Hypoxic response reaction evaluation method based on heart rate deceleration capacity and HRV dynamic evaluation

Abstract

The application discloses a hypoxic response reaction evaluation method based on heart rate deceleration force and HRV dynamic evaluation, and relates to the field of biomedical signal processing. The method synchronously collects electrocardiogram signals and environmental and blood oxygen parameters to dynamically correct the risk baseline; frequency domain, time domain and heart rate deceleration force characteristics in the electrocardiogram signals are extracted, and a double-channel cross determination logic is established; the determination result and the current blood oxygen saturation are input into an edge lightweight neural network to infer and output a risk stratification; according to the stratification result, a wearable actuator is controlled to start multi-target point stratified sequential stimulation containing heat and vibration, and an individualized heat therapy prescription is adaptively updated based on an intervention effectiveness index, so that the double-channel collaborative control of oxygen therapy and heat therapy is finally realized. The application effectively solves the defects of traditional early warning lag and the inability to dynamically correct in a network-free environment, constructs a closed loop from accurate monitoring to multi-target point collaborative intervention, and significantly improves the adaptability and anti-risk ability of the organism to the hypoxic environment.

Description

Technical Field

[0001] This invention belongs to the field of biomedical signal processing, specifically relating to a method for assessing hypoxia response based on dynamic evaluation of heart rate deceleration force and HRV. Background Technology

[0002] With the deepening of research in high-altitude medicine, real-time monitoring and early warning of the human physiological state under hypoxic conditions at high altitudes have become a key technology for ensuring the safety of personnel working at high altitudes. The autonomic nervous system, as the core regulator of cardiovascular function, directly affects the body's stress response when coping with extreme environments such as hypoxia and cold. Traditional electrophysiological analysis techniques are widely used to assess the interaction between the sympathetic and vagus nerves. Quantitative analysis of indicators such as heart rate variability provides important evidence for early screening of poor altitude adaptation and the risk of malignant arrhythmias.

[0003] Among these, hypoxia response assessment based on multidimensional physiological parameters is a core direction for improving health protection at high altitudes. This involves real-time acquisition of the body's autonomic nervous system compensatory state through in-depth analysis of electrocardiogram signals. This technology primarily extracts time-domain and frequency-domain features to reflect dynamic changes in cardiac regulatory function and attempts to combine external environmental parameters for baseline calibration. It then aims to achieve automated regulation of physiological states through physical interventions to enhance the body's tolerance to high-altitude environments. However, existing assessment schemes based on conventional physiological indicators are mostly limited to unipolar monitoring, failing to form an effective assessment and intervention loop. Furthermore, traditional methods rely excessively on the low-frequency to high-frequency power ratio as a regulatory basis, lacking deep integration of heart rate deceleration force indicators that are more specific to changes in vagal tone, resulting in a lag in the assessment results when identifying precursors to sympathetic storms. In addition, conventional systems often use a single threshold for judgment and lack local real-time decision-making capabilities, making it difficult to dynamically correct for individual differences and environmental changes in high-altitude, signal-free environments. The lack of multi-target precise intervention and real-time linkage mechanisms with physiological indicators makes it difficult for existing technologies to implement timely and individualized antagonistic regulation of complex physiological risks. Summary of the Invention

[0004] The present invention provides a hypoxia response assessment method based on heart rate deceleration force and HRV dynamic assessment, in order to solve the technical problems of the current method, which has obvious lag in the identification of sympathetic storm precursors, difficulty in dynamic correction based on individual differences and environmental changes in high-altitude areas without nets and complex climatic environments, and difficulty in implementing timely, effective and highly individualized antagonistic regulation of complex physiological risks.

[0005] To solve the above problems, the technical solution adopted by the present invention is as follows:

[0006] A method for assessing hypoxia response based on dynamic assessment of heart rate deceleration force and HRV includes:

[0007] Step 1: Synchronously acquire electrocardiogram signals, environmental parameters, and pulse oxygen saturation. Calculate the baseline autonomic threshold deviation using the environmental parameters and pulse oxygen saturation. Dynamically correct the risk judgment baseline for heart rate variability and heart rate deceleration force to obtain the environmentally corrected dynamic threshold.

[0008] Step 2: Extract frequency domain indices, time domain indices, and heart rate deceleration force characteristics from the electrocardiogram signal. The frequency domain indices include the ratio of low-frequency power to high-frequency power, the time domain indices include the root mean square of the difference between adjacent normal heartbeat intervals, and the heart rate deceleration force characteristics include the heart rate deceleration force value, the rate of change of heart rate deceleration force over time, and the coefficient of variation of heart rate deceleration force.

[0009] Step 3: Establish a dual-channel cross-determination logic based on heart rate deceleration force and heart rate variability, and make a logical determination by combining the frequency domain index, the time domain index, the heart rate deceleration force feature, and the environmentally corrected dynamic threshold.

[0010] Step 4: Combine the determination result of Step 3 with the current pulse oxygen saturation, input it into the lightweight micro machine learning neural network deployed in the microcontroller for inference, and output multi-level risk stratification results;

[0011] Step 5: Based on the multi-level risk stratification results, control the wearable actuator to start a stratified sequential stimulation program in the preset target area. The stratified sequential stimulation program includes dynamically adjusted thermal stimulation and vibration stimulation.

[0012] Step 6: After a single intervention, calculate the intervention efficacy index and use an online learning algorithm to adaptively update the stimulation parameters in the individualized thermotherapy prescription based on the intervention efficacy index;

[0013] Step 7: Based on the results of the multi-level risk stratification, the physical stimulation output of the wearable actuator and the oxygen supply flow of the gradient oxygen supply device are uniformly scheduled through the central controller to achieve dual-channel coordinated control of oxygen therapy and thermotherapy.

[0014] Further, the environmental parameters include altitude, temperature, and humidity; the deviation of the basic autonomic threshold is calculated using a preset linear regression model. The input of the linear regression model includes the missing values ​​corresponding to the altitude, temperature, relative humidity, and pulse oxygen saturation, and is assigned preset weighting coefficients. The output is the deviation of the basic autonomic threshold; in step 1, the environmentally corrected dynamic threshold is updated periodically according to a preset correction step size.

