A heart safety intelligent monitoring and forced protection system for hypoxic training

Abstract

The present application relates to the technical field of intelligent monitoring, in particular to a heart safety intelligent monitoring and forced protection system for hypoxic training, comprising a signal acquisition and preprocessing module, a multi-dimensional risk feature extraction module, a fuzzy logic fusion reasoning module, a forced intervention decision module and a mask gas path forced disengagement unit. The signal acquisition and preprocessing module performs variable step sub-band adaptive filtering on electrocardiogram signals according to motion state signals to generate reconstructed electrocardiogram signals; the multi-dimensional risk feature extraction module extracts dynamic QT deviation features, heart rate variability complexity features and weighted ventricular premature beat load features in parallel; the forced intervention decision module generates a trigger instruction when the index exceeds the threshold value and meets the confirmation period; the heart risk is evaluated through adaptive filtering and multi-feature fuzzy fusion, and the gas path is forced to switch to a normoxic environment in danger to eliminate the risk of secondary inhalation.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring technology, specifically to an intelligent monitoring and mandatory protection system for cardiac safety during hypoxia training. Background Technology

[0002] In the field of hypoxic training, simulating high-altitude environments to improve athletes' cardiopulmonary function or aid in weight loss has become a common practice. However, hypoxic environments significantly increase the risk of arrhythmias, myocardial ischemia, and even sudden cardiac death. Current conventional monitoring methods primarily rely on blood oxygen saturation as a single indicator, issuing alarms or requiring manual intervention by coaches when blood oxygen levels drop to a preset threshold. However, clinical practice shows that malignant cardiac events often occur rapidly, and their early electrophysiological abnormalities appear much earlier than a significant drop in blood oxygen saturation. Relying solely on blood oxygen levels has a long physiological lag, making it difficult to achieve truly timely warnings and protection. Furthermore, existing hypoxic training equipment typically only provides emergency methods such as manual mask removal or remote gas supply cutoff. There is a significant operational delay between abnormality detection and protective action, and residual hypoxic gas in the tubing after gas supply cutoff may still be inhaled by the trainee, resulting in unsatisfactory protective effects.

[0003] During hypoxic training, trainees often engage in treadmill exercise or stationary cycling, which introduces significant motion artifacts into ECG signal acquisition. These artifacts manifest as baseline drift and superimposed electromyographic noise, making it difficult for traditional fixed-cutoff-frequency filters to effectively remove noise while preserving key diagnostic features such as the QRS complex. Furthermore, the QT interval, used to assess cardiac repolarization risk, exhibits a physiological lag effect during rapid heart rate changes. Traditional heart rate correction formulas are prone to overcorrection or undercorrection under such dynamic conditions, leading to misjudgments. On the other hand, time-domain or frequency-domain indicators commonly used in heart rate variability analysis are poorly adapted to non-stationary signals, failing to capture the early nonlinear characteristics of the transition from ordered regulation to random disorder in cardiac autoregulation. Simple ventricular premature beat counting is also prone to generating numerous false positives during exercise due to physiological premature beats. Therefore, there is an urgent need for an intelligent monitoring method that can integrate multi-dimensional ECG characteristics, possess strong resistance to motion interference, and can identify early warning signs of cardiac electrical storm.

[0004] Current hypoxia training systems often employ hard threshold judgments for safety protection, triggering alarms or shutdowns when a certain indicator exceeds a fixed limit. This "jump-type" judgment method is prone to false triggers due to momentary interference and often delays in handling truly dangerous situations due to overly conservative threshold settings. More importantly, once a danger is detected, how to restore the trainee to normoxic breathing as quickly as possible is often addressed in existing solutions, which mostly only shut off the hypoxia source or issue audible and visual warnings. The trainee still needs to actively remove their mask or wait for instructor assistance, which is often ineffective in critical moments when consciousness may be impaired. Therefore, developing a mechanical actuator that can automatically, gently, and forcibly connect the trainee's airway to a normoxic environment, and working in conjunction with a multi-dimensional, intelligent risk assessment system, has significant clinical and practical value for improving the safety of hypoxia training. Summary of the Invention

[0005] The purpose of this invention is to provide a cardiac safety intelligent monitoring and forced protection system for hypoxia training. It assesses cardiac risk through adaptive filtering and multi-feature fuzzy fusion, and forcibly switches the airway to a normoxic environment in dangerous situations to eliminate the risk of secondary inhalation.

