A wearable device integrating asthma attack monitoring and automated drug delivery functions

Abstract

This application discloses a wearable device and method integrating asthma attack monitoring and automated drug delivery. The method includes: acquiring a user's real-time multimodal biosignal vector stream; constructing a personalized steady-state feature signature for the user, the personalized steady-state feature signature being used to characterize the user's healthy physiological baseline; determining a steady-state perturbation scalar in real-time, characterizing the degree of deviation between the current physiological state and the healthy physiological baseline; generating a quantified asthma attack risk index based on the steady-state perturbation scalar through an asynchronous risk escalation engine; and when the asthma attack risk index meets preset intervention triggering conditions, determining that a coherence desynchronization event has occurred, and controlling a miniaturized injection actuator within the wearable device to perform automated drug delivery. This application, by constructing a physiological state phase-locked loop model, achieves adaptive monitoring and closed-loop automated intervention for asthma attacks, improving emergency response speed and monitoring specificity.

Description

Technical Field

[0001] This invention relates to the field of wearable medical device technology, and in particular to a wearable device that integrates asthma attack monitoring and automatic injection drug delivery functions. Background Technology

[0002] Asthma is a common chronic respiratory disease, and acute attacks can cause severe breathing difficulties and even be life-threatening. For asthma patients, timely recognition of attack warning signs and rapid administration of emergency medications (such as bronchodilators) are crucial. Currently, patients typically rely on self-assessment or portable peak flow meters to determine their condition and manually administer medication using inhalers or syringes. However, during severe attacks, patients may be unable to effectively self-medicate due to panic, weakness, or confusion. Furthermore, while existing consumer-grade wearable devices such as smartwatches can monitor single or a few physiological indicators like heart rate and blood oxygenation, they lack the ability to specifically identify the complex physiological event of an asthma attack. They are susceptible to false alarms due to non-disease factors such as exercise and lack any physical intervention capabilities. Therefore, current technology lacks a highly integrated and intelligent wearable device that can combine accurate, adaptive asthma attack monitoring with automated emergency medication administration to achieve a truly closed-loop emergency response. Summary of the Invention

[0003] The purpose of this application is to provide a wearable device, method, and computer-readable storage medium that integrates asthma attack monitoring and automatic injection drug delivery functions, aiming to solve the technical problem that the prior art cannot achieve highly specific closed-loop monitoring and automatic emergency intervention for asthma attacks.

[0004] Firstly, this application provides a method integrating asthma attack monitoring and automated drug delivery, comprising: acquiring a user's multimodal biosignal vector stream; establishing a personalized steady-state feature signature characterizing the user's healthy physiological baseline; determining a steady-state perturbation scalar characterizing the degree of deviation between the physiological state corresponding to the multimodal biosignal vector stream and the personalized steady-state feature signature; generating a quantified asthma attack risk index based on the steady-state perturbation scalar; and executing automated drug delivery when the asthma attack risk index meets preset intervention triggering conditions. This application, by establishing a personalized physiological baseline model and quantifying the degree of deviation in real time, can achieve accurate identification of asthma attacks and form a closed loop from monitoring to intervention, thereby improving the timeliness and effectiveness of emergency assistance.

[0005] In one possible implementation of the first aspect, acquiring the user's multimodal biosignal vector stream includes: acquiring the user's heart rate and blood oxygen saturation signals via a photoplethysmography (PPG) sensor; acquiring the user's respiratory audio signals via a microelectromechanical system (MEMS) microphone; and acquiring the user's skin conductance signals via a skin conductance sensor. By fusing multiple physiological signals, the physiological stress response during an asthma attack can be captured more comprehensively, improving the dimensionality and reliability of the monitoring.

[0006] In one possible implementation of the first aspect, the method further includes: performing spectral analysis on the respiratory audio signal to extract wheezing features associated with asthma attacks. By introducing wheezing, a highly specific acoustic feature, this application can effectively distinguish abnormal physiological indicators caused by other factors (such as strenuous exercise), improving the accuracy of identification and reducing the false alarm rate.

[0007] In one possible implementation of the first aspect, establishing a personalized steady-state feature signature representing a user's healthy physiological baseline includes: continuously collecting multimodal biosignal data while the user is in a healthy state; and constructing a multidimensional statistical model based on the multimodal biosignal data to describe the statistical distribution characteristics of each signal component, as the personalized steady-state feature signature. This application, by adopting a personalized modeling approach, enables the device to adapt to the unique physiological characteristics of each user, avoiding errors caused by a one-size-fits-all threshold judgment and improving the accuracy of monitoring.

[0008] In one possible implementation of the first aspect, the multidimensional statistical model includes at least one mean vector and a covariance matrix. Utilizing the mean vector and covariance matrix not only characterizes the average level of each signal but also captures the correlation between signals, making the model's description of the physiological state more comprehensive.