[0015] Furthermore, the phase-ordered signal averaging algorithm is used to calculate the heart rate deceleration force value in real time. Specifically, this includes: using the cardiac cycle length as an alignment reference to identify heart rate deceleration segments; extracting a preset number of cardiac cycles before and after a selected deceleration point as a data segment; and performing alignment averaging on the overlapping data segments to obtain the heart rate deceleration force value. The calculation method of the heart rate deceleration force feature specifically includes: generating the rate of change of the heart rate deceleration force over time by calculating the first derivative of the continuous heart rate deceleration force values ​​within a preset sliding window; and generating the heart rate deceleration force variation coefficient by calculating the ratio of the standard deviation to the mean of the heart rate deceleration force values ​​within a predetermined period.

[0016] Furthermore, the dual-channel cross-judgment logic specifically includes: when the heart rate deceleration force value is lower than the environmentally corrected dynamic threshold, the rate of change of the heart rate deceleration force over time is continuously negative, and the ratio of low-frequency power to high-frequency power rises synchronously to above the environmentally corrected dynamic threshold, it is determined to be a precursor to a sympathetic storm; when the heart rate deceleration force value decreases, but the ratio of low-frequency power to high-frequency power remains stable or decreases, it is determined to be a period of vagal withdrawal compensation; when the root mean square of the difference between adjacent normal heartbeat intervals decreases continuously and the coefficient of variation of the heart rate deceleration force increases beyond a preset threshold, the warning level is automatically upgraded.

[0017] Furthermore, the lightweight micro-machine learning neural network adopts a convolutional neural network structure and is deployed in the microcontroller after model pruning and fixed-point quantization processing. The input vector of the lightweight micro-machine learning neural network is a three-dimensional tensor containing multiple consecutive sets of heart rate deceleration force values, the average ratio of low-frequency power to high-frequency power within a predetermined time period, and the current instantaneous value of pulse oxygen saturation. The output of the lightweight micro-machine learning neural network includes a heart rate deceleration force risk score, a predicted trend value of the ratio of low-frequency power to high-frequency power, and a probability of decreased pulse oxygen saturation. The microcontroller outputs the multi-level risk stratification result after weighted fusion of the multi-path inference results.

[0018] Furthermore, the multi-level risk stratification results include: Level 1 risk is vagal withdrawal and mild autonomic imbalance; Level 2 risk is sympathetic storm precursor with a continuous decrease in heart rate deceleration and significant sympathetic activation; Level 3 risk is high risk of ventricular arrhythmia with a malignant electrocardiographic trend or extremely dangerous heart rate deceleration.

[0019] Furthermore, the preset target area includes the vagus nerve skin distribution area on the back of the neck, the concha area, and the corresponding locations of the Neiguan (PC6) and Dazhui (GV14) acupoints. The step of controlling the wearable actuator to initiate a stratified sequential stimulation program in the preset target area based on the multi-level risk stratification results specifically includes: under a first-level risk response, initiating the thermal stimulation and vibration stimulation of the Neiguan acupoint, wherein the vibration stimulation is low-frequency vibration; under a second-level risk response, simultaneously activating the thermal stimulation of the concha area and the carotid sinus area and dynamically switching within a preset temperature range, and initiating the mid-frequency vibration stimulation; under a third-level risk response, based on the second-level intervention, increasing the thermal stimulation of the Dazhui acupoint to a predetermined high temperature level, and performing cyclic scanning vibration stimulation within a preset frequency range.

[0020] Furthermore, the intervention efficacy index is calculated by comprehensively considering multiple indicators, including the recovery rate of heart rate deceleration force before and after the intervention, the decrease rate of the ratio of low-frequency power to high-frequency power before and after the intervention, the root mean square recovery rate of the difference between adjacent normal heartbeat intervals before and after the intervention, and the stability of pulse oxygen saturation during the intervention. These indicators are then accumulated after being assigned preset weights. The adaptive updating of stimulation parameters in the individualized hyperthermia prescription using an online learning algorithm includes establishing an individual hyperthermia response profile, classifying individuals into high-response, medium-response, or low-response types based on the intervention efficacy index obtained after multiple consecutive interventions, and correcting the temperature gradient and vibration spectrum parameters in the individualized hyperthermia prescription using the gradient descent method.

[0021] Furthermore, the dual-channel coordinated control of oxygen therapy and thermotherapy based on the multi-level risk stratification results specifically includes: under the second level of risk, the system sets the initial oxygen flow rate to a first predetermined flow rate, calculates the blood flow changes caused by thermotherapy through a metabolic compensation algorithm, and dynamically adjusts the timing of pulse oxygen supply to deliver oxygen at the beginning of the inspiratory phase; under the third level of risk, the system forcibly executes full-channel parallel logic, switches the oxygen supply flow rate to a continuous oxygen supply mode higher than the first predetermined flow rate, and locks the flexible heating array in the wearable actuator that performs the thermal stimulation to the maximum safe operating temperature.

[0022] Furthermore, the method of the present invention also includes a high-altitude environment adaptation self-evaluation step: by analyzing the baseline drift trajectory of heart rate deceleration force and the frequency of triggering physical interventions in a predetermined past period, a high-altitude adaptation score is generated using an evaluation model; if the high-altitude adaptation score is lower than a preset safety score, a prompt instruction to reduce physical labor or perform assisted oxygen inhalation is output through the associated terminal.

[0023] Compared with the prior art, the beneficial effects of the present invention are:

[0024] (1) This invention integrates highly specific heart rate deceleration force characteristics (DC) with traditional heart rate variability (HRV) characteristics to construct a dual-channel cross-judgment logic. Combined with a judgment baseline that is dynamically corrected in real time based on environmental and blood oxygenation parameters, it eliminates individual misjudgments caused by external environmental fluctuations. At the same time, by combining the real-time application of the phase-order signal averaging algorithm, it effectively solves the lag problem of traditional spectrum analysis in non-stationary ECG signal processing, significantly advancing the identification and warning time of sympathetic storm precursors, and securing a critical physiological window for subsequent physical intervention.

