Real-time pulmonary ventilation impedance monitoring method and system based on multi-modal data fusion

Abstract

The embodiment of the invention provides a real-time pulmonary ventilation impedance monitoring method and system based on multi-modal data fusion, and the method comprises the steps: obtaining real-time multi-modal data from a plurality of sensors, the multi-modal data comprising a sound signal, an airflow signal and a pressure signal; performing fusion processing on the multi-modal data to generate comprehensive pulmonary ventilation impedance data; based on the comprehensive pulmonary ventilation impedance data, analyzing data distribution characteristics to detect anomalies; and when abnormal data distribution characteristics are detected, a fault early warning system is triggered to solve burst signal distortion. Through the scheme of the embodiment of the invention, the problem of how to carry out fault early warning on the monitoring system according to the abnormal data distribution characteristics so as to solve the problem of burst signal distortion can be solved.

Description

TECHNICAL FIELD

[0001] The present application relates to medical monitoring technology, particularly to a real-time lung ventilation impedance monitoring method and system based on multi-modal data fusion. BACKGROUND

[0002] The real-time lung ventilation impedance monitoring method and system based on multi-modal data fusion is a technology that integrates real-time information from multiple sensors (such as electrical impedance, acoustic, and image data) to monitor lung ventilation impedance changes with high precision, aiming to provide dynamic support for respiratory disease diagnosis and treatment; however, there is a key problem with this system, that is, when the monitoring data has abnormal distribution characteristics (such as signal mutation or noise interference), how to effectively build a fault warning mechanism to timely identify and respond to sudden signal distortion, thereby avoiding the risk of false positives or missed diagnosis. SUMMARY

[0003] Therefore, the embodiments of the present disclosure provide a real-time lung ventilation impedance monitoring method and system based on multi-modal data fusion, which at least partially solves the problems in the prior art.

[0004] The real-time lung ventilation impedance monitoring method and system based on multi-modal data fusion comprises: Obtaining real-time multi-modal data from multiple sensors, the multi-modal data including sound signals, airflow signals, and pressure signals; Fusing the multi-modal data to generate comprehensive lung ventilation impedance data; Analyzing data distribution characteristics based on the comprehensive lung ventilation impedance data to detect abnormalities; When abnormal data distribution characteristics are detected, triggering a fault warning system to address sudden signal distortion.

[0005] According to one embodiment, the analyzing data distribution characteristics to detect abnormalities further comprises: Calculating the mean μ and standard deviation σ of the comprehensive lung ventilation impedance data; Calculating the abnormal score of each data point based on the following formula: S = |Z μ| / σ, where Z is the value of the current data point, μ is the data mean, and σ is the standard deviation; If S is greater than a preset abnormal threshold Th, the data point is marked as abnormal; When the proportion of abnormal data points exceeds a preset proportion P%, triggering the fault warning system to address sudden signal distortion.

[0006] According to one embodiment, the triggering a fault warning system to address sudden signal distortion further comprises: Dividing the real-time data stream into a fixed time window T_window; calculating a proportion of abnormal points R_anomaly of the comprehensive pulmonary ventilation impedance data in each time window; performing conditional judgment based on the following formula: if R_anomaly>Q, then detecting a window abnormality, wherein Q is a preset proportion threshold; when detecting the window abnormality, activating an early warning signal and starting a sensor calibration module to solve sudden signal distortion.

[0007] According to an embodiment, the activating the early warning signal and starting the sensor calibration module further comprises: training a classifier model using historical data, wherein the input of the model is data distribution features; inputting data distribution features of a current time window into the classifier model to obtain an abnormality probability P_anomaly; performing conditional judgment based on the following formula: if P_anomaly>P_th, then confirming an abnormality and triggering an advanced early warning, wherein P_th is a preset probability threshold; after triggering the advanced early warning, performing sensor automatic calibration to correct signal distortion.

[0008] According to an embodiment, the performing sensor automatic calibration further comprises: when triggering the advanced early warning, analyzing signal quality scores Q_i of each sensor; calculating a drift amount D of current comprehensive pulmonary ventilation impedance data; performing conditional judgment based on the following formula: if D>D_th, then adjusting a multi-modal data fusion weight, wherein D=|μ_current-μ_reference|, μ_current is a current mean value, μ_reference is a reference mean value, and D_th is a preset drift threshold; after adjustment, re-fusing data and updating comprehensive pulmonary ventilation impedance data to solve sudden signal distortion.