[0009] In one possible implementation of the first aspect, determining the steady-state perturbation scalar includes: inputting the multimodal biosignal vector stream into a coherence deviation discriminator; the coherence deviation discriminator calculating the Mahalanobis distance of the multimodal biosignal vector stream in a multidimensional feature space defined by the personalized steady-state feature signature to generate the steady-state perturbation scalar. Using Mahalanobis distance can effectively measure the degree of anomaly in multidimensional data points, while considering the correlation between variables and being scale-insensitive, making it a very reliable anomaly detection method.

[0010] In one possible implementation of the first aspect, generating a quantified asthma attack risk index based on the steady-state perturbation scalar includes: inputting the steady-state perturbation scalar into an asynchronous risk upgrade engine; the asynchronous risk upgrade engine generating the asthma attack risk index based on the time series of the steady-state perturbation scalar. By analyzing the temporal evolution trend of the perturbation vector rather than its instantaneous value, the persistence and severity of abnormal states can be determined more reliably, effectively filtering out instantaneous noise.

[0011] In one possible implementation of the first aspect, the asynchronous risk upgrade engine further generates the asthma attack risk index based on the first and second time derivatives of the steady-state perturbation scalar. Introducing information on the rate of change and acceleration makes the risk assessment more forward-looking and enables a faster response to rapidly deteriorating physiological states.

[0012] In one possible implementation of the first aspect, the method further includes: initiating a user intervention confirmation procedure before performing automated injection administration; the user intervention confirmation procedure includes issuing an alarm and receiving a user's stop instruction within a preset time; and performing the automated injection administration when no user stop instruction is received within the preset time. The safety mechanism of this application, while ensuring emergency response, grants a conscious user the final veto right, enhancing the safety of use and user trust, and effectively preventing drug abuse due to misjudgment.

[0013] In one possible implementation of the first aspect, the method further includes: initiating the user intervention confirmation procedure in advance based on a prediction of the future evolution trajectory of the asthma attack risk index. This predictive feedforward control mechanism can buy users more valuable reaction and decision-making time, further improving the system's emergency response capability and security.

[0014] In one possible implementation of the first aspect, the method further includes: when the asthma attack risk index is lower than a preset steady-state threshold, using the multimodal biosignal vector stream with preset update weights to update the personalized steady-state feature signature, thereby achieving dynamic calibration of the personalized steady-state feature signature. This dynamic calibration mechanism ensures that the personalized model can continuously adapt to the user's long-term physiological changes, maintaining the long-term effectiveness and accuracy of the monitoring.

[0015] In a second aspect, this application provides a wearable device integrating asthma attack monitoring and automatic injection drug delivery functions, comprising: a multimodal sensor module for acquiring a user's multimodal biosignal vector stream; a processor; a memory storing a computer program; and a miniaturized injection actuator; wherein the processor is configured to execute the computer program to implement the method as described in any possible implementation of the first aspect.

[0016] In one possible implementation of the second aspect, the miniaturized injection actuator includes a replaceable microneedle cartridge module.

[0017] In one possible implementation of the second aspect, the microneedle cartridge module includes: a sealed drug reservoir; a micro piston; an injection needle initially concealed; and a drive mechanism driven by a shape memory alloy or a micro gas generator, for ejecting the injection needle and pushing the micro piston to inject the drug solution upon receiving an execution command.

[0018] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the method as described in any possible implementation of the first aspect. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a schematic diagram of the system architecture of a wearable device that integrates asthma attack monitoring and automatic injection drug delivery functions according to an embodiment of this application.

[0021] Figure 2 This is a flowchart illustrating a method for integrating asthma attack monitoring and automatic drug delivery according to an embodiment of this application.

[0022] Figure 3 This is a conceptual schematic diagram of a physiological state phase-locked loop model in one embodiment of this application.

[0023] Figure 4 This is a schematic diagram of the microneedle cartridge module in a miniaturized injection actuator in one embodiment of this application. Detailed Implementation

[0024] The technical solutions of the embodiments of this application will be described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0026] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0027] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0028] This application constructs a physiological state phase-locked loop (PLL) model, treating the user's healthy physiological state as a stable reference signal. Wearable devices monitor the user's multimodal physiological signals in real time as input signals and continuously calculate the phase difference between the two, i.e., the degree of deviation of the physiological state. When the deviation exceeds a certain threshold, i.e., a loss of lock occurs, the system determines an asthma attack and triggers automatic drug administration. Furthermore, through a continuous calibration process, the model continuously refines the description of the user's health baseline, thereby achieving personalized and adaptive monitoring.