[0025] (2) This invention innovatively combines micro machine learning neural networks with edge computing technology. After model pruning and fixed-point quantization, it is deployed in a microcontroller. It can realize local real-time solution of multi-level risk stratification under extremely low power consumption, completely eliminating the risk of early warning and intervention failure caused by poor cloud communication in high-altitude, network-free, and high-latency environments, and ensuring the long-term stable operation of the equipment.

[0026] (3) Based on the risk stratification results, this invention uses wearable actuators to perform stratified sequential stimulation, including thermotherapy and vibration, on specific acupoints and nerve distribution areas. This multi-target synergistic physical intervention mechanism can adjust the stimulation intensity, location and mode in real time and adaptively according to the risk level, directly enhance vagal nerve activity and improve local microcirculation, and achieve precise matching between stimulation intensity and the body's real-time physiological state.

[0027] (4) After a single closed-loop intervention, this invention quantifies the intervention efficacy index using multidimensional physiological indicators and uses an online learning algorithm to create a personal thermotherapy response profile for the user. This mechanism can automatically classify users according to their sensitivity to physical stimuli and dynamically correct intervention prescription parameters using a feedback mechanism, so that the system can automatically evolve the optimal individualized intervention strategy as the user's altitude adaptation stage changes.

[0028] (5) This invention pioneered the synergistic logic of physical stimulation and oxygen supply regulation. It not only solved the physiological contradiction of increased local oxygen consumption caused by thermotherapy through metabolic compensation algorithm, but also flexibly allocated resources according to different risk levels: in the low-risk stage, it saved limited oxygen consumption through non-drug physical means, while in the extremely high-risk stage of sympathetic storm, it forcibly implemented parallel high-flow output of all channels to ensure life safety to the greatest extent and significantly enhanced the body's homeostatic regulation ability under complex and extreme environmental stimulation.

[0029] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, embodiments of the present invention are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0030] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 A flowchart of an overall method for assessing hypoxia response based on dynamic assessment of heart rate deceleration force and HRV provided in an embodiment of the present invention;

[0032] Figure 2 This is a core principle framework diagram of the dual-channel cross-determination logic in an embodiment of the present invention;

[0033] Figure 3 This is a flowchart illustrating the multi-level risk stratification and early warning logic of real-time reasoning using edge artificial intelligence in this embodiment of the invention.

[0034] Figure 4 This is a flowchart illustrating the quantitative evaluation of intervention efficacy and the self-evolution of individualized prescriptions in embodiments of the present invention. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0036] Example 1

[0037] In methods for assessing hypoxia response based on dynamic evaluation of heart rate deceleration force and HRV, such as Figure 1 As shown, step 1, multi-parameter synchronous acquisition and dynamic baseline correction, specifically includes: using the ECG acquisition circuit of the wearable monitoring device to acquire single-lead or multi-lead ECG signals in real time at a preset sampling frequency. The preset sampling frequency is set to 500Hz to 1000Hz to ensure the extraction accuracy of ECG feature points, especially the R-wave peak. The sampling accuracy of the ECG acquisition circuit is set to a 24-bit width, and the weak bioelectrical signals are conditioned by a high-input-impedance instrumentation amplifier, and power frequency interference is eliminated using a right leg drive circuit. Simultaneously, environmental parameters are acquired through a miniature sensor array integrated into the wearable monitoring device. The miniature sensor array includes a piezoresistive altitude sensor, a thermistor temperature sensor, a capacitive humidity sensor, and an integrated pulse oximetry sensor.

[0038] The altitude sensor has a measurement accuracy better than 1 meter and a sampling frequency of 1 Hz; the temperature sensor has a measurement accuracy better than 0.1℃; and the humidity sensor has a measurement accuracy better than 3%. After acquiring the raw signal, the edge processing module performs a synchronization alignment operation, calibrating the timestamps of the environmental parameters with the timestamps of the electrocardiogram signal to ensure that all parameters are calculated in the same time dimension.

[0039] Specifically, in step 1, the edge processing module receives environmental parameters and calculates the deviation of the basic autonomic neural threshold under the current environment based on a preset linear regression model. The linear regression model is stored in the non-volatile memory of the edge processing module, and its mathematical expression is as follows:

[0040]

[0041] in, This indicates the deviation from the baseline autonomic nervous system threshold. This represents the preset baseline environmental correction constant; Altitude For ambient temperature, Relative humidity, This refers to pulse oxygen saturation. , , , These represent the preset weighting coefficients corresponding to altitude, ambient temperature, relative humidity, and blood oxygen deficiency, respectively. A common method for obtaining these preset weighting coefficients is as follows: First, collect historical multidimensional physiological and environmental sample data of healthy individuals exposed to different high-altitude environments. Second, use the baseline autonomic threshold deviation as the dependent variable and altitude, ambient temperature, relative humidity, and blood oxygen deficiency as independent variables, and train the model using a multiple linear regression algorithm. Third, minimize the mean square error between the model's predicted values ​​and the actual clinically calibrated values ​​using the least squares method, and calculate the corresponding regression coefficients after convergence; these coefficients are then set as the aforementioned preset weighting coefficients. , , , A custom window with preset weighting coefficients can also be set in the system window, allowing system maintenance personnel to manually adjust the preset weighting coefficients. Through this formula, the system dynamically weights the risk judgment baseline in subsequent calculations and updates it periodically according to a preset correction step size, which is set to 5 to 15 minutes, to eliminate individual physiological differences and misjudgments caused by external climate fluctuations and decreased oxygen partial pressure in the high-altitude environment.

[0042] In the above method, step 2, real-time calculation of autonomic nervous system features, specifically includes: the edge processing module extracting frequency domain indices, time domain indices, and heart rate deceleration force features from continuous electrocardiogram signals. The extraction of the frequency domain indices is achieved by the edge processing module performing a Fast Fourier Transform on a preset duration RR interval sequence. The preset duration is set to 300 seconds, and a Hanning window is used for truncation to reduce spectral leakage. The calculated high-frequency power index reflects vagal nerve activity, while the low-frequency power index reflects the combined regulation of the sympathetic and vagal nerves. The ratio of low-frequency power to high-frequency power serves as an important parameter for evaluating sympathetic-vagal balance.