[0009] According to an embodiment, the adjusting the multi-modal data fusion weight further comprises: monitoring environmental parameter changes ΔE, wherein ΔE includes temperature or humidity changes; updating an abnormality threshold Th based on environmental parameters; adjusting a fusion weight based on the following formula: W_i=Q_i / ΣQ_i, wherein W_i is a weight of the i-th sensor, and Q_i is a signal quality score; if a weight change rate after adjustment is greater than a preset threshold, then re-triggering data fusion to eliminate signal distortion.

[0010] According to an embodiment, the re-triggering data fusion further comprises: A multi-level early warning system is defined, where levels are based on anomaly severity; A severity score S_severity is calculated as R_anomaly * avg(S), where avg(S) is the average anomaly score; A condition is judged based on the following formula: if S_severity > S1, a low-level early warning is triggered; if S_severity > S2, a mid-level early warning is triggered; if S_severity > S3, a high-level early warning is triggered, where S1, S2, S3 are preset severity thresholds; Corresponding measures are executed for different early warning levels, including alarm notification or system shutdown, to address sudden signal distortion.

[0011] According to one embodiment, the executing corresponding measures further comprises: A warning response time T_response is recorded; System performance parameters are evaluated based on response effects; An anomaly threshold is adjusted based on the following formula: if T_response > T_max, Th = Th * 95, where T_max is a preset maximum response time threshold; Data distribution features are re-analyzed using the adjusted threshold to optimize fault early warning.

[0012] According to one embodiment, the re-analyzing data distribution features further comprises: A moving average filter is applied to smooth the comprehensive lung ventilation impedance data; A mean value μ_smooth and a standard deviation σ_smooth of the smoothed data are calculated; An anomaly is detected based on the following formula: if |ZMA| > M * σ_smooth, it is marked as an anomaly, where MA is the moving average value, and M is a preset multiplier; When an anomaly is detected, a feedback control loop is triggered to dynamically adjust fusion parameters to address sudden signal distortion.

[0013] According to one embodiment, the triggering feedback control loop further comprises: When the anomaly is confirmed, the normal state of individual data from each sensor is checked; A number of normal sensors N_normal is calculated; A condition is judged based on the following formula: if N_normal < N_min, a system-level anomaly is confirmed, where N_min is a preset minimum normal sensor threshold; After a system-level anomaly is confirmed, a multi-sensor redundancy switching is executed to address sudden signal distortion.

[0014] The embodiment of the present disclosure provides a real-time lung ventilation impedance monitoring method and system based on multi-modal data fusion, comprising: acquiring real-time multi-modal data from multiple sensors, wherein the multi-modal data comprises sound signals, airflow signals and pressure signals; performing fusion processing on the multi-modal data to generate comprehensive lung ventilation impedance data; analyzing data distribution characteristics based on the comprehensive lung ventilation impedance data to detect abnormalities; when abnormal data distribution characteristics are detected, triggering a fault warning system to solve sudden signal distortion. Through the scheme of the embodiment of the present disclosure, the problem of how to perform fault warning on the monitoring system according to abnormal data distribution characteristics to solve sudden signal distortion can be solved. BRIEF DESCRIPTION OF DRAWINGS

[0015] In the drawings, like reference numerals refer to same or similar components throughout the several views. These drawings are not necessarily to scale. It should be understood that these drawings are merely schematic representations, which should not limit the scope of the present application.

[0016] Figure 1 is a flowchart of a real-time lung ventilation impedance monitoring method and system based on multi-modal data fusion; Figure 2 is a further flowchart of analyzing data distribution characteristics to detect abnormalities; Figure 3 is a further flowchart of triggering a fault warning system to solve sudden signal distortion; Figure 4 is a further flowchart of activating a warning signal and starting a sensor calibration module; Figure 5 is a further flowchart of performing sensor automatic calibration; Figure 6 is a further flowchart of adjusting multi-modal data fusion weights; Figure 7 is a further flowchart of retriggering data fusion; Figure 8 is a further flowchart of performing corresponding measures; Figure 9 is a further flowchart of reanalyzing data distribution characteristics; Figure 10 is a further flowchart of triggering a feedback control cycle. DETAILED DESCRIPTION

