and when it is below 0.5, it is determined to be normal.

[0042] In this embodiment of the invention, the fitness of the rule set is calculated based on historical data, providing an evaluation standard that fits clinical practice for rule optimization; a roulette wheel selection method is used to retain the rule set with high fitness and generate new combinations through single-point crossover, continuing the effective characteristics of high-quality rules, and enriching the diversity of rules through gene recombination; mutation operations are performed on the new rule set with a preset probability to further introduce new features to avoid local finality and improve the generalization ability of the rule set; through multiple rounds of iteration until convergence, the final optimized rule set can more accurately adapt to the abnormal signs patterns of stroke patients, and combined with real-time parallel rule matching, the accuracy and efficiency of abnormal sign judgment are improved.

[0043] In a preferred embodiment of the present invention, step 4 above may include: Step 4.1: Receive the abnormal vital signs judgment results from the edge computing gateway. The judgment results include the abnormal status identifiers of each physiological parameter and the corresponding confidence scores. Specifically, the alarm system's processing module establishes a long connection with the edge computing gateway through the hospital's intranet TCP / IP protocol, and listens for and receives the judgment result data packets pushed by the gateway in real time. The data packets use a standardized JSON format and include a unique patient identifier, such as an encrypted hospital number, a data generation timestamp, and details of the vital signs assessment. For the four parameters—heart rate, systolic blood pressure, diastolic blood pressure, and cerebral oxygen saturation—blood pressure is further subdivided into systolic and diastolic pressure for accurate assessment. Abnormal status indicators are extracted for each parameter, such as normal, mildly abnormal, and significantly abnormal, along with corresponding confidence scores from 0 to 100. Higher scores indicate greater confidence in the assessment of the abnormality. For example, a patient's data might show: Heart rate: significantly abnormal (confidence 92), Systolic blood pressure: significantly abnormal (confidence 88), Cerebral oxygen saturation: normal (confidence 95). The received data undergoes integrity verification to confirm that all parameter status indicators and confidence scores are included and in the correct format. If any data is missing or formatted incorrectly, a retransmission request is immediately sent to the gateway to ensure the integrity of the assessment results.

[0044] Step 4.2: Analyze the judgment results. When at least two physiological parameters among heart rate, blood pressure, and cerebral oxygen saturation simultaneously exceed their respective first thresholds, and the duration of the abnormality reaches a preset duration, it is determined to meet the first abnormal condition. Specifically, this includes: firstly, extracting the first thresholds corresponding to each parameter. The first threshold for heart rate is set to >110 beats / minute, the first threshold for systolic blood pressure is set to >180 mmHg, and the first threshold for cerebral oxygen saturation is set to <65%. These thresholds are preset by the neurology expert group according to the stroke emergency guidelines and can be fine-tuned through the management terminal; checking whether at least two of the measured values ​​of heart rate, systolic blood pressure, and cerebral oxygen saturation exceed the corresponding first thresholds, and whether their abnormality is... The status is identified as significantly abnormal with a confidence score ≥80 to ensure the reliability of the abnormality judgment. For example, a heart rate of 120 beats / minute (exceeding the threshold) and systolic blood pressure of 190 mmHg (exceeding the threshold) or systolic blood pressure of 185 mmHg (exceeding the threshold) and cerebral oxygen saturation of 60% (exceeding the threshold) both meet the two exceeding conditions. The timer starts from the moment the parameter first exceeds the first threshold. The abnormal status is continuously monitored by receiving real-time data to see if it is maintained continuously without returning to the normal range. When the duration reaches the preset 30 seconds, which can be adjusted according to clinical needs, the minimum duration should not be less than 10 seconds to avoid misjudgment due to instantaneous fluctuations. Then, it is finally judged to meet the first abnormal condition, indicating a high risk of sudden stroke.

[0045] Step 4.3: When only a single physiological parameter is detected to exceed its corresponding second threshold, and other parameters are within the normal range, it is determined to meet the second abnormal condition. Specifically, this includes: extracting the second threshold corresponding to each parameter, the clinically defined mild abnormal threshold, setting the second threshold for heart rate to 100-110 beats / minute (not reaching the first threshold but higher than the upper limit of normal), setting the second threshold for systolic blood pressure to 150-180 mmHg, and setting the second threshold for cerebral oxygen saturation to 85%-90% (slightly lower than the lower limit of normal); traversing all physiological parameters, only finding one parameter whose measured value exceeds the corresponding second threshold but does not reach the first threshold, its abnormal state is marked as mild abnormal, with a confidence score between 60-80 points. For example, if a systolic blood pressure of 155 mmHg exceeds the second threshold but does not reach the first threshold, and the heart rate and cerebral oxygen saturation are within the normal range, and all other parameters are strictly verified to be within the clinically normal range except for this single abnormal parameter, such as a heart rate of 60-100 beats / minute and a systolic blood pressure of 90-140 mmHg, and the abnormal status is marked as normal with a confidence score ≥90 points, ensuring that no other potential abnormalities are missed, it is ultimately determined to meet the second abnormal condition.

