A care assistance system based on artificial intelligence

Abstract

The application relates to the technical field of medical care information science, and discloses a nursing auxiliary system based on artificial intelligence, which comprises a data verification module, a reference benchmark generation module, an alarm decision module and a monitoring and dispatching module. The system acquires bed displacement parameters, calculates displacement deviation variance, determines a weight adjustment factor, and then generates physiological benchmark reference data by using weighted recursive logic. Meanwhile, the system compares local mean values of physiological parameters with standard evolution vectors, so that the physiological benchmark reference data can be reset. The application dynamically modulates data weights based on physical displacement variance, fits the physiological relaxation process of a patient, solves the invalid alarm problem caused by the disconnection between physiological monitoring data and clinical context, and realizes the optimized scheduling of medical monitoring resources.

Description

Technical Field

[0001] This invention relates to an artificial intelligence-based nursing assistance system, belonging to the field of healthcare informatics technology. Background Technology

[0002] Current continuous vital signs monitoring systems collect physiological parameters such as heart rate, blood oxygen saturation, and respiratory rate from patients, and generate early warning signals in conjunction with preset static threshold logic. As a basic method for identifying the risk of clinical deterioration in the field of healthcare information processing, it plays a role in improving the efficiency of monitoring.

[0003] However, in practical applications in general wards, patients' physiological parameters are highly intertwined with the physical disturbances caused by daily nursing activities. Existing improved multi-focus monitoring equipment faces challenges in hardware integration or physical overlay of sensor channels and control logic at the data processing level. For example, Chinese invention patent application CN110444284A discloses an intelligent auxiliary nursing and monitoring system that achieves body movement monitoring through the integration of a pressure sensor array. However, the alarm mechanism still relies on static threshold logic with preset allowable ranges. Such existing technical logic has limitations in principle: it treats physiological parameters as isolated time series, failing to penetrate the dynamic coupling between physical displacement intensity and physiological functional compensation and recovery. In actual clinical situations, it ignores the nonlinear compensation and relaxation process of the patient's response to nursing intervention. The static discrimination logic misjudges step signals generated by physical disturbances as pathological deterioration. Since the existing monitoring logic mainly relies on the numerical changes of physical signals, it lacks semantic relation with the flow of clinical nursing events. Routine nursing actions such as turning over, assisting with sputum expectoration, or getting out of bed can easily induce short-term, drastic fluctuations in physiological indicators. This logical disconnect between physical signals and the nursing context leads to frequent invalid alarms triggered by the system, increasing the workload of nursing staff and reducing the clinical guidance value of alarm signals. To reduce the frequency of false alarms, strategies such as extending the judgment time window or relaxing the alarm threshold are usually adopted. However, this often leads to a delay in the system's response to real critical events. A deeper contradiction lies in the fact that the patient's body responds to nursing interventions with non-linear physiological compensation characteristics. After the physical action, the regression process of physiological indicators is accompanied by a significant relaxation time. Existing signal processing logic usually treats such fluctuations as transient noise that needs to be filtered out, forcing the expected trajectory to converge to the historical steady state. This approach ignores the benign baseline transition that may be brought about by effective medical intervention, resulting in a phase misalignment between the generated expected trajectory and the actual vital signs in the evolutionary trajectory.

[0004] Therefore, the technical problem to be solved by this invention is how to construct a mechanism that can semantically align discrete nursing event streams with continuous physiological time series, and adaptively reconstruct individual physiological compensation trajectories and transition baselines based on physical displacement features, so as to eliminate data phase misalignment and improve the specificity of early warning instructions. Summary of the Invention

[0005] To address the problems mentioned in the background art, the technical solution of the present invention is as follows: A nursing assistance system based on artificial intelligence, comprising:

[0006] The data verification module is used to obtain bed displacement parameters that characterize the body movement of the monitored object and to calculate the displacement offset variance within a set sliding time window.

[0007] The reference baseline generation module receives displacement offset variance and contextual labels representing current nursing behaviors, calculates weight adjustment factors based on the displacement offset variance, obtains physiological metabolic parameters associated with the contextual labels, and generates physiological baseline reference data through weighted recursive logic. The weighted recursive logic adjusts the physiological baseline reference data using weight adjustment factors. The update step size is adjusted to increase the retention weight of historical data in the physiological baseline reference data when the displacement offset variance increases; the reference baseline generation module also includes a baseline reset module, which is used to calculate the local mean of physiological parameters of physiological sampling parameters when the displacement offset variance is lower than a preset variance threshold for 5 consecutive sampling periods, and compare the residual between the local mean of physiological parameters and the standard evolution vector corresponding to the context label; if the residual is within the preset threshold range, the physiological baseline reference data is switched to the local mean of physiological parameters;

[0008] The alarm decision module is used to receive physiological baseline reference data and real-time physiological parameters, calculate the deviation index of real-time physiological parameters from physiological baseline reference data, extract the nursing level preset by context labels, and determine the intervention priority.