[0006] To achieve the above-mentioned technical objectives and effects, the present invention is implemented through the following technical solution: A cardiac safety intelligent monitoring and mandatory protection system for hypoxia training includes: The signal acquisition and preprocessing module is used to acquire the electrocardiogram (ECG) signal and motion state signal of the trainee, and to perform adaptive filtering on the ECG signal based on the motion state signal to generate a reconstructed ECG signal. A multi-dimensional risk feature extraction module is used to extract, at least in parallel, the dynamic QT deviation features, heart rate variability complexity features, and weighted premature ventricular contraction load features of the trainees based on the reconstructed electrocardiogram signal. The fuzzy logic fusion reasoning module is used to take the dynamic QT deviation features, heart rate variability complexity features and weighted premature ventricular contraction load features as input variables, and perform fusion calculations according to preset fuzzy rules and membership functions to generate an electrocardiogram risk assessment index that characterizes the risk level of cardiac malignant events. The mandatory intervention decision module is used to generate a mandatory intervention trigger command when the electrocardiogram risk assessment index exceeds a preset safety threshold and the duration exceeds a preset safety confirmation period; and, The forced disconnection unit for the mask airway is located between the hypoxia training mask and the hypoxia supply airway. In response to the forced intervention trigger command, it forcibly disconnects the airway connection between the hypoxia supply airway and the hypoxia training mask, and allows the air inlet of the hypoxia training mask to be fluidly connected to the external normal oxygen environment.

[0007] Furthermore, the signal acquisition and preprocessing module includes a variable step-size sub-band adaptive filter, which is configured to use the motion state signal as a reference input to perform sub-band decomposition on the electrocardiogram signal, and update the filter weights within each sub-band according to the normalized least mean square algorithm. The iteration step-size factor of the algorithm is dynamically adjusted by the intensity of the motion state signal, and the adjustment relationship satisfies:

[0008] in: Current discrete time The corresponding adaptive iteration step size factor; The initial step size factor preset when the trainee is in a stationary state; : Normalized adjustment coefficient, used to balance the sensitivity of step size to changes in motion intensity; Current moment The magnitude of the triaxial acceleration vector characterizes the intensity of the trainee's trunk movements.

[0009] Furthermore, when extracting the dynamic QT bias features, the multi-dimensional risk feature extraction module is configured to: base the dynamic QT bias on the individual trainee's baseline QT interval. and reference RR spacing Combined with real-time RR spacing and heart rate acceleration Calculate the individualized dynamic QT baseline using the following formula. :

[0010] in: :time The corresponding individualized dynamic QT interval baseline value; : The baseline QT interval of trainees under resting normoxic conditions; : The baseline RR interval of trainees under resting normoxic conditions; :time Real-time measured RR intervals; : Individualized heart rate variability index obtained through system self-calibration fitting; The lag compensating factor for QT interval changes in heart rate; Physiological lag time constant; The rate of change of heart rate per unit time, i.e., the acceleration of heart rate; The deviation rate of the measured QT interval relative to the individualized dynamic QT baseline is used as the dynamic QT deviation characteristic.

[0011] Furthermore, when extracting the heart rate variability complexity feature, the multi-dimensional risk feature extraction module is configured to perform detrended fluctuation analysis on the RR interval sequence within a preset length sliding window to calculate a scaling index within a scale range of 4 to 16 cardiac cycles. This serves as a feature for the complexity of heart rate variability.

[0012] Furthermore, when extracting the weighted premature ventricular contraction (PVC) load features, the multi-dimensional risk feature extraction module is configured to calculate the weighted load value based on the morphology and coupling interval of the detected PVCs. :

[0013] in: The weighted early ventricular load index accumulated within the preset time window; The total number of premature ventricular contractions detected within the preset time window; : No. The morphological weighting factor corresponding to each premature ventricular contraction is assigned a value based on the similarity of the QRS waveform; : No. The coupling weighting factor corresponding to each premature ventricular contraction is assigned a value according to the coupling pattern such as single, paired, or ventricular tachycardia. Furthermore, significantly higher weighting factor values ​​are assigned to polymorphic premature ventricular contractions (PVCs) or paired / chained PVCs.