[0029] Please see Figure 1 This illustration shows a system architecture diagram of a wearable device 100 integrating asthma attack monitoring and automated drug delivery functions according to an embodiment of this application. The wearable device 100 can take the form of a smartwatch, smart bracelet, or patch device. In one embodiment, the wearable device 100 includes a multimodal sensor module 110, a processor 120, a memory 130, a communication module 140, a user interface 150, and a miniaturized injection actuator 160.

[0030] The multimodal sensor module 110, memory 130, communication module 140, user interface 150, and miniaturized injection actuator 160 are all electrically connected to the processor 120. The memory 130 stores instructions executable by the processor 120, such as computer programs for implementing the methods described in this application. The processor 120 may be a central processing unit (CPU), microcontroller (MCU), digital signal processor (DSP), or field-programmable gate array (FPGA), etc., and executes the instructions stored in the memory 130 to control the overall operation of the wearable device 100.

[0031] The multimodal sensor module 110 is a collection of sensors used to non-invasively acquire physiological signal data from multiple dimensions. In one embodiment, the multimodal sensor module 110 includes a photoplethysmography (PPG) sensor 111, a microelectromechanical system (MEMS) microphone 112, and a skin conductance (GSR) sensor 113. The PPG sensor 111 acquires PPG signals by emitting a light beam of a specific wavelength (e.g., red and infrared light) and detecting changes in the intensity of reflected or transmitted light in subcutaneous blood vessels at locations such as the user's wrist, and calculates two key vital signs: heart rate (HR) and blood oxygen saturation (SpO2). The MEMS microphone 112 is optimized to capture faint sounds produced by the user's breathing, particularly high-frequency wheezing sounds highly specific during asthma attacks. The skin conductivity sensor 113 reflects changes in sweat gland secretion caused by sympathetic nervous system activity by applying a weak, imperceptible voltage to the skin and measuring changes in skin conductivity. These changes are significantly altered under physiological or psychological stress (such as difficulty breathing). Optionally, the multimodal sensor module 110 may also integrate a temperature sensor to monitor skin temperature as an auxiliary indicator.

[0032] The communication module 140 can be a Bluetooth module, a Wi-Fi module, or a cellular network module (such as NB-IoT or LTE-M), used to wirelessly transmit data monitored by the device, alarm information, and device status information to the user's smartphone application or cloud server. It can also be used to automatically send a distress signal containing the user's geographical location information (e.g., obtained through an integrated GPS module) to preset emergency contacts or emergency centers in emergency situations.

[0033] The user interface 150 serves as a medium for information interaction between the user and the device. In one embodiment, the user interface 150 includes a display screen 151, a vibration motor 152, and a speaker 153. The display screen 151, such as an OLED or LCD touchscreen, is used to display time, physiological data, device status, and a confirmation interface before triggering intervention. The vibration motor 152 and speaker 153 are used to issue a strong alarm to the user through touch and hearing to attract the user's attention in an emergency.

[0034] The miniaturized injection actuator 160 is the physical carrier for achieving automated drug delivery and is designed as a highly integrated and replaceable module. Please refer to [further details omitted]. Figure 4 In one embodiment, the miniaturized injection actuator 160 includes a replaceable microneedle cartridge module 400. The microneedle cartridge module 400 is designed to resemble a SIM card or SD card, allowing for easy ejection and replacement by the user from the main body of the device. The microneedle cartridge module 400 contains a sealed medication reservoir 410 for sterile storage of a single dose of emergency medication (such as adrenaline or salbutamol solution). A thin injection needle 420 is initially fully retracted within the cartridge, with its end connected to a miniature piston 430. At the core of the cartridge module is a drive mechanism 440, which can be implemented using various technologies. For example, a shape memory alloy (SMA) wire can be used; when electrically heated, it undergoes a phase change and contracts, generating a linear driving force that ejects the needle 420 to pierce the subcutaneous tissue and simultaneously pushes the piston 430 to complete the medication injection. Alternatively, a miniature gas generator (e.g., based on a chemical reaction or electric ignition) can be used to instantaneously generate a small amount of high-pressure gas upon receiving a trigger signal, driving the piston and needle. After injection, a reset spring mechanism 450 automatically retracts the needle 420 into the cartridge, preventing secondary injury and needlestick contamination, ensuring safe use.

[0035] The following will combine Figure 2 and Figure 3 This application describes a method for integrating asthma attack monitoring and automatic drug delivery functions, as provided in the embodiments of this application. Figure 2 This is the overall flowchart of the method. Figure 3 This is a conceptual diagram of a phase-locked loop model for physiological states.

[0036] S100: Establish personalized homeostatic signatures that characterize the user's baseline health and physiology.

[0037] After a user first uses the device or after an initialization period, a mathematical model is created for each user that accurately describes their physiological characteristics in a healthy state; this is known as a personalized homeostatic signature. This model constitutes Figure 3 The reference signal 310 of the physiological state phase-locked loop (PPL) serves as the benchmark for all subsequent anomaly detection. To establish this model, the wearable device 100 continuously collects data through the multimodal sensor module 110 during periods when the user is certain that they are in a healthy, calm state (e.g., healthy periods manually marked by the user via a mobile app).