[0017] It is to be noted that the terms such as first and second, and the like, merely identify one entity or action from another, without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0018] Next, referring to the accompanying drawings, the real-time lung ventilation impedance monitoring method and system based on multi-modal data fusion of the present application is described. First, the first step of the method is to obtain real-time multi-modal data from multiple sensors, including sound signals, airflow signals, and pressure signals. In specific operations, the system collects data in real time through a sensor array deployed on the patient's respiratory pathway, such as capturing respiratory sound signals using high-sensitivity microphones, recording airflow rate signals with thermal flow meters, and measuring airway pressure signals with piezoelectric sensors. These data are transmitted to the central processing unit at millisecond-level frequency, ensuring real-time and continuity. In one embodiment, for example, when monitoring chronic obstructive pulmonary disease (COPD) patients in an ICU ward, the sensors are integrated into a respiratory mask, the microphone captures the respiratory sound waveform, the flow meter quantifies the exhalation / inspiration flow, and the pressure sensor monitors the airway pressure difference. All signals are sent synchronously through a wireless transmission module to avoid data loss or delay, thereby providing a basis for subsequent fusion.

[0019] Next, the second step is to fuse the multi-modal data to generate comprehensive lung ventilation impedance data. In specific operations, the system uses a multi-modal data fusion algorithm, such as weighted Kalman filtering combined with a deep learning model, to fuse the normalized sound, airflow, and pressure signals by extracting features such as signal amplitude, frequency spectrum, and time domain correlation and applying adaptive weight distribution to generate a comprehensive impedance value that reflects lung ventilation resistance. For example, in one embodiment for monitoring asthma patients, the algorithm first performs FFT transformation on the sound signal to extract spectral features, calculates the volume change by integrating the airflow signal, and uses the pressure signal to calibrate the dynamic pressure difference. Then, the three signals are input into a pre-trained neural network model (such as CNN-LSTM) to output the comprehensive lung impedance value, which is updated in real time on the monitor. Specifically, this fusion process can suppress the noise of a single sensor, such as when the airflow signal is distorted due to motion interference, the sound and pressure signals provide redundant information to ensure the robustness and accuracy of the impedance data, laying the foundation for anomaly detection.

[0020] Then, the third step is to analyze the data distribution features based on the comprehensive lung ventilation impedance data to detect abnormalities; in specific operation, the system uses statistical analysis and machine learning models (such as Gaussian mixture model or support vector machine) to monitor the distribution parameters of impedance data in real time, including mean, variance, skewness and kurtosis, and introduces a sliding window mechanism to calculate dynamic thresholds, when the data points deviate from the normal distribution (such as outliers or mode mutation), the system is marked as abnormal; this solves the problem of how to give fault warning to the monitoring system according to the abnormal data distribution features to solve the sudden signal distortion, for example, by training the baseline model in real time (based on historical healthy data), when the impedance value distribution variance suddenly increases or double peak mode appears, the system can identify the signal distortion risk, such as sensor failure or environmental interference; in one embodiment, specifically for postoperative patient monitoring, the system establishes an impedance distribution baseline (mean is 5cmH2O / L / s, variance is less than 0.5) in normal breathing state, when the real-time data variance rises to 1.5 or more in 10 seconds window (such as pressure signal drift caused by sensor loosening), the model immediately triggers an abnormal mark, and confirms the distortion source combined with feature analysis (such as abnormal sound signal spectrum), thereby providing a basis for early warning.

[0021] Finally, the fourth step is to trigger a fault warning system to solve the sudden signal distortion when abnormal data distribution features are detected; in specific operation, the warning system includes a multi-level response mechanism, such as activating visual / auditory alarms, starting an adaptive correction module or switching to a redundant sensor to quickly repair distorted signals; this further solves the aforementioned problem, which prevents data interruption through real-time intervention, for example, when the anomaly is confirmed, the system first sends an alarm to the medical staff interface, while running a signal reconstruction algorithm (such as interpolation or filtering based on fusion data), or switching to a backup sensor; in one embodiment, for example, when treating a pneumonia patient in the emergency room, if the analysis shows that the impedance distribution is abnormal (such as mean shift to 8cmH2O / L / s), the system triggers an early warning: the alarm light flashes and notifies the mobile device, while automatically activating the correction module to compensate for the distorted airflow signal using historical fusion data to ensure monitoring continuity; specifically, this method can effectively handle sudden distortion (such as sensor contamination), by early warning to maintain or replace the equipment in time, thereby ensuring the reliability and safety of lung ventilation impedance monitoring.