[0043] The time-domain indicators include the root mean square of the difference between adjacent normal heartbeat intervals and the standard deviation of the normal heartbeat interval. Among them, the root mean square of the difference between adjacent normal heartbeat intervals is highly sensitive to short-term changes in vagal tone.

[0044] Specifically, in step 2, the edge processing module uses a phase-ordered signal averaging algorithm to calculate the heart rate deceleration force in real time. During algorithm execution, the edge processing module first uses the cardiac cycle length as an alignment reference to identify the heart rate deceleration segment. Definition For the first One heartbeat cycle, if satisfied If the selected deceleration point is identified, it is determined to be a heart rate deceleration segment. Subsequently, using the selected deceleration point as the center, 10 cardiac cycles before and after it are extracted as a data segment. Averaging of the numerous overlapping data segments is performed to effectively filter out respiratory sinus arrhythmias, artifacts, and random noise interference in the electrocardiogram signal. The value of the heart rate deceleration force is calculated from the waveform feature points after averaging, and is used to quantify the vagus nerve's control over heart rhythm.

[0045] In the above process, step 2 further includes generating the sliding window slope and short-term coefficient of variation of the heart rate deceleration force. The window length of the sliding window slope is set to 3 minutes, and the step size is set to 30 seconds. By calculating the first derivative of the continuous heart rate deceleration force values ​​within the current window, the sudden drop trend of vagal tone is identified. The calculation period of the coefficient of variation of the heart rate deceleration force is set to 10 minutes. By analyzing the ratio of the standard deviation to the mean of the heart rate deceleration force values ​​within a predetermined time period, the stability of the body's autonomic nervous regulation under continuous hypoxia is evaluated.

[0046] In the above method, step 3, combined with Figure 2 As shown, the dual-channel fusion dynamic sympathetic storm criterion specifically includes: the microcontroller establishing a dual-channel cross-judgment logic based on heart rate deceleration force and heart rate variability. The microcontroller monitors heart rate deceleration force and its derived indicators through the first channel, and monitors the frequency domain and time domain indicators of heart rate variability through the second channel. The system compares the output results of the two channels with the environmentally corrected dynamic threshold in real time.

[0047] Specifically, in step 3, the dynamic threshold of heart rate deceleration force is non-linearly corrected based on altitude. According to a preset gradient of altitude increase, for example, for every 1000 meters increase, the danger threshold of heart rate deceleration force is increased by 5% to 10% by a preset ratio. When the heart rate deceleration force value is lower than this dynamic threshold, the rate of change of heart rate deceleration force over time (i.e., the slope of the sliding window) is persistently negative, and the ratio of low-frequency power to high-frequency power simultaneously rises above the dynamic threshold, the system determines it as a precursor to a sympathetic storm.

[0048] Furthermore, the determination of sympathetic storm precursors also incorporates morphological characteristics of the electrocardiogram waveform, including the detection of T-wave alternation and abnormal ST-segment deviation. During the vagal withdrawal compensation period, i.e., when the heart rate deceleration force decreases but the ratio of low-frequency power to high-frequency power remains stable or shows a decreasing trend, the system determines that the body is in a stress protection state. At this time, only a warning marker is generated and stored in the local log, without triggering physical intervention. However, if the root mean square difference between adjacent normal heartbeat intervals is continuously decreasing and the coefficient of variation of heart rate deceleration force is abnormally increased, the system determines that the autonomic nervous regulatory reserve is exhausted, and the system automatically upgrades the warning level to level two or three. The system also records the heart rate rise slope in real time. If the slope exceeds the preset rate of change threshold of 15 beats per minute and lasts for more than 30 seconds, a level one warning is forcibly triggered.

[0049] In the above method, step 4, combined with Figure 3 As shown, edge AI real-time inference and multi-level risk stratification specifically include: optimizing the execution code of the phase-order signal averaging algorithm through a compiler, performing model pruning and fixed-point quantization, and deploying it in a low-power microcontroller with a digital signal processing instruction set. The microcontroller, through its built-in hardware multiplication and accumulation unit, achieves local real-time calculation of heart rate deceleration force characteristics without uploading the raw data to the cloud, thus adapting to high-altitude communication blind spots or high-latency environments.

[0050] Specifically, in step 4, a lightweight micro-machine learning neural network with a preset number of parameters is loaded. This network adopts a convolutional neural network structure, and the input vector is a three-dimensional tensor containing eight consecutive sets of heart rate deceleration force measurements, the average ratio of low-frequency power to high-frequency power over the previous 15 minutes, and the current instantaneous pulse oxygen saturation value. The convolutional layers are used to extract deep correlations between temporal features.

[0051] To operate within the limited memory resources of the microcontroller, the model undergoes pre-defined pruning to remove non-core neuron connections with weights approaching zero, ensuring that the model's weight parameters occupy less than 64KB of memory. Fixed-point quantization converts the 32-bit floating-point weights into 8-bit fixed-step integers, maintaining the microcontroller's dynamic power consumption during inference at a preset level below 50 milliwatts. The network output layer synchronously infers the heart rate deceleration risk score, the predicted trend of the ratio of low-frequency power to high-frequency power, and the probability of decreased pulse oxygen saturation.

[0052] The results of multi-path inference are weighted and fused within the microcontroller according to preset confidence weights, and finally output a multi-level risk stratification: Level 1 risk is defined as vagal withdrawal period, manifested as mild autonomic nervous system imbalance; Level 2 risk is defined as sympathetic storm precursor, manifested as a continuous decrease in heart rate deceleration force and significant activation of sympathetic activity; Level 3 risk is defined as high risk of ventricular arrhythmia, at which time a malignant ECG trend appears or the heart rate deceleration force reaches an extremely dangerous value, such as a heart rate deceleration force value of less than 2.0 ms.