[0038] The collected data forms a multidimensional time series. For example, at any time t, the collected data points can be represented as a vector. ,in It refers to blood oxygen saturation (SpO2). It is heart rate (HR). It is skin conductance (GSR). It is the wheezing power feature extracted from the respiratory audio signal. Processor 120 processes the large amount of data vectors collected from healthy individuals. Statistical analysis is performed to construct a multidimensional statistical model as the personalized steady-state feature signature. In a preferred embodiment, the multidimensional statistical model is defined as a multidimensional Gaussian distribution, whose core parameter is a mean vector. and a covariance matrix The mean vector This represents the average level of various physiological indicators when the user is healthy, while the covariance matrix... This describes the range of fluctuations of these indicators (diagonal elements) and the interrelationships between them (off-diagonal elements). This ability to capture correlations between variables is crucial because physiological systems are highly coupled; normal fluctuations in one indicator may be accompanied by corresponding changes in another, and the covariance matrix can reflect this normal coupling relationship.

[0039] For example, assuming that after collecting and analyzing user A's health status data for a period of one week, processor 120 constructs the following personalized steady-state feature signature. The signature consists of a mean vector. and a covariance matrix Composition. Assume the feature vector contains four dimensions: blood oxygen saturation. (Unit: %), Heart Rate (HR) (unit: beats / minute, bpm), Skin Conductance (GSR) (unit: microSiemens, μS), and Wheezing Characteristic Power (Unit: any unit, au).

[0040] The mean vector It may be calculated as: This means that user A, in a healthy and calm state, has an average blood oxygen saturation of 98.5%, an average heart rate of 72.0 bpm, an average skin conductance of 5.5 μS, and almost no wheezing in their breathing (background noise level of 0.2 au).

[0041] The covariance matrix It may be calculated as: The diagonal elements of this covariance matrix represent the variance of each physiological indicator. For example, 0.25 in the first row and first column indicates... The variance is 0.25, meaning its standard deviation is... This indicates that user A's blood oxygen level is usually around [value missing]. The heart rate fluctuates slightly within a certain range. The 16.0 in the second row and second column indicates that the variance of HR is 16.0 and the standard deviation is 4.0 bpm, meaning that the heart rate is typically within a certain range. Fluctuations within the bpm range. Off-diagonal elements represent the covariance between indicators. For example, -1.5 in the first row and second column indicates... A slight negative correlation exists between heart rate and HR, which is physiologically plausible (e.g., a slight decrease in blood oxygenation may be accompanied by a compensatory slight increase in heart rate). The 2.0 in the second row, third column indicates a positive correlation between heart rate and skin conductance, reflecting the co-regulation of the autonomic nervous system. This complete... This creates a deep and quantitative profile of user A's health status, which becomes a solid foundation for subsequent judgments.

[0042] S200: Acquire the user's multimodal biosignal vector stream.

[0043] During normal device operation, S200 is continuously executed. The multimodal sensor module 110 continuously collects various physiological signals from the user at a preset sampling frequency and sends these raw data streams to the processor 120. The processor 120 preprocesses the received raw data, including filtering, noise reduction, and feature extraction. For example, a bandpass filter is applied to the PPG signal to eliminate motion artifacts and baseline drift, and then the beat-by-beat heart rate and windowed mean oxygen saturation are calculated using a peak detection algorithm. For the audio signal collected by the MEMS microphone, a high-pass filter is applied to filter out heartbeat sounds and low-frequency ambient noise, and then a short-time Fourier transform (STFT) is performed to convert it to the frequency domain. The signal energy or power spectral density in a specific high-frequency band (e.g., 400Hz-2000Hz, which is the typical frequency band for wheezing) is calculated as a characteristic of wheezing. The GSR signal is low-pass filtered to remove high-frequency noise. After preprocessing and feature extraction, at each time point... Processor 120 will generate a personalized steady-state feature signature. Real-time multimodal biological signal vectors with completely consistent dimensions and units These continuously generated vectors It constitutes Figure 3 The input signal 320 is the multimodal biological signal vector flow.

[0044] For example, during the continuous monitoring process of wearable device 100, suppose at a certain moment... The processor 120 collects and processes data through the multimodal sensor module 110, generating a real-time multimodal biosignal vector. The structure of this vector is consistent with the model structure defined in S100. Assume that user A is currently in the early stages of an asthma attack, and their physiological indicators are beginning to show abnormalities.

[0045] At any moment The obtained vector may be: Each component of this vector represents a specific physiological measurement value: The first component, This value is significantly lower than User A's baseline health value of 98.5%, which is a sign of hypoxia.