[0046] Step 4.4: Based on the first abnormal situation determination result, generate an advanced alarm instruction containing detailed patient information, abnormal parameter data, and an urgency level indicator. Specifically, this includes: retrieving basic data associated with the patient's unique identifier from the Hospital Information System (HIS), including name, gender, age, inpatient bed number, admission diagnosis (e.g., grade 3 hypertension, atrial fibrillation, history of stroke), ensuring that medical staff can quickly identify the patient's background; listing in detail the parameter name, real-time value, and magnitude of exceeding the first threshold (e.g., "systolic blood pressure 195 mmHg, exceeding the first threshold by 15 mmHg," the start time and duration of the abnormality, along with parameter change curve data within the last 5 minutes to help determine the speed of abnormal development; clearly marking the three alarm levels in the instruction and setting a priority field (value 1 represents the highest priority), while also adding clinical prompts such as immediate bedside assessment if the patient exhibits signs of a sudden stroke; and using the hospital's unified alarm protocol format, including instruction type, generation time, patient information area, abnormal data area, and urgency level indicator, ensuring that the on-duty doctor's mobile terminal can correctly parse and display the alarm.

[0047] Step 4.5: Based on the second abnormal situation determination result, generate a low-level alarm instruction containing the abnormal parameter type and current value. Specifically, this includes: clearly labeling the parameter name exceeding the second threshold, the real-time measured value, the comparison with the second threshold, and the reference value of the normal range for the parameter, allowing nurses to quickly understand the abnormality and set a reasonable retest time according to the parameter type. For example, for abnormal blood pressure, it is recommended to retest within 30 minutes, and for abnormal heart rate, it is recommended to retest within 15 minutes. Precautions for retesting are also provided. The instruction clearly identifies the first-level alarm, with the priority field set to 3, the lowest priority, to avoid conflict with higher-level alarms. A lightweight data structure is used, containing only the patient's bed number, abnormal parameter information, retest suggestion, and alarm level. This information is pushed to the responsible nurse's terminal through the message queue of the nursing workstation system.

[0048] In this embodiment of the invention, by receiving the judgment result containing parameter abnormality identifiers and confidence scores, a precise basis for graded alarms is provided; by clarifying the judgment criteria for the first and second abnormal situations, a scientific distinction is made between the degree of stroke risk; based on different situations, high and low level alarm instructions containing corresponding information are generated to ensure that high-level alarms can convey patient details and urgency to quickly trigger intervention, and that low-level alarms can concisely indicate abnormal parameters to reasonably arrange retesting, thereby improving the pertinence and clinical applicability of the alarms.

[0049] In a preferred embodiment of the present invention, step 5 above may include: Step 5.1: Based on advanced alarm commands, generate an alarm message containing the patient's bed number, real-time values ​​of abnormal parameters, and historical trend data; based on low-level alarm commands, generate a notification message containing the type of abnormal parameter and a suggested retest time. Specifically, based on the advanced alarm commands, first, clarify the patient's specific bed information, such as bed 12 in ward 3 of the neurology department; second, list all abnormal parameters exceeding the first threshold, including parameter names, such as heart rate, systolic blood pressure, and cerebral oxygen saturation, real-time measurement values ​​and units, and indicate the extent to which each parameter exceeds the corresponding threshold, such as exceeding the threshold by 25 beats / minute; simultaneously, extract the parameter change data of the patient in the past 30 minutes and generate a trend description, such as heart rate within 10 minutes. The heart rate suddenly spiked from 90 beats per minute to 135 beats per minute, and the message was embedded as a line graph, plotted using historical data cached by the edge gateway. An emergency marker was added at the beginning of the message. Based on low-level alarm commands, the specific type of abnormal parameter was clearly indicated, such as systolic blood pressure and heart rate, along with the current measurement value and unit, e.g., 155 mmHg, 105 beats per minute, and compared to the normal range of the parameter, e.g., 90-140 mmHg. Recommended retesting times were set according to the parameter type, e.g., 30 minutes for abnormal blood pressure and 15 minutes for abnormal heart rate, along with retesting precautions, such as instructing the patient to rest quietly for 5 minutes before retesting. A notification was added at the beginning of the message.