[0009] The monitoring and scheduling module is used to output intervention instructions to related nursing terminals based on intervention priority.

[0010] Preferably, when calculating the weight adjustment factor, the reference benchmark generation module calculates the difference between the displacement offset variance at the current sampling time and the previous sampling time, and calculates the ratio of the difference to the preset upper limit of the deviation to determine the value of the weight adjustment factor; wherein, the weight adjustment factor is positively correlated with the difference, and the upper limit of the weight adjustment factor is 1.

[0011] Preferably, the reference baseline generation module is also used to retrieve metabolic equivalent data associated with context labels, convert the metabolic equivalent data into steady-state expected increments of physiological characteristics, and adjust the evolution slope of the physiological baseline reference data by superimposing the steady-state expected increments onto the physiological baseline reference data.

[0012] Preferably, the weighted recursive logic executed by the reference benchmark generation module follows the following formula: ,in, This serves as the physiological baseline reference data for the current moment. This serves as a physiological baseline reference data from the previous moment. These are the steady-state values ​​of physiological characteristics determined by contextual labels. This is the weighting adjustment factor; the reference benchmark generation module uses the weighting adjustment factor... Limiting the steady-state values ​​of physiological characteristics Physiological reference data The correction range.

[0013] Preferably, when comparing residuals, the benchmark reset module extracts physiological parameters within 10 seconds from the moment when the displacement offset variance is lower than the preset variance threshold, calculates the evolution trend of the local mean of physiological parameters relative to the mean of physiological parameters within 1 hour before the context label is activated, and determines whether the evolution trend is within the confidence interval defined by the standard evolution vector.

[0014] Preferably, the alarm decision module is used to calculate the instantaneous residual of real-time physiological parameters deviating from the physiological baseline reference data, and to perform time-domain integration on the instantaneous residual to generate a deviation index that reflects the degree of accumulation of physiological risk.

[0015] Preferably, when determining the intervention priority, the alarm decision module calculates the product of the deviation index and the risk weight corresponding to the nursing level, and prioritizes the alarm tasks of each monitoring node in the ward according to the product.

[0016] Preferably, the monitoring and scheduling module is used to identify high-risk nodes according to priority and push an early warning interface with strong interruption display attributes to the nursing terminals associated with the high-risk nodes.

[0017] Preferably, the monitoring and scheduling module is also used to set the alarm status of monitoring nodes other than high-risk nodes to a silent observation state.

[0018] Preferably, the intervention instructions output by the monitoring and scheduling module include: clinical event type identified by contextual tags, treatment time limit identified by intervention priority, and trend graph reflecting abnormal fluctuations in real-time physiological parameters.

[0019] Compared with the prior art, the beneficial effects of the present invention are:

[0020] 1. In AI-assisted nursing care, the physiological parameter receiving unit and the event anchoring unit work together to extract the variance of the bed's centroid offset parameter within a set sliding time window. This establishes a physical correlation mapping between nursing event flow data and physiological time series data. This logical verification method uses the objective variance of the bed's physical displacement as a prerequisite for context activation. Without introducing subjective judgment, it achieves the structural identification of physiological fluctuations induced by nursing behavior. This mechanism ensures that the system's judgment of indicators such as heart rate or blood oxygen exceeding limits has clear clinical background support, eliminates semantic deviations between physiological parameter fluctuations and nursing operation instructions, and enables the early warning decision flow to accurately filter alarm signals generated by routine nursing activities.

[0021] 2. The system introduces a dynamic forgetting filter mechanism controlled by the variance of the physical centroid. The expected physiological trajectory is generated iteratively through an asymmetric difference equation. During physical care procedures, the value logic of the dynamic forgetting factor enables the system to cut off the immediate impact of fluctuation increments on the baseline at the control flow level, maintaining the historical inertia of the expected trajectory. In the compensatory recovery phase after the procedure, the decay function drives the dynamic forgetting factor to gradually decrease, so that the expected physiological trajectory exhibits a nonlinear hysteresis fallback pattern that conforms to the characteristics of biological compensation. This method of dynamically modulating data weights based on the variance of objective physical displacement fits the individualized physiological relaxation process of the patient, solving the alarm phase misalignment problem caused by the inability of traditional static compensation models to match the compensatory hysteresis effect of the patient's body.