[0014] Furthermore, the fuzzy logic fusion reasoning module includes at least one of the following interlocking rules: when the dynamic QT deviation feature is characterized as high risk and the heart rate variability complexity feature is characterized as decompensated, regardless of the weighted premature ventricular contraction load feature, the electrocardiogram risk assessment index is forcibly pushed up to the numerical range corresponding to the severe risk level.

[0015] Furthermore, the forced intervention decision module is also configured to: when the heart rate variability complexity feature is lower than the first critical threshold and the dynamic QT deviation feature is higher than the second critical threshold, the safety confirmation period is automatically shortened to less than one second.

[0016] Furthermore, the mask airway forced disconnection unit is an electrically controlled pilot-operated disconnection valve assembly, which is configured to perform a sequential action after receiving the forced intervention trigger command, including: in the first stage, opening the pilot relief valve to release the residual pressure gas in the low-oxygen supply pipeline to the outside atmosphere; after the first stage of depressurization is completed, in the second stage, driving the locking mechanism to release, thereby releasing the lock on the low-oxygen supply pipeline and the low-oxygen training mask, and achieving pressureless separation.

[0017] Furthermore, it also includes a blood oxygen saturation acquisition module for acquiring the trainee's blood oxygen saturation value SpO2; the fuzzy logic fusion reasoning module uses the blood oxygen saturation value as an input variable and includes a rule: when the blood oxygen saturation value is lower than a preset danger threshold, the electrocardiogram risk assessment index is directly set to the highest risk level.

[0018] On the other hand, this invention proposes a method for intelligent monitoring and mandatory protection of cardiac safety during hypoxia training, comprising the following steps: The electrocardiogram (ECG) signal and motion state signal of the trainee are collected, and the ECG signal is adaptively filtered based on the motion state signal to generate a reconstructed ECG signal; Based on the reconstructed electrocardiogram signal, the dynamic QT deviation features, heart rate variability complexity features, and weighted premature ventricular contraction load features of the trainees are extracted in parallel. The dynamic QT deviation characteristics, heart rate variability complexity characteristics, and weighted premature ventricular contraction load characteristics are input into a preset fuzzy logic inference engine, and an electrocardiogram risk assessment index is generated through fuzzy fusion calculation. When the ECG risk assessment index is determined to exceed a preset safety threshold and the duration exceeds a preset safety confirmation period, a mandatory intervention trigger command is generated. In response to the forced intervention trigger command, the airway connection between the hypoxia supply line and the hypoxia training mask is forcibly cut off, and the air inlet of the hypoxia training mask is made fluidly connected to the external normal oxygen environment.

[0019] The beneficial effects of this invention are: This invention addresses the issues of motion artifact interference and QT correction inaccuracies during sudden heart rate changes by using a variable-step-size subband adaptive filter that dynamically adjusts the step-size factor based on the magnitude of the acceleration vector, in conjunction with a dynamic QT baseline model that incorporates heart rate acceleration lag compensation. This improves the accuracy of repolarization anomaly identification. Specifically, the filter step-size factor adjusts inversely with exercise intensity, suppressing divergence under strong non-stationary noise and ensuring the morphological fidelity of the reconstructed ECG signal's QRS complex. The dynamic QT baseline model compensates for physiological lag with the heart rate acceleration term, preventing misjudgment of pathological prolongation due to QT not shortening in time during heart rate acceleration. Together, these two mechanisms ensure the accuracy of assessing repolarization reserve depletion under conditions of combined exercise and hypoxia.

[0020] This invention uses the scaling index α1, extracted from detrended fluctuation analysis across 4 to 16 cardiac cycle scales, to characterize the nonlinear degradation of autoregulation. This index complements the premature ventricular contraction (PVC) load index, calculated based on morphological and coupling weights, enabling early identification of precursors to malignant arrhythmias from the perspectives of chaotic dynamics and abnormal depolarization load. A decrease in the α1 index reflects the degradation of autoregulation from fractal order to random disorder, with an early warning time preceding changes in blood oxygen saturation. The weighted PVC load index assigns high morphological weights to polymorphic PVCs and high coupling weights to paired or ventricular tachycardias, quantifying the severity of depolarization abnormalities and suppressing false positives of physiological premature beats. The parallel extraction of both indices forms a multidimensional perception of the precursor state of electrical storm.