[0050] Step 5.2: Based on the patient's identity information, query the corresponding responsible medical staff information to determine the target push terminal for the advanced alarm message and the device identifier of the ward's on-site alarm device. Specifically, this includes: the system automatically links to the patient information database in the hospital's HIS system based on the patient's unique identifier contained in the advanced alarm command, such as the encrypted hospital number, and extracts the information of the responsible medical team corresponding to the patient, including the ward to which the bed belongs, the attending physician, the responsible nurse, the ward head nurse, and the on-duty physician; the dedicated mobile clinical terminal PDA assigned to the attending physician and the on-duty physician; the terminal device number is bound to the medical staff's employee number, such as physician employee number 0123 corresponding to PDA number PD-2023056; it also includes the hospital's dedicated alarm APP installed on the doctor's personal mobile phone to ensure that the doctor can receive the message even when leaving the ward; and querying the ward equipment management system based on the patient's bed number to obtain the on-site alarm device information corresponding to the bed, including the audible and visual alarm at the ward entrance, the bed indicator light at the ward nurse station, and the centralized alarm host, to ensure that the alarm signal can accurately trigger the device at the corresponding location.

[0051] Step 5.3: Advanced alarm messages are pushed in real-time to designated medical staff terminals via the hospital's intranet communication system, and simultaneously, audible and visual alarm trigger commands are sent to the ward's on-site alarm devices. Specifically, this includes: pushing the message to the target doctor's PDA in real-time via the WebSocket protocol, with a push timeout of 1 second; if the timeout occurs, the system will immediately retry 3 times; simultaneously, a push command is sent to the doctor's mobile app via the hospital's SMS gateway, including a message summary, such as "Ward 3, Bed 12, significant abnormal heart rate and blood pressure," and a redirect link; upon receiving the message, the doctor's PDA immediately triggers the highest-level alert, with a pop-up window highlighting the patient's bed number, abnormal parameters, and emergency indicator. The mobile app simultaneously sends a push notification, the screen lights up and plays a preset prompt tone until the doctor clicks the "viewed" button; the system sends a trigger command to the audible and visual alarm at the ward entrance via TCP protocol, the alarm immediately activates, the red LED flashes at a frequency of 2 times / second, and emits an intermittent beeping sound; the bed indicator light at the nurse station simultaneously lights up red, the screen of the centralized alarm host displays an advanced alarm for bed 12 in ward 3 and emits a prompt tone, reminding the nurse to quickly locate the patient's position; if no confirmation of arrival is received from any doctor within 5 minutes, the system automatically expands the push scope and repeatedly pushes the alarm message to the terminals of all on-duty doctors in the ward.

[0052] Step 5.4 involves distributing low-level alarm messages to the terminal devices of the patient's responsible nurses through the nursing workstation management system. Specifically, the nursing workstation management system extracts the patient's bed number and responsible nurse information from the low-level alarm command and distributes the notification message to the nurse's mobile nursing PDA and workstation computer via the system's internal message queue. Upon receiving the message, the mobile nursing PDA triggers a medium-intensity alert, highlighting the notification in yellow in the message list. The message displays that the systolic blood pressure of patient number 12 is 155 mmHg, and a retest is recommended in 30 minutes. A notification window with the same content pops up in the lower right corner of the nursing system interface on the nurse's workstation computer. The window remains open for 30 seconds before automatically shrinking to the taskbar, where it can be viewed again. The message includes two shortcut buttons: "Read" and "Retest Scheduled." The system automatically records the confirmation time after the nurse clicks the button. If the responsible nurse is busy with other tasks and does not view the message promptly, the system forwards it to the terminal devices of the assistant nurses in the same group after 10 minutes, indicating that the original responsible nurse has not confirmed it, ensuring that the message is not missed. Low-level alarm messages have a lower priority than high-level alarms in the nursing system, but higher than routine work reminders, to avoid being overwritten by other information.