[0022] Next, the analysis of data distribution features to detect abnormalities of the present application is further described. The steps include: first, calculating the mean μ and standard deviation σ of the comprehensive lung ventilation impedance data; second, calculating the anomaly score of each data point based on the formula S = |Z-μ| / σ; third, if S is greater than a preset anomaly threshold Th, the data point is marked as abnormal; fourth, when the proportion of abnormal data points exceeds a preset proportion P%, a fault warning system is triggered to solve the sudden signal distortion.

[0023] The meanings of these steps are as follows: the first step describes the central tendency and dispersion of the data by calculating the mean μ and standard deviation σ, providing a statistical benchmark for subsequent anomaly detection. The second step evaluates the standardized deviation of each data point Z using the formula S = |Z - μ| / σ, where Z is the current impedance value (range usually 0-100 ohms), μ is the calculated mean (range depends on system calibration), σ is the standard deviation (positive real number), and the formula represents the ratio of absolute deviation to standard deviation, which quantifies the degree of anomaly; Th is usually set to 2-3 (optimal value 3, corresponding to 99.7% confidence interval), and P% is set to 1-10% (optimal value 5%). This formula is based on the assumption of normal distribution, which can efficiently identify significant outliers and avoid noise interference. The third step directly marks abnormal points according to the comparison of S and Th, ensuring real-time response. The fourth step monitors the overall abnormality ratio to prevent local faults from spreading and trigger the early warning system.

[0024] Specifically, in one embodiment, the multi-modal data fusion system collects real-time lung ventilation impedance, such as sensor data mean μ = 50 ohms, σ = 5 ohms. For example, a data point Z = 65 ohms, calculate S = |65-50| / 5 = 3; if the preset Th = 2.5, then S > Th, marked as abnormal. When the abnormality ratio exceeds P% = 5% (e.g. 6 out of 100 points are abnormal), the system triggers an alarm and handles signal distortion.

[0025] Next, the trigger failure warning system of the present application is described to further address sudden signal distortion. First, the real-time data stream is divided into fixed time windows T_window, which cuts the continuous data stream into equal-length segments for local anomaly analysis. For example, in one embodiment, T_window is set to 10 seconds, and 100 impedance data points are collected per second, so that each window contains 1000 data points, suitable for the fast response of the real-time lung ventilation impedance monitoring system. Second, in each time window, the proportion of abnormal points R_anomaly of the comprehensive lung ventilation impedance data is calculated, which is defined as the number of abnormal points divided by the total number of points, for quantifying the degree of signal distortion. Specifically, abnormal points can be identified by a pre-set threshold or a machine learning model. Then, a conditional judgment is made based on the formula: if R_anomaly>Q, then the window is detected as abnormal, where Q is a pre-set proportion threshold. The parameter R_anomaly represents the proportion of abnormal points, ranging from 0 to 1; Q is the threshold, also ranging from 0 to 1, and the optimal value is usually set to 0.05 to 0.1, for example, Q=0.08, indicating that more than 8% of abnormal points trigger it, to balance the sensitivity and false positive rate. The formula means that when the proportion of abnormalities is too high, the window is determined to be abnormal, which is set in this way because high proportion distortion may be caused by sensor drift or external interference, which needs to be intervened in time to prevent misdiagnosis. For example, in one embodiment of lung ventilation impedance monitoring, if the impedance value in the window suddenly changes due to patient movement, increasing the number of mutation points, R_anomaly is calculated as 0.12 (greater than Q=0.1), then the system detects an anomaly. Finally, when a window anomaly is detected, a warning signal is activated and a sensor calibration module is started to address sudden signal distortion, which eliminates noise by recalibrating the sensor to restore data accuracy. Specifically, in the example, if the impedance signal is distorted due to poor electrode contact, the calibration module automatically adjusts the gain after the warning is activated, ensuring the reliability of multi-modal data fusion. The entire process is completed within 200 milliseconds, ensuring the continuity of real-time monitoring.