[0046] The second component, HR = 95.0 bpm. This value is significantly higher than user A's healthy baseline mean of 72.0 bpm, indicating that the heart is beating faster to compensate for insufficient oxygen supply, which is a typical stress response.

[0047] The third component, GSR = 15.0 μS. This value is much higher than user A's healthy baseline mean of 5.5 μS, reflecting that due to difficulty breathing and anxiety, the sympathetic nervous system is highly activated, leading to increased sweat gland secretion.

[0048] The fourth component au. This value is significantly higher than the background noise level of 0.2au at a healthy baseline, clearly indicating narrowing of the user's airway and producing typical wheezing sounds.

[0049] This vector As a whole, it paints a physiological picture that is quite different from the user's health status. It is no longer centered around the health baseline mean. The fluctuations are not minor, but rather deviations occurring simultaneously across multiple dimensions that clearly align with the pathophysiological changes characteristic of an asthma attack. This specific, quantified real-time data vector will serve as input for the next computational step, assessing the degree of anomaly in the current state.

[0050] S300: Determines the steady-state perturbation scalar that characterizes the degree of deviation.

[0051] This step corresponds to Figure 3 The function of the phase detector 330 in the physiological state phase-locked loop model. The coherence deviation discriminator algorithm module runs internally in the processor 120. The input to this module is the real-time signal vector generated by S200. Personalized steady-state signature established with S100 The core task of the coherence deviation discriminator is to calculate... Compared to The distance to the centers of the described health status distribution. To effectively measure this multi-dimensional distance that considers the correlation between variables, this embodiment uses Mahalanobis distance as the core calculation method. The formula for Mahalanobis distance is: in, It is the inverse of the covariance matrix. The result of Mahalanobis distance calculation. This is a scalar value representing the statistical distance of the current physiological state vector from its health center; the larger the value, the more abnormal the state. This value is defined as the steady-state perturbation scalar. In this embodiment, this scalar value is used as the core metric of the steady-state perturbation signal. It is understood that in other possible implementations, other mathematical methods can also be used to quantify the degree of deviation. This output, whether a scalar or a vector, will be passed to the next processing stage.

[0052] For example, the personalized steady-state feature signature of user A as defined in S100 is used. and the real-time signal vector obtained in S200 Let's take a closer look at the process of calculating S300.

[0053] First, calculate the difference vector. : This difference vector visually shows the deviation of each physiological indicator.

[0054] Next, we need to calculate the inverse of the covariance matrix. This is a standard linear algebra operation, and its result is a... The matrix (for simplification, the complex values ​​of the inverse matrix are not shown).

[0055] Then, the difference vector and inverse covariance matrix Substituting into the formula for the square of the Mahalanobis distance: This is a matrix multiplication operation: Ultimately, a scalar value is obtained.

[0056] Assuming that after calculation, we get We obtain a certain numerical form of the matrix and perform the matrix multiplication described above, ultimately yielding a specific squared Mahalanobis distance value, for example... .

[0057] Finally, taking its square root yields the Mahalanobis distance: The value 6.77 is the calculated steady-state perturbation scalar, a dimensionless statistical distance. Under the assumption of a multidimensional normal distribution, this value can be compared with the critical value of the chi-square distribution to determine its statistical significance. Such a large Mahalanobis distance value (usually greater than 3 is considered an outlier) strongly indicates that the current physiological state vector... It is highly unlikely that this deviation comes from user A's health and physiological status distribution. This quantified deviation value of 6.77 will be used as a core input and passed to the asynchronous risk escalation engine for the next step of risk assessment.

[0058] S400: An asthma attack risk index generated and quantified based on steady-state perturbation scalars.

[0059] This step first applies a loop filter 340 to the steady-state perturbation sequence output by the S300. The filter is designed to smooth out disturbance signals and filter out isolated, transient peaks caused by sudden physical movements, brief emotional fluctuations, etc., to avoid false alarms. For example, a moving average filter or a low-pass filter can be used.

[0060] The filtered, smoothed perturbation signal D'_{M}(t) is fed into an asynchronous risk escalation engine, which corresponds to... Figure 3 The engine utilizes a voltage-controlled oscillator (VCO) 350. Its function is to transform the current deviation into a more informative, dynamically changing asthma attack risk index R(t), whose value is normalized to [0,1], where 0 represents perfect health and 1 represents extremely high risk or a confirmed attack. The calculation of the risk index depends not only on the current perturbation value. The size of the sample will also be analyzed, along with its temporal evolution trend.

[0061] In one embodiment, the risk index The computation function is a comprehensive model: in, It is the first time derivative of the disturbance value, representing the "velocity" of the deviation; It is the second time derivative, representing the acceleration of the deviation. These are preset weighting coefficients. This model implies that a state where the deviation is small but rapidly increasing (large velocity and acceleration terms) may have its risk index rapidly amplified, thus enabling a rapid response to a sudden deterioration of the condition. Understandably, this assessment method, which integrates current value, velocity, and acceleration, allows the risk index R(t) to more sensitively and proactively reflect the true degree of danger of the condition.