[0053] Step 5.5 records the sending status and reception confirmation of all alarm messages and updates them to the patient monitoring log database. This includes recording the generation time of high-level and low-level alarm messages, the target terminal list, the first sending time, the sending success or failure status, and the actual reception time of each terminal. The system captures real-time operations by medical staff on the terminals. When a doctor clicks the "viewed" or "arrived on-site" button for a high-level alarm message, or a nurse clicks the "read" or "retest arranged" button for a low-level message, the system immediately records the operation time, operator identity, and operation type, and simultaneously generates an operation log. All records are integrated into structured data and written to the alarm record table in the patient monitoring log database. This table includes fields such as patient ID, alarm level, alarm time, abnormal parameters, push terminal list, response status of each terminal, and final confirmation person and time. The database supports real-time querying, allowing medical staff to trace the alarm history of any patient through the system. Simultaneously, daily alarm response statistics reports are automatically generated, including indicators such as average response time and the number of alarms not confirmed in a timely manner, providing data support for optimizing the treatment process. All records are retained for more than 3 years to meet medical quality management and traceability requirements.

[0054] In this embodiment of the invention, messages containing corresponding key information are generated based on alarm commands of different levels. This ensures that high-level alarms provide bed availability, real-time parameters, and trend data to support rapid decision-making, while low-level alarms clearly define the abnormality type and retest time to optimize nursing procedures. Accurate querying of responsible medical personnel information determines the push terminal and alarm device, ensuring precise alarm delivery. The hospital intranet and nursing system enable efficient message push and on-site alarm linkage, improving response speed. Simultaneously, sending and receiving status are recorded and updated to the monitoring log, achieving full traceability. Overall, this meets the needs for rapid intervention in emergency situations, optimizes routine nursing work arrangements, and improves the standardization of vital sign monitoring for stroke patients.

[0055] like Figure 2 As shown, embodiments of the present invention also provide a real-time hierarchical alarm system for sudden stroke signs based on edge computing, comprising: The acquisition module is used to collect multimodal physiological data in real time through the smart monitoring wristband worn by the patient. The multimodal physiological data includes heart rate, blood pressure and cerebral oxygen saturation; and transmit the multimodal physiological data to the edge computing gateway deployed in the ward. The computing module is used by the edge computing gateway to construct a multidimensional convex hull that defines the spatial range of normal physiological signs based on normal data in historical multimodal physiological sign data. It calculates the distance from real-time multimodal physiological sign data points to the multidimensional convex hull, thereby initializing and generating a population containing multiple rule sets. Based on the fitness of each rule set in historical data, it performs selection, crossover, and mutation operations to iteratively evolve the rule sets, obtaining optimized rule sets. The optimized rule sets are then used to perform real-time parallel comparison and logical judgment on multimodal physiological sign data to obtain the judgment results. The alarm module is used to determine whether a first abnormal situation is met and generate a high-level alarm if multiple physiological parameters exceed the corresponding first threshold and satisfy the combinational logic relationship based on the judgment result; if only a single physiological parameter exceeds the corresponding second threshold, it is determined to be a second abnormal situation and generates a low-level alarm. The processing module is used to push alarm messages containing patient location identifiers and real-time vital sign details to designated medical and nursing terminals based on high-level alarms, and simultaneously trigger the on-site alarm devices in the ward to issue audible and visual alarm signals; based on low-level alarms, it sends notification information prompting retesting to designated nursing terminals.