[0022] 3. The semantic transition verification module and the baseline generation unit work together to construct an asymmetric baseline transition path based on a behavioral label semantic dictionary. After the physical action is completed, the system uses the consistency comparison between the local surface smoothing mean and the evolution vector to identify the physiological baseline reset guided by medical intervention. This breaks the mechanical inertia of the monitoring system to return to the historical initial steady state, enabling the system to actively approach the new steady-state benchmark generated after the intervention. This dynamic switching method of the benchmark target based on intervention semantics avoids false alarms of intervention instructions triggered by the improvement of patient vital signs, and realizes the reproduction of clinical context in the process of medical and health care information processing. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the operational logic and alarm decision-making process of the artificial intelligence-based nursing assistance system of the present invention.

[0024] Figure 2 This is a diagram showing the device interaction and system architecture of the artificial intelligence-based nursing assistance system of the present invention.

[0025] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0026] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0027] An artificial intelligence-based nursing assistance system includes:

[0028] The data verification module is used to obtain bed displacement parameters that characterize the body movement of the monitored object and to calculate the displacement offset variance within a set sliding time window.

[0029] The reference baseline generation module receives displacement offset variance and contextual labels representing current nursing behaviors, calculates weight adjustment factors based on the displacement offset variance, obtains physiological metabolic parameters associated with the contextual labels, and generates physiological baseline reference data through weighted recursive logic. The weighted recursive logic adjusts the physiological baseline reference data using weight adjustment factors. The update step size is adjusted to increase the retention weight of historical data in the physiological baseline reference data when the displacement offset variance increases; the reference baseline generation module also includes a baseline reset module, which is used to calculate the local mean of physiological parameters of physiological sampling parameters when the displacement offset variance is lower than a preset variance threshold for 5 consecutive sampling periods, and compare the residual between the local mean of physiological parameters and the standard evolution vector corresponding to the context label; if the residual is within the preset threshold range, the physiological baseline reference data is switched to the local mean of physiological parameters;

[0030] The alarm decision module is used to receive physiological baseline reference data and real-time physiological parameters, calculate the deviation index of real-time physiological parameters from physiological baseline reference data, extract the nursing level preset by context labels, and determine the intervention priority.

[0031] The monitoring and scheduling module is used to output intervention instructions to related nursing terminals based on intervention priority.

[0032] Preferably, when calculating the weight adjustment factor, the reference benchmark generation module calculates the difference in displacement offset variance at the current sampling time relative to the previous sampling time, and calculates the ratio of this difference to a preset upper limit of deviation to determine the value of the weight adjustment factor; wherein, the weight adjustment factor is positively correlated with the difference, and the weight adjustment factor... The upper limit is 1.

[0033] Preferably, the reference baseline generation module is also used to retrieve metabolic equivalent data associated with context labels, convert the metabolic equivalent data into steady-state expected increments of physiological characteristics, and then superimpose these steady-state expected increments onto the physiological baseline reference data. Adjust the slope of the evolution of physiological baseline reference data.

[0034] Preferably, the weighted recursive logic executed by the reference benchmark generation module follows the following formula: ,in, This serves as the physiological baseline reference data for the current moment. This serves as a physiological baseline reference data from the previous moment. These are the steady-state values ​​of physiological characteristics determined by contextual labels. This is the weighting adjustment factor; the reference benchmark generation module uses the weighting adjustment factor... Limiting the steady-state values ​​of physiological characteristics Physiological reference data The correction range.

[0035] Preferably, when comparing residuals, the benchmark reset module extracts physiological parameters within 10 seconds from the moment when the displacement offset variance is lower than the preset variance threshold, calculates the evolution trend of the local mean of physiological parameters relative to the mean of physiological parameters within 1 hour before the context label is activated, and determines whether the evolution trend is within the confidence interval defined by the standard evolution vector.

[0036] Preferably, the alarm decision module is used to calculate the instantaneous residual of real-time physiological parameters deviating from the physiological baseline reference data, and to perform time-domain integration on the instantaneous residual to generate a deviation index that reflects the degree of accumulation of physiological risk.