[0053] In the above method, step 5, the closed-loop multi-target thermotherapy and vibration combined intervention based on the evaluation results, specifically includes: a microcontroller controlling a wearable actuator to perform physical stimulation at corresponding locations in the dermal distribution area of ​​the vagus nerve in the neck and back, the concha area, the Neiguan acupoint, and the Dazhui acupoint. The wearable actuator integrates a flexible heating array and a micro-piezoelectric ceramic vibration module. The flexible heating array uses carbon fiber heating cloth or subcutaneous flexible printed circuitry, and adjusts the heating power via a PWM signal, with a thermal response speed of less than 10 seconds.

[0054] Specifically, based on the risk level generated in step 4, the system automatically initiates a tiered sequential stimulation program. Under the first-level risk response, the system activates only the thermal stimulation module of the Neiguan acupoint, maintaining a constant temperature of 38 to 40°C (a first preset temperature), while simultaneously activating low-frequency vibration at a first preset frequency of 20 to 40 Hz. This stage aims to enhance vagal nerve efferentiation through gentle acupoint stimulation.

[0055] Under Level 2 risk response, the system simultaneously activates the thermotherapy modules in the concha and carotid sinus regions, dynamically switching the temperature between 40 and 42°C to simulate the pulsating pattern of biofeedback. The mid-frequency vibration frequency is set to a second preset frequency of 40Hz to 60Hz. Through transcutaneous vagus nerve activation via the ear-neck dual pathway, a powerful physiological effect antagonizing the sympathetic storm is generated.

[0056] Under Level 3 risk response, the system, building upon Level 2 intervention, adds thermal shock intervention to the Dazhui acupoint, raising the temperature to a second preset temperature of 42-43°C. The vibration frequency is cyclically scanned within a preset sweep frequency range of 20Hz to 80Hz to generate the maximum intensity of neuromodulation. Simultaneously, the microcontroller automatically triggers an audible and visual evacuation warning for the wearer and sends a high-flow oxygen supply request to the associated oxygen supply device.

[0057] During the intervention, the system monitors in real time the recovery rate of heart rate deceleration force and the rate of decrease in the ratio of low-frequency power to high-frequency power. If the recovery rate is lower than expected, the system will automatically increase the thermotherapy temperature or increase the vibration amplitude, forming a closed-loop control that adapts to the autonomic nervous system state and multi-target physical stimulation parameters.

[0058] In the above method, step 6, combined with Figure 4 As shown, the quantitative assessment of intervention efficacy and the self-evolution of individualized prescriptions specifically include: after each closed-loop intervention, the system automatically enters the assessment mode. The microcontroller calculates the intervention efficacy index, which serves as the core indicator for evaluating the effectiveness of the intervention program. Specifically, in step 6, the calculation formula for the intervention efficacy index executed by the microcontroller is as follows:

[0059]

[0060] in, For intervention effectiveness index, and These represent the measured heart rate deceleration force values ​​before and after the intervention, respectively. and These represent the ratios of low-frequency power to high-frequency power before and after the intervention, respectively. and These represent the root mean square of the difference between adjacent normal heartbeat intervals before and after the intervention, respectively. The standard deviation of pulse oxygen saturation during the intervention period is used to characterize blood oxygen stability. Taking its reciprocal indicates that smaller blood oxygen fluctuations contribute more positively to the efficacy score (if...). If the value is 0, then the preset maximum saturation value is assigned to that item. to Each of the above items is assigned a preset weight, and the following conditions are met: The common method for obtaining the pre-assigned weights is as follows: Based on the importance of each physiological indicator to the clinical manifestations of hypoxia tolerance and sympathetic-vagal balance, a judgment matrix is ​​constructed using the analytic hierarchy process (AHP) or the expert Delphi method. The maximum eigenvalue and corresponding eigenvector of each indicator are calculated. After passing the consistency test, normalization is performed to determine the weights. to The specific numerical values ​​can be used; alternatively, principal component analysis can be performed on historical effective intervention sample data to automatically calculate and assign preset weights based on the variance contribution rate of each indicator in reflecting the recovery of autonomic nervous function. Similar to preset weighting coefficients, a custom settings window can also be used to set preset weight assignments.

[0061] The system establishes an individual hyperthermia response profile in local non-volatile memory, continuously recording the intervention efficacy index and physiological characteristic changes before and after intervention during multi-day high-altitude exposure. Based on feedback data from the most recent 10 interventions, an online learning algorithm uses stochastic gradient descent to correct the temperature gradient, vibration spectrum distribution, and stimulation timing parameters in the hyperthermia prescription. The correction step size is set to 2% to 5% of the original parameters to ensure smooth prescription evolution.

[0062] Furthermore, the thermotherapy response classification logic includes: if the intervention efficacy index after 5 consecutive interventions is greater than the preset first criterion of 1.5, the system determines the user to be a high-response type and automatically reduces the stimulation intensity under subsequent risk levels to save power and prevent skin thermal fatigue; if several pre-efficiency indices are in the preset range of 1.0 to 1.5, the user is determined to be a medium-response type and the currently evolved optimized prescription is maintained; if several pre-efficiency indices are less than the second preset criterion of 0.8, the user is determined to be a low-response type, and the system will automatically increase the vibration power to the preset 120% ratio and trigger the oxygen supply request in advance at the second-level risk level, while starting the compensation plan, such as extending the stimulation time of the Dazhui acupoint.

[0063] In the above method, step 7, the dual-channel coordinated control of oxygen therapy and thermotherapy, specifically includes: the microcontroller interconnecting with the gradient oxygen supply device via Bluetooth Low Energy protocol or a low-power LAN wireless interface. The central controller, based on the risk stratification results, uniformly schedules the coordination logic between the physical stimulation output and the oxygen supply actuator.