[0021] This invention integrates multi-dimensional features with fuzzy logic interlocking rules to output a risk assessment index. The mandatory intervention decision module automatically shortens the safety confirmation cycle when both critical characteristics are simultaneously met. Combined with the sequential action of the electrically controlled pilot-operated disconnect valve assembly—depressurizing before disconnecting—it forms a low-delay closed-loop protection and eliminates the risk of secondary inhalation due to residual hypoxia. The interlocking rules forcibly increase the risk level when both high QT and α1 decompensation exist, avoiding misjudgment based on a single threshold. The decision module responds when α1 is below the first critical threshold and ΔQTc... dynamic When the threshold is higher than the second critical threshold, the confirmation cycle is reduced to within one second to achieve rapid response in high-confidence scenarios; the pilot relief valve of the disconnect valve group is opened first to release the residual pressure in the pipeline and then the locking mechanism is released to ensure that no pressure gas is sprayed onto the mask at the moment of separation, and the trainee can directly inhale air after connecting with the normal oxygen environment.

[0022] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Detailed Implementation

[0023] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example 1

[0025] The intelligent cardiac safety monitoring and forced protection system for hypoxia training described in this embodiment includes: The system comprises a signal acquisition and preprocessing module, a multi-dimensional risk feature extraction module, a fuzzy logic fusion reasoning module, a forced intervention decision-making module, and a mask airway forced disconnection unit. Data exchange and control command transmission between these subsystems occur via wired or wireless communication buses.

[0026] The signal acquisition and preprocessing module specifically includes: a wearable ECG electrode array, a pulse oximeter, a triaxial accelerometer, and a variable step-size sub-band adaptive filter. The wearable ECG electrode array preferably uses flexible fabric dry electrodes, arranged inside an elastic chest band according to standard clinical chest lead positions, for acquiring single-lead or multi-lead raw ECG signals.

[0027] The pulse oximetry probe uses a reflective photoplethysmography (PPG) sensor, clipped to the trainee's earlobe or fingertip, to synchronously acquire blood oxygen saturation signals, denoted as... The triaxial accelerometer is integrated on the back of the substrate of the wearable ECG electrode array, and is used to sense the motion acceleration vector of the trainee's torso in three dimensions: front-back, left-right, and up-down, denoted as [missing information].

[0028] in These represent the acceleration components in the forward, backward, left-right, and vertical directions, respectively, with the superscript T indicating vector transpose.

[0029] It should be understood that during hypoxic training, trainees often engage in treadmill exercise or stationary cycling, at which time the electrocardiogram signal is severely affected by motion artifacts, manifesting as significant baseline drift and superimposed high-frequency electromyographic noise. Traditional fixed-cutoff frequency filters struggle to achieve a balance between noise removal and preservation of high-frequency details in the QRS complex.

[0030] To address this issue, in this embodiment, the variable step-size subband adaptive filter is configured to use the output signal of the triaxial accelerometer as a noise reference input to adaptively cancel the original electrocardiogram (ECG) signal. The specific algorithm implementation is as follows: First, the original ECG signal... and acceleration reference signal Perform multiphase subband decomposition separately to obtain Individual with signal and ,in This represents the subband index. Within each subband, a normalized least mean square adaptive filter is run independently. The weight vector of this filter... Update according to the following iterative formula:

[0031] In the above formula: Indicates the first Subband at discrete time The filter weight vector is initialized to a zero vector; Indicates the first Height belt at all times The input signal vector is obtained by subband decomposition of the acceleration reference signal; Let be the prior error signal of this sub-band, where For the first The expected signal of each sub-band, i.e., the noisy ECG sub-band component; The regularization factor takes the value of a very small positive number, for example... This is used to prevent numerical overflow caused by a denominator of zero; Represents the input signal vector The square of the Euclidean norm; For the first Height belt at all times The adaptive step size factor.

[0032] The key improvement in this embodiment lies in the introduction of a variable step size factor that adapts to motion intensity. Its magnitude is dynamically adjusted by the output vector of the accelerometer, and the specific relationship is as follows:

[0033] In the above formula: For the first The preset initial convergence step size of the subband in the static state, typically takes the value of ; This is the normalized sensitivity coefficient, typically taking the value of [value missing]. ; The magnitude of the triaxial acceleration composite vector reflects the intensity of the trainee's trunk movements.