[0037] Preferably, when determining the intervention priority, the alarm decision module calculates the product of the deviation index and the risk weight corresponding to the nursing level, and prioritizes the alarm tasks of each monitoring node in the ward according to the product.

[0038] Preferably, the monitoring and scheduling module is used to identify high-risk nodes according to priority and push an early warning interface with strong interruption display attributes to the nursing terminals associated with the high-risk nodes.

[0039] Preferably, the monitoring and scheduling module is also used to set the alarm status of monitoring nodes other than high-risk nodes to a silent observation state.

[0040] Preferably, the intervention instructions output by the monitoring and scheduling module include: clinical event type identified by contextual tags, treatment time limit identified by intervention priority, and trend graph reflecting abnormal fluctuations in real-time physiological parameters.

[0041] Example 1: This invention provides an artificial intelligence-based nursing assistance system that operates in a general internal medicine and surgical ward environment characterized by a high bed-to-nurse ratio. The physiological parameter receiving unit continuously acquires multidimensional physiological time-series data characterizing the fluctuations in the vital signs of the monitored subjects. Simultaneously, the event anchoring unit retrieves nursing event stream data with definite timestamps and behavioral tags from the hospital information system. Because patients experience short-term, dramatic fluctuations in heart rate and respiratory rate during routine nursing procedures such as turning over or assisting with sputum clearance, this physical interference often leads to a massive number of invalid alarms in traditional monitoring systems. This results in the actual pathological risks being masked by data noise, affecting the accuracy of nursing resource allocation. The data verification module obtains bed displacement parameters characterizing the monitored subject's body movement, calculates the displacement offset variance within a set sliding time window, and determines a weighting adjustment factor based on the displacement offset variance. The weighting adjustment factor is related to the displacement offset variance. There is a positive correlation mapping relationship, and the weight adjustment factor The upper limit is 1; the reference benchmark generation module receives the displacement offset variance. Physiological baseline reference data are generated through weighted recursive logic, using contextual labels that characterize current nursing behaviors. This weighted recursive logic follows the formula ,in This serves as the physiological baseline reference data for the current moment. This serves as a physiological baseline reference data from the previous moment. As a weighting adjustment factor, The target convergence baseline.

[0042] When the displacement variance When the weight adjustment factor increases to above the preset threshold, As the value approaches 1, the system increases the weighting of historical data and cuts off the disturbance of instantaneous fluctuations to the baseline, thereby maintaining the physiological baseline reference data. The operational inertia is used to avoid false alarms triggered directly by physical displacement; the nursing action ends and the displacement deviation variance is... When the variance is below a preset threshold for five consecutive sampling periods, the benchmark reset module initiates the semantic transition verification procedure, which calculates the displacement variance. Local mean of physiological parameters within 10 seconds of falling back to the static dead zone And compare the local mean values ​​of physiological parameters The residual between the standard evolution vectors corresponding to the context labels; if the residual is within a preset threshold range, it indicates that the current physiological parameter change direction conforms to the benign transition characteristics brought about by effective medical intervention, and the system will converge the target to the baseline. Switch to local mean of physiological parameters Otherwise, converge the target to the baseline. Maintain the average physiological parameters within 1 hour prior to the occurrence of the nursing event. The weighting adjustment factor is controlled by a decay function that gradually decreases, driving the physiological baseline reference data. Converging to the updated target baseline in a nonlinear hysteresis manner Dynamic approximation is performed to fit an individualized physiological compensatory regression trajectory; the alarm decision module receives physiological baseline reference data. Combined with real-time physiological parameters, the deviation of real-time physiological parameters from physiological baseline reference data is calculated. The deviation index is calculated, and the nursing level preset by the context label is extracted to determine the intervention priority. The monitoring and scheduling module outputs intervention instructions to the associated nursing terminal based on the intervention priority, including the clinical event type identified by the context label and the trend map reflecting the abnormal fluctuation of real-time physiological parameters. By logically coupling discrete clinical semantic labels with continuous physiological time series, the data update step size is dynamically adjusted using objective physical displacement variance. At the same time, the phase difference alarm caused by physiological lag effect is eliminated, and the context purification of the medical and health care information flow with high specificity is achieved.