[0064] Specifically, in step 7, the gradient oxygen supply device achieves stepless adjustment within a preset flow rate range of 0 to 5 liters per minute. Under level 2 risk, the initial oxygen flow rate is set to a first predetermined flow rate of 1.5 liters per minute. The system uses a metabolic compensation algorithm to calculate in real time the increase in local blood flow and slight increase in oxygen consumption caused by thermotherapy, and dynamically fine-tunes the duty cycle of pulse oxygen supply to ensure precise delivery at the beginning of the inspiratory phase.

[0065] Under Level 3 risk, the parallel logic across all channels is enforced. At this point, the oxygen supply device switches to a continuous high-flow-rate output mode of over 3.5 liters per minute, while simultaneously locking all hyperthermia arrays at the maximum safe operating temperature of 43°C. The central controller, through a priority scheduling algorithm, ensures that the transmission of ECG monitoring and intervention commands has the highest execution priority under high load conditions, thereby guaranteeing the complementarity of physiological effects in counteracting severe sympathetic storms.

[0066] This method also includes a self-assessment function for high-altitude environmental adaptation. The edge processing module analyzes the baseline drift trajectory of heart rate deceleration force and the frequency of triggered physical interventions over the past 24 hours, and uses a fuzzy logic evaluation model to generate a high-altitude adaptation score within a range of 0 to 100. If the score remains below the preset safety score of 60, the system will, through the accompanying voice broadcast module or mobile APP, advise the user to immediately stop strenuous physical activity, rest in bed, or receive mandatory assisted oxygen, thus achieving coverage from instantaneous warning to long-term health trend management.

[0067] Example 2

[0068] Based on Example 1, this example further expands and explains the data processing details and hardware interaction logic in the hypoxia response assessment method based on heart rate deceleration force and HRV dynamic assessment.

[0069] In step 1, the digital processing of the ECG signal also involves the application of an adaptive notch filter. After completing 24-bit sampling, the edge processing module first filters out baseline drift in the ECG signal using a set of two-stage cascaded infinite impulse response filters. Subsequently, by analyzing the spectral distribution in the signal in real time, it automatically locates the electromagnetic interference frequencies generated by the high-altitude power system or the internal digital circuitry of the wearable device, and dynamically adjusts the center frequency of the notch filter. This dynamic correction mechanism in this embodiment ensures that the positioning error of the R-wave peak is controlled within 2 milliseconds even in complex electromagnetic environments.

[0070] Regarding the phase-ordered signal averaging algorithm in step 2, its specific calculation process is refined into a data stream processing task. The DMA controller inside the microcontroller is configured to automatically transfer the acquired RR interval data in the background, forming a circular buffer. When the number of deceleration points in the buffer reaches 32, a PRSA calculation task is triggered. When identifying the heart rate deceleration segment, the system also introduces a mechanism for identifying and eliminating premature beats. If the rate of change of adjacent RR intervals exceeds 20%, it is determined to be an ectopic beat and is not used as an anchor point for PRSA, thus ensuring that the calculated heart rate deceleration force value purely reflects the regulatory component of the vagus nerve.

[0071] In the dual-channel fusion judgment logic of step 3, to address false positive rejection in extreme high-altitude environments, the system introduces accelerometer data as an auxiliary tool. When the accelerometer detects that the user is in a state of vigorous exercise, the system automatically increases the alarm threshold for the ratio of low-frequency power to high-frequency power and decreases the judgment weight of heart rate deceleration force. This is because exercise itself causes physiological sympathetic activation, and multi-sensor fusion can avoid triggering false alarms due to normal physiological fluctuations caused by exercise. Only when the exercise stops, if the autonomic nervous system indicators still do not recover within the preset recovery period, does the system officially determine an abnormal state.

[0072] In step 4, the edge AI inference process utilizes a high-altitude-specific physiological dataset for transfer learning during the training of the micro-machine learning neural network. Before model deployment, the network structure is trained offline using a massive amount of human ECG and environmental samples collected within a simulated hypoxic chamber. Depthwise separable convolutions are used in the convolutional layers to reduce the number of parameters. For the weighted fusion of inference results, the system employs a dynamic weight allocation strategy based on information entropy. When the signal quality index (SQI) acquired by the sensors decreases, for example, due to poor ECG lead contact caused by dry skin, the system automatically reduces the weight of the heart rate deceleration channel and increases the weight of environmental parameters and the probability of decreased blood oxygenation to maintain the robustness of the evaluation results.

[0073] For the intervention process in step 5, the drive circuit of the flexible heating array employs constant current control technology. Since the battery voltage of the wearable device gradually decreases during discharge, pulse width modulation combined with a constant current source ensures that the heating temperature remains constant regardless of voltage fluctuations. Under Level 3 risk response, to prevent low-temperature burns when the user is confused or has reduced skin sensitivity, the actuator incorporates an independent hardware temperature limiting circuit. If the local temperature sensor reading exceeds 44°C, the hardware circuit will forcibly cut off the heating power supply, even if the microcontroller malfunctions.

[0074] In step 6, the personalized prescription evolution process, the online learning algorithm also considers the influence of circadian rhythms. Based on a built-in real-time clock, the system divides the day's physiological state into wakefulness, sleep, and transition periods. For the same user, the baseline vagal tone is naturally higher during sleep; therefore, the system maintains independent thermotherapy response profiles for different time periods. For example, when the same level of risk is detected during sleep, the system prioritizes high-frequency vibration over high-temperature stimulation to avoid disrupting the user's deep sleep while simultaneously regulating the autonomic nervous system.

[0075] In the coordinated control of step 7, the wireless connection of the gradient oxygen supply device adopts a protocol with an acknowledgment mechanism. Each oxygen flow adjustment command issued by the microcontroller requires the oxygen supply device to return the current actual flow rate and battery level within 50 milliseconds. If the connection is lost, the microcontroller will immediately activate an emergency vibration alert on the wearable actuator to inform the user of device coordination failure. Simultaneously, the oxygen therapy compensation algorithm calculates the increment of the basal metabolic rate based on the user's current rate of body temperature rise and performs feedforward adjustment at a rate of 10% increase in oxygen flow rate for every 1°C increase.