[0026] Next, the activation of the early warning signal and the initiation of the sensor calibration module of the present application are further described. First, this module includes four core steps: Step 1, training a classifier model using historical data, with input as data distribution features; Step 2, inputting the data distribution features of the current time window into the classifier model to obtain the anomaly probability P_anomaly; Step 3, conditional judgment based on the formula: if P_anomaly > P_th, confirm the anomaly and trigger the advanced warning; Step 4, after triggering the advanced warning, perform automatic calibration of the sensor to correct signal distortion. The meaning of each step is as follows: training a classifier model aims to learn the data distribution features (such as mean, variance or skewness) of normal and abnormal patterns from historical data to establish a prediction basis; obtaining the anomaly probability is to evaluate the anomaly likelihood of the current data window in real time through the model, outputting a probability value; the conditional judgment formula is used to verify the anomaly confidence, triggering the warning only when the probability exceeds the threshold value, avoiding false positives; performing automatic calibration is to adjust in real time for sensor distortion (such as drift or noise) to ensure the accuracy of subsequent data.

[0027] For the formula part, P_anomaly represents the anomaly probability, ranging from 0 to 1 (probability value), with higher values indicating greater anomaly likelihood; P_th is the preset probability threshold, also ranging from 0 to 1, with optimal values typically set to 0.8 to 0.9 (e.g. 0.85) to balance sensitivity and specificity. The meaning of this formula is: by comparing P_anomaly with P_th, it is determined whether to trigger the warning, aiming to ensure that the warning is activated only when the anomaly probability is high enough, thereby reducing false positives and improving system reliability. The reason for such setting is that in the medical monitoring scenario, a high threshold can prevent unnecessary alarms, avoiding interference with clinical decision-making, while ensuring a rapid response to real anomalies.

[0028] In one embodiment, specifically related to a real-time lung ventilation impedance monitoring system based on multi-modal data fusion: for example, the system collects historical lung ventilation impedance data (such as impedance distribution features and respiratory frequency fusion features), and trains a support vector machine (SVM) classifier model. After inputting the impedance distribution features (such as standard deviation of 0.5) of the current time window, the model outputs P_anomaly = 0.88. Since P_th is set to 0.85, P_anomaly > P_th is satisfied, the system triggers the advanced warning, indicating potential lung ventilation abnormalities (such as obstructive events); subsequently, the automatic calibration module is initiated, correcting signal distortion (e.g. impedance drift due to temperature changes) by adjusting the sensor gain parameter, restoring monitoring accuracy. This ensures the robustness of real-time monitoring and avoids the risk of misdiagnosis.

[0029] Next, the execution of the sensor automatic calibration of the present application is further described. The steps include: 1. When the advanced warning is triggered, analyze the signal quality score Q_i of each sensor; 2. Calculate the drift amount D of the current comprehensive lung ventilation impedance data; 3. Make a conditional judgment based on the formula: if D>D_th, adjust the multi-modal data fusion weight; 4. Re-fuse the data after adjustment and update the comprehensive lung ventilation impedance data to solve the sudden signal distortion.

[0030] The meaning of each step is as follows: First, analyze the signal quality score Q_i to evaluate the reliability of the sensor data, Q_i is calculated based on signal-to-noise ratio or stability, ensuring that abnormal signals are identified. Second, calculate the drift amount D to quantify the data offset, D is the absolute difference formula D=|μ_current-μ_reference|, where μ_current is the current comprehensive lung ventilation impedance mean, reflecting the real-time state; μ_reference is the reference mean, representing the baseline or calibration value; D_th is the preset drift threshold, used to determine whether adjustment is needed. The meaning of this formula is to measure the absolute deviation of the current data from the reference value, which is simple and efficient in detecting drift. This is because absolute value calculation is fast and suitable for real-time systems, avoiding complex operations that delay response. Parameter range: μ_current and μ_reference are usually based on impedance units (such as 0-100Ω), and the optimal value depends on the physiological state of the patient, for example, the reference mean μ_reference of a healthy adult is 50Ω; D_th is preset to 5Ω, based on clinical trial settings, ranging from 2-10Ω, ensuring sensitivity while avoiding false positives. Finally, adjust the weight to optimize the fusion algorithm, re-fuse and update the data to solve the distortion problem.