[0043] Example 2: This experiment used a digital simulation platform to acquire physiological monitoring data and simulated a high-bed-to-nurse ratio ward environment. The input data consisted of raw sampling sequences labeled with heart rate and respiratory rate, superimposed with background noise of 20dB signal-to-noise ratio and power frequency interference of 50Hz. The sampling frequency was... The frequency was set to 250Hz to maintain the processor's data throughput pressure below 50% of its rated load range, and the sliding time window width was set to 5s to ensure instantaneous response sensitivity to sudden body movements; the test groups included a control group using static threshold alarm logic and a control group using no displacement offset variance. In a control group with partial missing weighted recursive logic and a sample group using the present invention, when simulating patient turning movements, the sample group using the present invention monitored changes in bed displacement parameters that caused variance in displacement. The value increased from 0.05 to 2.15, based on the displacement offset variance. Weighting adjustment factor Adjusted from 0.12 to 0.96, physiological baseline reference data. The value was maintained around 76.5, thus achieving logical decoupling from the spurious peak of 112.4 generated in the real-time physiological parameters.

[0044] Data comparison showed that the control group generated 4 false alarms during the turning-over disturbance. The partial missing control group lacked physical displacement compensation, resulting in inaccurate physiological baseline reference data. The deviation index of the sample group of this invention was below 1.2 times the alarm threshold, and normal monitoring was restored within 5 seconds after the body movement ended. This result confirms the displacement deviation variance. The positive correlation mapping logic with the weighting adjustment factor can suppress false alarms induced by physical displacement; in the physiological metabolic gradient pressure test, the system processes the pathological sequence in which the patient's respiratory rate increases stepwise from 16 breaths / min to 28 breaths / min. The sample group of this invention shows that the displacement offset variance... When the value is at a low level of 0.06 and the weighting adjustment factor is stable within the range of 0.15, the physiological baseline reference data is... Tracking real-time physiological parameter changes, when the respiratory rate reached 24 breaths / min, the deviation index exceeded 1.2 times the threshold, and the system output an intervention command within 12.5 seconds. This response time is lower than the industry standard delay of 30 seconds. Test results for parameter boundaries show that when the displacement deviation variance... After the baseline reset module reverts to its previous state, it calculates the local mean of physiological parameters over 10 seconds. The value was 82.5. The standard evolution vector residual associated with this value and the behavioral label was 2.1, which was within the preset threshold of 5.0. The system then performed a baseline reset to make the physiological baseline reference data... Approaching 82.5 in a non-linear manner, the experiment showed that the false alarm rate of the sample group of the present invention was reduced by 92.5% and the risk identification accuracy was maintained above 98.6%.

[0045] Example 3: In the monitoring scenario of early postoperative ambulation, the monitoring node processes the postural physiological fluctuations that occur when transitioning from a bedridden to a sitting position, and the physiological parameter receiving unit acquires multidimensional physiological time series data characterizing the patient's heart rate time series. Simultaneously, the state verification unit monitors the instantaneous displacement of the bed's centroid coordinates; the data verification module performs sliding variance calculation on the coordinate sequence to calculate the displacement offset variance reflecting the body dynamic intensity. During this process, the reference benchmark generation module parses the contextual label "sit up" and retrieves the corresponding physiological metabolic parameters from the mapping database. The reference benchmark generation module then retrieves metabolic equivalent data, quantifies it based on nursing intensity and oxygen consumption, and uses a linear mapping relationship to convert the metabolic equivalent data into the expected steady-state increment ΔH of physiological characteristics. The expected steady-state increment ΔH satisfies the formula... Wherein, variable k is a proportionality coefficient determined based on the subject's body mass index and baseline heart rate. The metabolic equivalent of the task corresponding to the context label. This is the resting metabolic equivalent; the system adds ΔH to the steady-state value of physiological characteristics. Adjust physiological baseline reference data The evolution slope causes the expected trajectory to dynamically approach a new steady-state value that conforms to the metabolic load characteristics after the physical operation is completed; the weighting adjustment factor... The values ​​are limited to the closed-loop range of 0.05 to 0.98, and the system is set to monitor the displacement deviation variance with a 600-second safety timer. If it is at a high level for a period of time, If the duration exceeds the preset variance threshold for an extended period, reaching the preset threshold of the safety timer, then... Reduced to 0.1 to restore real-time physiological parameters to physiological baseline reference data Adjusting weights; the reference benchmark generation module utilizes an exponential decay function. The update step size is dynamically adjusted, with variable τ being a time constant calibrated based on the physiological relaxation characteristics of the subjects, so that the physiological baseline reference data... It enters the steady-state error band within 5 seconds after the body motion disappears.