[0034] It should be understood that when the trainee's range of motion increases, Increase, leading to step size Automatic shrinkage avoids the divergence problem of adaptive filters in non-stationary high-noise environments, ensuring high accuracy of subsequent QRS group detection and ST segment measurement.

[0035] In some preferred embodiments, to further reduce the computational complexity of the embedded hardware implementation, the variable step-size subband adaptive filter can adopt a sign-based subband adaptive filtering structure. This means that only the sign information of the error signal and the input signal is used in the weight update iteration, thereby achieving approximate convergence performance with extremely low hardware multiplier overhead. After the above subband filtering processing, the reconstructed signals of each subband are synthesized to obtain a high-fidelity reconstructed electrocardiogram signal. The signal is sent to the multi-dimensional risk feature extraction module. The multi-dimensional risk feature extraction module contains three dedicated computing units that operate in parallel: a first computing unit for dynamic QT analysis; a second computing unit for DFA complexity analysis; and a third computing unit for weighted early childhood load assessment.

[0036] The traditional QT interval correction formula based on dynamic QT analysis with individualized lag post-correction is, in this embodiment, the Bazett formula:

[0037] Significant overcorrection or undercorrection defects exist during rapid changes in heart rate. More importantly, the QT interval exhibits a physiological lag of approximately one to three minutes in adapting to heart rate changes. To accurately capture the subtle pathological prolongation of the QT interval under hypoxic exposure, this embodiment constructs the following individualized dynamic QT baseline model:

[0038] In the above formula: Indicates at time The calculated individualized dynamic QT interval baseline value; The baseline QT interval was measured in trainees under conscious, resting, and normoxic conditions. The baseline RR interval was measured in trainees under conscious, resting, and normoxic conditions. Indicates at time The instantaneous RR spacing was obtained by actual measurement; It is an individualized time-varying index; The hysteresis compensation factor for the QT interval on heart rate acceleration is taken as the empirically optimal value of 0.02 in this embodiment; The time constant is 60s in this embodiment; The slope of the linear regression of the average heart rate over the past fifteen seconds represents the acceleration of the heart rate change.

[0039] In this embodiment, the system forces the user into an individualized self-calibration mode upon initial startup, requiring the user to sit quietly for five minutes. The system automatically calculates the median QT interval and median RR interval of all valid heartbeats during this period. and The data is stored in a register. Subsequently, during the warm-up phase, the system collects paired data points as the trainee's heart rate gradually increases from 60 beats per minute to 120 beats per minute. Fitting the power function using the least squares method Individualized time-varying index was obtained. And write it into the trainee's file register.

[0040] It should be understood that, The value typically ranges from 0.30 to 0.45, varying among individuals due to differences in autonomic nervous system regulation. To ensure individualized chronotropic index... In a preferred embodiment of this example, regarding the robustness of the fit, the first computational unit performs the power function... Before fitting, median filtering is used to remove abnormal deviations caused by missed motion artifacts or premature beats. Specifically, for points on the scatter plot... If the absolute value of its residual is greater than three times the standardized residual, it is marked as an outlier and removed. Subsequently, the Levenberg-Marquardt algorithm is used iteratively to solve the nonlinear least squares problem for the remaining valid data points to obtain the optimal solution. Value. This process ensures The values ​​are unaffected by transient fluctuations, significantly improving the stability of the individualized model. This is achieved after calculating the dynamic baseline. Then, the real-time QTc dynamic deviation rate is further calculated:

[0041] In the above formula: Indicates at time The percentage of dynamic QTc deviation; Indicates at time The QT interval was obtained from the electrocardiogram waveform.

[0042] Clinical studies have shown that when cardiac autoregulation decompensates, the temporal structure of the RR interval sequence changes from a long-term correlated " The "noise" degenerates into completely random "white noise". This embodiment introduces a short timescale index. As a precursor and early warning indicator for malignant arrhythmias during hypoxic training.

[0043] In this embodiment, the second computing unit maintains a circular buffer with a length of 256 heartbeats to store the latest RR interval sequence. Each time a new R-wave is detected, a DFA calculation is triggered. The calculation steps are as follows: The first step is to construct the cumulative sum of the RR interval series, i.e., the "walking outline":

[0044] In the above formula: Indicates the first A cumulative sum; Represents the first in the sequence One RR interval; This represents the arithmetic mean of all RR intervals within the current analysis window.