[0031] For example, in one embodiment, the lung ventilation monitoring system is deployed in an ICU, and when the patient suddenly coughs causing signal distortion, an advanced warning is triggered. Specifically, analyzing the sensor Q_i finds that the chest impedance sensor Q_i has decreased; calculating D=|55Ω-50Ω|=5Ω, if D_th is 5Ω, then adjust the weight to reduce the contribution of this sensor; after re-fusion, update the comprehensive data to eliminate distortion and ensure accurate real-time impedance monitoring.

[0032] Next, the adjustment of the multi-modal data fusion weight of the present application is further described. The process includes four steps: First, monitor the environmental parameter change ΔE, where ΔE includes temperature or humidity changes; second, update the abnormal threshold Th based on the environmental parameters; third, adjust the fusion weight based on the formula: W_i=Q_i / ΣQ_i, where W_i is the weight of the i-th sensor and Q_i is the signal quality score; finally, if the weight change rate after adjustment is greater than the preset threshold, re-trigger data fusion to eliminate signal distortion.

[0033] Monitoring the change in environmental parameters ΔE involves real-time detection of fluctuations in external factors such as temperature or humidity, ensuring that the system adapts to environmental disturbances and avoids sensor drift. Based on the environmental parameters, the abnormal threshold Th is updated, and the detection threshold is adjusted to compensate for environmental influences, improving the robustness of anomaly identification. When adjusting the fusion weights, a normalization formula is used to assign weights: W_i represents the weight of sensor i (range 0 to 1, optimal value close to 1 indicates dominant role), Q_i is the signal quality score (range 0-100, optimal value high indicates high-quality signal), and ΣQ_i is the sum of all sensor Q_i; this formula ensures that the sum of weights is 1, and high Q_i sensors are preferentially assigned to optimize fusion accuracy. If the weight change rate is too large (e.g., exceeding a preset threshold of 10%), re-triggering fusion is triggered to prevent cumulative distortion.

[0034] For example, in one embodiment, when monitoring lung ventilation impedance in real time, the operating room temperature suddenly rises ΔE=5°C, and the system updates Th to relax the impedance anomaly range, compensating for pressure sensor errors caused by thermal expansion; then, Q_i is calculated (e.g., temperature sensor Q_i drops to 60, impedance sensor Q_i remains at 90), and after adjusting the weights, the temperature sensor weight W_i is reduced; specifically, if the weight change rate reaches 15%, re-fusion data is used to eliminate distortion and avoid misdiagnosis of ventilation disorders. This setting improves the reliability of multi-modal data in medical monitoring through dynamic weight distribution.

[0035] Next, the re-triggering data fusion of the present application is further described. First, a multi-level warning system is defined, which achieves risk stratification management by setting levels based on the severity of anomalies (e.g., low, medium, and high), to distinguish different levels of urgency. Second, the severity score \(S_{\text{severity}}=R_{\text{anomaly}}\times\text{avg}(S)\) is calculated, where \(R_{\text{anomaly}}\) represents the anomaly occurrence rate (range 0 to 1, optimal value about 0.7 to balance sensitivity), and \(\text{avg}(S)\) is the average anomaly score (range 0 to 100, optimal value about 60 to reflect moderate risk). This formula multiplies the anomaly frequency and average intensity to generate a comprehensive score; this setting is because a single indicator can overlook the overall impact of sudden distortion events, ensuring that fusion data can comprehensively evaluate potential hazards. Then, based on the formula, conditional judgment is performed: if \(S_{\text{severity}}>S1\) triggers a low-level warning, if >S2 triggers a medium-level warning, and if >S3 triggers a high-level warning, where S1, S2, and S3 are preset thresholds (e.g., S1=40, S2=50, and S3=60). Finally, corresponding measures are taken for different warning levels, such as sending an alarm notification for low-level warning, intensifying the notification for medium-level warning, and system shutdown for high-level warning, to immediately address signal distortion problems.