[0062] Continuing with the S300 example, assuming the coherence deviation discriminator outputs a series of time-varying steady-state perturbation scalar values, focus on time... arrive Data from three consecutive time points, assuming a sampling interval of 1 second.

[0063] At any moment The calculated steady-state disturbance scalar is .

[0064] At any moment The calculated steady-state disturbance scalar is .

[0065] At any moment The calculated steady-state disturbance scalar is .

[0066] First, a loop filter (e.g., a simple 3-point averaging filter) smooths these values, assuming that the smoothed values ​​D'_{M}(t) are close to the original values.

[0067] Next, the asynchronous risk upgrade engine calculates approximate values ​​for the first and second derivatives.

[0068] First derivative (velocity) in The approximate value is: (Unit: distance / second) This positive value indicates that the deviation is increasing rapidly.

[0069] The second derivative (acceleration) in The approximate value is: (Unit: distance / second²) This positive value indicates that the rate of deviation is accelerating.

[0070] In one embodiment, the risk index The model includes a set of personalized calibration factors. These personalized calibration factors are determined during the calibration process described in the adaptive calibration method for model parameters and thresholds, by analyzing the calibration dataset. It is determined by the statistical characteristics.

[0071] Specifically, the three calibration factors are: Feature perturbation scale ( ): A dimensionless scalar used to normalize steady-state perturbation values. For example, it can be set as the steady-state perturbation scalar for all "positive samples" in the calibration dataset. The 95th percentile.

[0072] Characteristic perturbation rate ( (): Dimension is [time] -1 , which is used to normalize the first derivative of the perturbation. For example, it can be set to the 95th percentile of the first derivative of all “positive samples” in the calibration dataset.

[0073] Characteristic perturbation acceleration ( (): Dimension is [time] -2 , which is used to normalize the second derivative of the perturbation. For example, it can be set to the 95th percentile of the second derivative of all “positive samples” in the calibration dataset.

[0074] Therefore, the core calculation function of the risk index was revised to: The overall score obtained This is a raw score that can take any real value. To convert it into a risk index within the interval [0, 1] that is easy to understand and has a threshold for setting, this embodiment uses a sigmoid function as a non-linear mapping function. This function can smoothly normalize the overall score, and its mathematical form is: Continuing with the example of S300, and assuming that the following calibration factors were determined during the calibration procedure: , , . but The calculation becomes:

[0075]

[0076] Substitute into the sigmoid function:

[0077] Calculated risk index The index is as high as 0.71, not only because of the large degree of deviation, but also because of the high speed and acceleration of the deviation. The system's comprehensive judgment indicates that this is a very dangerous and rapidly deteriorating state. This high-risk index will trigger the next intervention decision.

[0078] S500: When the risk index meets the preset intervention trigger conditions, automatic injection of medication is performed.

[0079] Processor 120 continuously monitors the asthma attack risk index R(t) generated by S400. The system has at least one preset intervention trigger threshold. When R(t) first exceeds this threshold (e.g., The system determined that a coherent desynchronization event had occurred, meaning that the physiological state had deviated significantly from the healthy baseline. The PPL model entered a Loss of Lock 360 state, which is technically interpreted as an asthma attack.

[0080] To ensure safety, the system initiates a user intervention confirmation procedure before direct injection. Processor 120, through user interface 150, drives vibration motor 152 to vibrate at high intensity, speaker 153 sounds an alarm, and display screen 151 shows a prominent warning interface and a large "Cancel" button. The system starts a countdown, for example, 15 seconds. Within these 15 seconds, if the user is conscious and believes it is a false alarm (e.g., having just completed a set of extreme exercise), they can press the "Cancel" button to stop the subsequent operation. If the countdown ends and processor 120 does not receive any abort command, it will consider the user to be in an unresponsive emergency state. At this time, processor 120 sends an encrypted execution command to miniaturized injection actuator 160. Upon receiving the command, actuator 440 is immediately activated, completing the entire process of needle ejection, drug injection, and needle retraction within a short time, completing a drug intervention for the user. After injection, communication module 140 immediately sends the emergency event, injection completion status, and the user's current GPS location information to a pre-set emergency contact.

[0081] In a preferred embodiment, to enable the invention to provide optimal monitoring performance for each user, the wearable device 100 supports a personalized calibration process to determine key parameters of the asynchronous risk escalation engine in S400, including weighting coefficients. and the intervention trigger threshold described in S500 This calibration procedure is used to ensure the risk assessment model is adapted to the individual physiological characteristics of the user. Those skilled in the art will understand that this calibration procedure can be completed under guidance during the initial device pairing or under the supervision of a medical professional. The calibration procedure may include: To train and validate the risk assessment model, a labeled dataset containing samples of the user's health status and simulated asthma attacks needs to be constructed. .