[0076] Furthermore, the method of this invention also possesses a self-learning sensor fault self-diagnosis function. The system monitors the output of the environmental sensor array in real time. If it detects a sudden change in the altimeter reading that is inconsistent with physical laws within a short period of time, the system automatically switches to an auxiliary altitude model based on GPS latitude and longitude information. Through this multiple redundancy and self-healing mechanism, the continuous availability of the evaluation method is ensured even in extremely harsh environments.

[0077] Example 3

[0078] This embodiment further details the implementation process and data change details of the method of the present invention in a specific application scenario.

[0079] The user is set to be exposed in a high-altitude environment at 4500 meters. Step 1 begins, and the sensor array collects data showing an ambient temperature of 5°C, relative humidity of 20%, and baseline blood oxygen saturation of 85%. The edge processing module, based on a linear regression model, adjusts the warning baseline for heart rate deceleration force from 5.0ms in a healthy normal temperature environment to 6.5ms, using this as the starting and ending points for hazard assessment in this environment.

[0080] Proceeding to step 2, the system performs real-time ECG signal analysis. Within a 5-minute monitoring window, the system extracts the user's average heart rate as 95 beats per minute, with a significant decrease in high-frequency power and an increase in the ratio of low-frequency power to high-frequency power to 3.5. The PRSA algorithm calculates a heart rate deceleration rate of 5.8 ms. Simultaneously, the sliding window displays ΔDC / Δt as a continuously negative value over the past 2 minutes, with a slope of -0.2 ms / min.

[0081] Upon entering step 3, since the heart rate deceleration force of 5.8ms is lower than the corrected baseline of 6.5ms, and the ratio of low-frequency power to high-frequency power of 3.5 exceeds the dynamic threshold of 2.5 at this altitude, the system determines that the user has entered the "sympathetic storm precursor" stage, and the warning level is determined to be level two.

[0082] In step 4, the lightweight neural network within the microcontroller performs inference on the input vector. Since the blood oxygen saturation showed a trend of decreasing from 85% to 81%, the confidence level of the model's output probability of decreased blood oxygen was 0.85, and the final weighted fusion risk stratification was determined to be level two risk.

[0083] Step 5 responds immediately. The wearable actuator initiates thermotherapy in the concha and carotid sinus areas, automatically adjusting the PWM duty cycle to raise the temperature to 41°C within 60 seconds, accompanied by 50Hz mid-frequency vibration. The user's Neiguan acupoint is maintained at 39°C with warm stimulation. At this time, Step 7 notifies the gradient oxygen supply device via Bluetooth to activate pulse oxygen supply, with the oxygen flow rate set to 2.0 liters per minute.

[0084] After 10 minutes of intervention, the system performed an efficacy assessment in step 6. Monitoring showed that the heart rate deceleration rate recovered to 7.2 ms, the ratio of low-frequency power to high-frequency power decreased to 2.1, and blood oxygen saturation recovered and stabilized at 88%. The calculated IEI index for this intervention was 1.45.

[0085] Based on this result, the online learning algorithm recorded in the user's profile that the user had a moderate to high sensitivity to 41°C thermotherapy stimulation of the concha. The prescription self-evolution module decided to fine-tune the heating rate of the thermotherapy to make it more rapid the next time a similar risk occurred, and to try to extend the sweep range of the mid-frequency vibration to a higher frequency band to see if it could further shorten the autonomic nerve recovery time.

[0086] Throughout the implementation process, all computational logic, decision-making reasoning, and actuator control are completed at the edge of the wearable device, and localized data processing ensures real-time response. The system's internal task scheduler employs hard real-time priority allocation, ensuring that the ECG feature extraction task receives a processor cycle every 200 milliseconds, while the slow evolution task of the intervention prescription runs in the background with low priority, making full use of the microcontroller's computing power.

[0087] This invention achieves a closed-loop process for responding to hypoxia through deep synergy of multiple steps and parameters, from precise monitoring to proactive intervention. Compared to traditional single-point early warning, this approach, based on dynamic changes in physiological indicators and individual efficacy feedback, significantly enhances the body's ability to regulate autonomic homeostasis in the complex environment of high altitudes, effectively preventing malignant physiological events induced by sympathetic storms.