[0046] To establish a quantitative benchmark for the mapping relation library, this embodiment provides a method for calibrating metabolic parameters based on statistical expectation. Specifically, it involves pre-collecting the average oxygen consumption increment of 100 groups of subjects after performing a sit-up exercise and converting it into the steady-state expected increment of the heart rate baseline. The reference datum generation module calculates the displacement offset variance. The difference between the current time and the previous time, and using the formula Determine the weighting adjustment factor The value of, among which The variance of displacement at the current sampling time. This represents the variance of the displacement at the previous sampling time. The system has a preset upper limit for displacement deviation, set to 5.0; the system uses a weighting adjustment factor to adjust the physiological baseline reference data. The update step size, in the displacement offset variance During the sampling period in which abrupt changes occur, the weighting adjustment factor approaches 0.98, and the system increases the weighting of historical data retention and suppresses the disturbance of real-time physiological parameters to the baseline; when physical motion tends to stabilize and the displacement deviation variance decreases... After five consecutive sampling cycles, the baseline reset module returns to a static dead zone of 0.1, and then initiates a timeliness guarantee procedure to correct the baseline phase. The system extracts physiological sampling parameters within 10 seconds after the start of the static dead zone and calculates the local mean of these physiological parameters. And retrieve the standard evolution vector corresponding to the sit-up label. .

[0047] The semantic transition verification module executes residual comparison logic and calculates the local mean of physiological parameters. Smoothed mean of physiological parameters one hour prior to the nursing event The difference ΔM is determined, and it is determined whether ΔM is within the standard evolution vector. Within the defined confidence interval; if the judgment result is yes, the system confirms that the current physiological fluctuation is a steady-state transition caused by benign postural changes, and then synchronously converges the target to the baseline. Updated to local mean of physiological parameters To address the smoothness requirement of the system's transition from the disturbance state to the monitoring state, the weight adjustment factor... A compensation loop controlled by an attenuation function is connected, which follows the formula... Where k is the sampling point number deviating from the static dead zone, and τ is a preset time constant set to 0.2. This time constant is calibrated by recording 90% of the time it takes for the subject's heart rate to drop from the excited state to the steady state; with the weighting adjustment factor The exponential decrease, physiological baseline reference data The system dynamically approximates the updated target convergence baseline using a nonlinear hysteresis pattern; the alarm decision module continuously calculates real-time physiological parameters and physiological baseline reference data. The deviation index remained below the alarm threshold throughout the entire body position change process, eliminating false alarms caused by non-pathological changes in vital signs. This solution establishes a dynamic relationship between physical displacement intensity and physiological baseline inertia at the system logic level, utilizing displacement deviation variance. As a damping factor, it dynamically modulates the penetration weight of physiological time series in the weighted recursive process; by implanting a calibration procedure with physical traceability of the decay function, it solves the benchmark instability problem caused by the action switching gap of traditional nursing auxiliary equipment; this decoupling in the feature dimension and reconstruction in the semantic dimension based on multi-source heterogeneous data ensures the continuity of monitoring tasks in the dynamic clinical context, and reflects the technical path of improving the system monitoring specificity through multi-source data fusion in healthcare informatics.

[0048] Example 4: In the offline calibration scenario simulating the human metabolic load gradient, the image relation library retrieved by the reference generation module is constructed as follows: Subject groups with different body mass index gradients and age gradients are selected. Reference heart rate data is collected in a controlled laboratory environment, and real-time oxygen consumption is measured using an indirect calorimeter when subjects perform context-labeled nursing behaviors such as sitting up, turning over, and assisted expectoration. The data processing unit converts the measured metabolic equivalent data into the steady-state expected increment ΔH of the corresponding physiological characteristic parameters and establishes an index relationship between the steady-state expected increment ΔH and the context label, forming a mapping relationship table supported by physical dimensions, where ΔH is the expected offset of the physiological characteristic parameter. The physical calibration and calculation logic of the proportionality coefficient is as follows: The system retrieves the subject's body mass index; if the index is in the range of 18.5 to 24.0... If the index is within the range of 24.0 to 28.0, the proportional coefficient is set to 0.8; if the index is within the range of 24.0 to 28.0, the proportional coefficient is automatically adjusted to 1.2; if the index exceeds 28.0, the proportional coefficient is set to 1.5. This proportional coefficient is used to convert the difference between the task metabolic equivalent and the resting metabolic equivalent into a specific fluctuation value of heart rate. For example, when the difference in metabolic equivalent is 3.0 and the proportional coefficient is 1.2, the corresponding expected increment of the steady state of physiological characteristics is determined to be 3.6 heart rate beats per minute, thereby realizing the quantitative mapping from the abstract metabolic equivalent to the physical heart rate increment. For the acquisition link of the bed center of mass offset parameter, pressure sensors with a range of 0 to 200 kg and a measurement accuracy of 0.01 kg are arranged at the four corners of the bed. Coordinate normalization calibration is performed under static conditions of empty bed and load of 50 kg to 150 kg to calculate the upper limit of bed displacement deviation. Based on the baseline data, the physical boundary of the denominator term of the weighting adjustment factor is established.