[0036] In one embodiment, specifically related to real-time lung ventilation impedance monitoring system based on multi-modal data fusion, such as fusion of airflow sensor and thoracic impedance data, sudden signal distortion is detected. Assuming avg(S) is calculated as 70 (indicating high severity anomaly, such as dramatic impedance fluctuation), R_anomaly is 0.8 (anomaly frequently occurs), then S_severity = 56. If preset S2 = 50 and S3 = 60, the system triggers a medium-level warning, performs alarm notification operation, reminding medical staff to check device connection, avoiding ventilation interruption risk.

[0037] Next, the present application is further described to take corresponding measures. The steps include: recording warning response time T_response; evaluating system performance parameters based on response effect; adjusting anomaly threshold; reanalyzing data distribution characteristics using adjusted threshold to optimize fault warning.

[0038] Firstly, record warning response time T_response, i.e. the time interval from system issuing warning to actual response completion, for quantifying response efficiency. For example, in one embodiment, when patient's lung ventilation impedance anomaly is detected, the system triggers warning, medical staff intervenes for treatment; specifically, if sputum suction operation is completed within 10 seconds after warning is issued, T_response is recorded as 10 seconds.

[0039] Secondly, evaluate system performance parameters based on response effect, involving analyzing whether lung ventilation impedance data returns to normal after response to calculate false positive rate or false negative rate, etc. For example, specifically, if impedance value stabilizes in normal range after response, indicating that warning is effective, system performance parameters such as precision score improve; otherwise, if ineffective, optimization is needed.

[0040] Thirdly, adjust anomaly threshold, using formula: if T_response > T_max, then Th = Th * 0.95. Wherein, T_response is response time (range: 0 to several minutes, optimal value lower than T_max to minimize delay), T_max is preset maximum response time threshold (range: 30-60 seconds, optimal value set according to clinical needs, such as 40 seconds to ensure timely intervention), Th is anomaly threshold (initial value based on historical data distribution, range variable, optimal value balances sensitivity and specificity). The formula means: when response time is too long, reduce Th by 5% to make system more sensitive, thereby reducing future response delay. So set, because long response time may be caused by too high threshold leading to insufficient warning, by slightly reducing Th, anomaly detection can be enhanced. For example, in one embodiment, T_max is set as 40 seconds; if T_response is 50 seconds, Th is reduced from initial 100 units to 95 units, promoting earlier warning.

[0041] Finally, the data distribution characteristics are re-analyzed using the adjusted threshold, i.e., based on the new Th value, to process the multi-modal data (e.g., respiratory sound and pressure signals) to update the abnormal patterns and optimize the early warning algorithm. For example, specifically, in lung ventilation monitoring, the adjusted analysis shows that the impedance fluctuation characteristics are more easily identified, and the fault warning accuracy is improved by 10%.

[0042] Next, the re-analyzed data distribution characteristics of the present application are further described.

[0043] First, a moving average filter is applied to smooth the integrated lung ventilation impedance data to eliminate short-term noise and reveal long-term trends. For example, in one embodiment, a moving average filter with a window size of 5 is used to filter high-frequency interference on the impedance data fused from multi-modal sensors (e.g., respiratory flow and chest movement sensors).

[0044] Second, the mean μ smooth and standard deviation σ smooth of the smoothed data are calculated to quantify the center position and dispersion of the data. Specifically, in real-time monitoring, μ smooth and σ smooth are calculated based on a 30-second time window, for example, μ smooth reflects the average impedance level, and σ smooth measures the fluctuation range.

[0045] Next, the abnormality is detected based on the formula: if |ZMA|>M*σ smooth, it is marked as abnormal, where ZMA is the moving average integrated lung ventilation impedance value, and M is a preset multiplier. The parameter |ZMA| represents the current smoothed impedance value, ranging from the actual measured value (e.g., 0-100Ω); M ranges from 1-5, with an optimal value of 2.5, used to adjust the sensitivity. The meaning of this formula is to identify significant deviations through statistical thresholding, i.e., when the data point exceeds M times the standard deviation, it is considered abnormal. This formula is set based on the principle of normal distribution to ensure reliable detection of signal distortion in sudden events (e.g., patient coughing), while avoiding false positives. For example, in one embodiment, when M=2.5, if the smoothed impedance |ZMA| is 50Ω and σ smooth is 10Ω, the threshold M*σ smooth=25Ω; if |ZMA| exceeds 75Ω, it is marked as abnormal, corresponding to a patient apnea event.