[0088] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for assessing hypoxia response based on dynamic evaluation of heart rate deceleration force and HRV, characterized in that, The method is executed by a system comprising an edge processing module, a microcontroller, and a central controller, including: Step 1: Synchronously acquire electrocardiogram signals, environmental parameters, and pulse oxygen saturation. Calculate the baseline autonomic threshold deviation using the environmental parameters and pulse oxygen saturation. Dynamically correct the risk judgment baseline for heart rate variability and heart rate deceleration force to obtain the environmentally corrected dynamic threshold. Step 2: The edge processing module extracts frequency domain indicators, time domain indicators, and heart rate deceleration force features from the electrocardiogram signal. The frequency domain indicators include the ratio of low-frequency power to high-frequency power, the time domain indicators include the root mean square of the difference between adjacent normal heartbeat intervals, and the heart rate deceleration force features include the heart rate deceleration force value, the rate of change of heart rate deceleration force over time, and the coefficient of variation of heart rate deceleration force. Step 3: The microcontroller establishes a dual-channel cross-judgment logic based on heart rate deceleration force and heart rate variability, and performs logical judgment on the frequency domain index, the time domain index, the heart rate deceleration force feature, and the environmentally corrected dynamic threshold. The dual-channel cross-judgment logic specifically includes: when the heart rate deceleration force value is lower than the environmentally corrected dynamic threshold, the rate of change of the heart rate deceleration force over time is continuously negative, and the ratio of low-frequency power to high-frequency power rises to above the environmentally corrected dynamic threshold, it is judged as a precursor to a sympathetic storm; when the heart rate deceleration force value decreases, but the ratio of low-frequency power to high-frequency power remains stable or decreases, it is judged as a vagal withdrawal compensation period; when the root mean square of the difference between adjacent normal heartbeat intervals continuously decreases and the coefficient of variation of the heart rate deceleration force increases beyond a preset threshold, the warning level is automatically upgraded. Step 4: The microcontroller combines the determination result of Step 3 with the current pulse oxygen saturation and inputs it into the lightweight micro machine learning neural network deployed in the microcontroller for inference, and outputs multi-level risk stratification results; Step 5: The microcontroller controls the wearable actuator to start a hierarchical sequential stimulation program in a preset target area based on the multi-level risk stratification results. The hierarchical sequential stimulation program includes dynamically adjusted thermal stimulation and vibration stimulation. Step 6: After a single intervention, the microcontroller calculates the intervention efficacy index and uses an online learning algorithm to adaptively update the stimulation parameters in the individualized intervention prescription based on the intervention efficacy index. The intervention efficacy index is calculated by comprehensively considering multiple indicators, including the recovery rate of the heart rate deceleration force before and after the intervention, the decrease rate of the ratio of low-frequency power to high-frequency power before and after the intervention, the root mean square recovery rate of the difference between adjacent normal heartbeat intervals before and after the intervention, and the stability of pulse oxygen saturation during the intervention. These indicators are then summed after being assigned preset weights. The online learning algorithm is used to adaptively update the stimulation parameters in the individualized intervention prescription, including establishing a personal thermo-stimulation response profile, classifying individuals into high-response, medium-response, or low-response types based on the intervention efficacy index obtained after multiple consecutive interventions, and using the gradient descent method to correct the temperature gradient and vibration spectrum parameters in the individualized intervention prescription. Step 7: The central controller, based on the results of the multi-level risk stratification, uniformly schedules the physical stimulation output of the wearable actuator and the oxygen supply flow of the gradient oxygen supply device, thereby achieving dual-channel coordinated control of oxygen supply regulation and thermal stimulation. The method also includes a high-altitude environment adaptation self-evaluation step: by analyzing the heart rate deceleration force baseline drift trajectory and the frequency of triggering physical interventions in a predetermined past period, an evaluation model is used to generate a high-altitude adaptation score; if the high-altitude adaptation score is lower than a preset safety score, a prompt instruction to reduce physical labor or perform supplemental oxygen inhalation is output through the associated terminal.

2. The method according to claim 1, characterized in that, The environmental parameters include altitude, temperature, and humidity; the deviation of the basic autonomic nervous system threshold is calculated using a preset linear regression model. The input of the linear regression model includes the missing values ​​corresponding to the altitude, temperature, relative humidity, and pulse oxygen saturation, and is assigned preset weighting coefficients. The output is the deviation of the basic autonomic nervous system threshold. In step 1, the environmentally corrected dynamic threshold is updated periodically according to a preset correction step size.

3. The method according to claim 1, characterized in that, The heart rate deceleration force value is calculated in real time using a phase-ordered signal averaging algorithm. Specifically, this includes: using the cardiac cycle length as an alignment reference to identify the heart rate deceleration segment; extracting a preset number of cardiac cycles before and after the selected deceleration point as a data segment; and performing alignment averaging on the overlapping data segments to obtain the heart rate deceleration force value. The calculation method of the heart rate deceleration force feature specifically includes: generating the rate of change of the heart rate deceleration force over time by calculating the first derivative of the continuous heart rate deceleration force values ​​within a preset sliding window, and generating the coefficient of variation of the heart rate deceleration force by calculating the ratio of the standard deviation to the mean of the heart rate deceleration force values ​​within a predetermined period.

4. The method according to claim 1, characterized in that, The lightweight micro machine learning neural network adopts a convolutional neural network structure and is deployed in the microcontroller after model pruning and fixed-point quantization processing. The input vector of the lightweight micro machine learning neural network is a three-dimensional tensor containing multiple consecutive sets of the heart rate deceleration force values, the average ratio of the low-frequency power to the high-frequency power within a predetermined time period, and the current instantaneous value of the pulse oxygen saturation. The output of the lightweight micro machine learning neural network includes a heart rate deceleration risk score, a predicted value of the ratio of low-frequency power to high-frequency power, and a probability of decreased pulse oxygen saturation. The microcontroller outputs the multi-level risk stratification result after weighted fusion of the multi-path inference results.

5. The method according to claim 4, characterized in that, The results of the multi-level risk stratification include: The first level of risk is the vagal withdrawal period and mild autonomic nervous system imbalance; The second level of risk is the precursor to a sympathetic storm, with a persistent decrease in heart rate deceleration and significant sympathetic activation. Level 3 risk is the appearance of malignant electrocardiographic trends or extremely dangerous heart rate deceleration.

6. The method according to claim 5, characterized in that, The preset target area includes the vagus nerve skin distribution area on the back of the neck, the concha area, and the corresponding locations of the Neiguan and Dazhui acupoints; the step of controlling the wearable actuator to initiate a stratified sequential stimulation program in the preset target area based on the multi-level risk stratification results specifically includes: Under the first-level risk response, the thermal stimulation and vibration stimulation of the Neiguan acupoint are initiated, wherein the vibration stimulation is low-frequency vibration; Under the second-level risk response, the thermal stimulation of the concha and carotid sinus areas is activated simultaneously and dynamically switched within a preset temperature range, and the vibration stimulation of the mid-frequency range is initiated. Under the third-level risk response, based on the second-level intervention, the thermal stimulation of the Dazhui acupoint is increased to a predetermined high temperature level, and the vibration stimulation is performed in a cyclic scanning manner within a preset frequency range.

7. The method according to claim 5, characterized in that, The method of achieving dual-channel coordinated control of oxygen supply regulation and thermotherapy based on the multi-level risk stratification results specifically includes: Under the second level of risk, the system sets the initial oxygen flow rate to the first predetermined flow rate and calculates the blood flow changes caused by thermal stimulation through a metabolic compensation algorithm, dynamically adjusting the timing of pulse oxygen delivery to deliver oxygen at the beginning of the inspiratory phase. Under the third level of risk, the full-channel parallel logic is forcibly executed to switch the oxygen supply flow rate to a continuous oxygen supply mode higher than the first predetermined flow rate, and the flexible heating array in the wearable actuator that performs the thermal stimulation is locked to the maximum safe operating temperature.

Images