[0056] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An edge computing-based real-time stroke sign burst grading alarm method, characterized in that, The method comprises: Step 1, real-time acquisition of multi-modal physiological sign data through a smart monitoring bracelet worn by the patient, the multi-modal physiological sign data including heart rate, blood pressure and cerebral oxygen saturation; Step 2, transmission of the multi-modal physiological sign data to an edge computing gateway deployed in the ward; Step 3, the edge computing gateway constructs a multi-dimensional convex hull defining the normal sign space range based on normal data in historical multi-modal physiological sign data, calculates the distance from the real-time multi-modal physiological sign data point to the multi-dimensional convex hull, and initializes the generation of a population containing multiple rule sets; the fitness of each rule set in the historical data is selected, crossed and mutated to iteratively evolve the rule set, and an optimized rule set is obtained, and the multi-modal physiological sign data is compared and logically judged in real time through the optimized rule set to obtain a judgment result; Step 4, based on the judgment result, if multiple physiological parameters exceed the corresponding first threshold and meet the combined logical relationship, it is determined that the first abnormal situation is met and a high-level alarm is generated; if only a single physiological parameter exceeds the corresponding second threshold, it is determined that the second abnormal situation is met and a low-level alarm is generated; Step 5, based on the high-level alarm, an alarm message containing the patient's location identifier and real-time sign details is pushed to the designated medical terminal, and a synchronous trigger is sent to the on-site alarm device in the ward to issue an audible and visual alarm signal; based on the low-level alarm, notification information prompting re-measurement is sent to the designated nursing terminal. 2.The stroke burst sign real-time grading alarm method based on edge computing according to claim 1, characterized in that, Real-time acquisition of multi-modal physiological sign data through a smart monitoring bracelet worn by the patient, the multi-modal physiological sign data including heart rate, blood pressure and cerebral oxygen saturation, comprising: Step 1.1, acquisition of the original photoplethysmogram signal through the photoelectric sensor integrated in the bracelet, acquisition of the radial artery pulse signal through the pressure sensor, and acquisition of the brain blood oxygen parameter original signal through the near-infrared spectrum sensor; Step 1.2, preprocessing of the collected original signals, including noise reduction filtering processing of the photoplethysmogram signal, baseline correction processing of the radial artery pulse signal, and motion artifact elimination processing of the near-infrared spectrum signal; Step 1.3, based on the preprocessed signals, calculating the heart rate value using a pulse wave feature extraction algorithm, calculating the blood pressure value using a pulse wave transmission time algorithm, and calculating the cerebral oxygen saturation value using a dual-wavelength near-infrared spectrum analysis method; Step 1.4, time alignment and packaging of the calculated heart rate, blood pressure and cerebral oxygen saturation values to obtain multi-modal physiological sign data in a standardized format. 3.The stroke burst sign real-time grading and alarming method based on edge computing according to claim 2, characterized in that, Transmission of the multi-modal physiological sign data to an edge computing gateway deployed in the ward, comprising: Step 2.1, the smart monitoring bracelet establishes a secure communication connection with the edge computing gateway in the ward through the Bluetooth Low Energy protocol, and encrypts and packages the standardized data frame; Step 2.2, after receiving the encrypted data frame, the edge computing gateway decrypts and verifies the data packet format, checks the data integrity and validity; Step 2.3, adding a timestamp and a patient identifier to the physiological sign data that passes the verification, and converting it to a standardized data format suitable for edge computing processing; Step 2.4, temporarily store the processed normalized data in the local cache area of the edge computing gateway. 4.The stroke burst sign real-time grading and alarming method based on edge computing according to claim 3, characterized in that, The edge computing gateway constructs a multi-dimensional convex hull defining the normal range of the physical sign space based on the normal data in the historical multi-modal physiological sign data, calculates the distance from the real-time multi-modal physiological sign data point to the multi-dimensional convex hull, and initializes the generation of a population containing multiple rule sets, including: Step 3.1, extract historical multi-modal physiological sign data from the local cache area, and filter out data samples representing normal physiological state according to the clinically defined normal physiological parameter range value; Step 3.2, based on the data samples representing normal physiological state, determine the boundary points of normal data in high-dimensional feature space using a convex hull generation algorithm, and form the smallest convex polygon enclosing all normal data by connecting these boundary points, thereby defining the spatial range of normal physical signs; Step 3.3, calculate the shortest Euclidean distance from the real-time collected multi-modal physiological sign data point to each surface of the multi-dimensional convex hull as a quantitative indicator of the degree of deviation from the normal state of the data point; Step 3.4, based on the distance index, generate initial rule sets with different parameter thresholds and logical combinations, where each rule contains specific physiological parameter threshold range, logical operation relationship and corresponding weight coefficient; Step 3.5, encode the generated multiple rule sets as genetic algorithm population individuals to complete the initialization of the population.