[0082] While the user is in a relaxed, still, and healthy state, the system continuously collects multimodal biosignal vector streams for at least 30 minutes. These data are labeled as "negative samples" (labeled 0).

[0083] Since real asthma attacks are difficult to reproduce on demand, this method employs a controlled physiological stress test to generate "positive samples" (labeled 1) simulating attacks. For example, users are guided to perform a short, standardized breath-holding test (e.g., holding their breath for 30-45 seconds) or rapid step exercises within a safe range. These activities induce a range of physiological responses similar to early asthma attacks, including a transient decrease in blood oxygenation and a significant increase in heart rate and skin conductance. The multimodal biosignal vector streams acquired by the system during these stress tests are processed and labeled as "positive samples." Sufficient positive sample data can be obtained by repeating the stress test multiple times.

[0084] Secondly, we need to find an optimal set of weighting coefficients. This makes the risk index function It can distinguish between a healthy state and a simulated seizure state to the greatest extent possible.

[0085] For calibration dataset For each time series data point, the system first calculates its corresponding steady-state disturbance scalar sequence. and the first derivative of the sequence and second derivative These three time series constitute the feature set used for optimization.

[0086] Define an objective function to evaluate model performance, such as the log-likelihood loss function in logistic regression or the hinge loss function in support vector machines. This objective function measures the performance of a model given a set of weights. In this case, the difference between the model prediction and the true label (0 or 1).

[0087] A systematic search or optimization algorithm is used to find the optimal weight combination that minimizes the objective function.

[0088] Option 1 (Grid Search): Define a reasonable search range and step size for each weight coefficient (e.g., The search is performed within the interval [0.1, 1.0] with a step size of 0.1. The system will traverse all possible weight combinations, calculate the objective function value for each combination, and finally select the weight combination that minimizes the objective function value as the optimal solution.

[0089] Option 2 (Gradient Descent): Treat the risk index model as a simple neural network layer, and use gradient descent or its variants (such as the Adam optimizer) to minimize the loss function. The optimal weight coefficients are automatically learned through multiple iterations.

[0090] For example, using a grid search algorithm, the system might discover that for user B, the optimal weight combination is... This indicates that the "current value" and "speed" of the user's physiological state deviation are more important for risk assessment.

[0091] After determining the optimal weighting coefficients, a reasonable intervention trigger threshold needs to be set for this user. The optimized weights are then applied to the calibration dataset. The above calculates the corresponding risk index for each time point in the dataset. .

[0092] Using all calculated risk indices as the classifier's output scores, the true positive rate (TPR, i.e., sensitivity) and false positive rate (FPR, i.e., 1-specificity) at each threshold are calculated by sliding different thresholds within the [0, 1] interval. Plotting these (FPR, TPR) pairs on a two-dimensional coordinate system generates a personalized ROC curve for the user. Based on preset clinical requirements or user preferences, an optimal operating point is selected on the ROC curve to determine the threshold. .

[0093] If the goal is to avoid missing any potential seizures (e.g., for users with a history of severe seizures), the highest point on the curve where the TPR is close to 1.0 (e.g., TPR > 0.99) can be selected, even if this may result in a slightly higher false positive rate. The risk index value corresponding to this point is then set as... .

[0094] You can choose the point on the ROC curve closest to the top left corner (0, 1) (the point with the largest Youden index), which represents the best balance between sensitivity and specificity.

[0095] If users are very concerned about false triggers (e.g., severe drug side effects), they can choose a point on the curve with an extremely low FPR (e.g., FPR < 0.01).

[0096] For example, by analyzing the ROC curve of user B, it was found that the system achieved a sensitivity of 99.5% and a specificity of 98.0% when the risk index was 0.88. Therefore, the system can trigger the intervention threshold for this user. Set to 0.88.

[0097] Through the above calibration process, this invention customizes an optimal set of model parameters and decision thresholds for each user, ensuring high sensitivity and high specificity of monitoring.

[0098] To further improve response speed, this system can integrate a predictive feedforward control module. This module, based on the historical time series of the risk index R(t), uses a Kalman filter or a lightweight recurrent neural network (RNN) model to predict the evolution trajectory of R(t) within the next few seconds (e.g., the next 15 seconds). If the predictive model outputs results indicating that the risk index will exceed the intervention threshold with a very high probability (e.g., 95%) within a very short time (e.g., within 10 seconds), then... Even if the current R(t) has not yet reached the threshold, the system can initiate the user intervention confirmation procedure in advance (alarms and countdowns in S500). This is equivalent to giving the user valuable early warning time, allowing them to react or prepare before the situation completely gets out of control, reflecting an evolution from passive response to proactive prediction.