[0045] The second step is to sequence Divided into lengths of Non-overlapping windows. In this embodiment, the window length... The range of values ​​is Each cardiac cycle is traversed with a step size of 1. This scale range is specifically designed for respiratory sinus arrhythmias and transient sympathetic-vagal modulation fluctuations, and has high sensitivity for predicting ventricular fibrillation.

[0046] The third step is to process the data in each window. Linear least squares fitting is performed on the segments to obtain the local trend sequence. .

[0047] The fourth step is to calculate the scale. The root mean square oscillation function is as follows:

[0048] In the above formula: Indicated in scale The root mean square fluctuation amplitude; This represents the total number of heartbeats in the current analysis window; For the cumulative sum sequence in the th The value of the point; In scale The local trend values ​​are obtained through linear fitting.

[0049] Step 5, in a double logarithmic coordinate system The slope of the fitted line is obtained using the least squares method, and this slope is the short-time scaling exponent. .

[0050] Under healthy, homeostatic conditions... Usually stable at Interval. In some embodiments, the second computing unit sets a warning threshold: when two consecutive DFA calculation cycles (approximately two minutes) pass, The descent gradient of the value exceeds And currently Value fell below At this time, regardless of whether there are any visible abnormalities in the electrocardiogram waveform, the system determines that the heart's autoregulatory function has entered a state of imminent collapse and generates high-priority internal flags. .

[0051] The morphological clustering-based weighted premature ventricular contraction (PVC) burden assessment uses simple PVC counting, such as triggering an alarm if the PVC rate exceeds six per minute, which is prone to misjudgment during exercise. This embodiment introduces a weighted counting mechanism that considers the morphological severity of PVCs. A lightweight convolutional neural network classifier is integrated within the third computational unit to extract morphological features from each detected wide and abnormal QRS complex and compares them with cosine similarity templates of PVCs in the trainee's history. Weighted PVC burden index. The calculation formula is as follows:

[0052] In the above formula: This represents the weighted early ventricular load index accumulated within a preset time window; This indicates the total number of premature ventricular contractions detected within the preset time window; Indicates the first The morphological weighting factor for each premature ventricular contraction; Indicates the first Coupling weighting factor for each premature ventricular contraction.

[0053] Among them, morphological weight factor The assignment rule is as follows: if the similarity between the current premature ventricular contraction (PVC) waveform and the previously recorded PVC template waveform is greater than 0.85, it is determined to be a monomorphic PVC. If the similarity is less than 0.85 and it differs from the sinus template, it is judged as a newly emerging polymorphic premature ventricular contraction. .

[0054] Coherence weighting factor The assignment rule is: single premature ventricular contraction The second premature ventricular contraction in a pair of premature ventricular contractions Each heartbeat in ventricular tachycardia .

[0055] The accumulation period is set to a sliding time window of the past sixty seconds. At that time, the system generates a flag bit. .

[0056] The fuzzy logic fusion inference module receives the continuous feature values ​​and flag bits output by the aforementioned computing unit and performs intelligent risk fusion decision-making. This inference engine avoids the "false triggering" of hard thresholds and instead simulates the diagnostic thinking of a clinician.

[0057] In this embodiment, the inference engine defines a three-input, single-output Mamdani-type fuzzy system: Input variable 1: QTc dynamic deviation rate The domain of discourse is fuzzy set is Among them, the high-risk membership function is in It then rose rapidly.

[0058] Input variable 2: DFA short-time scaling exponent The domain of discourse is The fuzzy set is {loss of compensation (L), warning (M), normal (H)}.

[0059] Input variable 3: Weighted early ventricular load The domain of discourse is The fuzzy set is {low(L), medium(M), high(H)}.

[0060] Output variable 4: Electrocardiogram risk assessment index The domain of discourse is The fuzzy set is {Safety(S), Attention(A), Severity(C)}.

[0061] The fuzzy rule base in this embodiment includes the following core interlocking rules: Rule 1: IF (QTc is high risk) AND ( is decompensated) THEN ( (is severe) [weight 1.0] Rule 2: IF (PVC is high) AND ( is decompensated) THEN ( (is severe) [weight 1.0] Rule 3: IF (QTc is high) AND ( is alert) AND (PVC is in) THEN ( (is severe) [weight 0.8] Rule 4: IF( THEN( (is severe) [weight 1.0] Defuzzification is performed using the centroid method to obtain continuous values ​​between 0 and 1. This continuous value is sent to the mandatory intervention decision module.