5. The stroke burst sign real-time grading and alarming method based on edge computing according to claim 4, characterized in that, Based on the fitness of each rule set in the historical data, selection, crossover and mutation operations are performed to iteratively evolve the rule sets, resulting in optimized rule sets. The multi-modal physiological sign data is compared and logically judged in real time in parallel through the optimized rule sets to obtain the judgment result, including: Step 3.6, based on the historical multi-modal physiological sign data, calculate the fitness of each rule set in the population, which is evaluated by the warning accuracy and false alarm rate of the rule set on the historical data; Step 3.7, according to the fitness evaluation result, select high-quality rule set individuals from the current population using roulette wheel selection method, retain rule sets with high fitness, and perform single-point crossover operation on the selected rule set individuals to exchange part of the rule conditions between different rule sets, generating new rule set combinations; Step 3.8, perform mutation operation on the newly generated rule sets to randomly adjust the physiological parameter threshold, logical relationship or weight coefficient in the rule with a preset probability; Step 3.9, repeat steps 3.6 to 3.8 until a preset number of iterations or fitness convergence is reached, resulting in optimized rule sets. The real-time collected multi-modal physiological sign data is matched in parallel through the optimized rule sets, and the abnormal sign judgment result is generated according to the rule condition satisfaction. 6.The stroke burst sign real-time grading and alarming method based on edge computing according to claim 5, characterized in that, Based on the judgment result, if multiple physiological parameters exceed the corresponding first threshold and meet the combined logical relationship, it is determined to be consistent with the first abnormal situation and a high-level alarm is generated; If only a single physiological parameter exceeds the corresponding second threshold, it is determined to be the second abnormal situation and a low-level alarm is generated, including: Step 4.1, receive the abnormal sign judgment result from the edge computing gateway, which includes the abnormal state identification and corresponding confidence score of each physiological parameter; Step 4.2, analyze the judgment result, when at least two physiological parameters of heart rate, blood pressure and cerebral oxygen saturation are detected to exceed their respective first threshold values at the same time, and the parameter abnormality duration reaches the preset length of time, it is determined to meet the first abnormal situation; Step 4.3, when only a single physiological parameter is detected to exceed its corresponding second threshold value, and the other parameters are within the normal range, it is determined to meet the second abnormal situation; Step 4.4, based on the first abnormal situation judgment result, generate a high-level alarm instruction containing patient detailed information, abnormal parameter data and emergency level identification; Step 4.5, based on the second abnormal situation judgment result, generate a low-level alarm instruction containing the type and current value of the abnormal parameter.

7. The edge computing-based stroke burst sign real-time grading alarm method according to claim 6, characterized in that, Based on the high-level alarm, an alarm message containing the patient location identification and real-time physical sign details is pushed to the designated medical terminal, and a synchronous sound and light alarm signal is triggered by the on-site alarm device in the ward; Based on the low-level alarm, a notification message prompting retesting is sent to the designated nursing terminal, including: Step 5.1, based on the high-level alarm instruction, generate an alarm message containing the patient bed number, real-time value of the abnormal parameter and historical trend data; Based on the low-level alarm instruction, generate a notification message containing the type of abnormal parameter and the recommended retesting time; Step 5.2, based on the patient identity information, query the corresponding responsible medical staff information to determine the target push terminal of the high-level alarm message and the device identification of the on-site alarm device in the ward; Step 5.3, through the hospital intranet communication system, the high-level alarm message is pushed to the designated medical terminal in real time, and a synchronous sound and light alarm trigger instruction is sent to the on-site alarm device in the ward; Step 5.4, through the nursing workstation management system, the low-level alarm message is distributed to the terminal device of the patient's responsible nursing staff; Step 5.5, record the sending status and receiving confirmation of all alarm messages, and update to the patient monitoring log database.

8. An edge computing based real-time stroke sign grading alarm system, which implements the method of any one of claims 1 to 7, characterized in that, including: An acquisition module for real-time acquisition of multi-modal physiological sign data through a smart monitoring bracelet worn by a patient, the multi-modal physiological sign data including heart rate, blood pressure and cerebral oxygen saturation; Transmit the multi-modal physiological sign data to the edge computing gateway deployed in the ward; A computing module for the edge computing gateway to construct a multi-dimensional convex hull defining the normal sign space range based on normal data in historical multi-modal physiological sign data, calculate the distance from the real-time multi-modal physiological sign data point to the multi-dimensional convex hull, and initialize to generate a population containing multiple rule sets; Based on the fitness of each rule set in the historical data, selection, crossover and mutation operations are performed to iteratively evolve the rule set, and an optimized rule set is obtained, which is used for real-time parallel comparison and logical judgment of the multi-modal physiological sign data to obtain a judgment result; An alarm module for generating a high-level alarm if multiple physiological parameters exceed the corresponding first threshold values and meet the combined logical relationship based on the judgment result; if only a single physiological parameter exceeds the corresponding second threshold value, a low-level alarm is generated. The processing module is used for pushing an alarm message containing a patient position identifier and real-time sign details to a designated medical terminal based on a high-level alarm, and triggering a sound and light alarm signal of a field alarm device in a ward synchronously; and sending a notification information prompting retest to a designated nursing terminal based on a low-level alarm.

9. A computing device, comprising: The method comprises the steps of: one or more processors; a storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, so that the one or more processors implement the method as claimed in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a program, and the program is executed by the processor to implement the method as claimed in any one of claims 1 to 7.

Images