[0049] When the system is deployed in a medical bed environment with specific mechanical vibration characteristics, the data verification module performs on-site initialization calibration procedures to determine the static dead zone threshold and the physical parameters of the attenuation function. The system continuously collects displacement signals for 1000 sampling periods in a resting state without patients in bed and calculates its variance distribution. The upper limit of the 99.7% confidence interval of this distribution is selected as the initial threshold for determining the static dead zone. A single-pulse interference action is simulated, and the weighting adjustment factor is recorded. The response curve from the peak back to the initial steady state is fitted with the exponential decay equation using the least squares method. To solve for the time constant τ, where The weight adjustment factor at the current moment. τ is the weighting adjustment factor at the end of the interference, k is the sampling point number deviating from the static dead zone, and τ is the decay coefficient obtained by recording 90% of the time it takes for the subject's heart rate to fall from the excited state to the steady state. The value of τ is within the range that ensures the physiological baseline reference data. Within 5 seconds after the end of the physical activity, the system enters the physical range of the steady-state error band; after calibration, the system enters continuous monitoring mode and locks the calibration parameters and physiological reference data. To converge to the target baseline in a defined dynamical form Approaching.

[0050] Example 5: In the digital monitoring unit, the physiological parameter receiving unit acquires multidimensional physiological time-series data of the subject when performing a preset turning movement. A sliding window with a sampling frequency of 250Hz is used to capture the heart rate waveform within 10 seconds after the movement trigger moment. The first derivative of the heart rate waveform is calculated to determine the instantaneous evolution trend of the physiological parameters. The data processing unit extracts the extreme values, mean, and variance from the instantaneous evolution trend as feature components. Weighted linear regression is then performed to encapsulate the feature components into a standard evolution vector corresponding to the context label. And store it in a mapping relation library, where the standard evolution vector Each dimension corresponds to the steady-state expected increment ΔH of the physiological characteristic parameter in a specific nursing context, where For multidimensional physiological time series data, Let ΔH be the standard evolution vector, and let ΔH be the steady-state expected increment.

[0051] When an AI-based nursing assistance system is first deployed on a specific physical hardware platform, the data verification module performs an on-site initialization and calibration process to determine the upper limit of bed displacement deviation. Using the quantization threshold of the static dead zone, the system continuously collects pressure signals for 2000 sampling cycles under static conditions: the bed is unloaded and the load is a standard 100kg load. The displacement deviation variance of the pressure signal within the sliding time window is calculated. The distribution characteristics were analyzed; during the system deployment phase, an initialization calibration procedure was executed, using pressure sensors at the four corners of the medical bed to collect pressure fluctuation signals under no-load and standard load conditions. The displacement deviation variance distribution of the pressure fluctuation signals within a 5-second sliding window was calculated, and the upper limit of the 99% confidence interval of the distribution was selected as the preset variance threshold. The instantaneous peak variance generated by the simulated turning action was recorded and defined as the upper limit of the displacement deviation. Standard evolution vector The baseline reset module calculates the local mean of physiological parameters by including a three-dimensional feature vector containing the rate of change of heart rate, deviation of respiratory rate, and slope of recovery of blood oxygen saturation. Associated with contextual tags Determine whether the residuals are within a preset threshold range using Euclidean distance to determine the target convergence baseline. The logic is reset by simulating the maximum intensity of limb flipping movements and recording the variance of displacement. The instantaneous peak value is defined as the deviation of the bed displacement from the upper limit. Thus, the weighting adjustment factor The calculation provides a physical boundary reference, and the calculation formula is as follows: ,in As a weighting adjustment factor, The variance of displacement at the current sampling time. This represents the variance of the displacement at the previous sampling time. The upper limit of bed displacement deviation, physiological reference data. Based on this, converge the baseline towards the target. Approaching.