[0062] The mandatory intervention decision module executes the following logical judgment: when Furthermore, this state persists for longer than the preset security confirmation period. (This time window is to filter out transient spike interference caused by violent coughing or brief electrode detachment.) When this time window is in effect, the decision-maker outputs a high-level, non-maskable hardware trigger command to the mask airway forced disconnection unit through the IO port.

[0063] In this embodiment, the security confirmation period It is typically set to three seconds. In a preferred embodiment, when both conditions are met... and During the security confirmation period The time limit is automatically shortened to one second to enable a rapid response in extremely critical situations and further gain valuable rescue time.

[0064] In summary, this invention proposes an intelligent monitoring and forced protection system for cardiac safety during hypoxia training, comprising a signal acquisition and preprocessing module, a multi-dimensional risk feature extraction module, a fuzzy logic fusion reasoning module, a forced intervention decision module, and a mask airway forced disconnection unit. The signal acquisition and preprocessing module performs variable-step-size sub-band adaptive filtering on the electrocardiogram (ECG) signal based on motion state signals to generate a reconstructed ECG signal; the multi-dimensional risk feature extraction module extracts dynamic QT deviation features, heart rate variability complexity features, and weighted premature ventricular contraction load features in parallel; the fuzzy logic fusion reasoning module fuses and outputs an ECG risk assessment index based on interlocking rules; the forced intervention decision module generates a trigger command when the index exceeds a threshold and the confirmation cycle is met; the mask airway forced disconnection unit responds to the command by forcibly cutting off the hypoxia air supply line with a sequential action of first depressurizing and then separating, and connecting the mask air inlet to the normoxic environment. This invention solves the problems of motion artifact interference and heart rate correction lag, achieving low-latency closed-loop forced protection.

[0065] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A cardiac safety intelligent monitoring and forced protection system for hypoxia training, characterized in that, include: The signal acquisition and preprocessing module is used to acquire the electrocardiogram (ECG) signal and motion state signal of the trainee, and to perform adaptive filtering on the ECG signal based on the motion state signal to generate a reconstructed ECG signal. A multi-dimensional risk feature extraction module is used to extract, at least in parallel, the dynamic QT deviation features, heart rate variability complexity features, and weighted premature ventricular contraction load features of the trainees based on the reconstructed electrocardiogram signal. The fuzzy logic fusion reasoning module is used to take the dynamic QT deviation features, heart rate variability complexity features and weighted premature ventricular contraction load features as input variables, and perform fusion calculations according to preset fuzzy rules and membership functions to generate an electrocardiogram risk assessment index that characterizes the risk level of cardiac malignant events. The mandatory intervention decision module is used to generate a mandatory intervention trigger command when the electrocardiogram risk assessment index exceeds a preset safety threshold and the duration exceeds a preset safety confirmation period; and, The forced disconnection unit for the mask airway is located between the hypoxia training mask and the hypoxia supply airway. In response to the forced intervention trigger command, it forcibly disconnects the airway connection between the hypoxia supply airway and the hypoxia training mask, and allows the air inlet of the hypoxia training mask to be fluidly connected to the external normal oxygen environment.

2. The system as described in claim 1, characterized in that, The signal acquisition and preprocessing module includes a variable step-size sub-band adaptive filter, which is configured to use the motion state signal as a reference input to perform sub-band decomposition on the electrocardiogram signal, and update the filter weights within each sub-band according to the normalized least mean square algorithm. The iteration step-size factor of the algorithm is dynamically adjusted by the intensity of the motion state signal, and the adjustment relationship satisfies: in: Current discrete time The corresponding adaptive iteration step size factor; The initial step size factor preset when the trainee is in a stationary state; : Normalized adjustment coefficient, used to balance the sensitivity of step size to changes in motion intensity; Current moment The magnitude of the triaxial acceleration vector characterizes the intensity of the trainee's trunk movements.