[0099] To ensure personalized steady-state signature features To ensure long-term effectiveness, the system can also include a dynamic calibration mechanism. This mechanism corresponds to the "calibration" process in the PPL model. When the system determines that the user is in a healthy "locked-out" state (i.e., the risk index R(t) remains below a very low steady-state threshold, such as 0.1), the processor 120 will update the weights at a very small learning rate. (For example, ), using the latest collected health data Update slowly Parameters (mean vector) Covariance Matrix The update rule can use the exponential moving average method: The covariance matrix can also be updated using a similar recursive method. This continuous update process ensures that the model can automatically adapt to long-term physiological baseline drift caused by factors such as seasonal changes, weight gain or loss, and improved physical fitness, thereby maintaining the long-term accuracy of monitoring and achieving true adaptive personalization.

[0100] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0101] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and not to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

[0103] Through the above description of the embodiments, those skilled in the art can clearly understand that, for the sake of convenience and brevity, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0104] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another apparatus, or some features may be ignored or not executed. Furthermore, the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0105] The units described as separate components may or may not be physically separate. A component shown as a unit can be one physical unit or multiple physical units; that is, it can be located in one place or distributed in multiple different places. Depending on actual needs, some or all of the units can be selected to achieve the purpose of this embodiment.

[0106] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit described above can be implemented in hardware.

[0107] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method integrating asthma attack monitoring and automated drug delivery, characterized in that, include: Acquire the user's multimodal biosignal vector stream; Establish a personalized steady-state feature signature that represents the user's healthy physiological baseline, wherein the personalized steady-state feature signature is used to represent the user's healthy physiological baseline; A steady-state perturbation scalar is determined to characterize the deviation between the physiological state corresponding to the multimodal biological signal vector flow and the personalized steady-state feature signature; Based on the steady-state perturbation scalar, a quantitative asthma attack risk index is generated; When the asthma attack risk index meets the preset intervention trigger conditions, automatic injection of medication is performed.

2. The method according to claim 1, characterized in that, The acquisition of the user's multimodal biosignal vector stream includes: The user's heart rate and blood oxygen saturation signals are acquired through a photoplethysmography (PPG) sensor. The user's breathing audio signal is acquired through a microelectromechanical system microphone; The user's skin conductivity signal is acquired through a skin conductivity sensor.

3. The method according to claim 2, characterized in that, The method further includes: Spectral analysis was performed on the respiratory audio signal to extract wheezing features associated with asthma attacks.

4. The method according to claim 1, characterized in that, The establishment of a personalized homeostatic signature characterizing the user's health physiological baseline includes: Multimodal biosignal data are continuously collected while the user is in a healthy state; Based on the multimodal biological signal data, a multidimensional statistical model is constructed to describe the statistical distribution characteristics of each signal component, serving as the personalized steady-state feature signature.

5. The method according to claim 1, characterized in that, The determination of the steady-state disturbance scalar includes: The multimodal biological signal vector stream is input into a coherence deviation discriminator; The coherence deviation discriminator calculates the Mahalanobis distance of the multimodal biological signal vector flow in the multidimensional feature space defined by the personalized steady-state feature signature, and uses the Mahalanobis distance as the steady-state perturbation scalar.

6. The method according to claim 1, characterized in that, The generation of a quantified asthma attack risk index based on the steady-state perturbation scalar includes: The steady-state perturbation scalar is input into an asynchronous risk escalation engine; The asynchronous risk upgrade engine generates the asthma attack risk index based on the time series of the steady-state perturbation scalar.

7. The method according to claim 6, characterized in that, The asynchronous risk upgrade engine also generates the asthma attack risk index based on the first and second time derivatives of the steady-state perturbation scalar.

8. The method according to claim 1, characterized in that, The method further includes: Before performing automated injection and drug delivery, initiate a user intervention confirmation procedure; The user intervention confirmation procedure includes issuing an alarm and receiving a user's stop command within a preset time. If no user's stop instruction is received within the preset time, the automatic injection drug delivery is performed.

9. The method according to claim 1, characterized in that, The method further includes: Based on the prediction of the future evolution trajectory of the asthma attack risk index, the user intervention confirmation procedure is initiated in advance. And / or, when the asthma attack risk index is lower than a preset steady-state threshold, the multimodal biosignal vector stream is used to update the personalized steady-state feature signature with a preset update weight, so as to achieve dynamic calibration of the personalized steady-state feature signature.

10. A wearable device integrating asthma attack monitoring and automatic drug delivery functions, characterized in that, include: A multimodal sensor module is used to acquire the user's multimodal biosignal vector stream; processor; A memory, wherein a computer program is stored; Miniaturized injection actuator; When the processor is configured to execute the computer program, it implements the method as described in any one of claims 1-9.

Images