[0052] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A nursing assistance system based on artificial intelligence, characterized in that, include: The data verification module is used to obtain bed displacement parameters that characterize the body movement of the monitored object and to calculate the displacement offset variance within a set sliding time window. The reference baseline generation module receives displacement offset variance and contextual labels representing current nursing behaviors, calculates weight adjustment factors based on the displacement offset variance, obtains the steady-state values ​​of physiological characteristics associated with the contextual labels, and generates physiological baseline reference data through weighted recursive logic. The weighted recursive logic adjusts the physiological baseline reference data using the weight adjustment factors. The update step size is adjusted to increase the retention weight of historical data in the physiological baseline reference data when the displacement offset variance increases; the reference baseline generation module also includes a baseline reset module, which is used to calculate the local mean of the physiological parameters when the displacement offset variance is lower than a preset variance threshold for 5 consecutive sampling periods, and compare the residual between the local mean of the physiological parameters and the standard evolution vector corresponding to the context label; if the residual is within the preset threshold range, the physiological baseline reference data is switched to the local mean of the physiological parameters; The alarm decision module is used to receive physiological baseline reference data and real-time physiological parameters, calculate the deviation index of real-time physiological parameters from physiological baseline reference data, extract the nursing level preset by context labels, and determine the intervention priority. The monitoring and scheduling module is used to output intervention instructions to related nursing terminals based on intervention priority; Furthermore, the weighted recursive logic executed by the reference benchmark generation module follows the following formula: ,in, This serves as the physiological baseline reference data for the current moment. This serves as a physiological baseline reference data from the previous moment. These are the steady-state values ​​of physiological characteristics determined by contextual labels. The weighting adjustment factor; the reference benchmark generation module uses the weighting adjustment factor. Limiting the steady-state values ​​of physiological characteristics Physiological reference data The correction range.

2. The nursing assistance system based on artificial intelligence according to claim 1, characterized in that, When calculating the weight adjustment factor, the reference benchmark generation module calculates the difference between the displacement offset variance at the current sampling time and the previous sampling time, and calculates the ratio of the difference to the preset upper limit of the deviation to determine the value of the weight adjustment factor; wherein, the weight adjustment factor is positively correlated with the difference, and the upper limit of the weight adjustment factor is 1.

3. The nursing assistance system based on artificial intelligence according to claim 1, characterized in that, The reference baseline generation module is also used to retrieve metabolic equivalent data associated with context labels and convert the metabolic equivalent data into steady-state expected increments of physiological characteristics. By superimposing the steady-state expected increments onto the physiological baseline reference data, the evolution slope of the physiological baseline reference data is adjusted.

4. The nursing assistance system based on artificial intelligence according to claim 1, characterized in that, The benchmark reset module extracts the displacement offset variance when comparing residuals. For physiological parameters within 10 seconds from the time when the variance threshold is lower than the preset variance threshold, calculate the evolution trend of the local mean of the physiological parameters relative to the mean of the physiological parameters within 1 hour before the activation of the context label, and determine whether the evolution trend is within the confidence interval defined by the standard evolution vector.

5. The nursing assistance system based on artificial intelligence according to claim 1, characterized in that, The alarm decision module is used to calculate the instantaneous residual of real-time physiological parameters deviating from the physiological baseline reference data, and to perform time-domain integration on the instantaneous residual to generate a deviation index that reflects the degree of accumulation of physiological risk.

6. The nursing assistance system based on artificial intelligence according to claim 1, characterized in that, When determining intervention priorities, the alarm decision module calculates the product of the deviation index and the risk weight corresponding to the nursing level, and prioritizes the alarm tasks of each monitoring node in the ward based on the product.

7. The nursing assistance system based on artificial intelligence according to claim 6, characterized in that, The monitoring and scheduling module is used to identify high-risk nodes based on priority and push warning interfaces with strong interruption display attributes to the nursing terminals associated with the high-risk nodes.

8. The nursing assistance system based on artificial intelligence according to claim 7, characterized in that, The monitoring and scheduling module is also used to set the alarm status of monitoring nodes other than high-risk nodes to a silent observation state.

9. The nursing assistance system based on artificial intelligence according to claim 1, characterized in that, The intervention instructions output by the monitoring and scheduling module include: clinical event types identified by context labels, treatment time limits identified by intervention priorities, and trend graphs reflecting abnormal fluctuations in real-time physiological parameters.

Images