3. The system as described in claim 1 or 2, characterized in that, The multi-dimensional risk feature extraction module is configured to extract the dynamic QT bias features based on the individual trainee's baseline QT interval. and reference RR spacing Combined with real-time RR spacing and heart rate acceleration Calculate the individualized dynamic QT baseline using the following formula. : in: :time The corresponding individualized dynamic QT interval baseline value; : The baseline QT interval of trainees under resting normoxic conditions; : The baseline RR interval of trainees under resting normoxic conditions; :time Real-time measured RR intervals; : Individualized heart rate variability index obtained through system self-calibration fitting; The lag compensating factor for QT interval changes in heart rate; Physiological lag time constant; The rate of change of heart rate per unit time, i.e., the acceleration of heart rate; The deviation rate of the measured QT interval relative to the individualized dynamic QT baseline is used as the dynamic QT deviation characteristic.

4. The system as described in claim 1 or 2, characterized in that, When extracting the heart rate variability complexity feature, the multi-dimensional risk feature extraction module is configured to perform detrended fluctuation analysis on the RR interval sequence within a preset length sliding window, using a scaling index calculated over a scale range of 4 to 16 cardiac cycles. This serves as a feature for the complexity of heart rate variability.

5. The system as described in claim 1 or 2, characterized in that, When extracting the weighted premature ventricular contraction (PVC) load features, the multi-dimensional risk feature extraction module is configured to calculate the weighted load value based on the morphology and coupling interval of the detected PVCs. : in: The weighted early ventricular load index accumulated within the preset time window; The total number of premature ventricular contractions detected within the preset time window; : No. The morphological weighting factor corresponding to each premature ventricular contraction is assigned a value based on the similarity of the QRS waveform; : No. The coupling weighting factor corresponding to each premature ventricular contraction is assigned a value according to the coupling pattern such as single, paired, or ventricular tachycardia. Furthermore, significantly higher weighting factor values ​​are assigned to polymorphic premature ventricular contractions (PVCs) or paired / chained PVCs.

6. The system as described in claim 1 or 2, characterized in that, The fuzzy logic fusion reasoning module includes at least one of the following interlocking rules: when the dynamic QT deviation feature is characterized as high risk and the heart rate variability complexity feature is characterized as decompensated, regardless of the weighted premature ventricular contraction load feature, the electrocardiogram risk assessment index is forcibly pushed up to the numerical range of the corresponding severe risk level.

7. The system as described in claim 1 or 2, characterized in that, The mandatory intervention decision module is also configured to automatically shorten the safety confirmation period to less than one second when the heart rate variability complexity feature is lower than the first critical threshold and the dynamic QT deviation feature is higher than the second critical threshold.

8. The system as described in claim 1 or 2, characterized in that, The forced disengagement unit for the mask airway is an electrically controlled pilot-operated disengagement valve assembly, which is configured to perform a sequential action after receiving the forced intervention trigger command, including: in the first stage, opening the pilot relief valve to release the residual pressure gas in the hypoxia supply line to the outside atmosphere; after the first stage of depressurization is completed, in the second stage, driving the locking mechanism to release, thereby releasing the lock on the hypoxia supply line and the hypoxia training mask, and achieving pressureless separation.

9. The system as described in claim 1 or 2, characterized in that, It also includes a blood oxygen saturation acquisition module for acquiring the trainee's blood oxygen saturation value SpO2; the fuzzy logic fusion reasoning module uses the blood oxygen saturation value as an input variable and includes a rule: when the blood oxygen saturation value is lower than a preset danger threshold, the electrocardiogram risk assessment index is directly set to the highest risk level.

10. A method for intelligent monitoring and mandatory protection of cardiac safety during hypoxia training, characterized in that, Includes the following steps: The electrocardiogram (ECG) signal and motion state signal of the trainee are collected, and the ECG signal is adaptively filtered based on the motion state signal to generate a reconstructed ECG signal; Based on the reconstructed electrocardiogram signal, the dynamic QT deviation features, heart rate variability complexity features, and weighted premature ventricular contraction load features of the trainees are extracted in parallel. The dynamic QT deviation characteristics, heart rate variability complexity characteristics, and weighted premature ventricular contraction load characteristics are input into a preset fuzzy logic inference engine, and an electrocardiogram risk assessment index is generated through fuzzy fusion calculation. When the ECG risk assessment index is determined to exceed a preset safety threshold and the duration exceeds a preset safety confirmation period, a mandatory intervention trigger command is generated. In response to the forced intervention trigger command, the airway connection between the hypoxia supply line and the hypoxia training mask is forcibly cut off, and the air inlet of the hypoxia training mask is made fluidly connected to the external normal oxygen environment.

Images