[0046] Finally, when an abnormality is detected, a feedback control loop is triggered to dynamically adjust the fusion parameters (e.g., sensor weights) to address sudden signal distortion. Specifically, for example, after the abnormality is marked, the system automatically reduces the fusion weight of the noise sensor (e.g., the chest sensor affected by motion interference) to improve data quality.

[0047] Next, the trigger feedback control loop of the present application is further described. When an abnormality is confirmed, the normal state of each sensor individual data is checked; the meaning of this step is to verify whether each sensor data is within the predefined normal range, in order to distinguish between local faults and potential system problems. For example, in one embodiment, in a real-time lung ventilation impedance monitoring system based on multi-modal data fusion, the patient's breathing is monitored using pressure sensors and flow sensors; when an impedance abnormality is detected, the software checks whether each sensor reading deviates from the normal physiological range, such as whether the flow value is within 0.5-5 L / s.

[0048] The number of normal sensors N_normal is calculated; the meaning of this step is to count the total number of sensors currently working normally, which is used to evaluate the system reliability. For example, specifically in lung ventilation monitoring, the system deploys 5 sensors (such as 3 pressure sensors and 2 flow sensors); if 4 of them report normal data, then N_normal = 4.

[0049] Conditional judgment is made based on the formula: if N_normal < N_min, then a system-level abnormality is confirmed; where the parameter N_normal represents the number of normal sensors, ranging from 0 to the total number of sensors N_total (such as N_total = 5), with the optimal value close to N_total to ensure redundancy; the parameter N_min is a preset minimum normal sensor threshold, representing the minimum number of sensors required for the system to operate, usually ranging from 1 to N_total-1 (such as 2 or 3), with the optimal value set to ceil(N_total / 2) (for example, N_min = 3) to balance reliability and fault tolerance. The meaning of the formula N_normal < N_min is to determine whether the number of normal sensors is insufficient, thereby triggering the confirmation of a system-level abnormality; this formula is set in this way to avoid misjudgment when there are many sensor failures (such as sudden electromagnetic interference causing signal distortion), ensuring that the system only intervenes in high-risk situations. For example, in one embodiment, N_min = 3, if N_normal = 2 (such as two sensors fail), the formula is true, and the system determines a system-level abnormality.

[0050] After confirming the system-level abnormality, multi-sensor redundancy switching is performed to solve the sudden signal distortion; the meaning of this step is to activate the backup sensors or reconfigure the data source to restore the monitoring continuity. For example, specifically in lung ventilation impedance monitoring, the system automatically switches to a redundant sensor array (such as an additional backup sensor) to handle the signal distortion problem, ensuring that real-time impedance data fusion is not affected.

[0051] The real-time lung ventilation impedance monitoring method and system based on multi-modal data fusion of the application includes: first, real-time multi-modal data such as sound signals, airflow signals and pressure signals are acquired in real time through multiple sensors, which are integrated in the monitoring device, such as microphones, airflow meters and pressure sensors, to ensure the continuity and synchronicity of data acquisition, so as to capture the dynamic changes in the lung ventilation process. Secondly, the multi-modal data is fused and processed, and a weighted fusion algorithm or a deep learning model (such as a convolutional neural network) is used to integrate the complementary information of sound, airflow and pressure signals to generate comprehensive lung ventilation impedance data; this step generates a high-precision comprehensive data set by eliminating signal redundancy and enhancing key features, such as calculating the impedance change rate, thereby laying a foundation for subsequent analysis. Then, based on the comprehensive lung ventilation impedance data, the data distribution characteristics are analyzed to detect abnormalities, specifically including using statistical analysis methods (such as standard deviation calculation, probability density function modeling or clustering algorithms) to identify the shift, peak anomaly or outlier of data distribution, which can indicate sudden signal distortion, such as sensor noise interference or physiological state anomaly. Finally, when abnormal data distribution characteristics are detected, a fault warning system is triggered to solve the sudden signal distortion; the warning system includes a real-time alarm module and an automatic correction mechanism, such as signal filtering or redundant sensor switching to restore data integrity, while alerting the user, ensuring the reliability and timeliness of the monitoring.

[0052] In solving the technical problem of how to detect abnormal data distribution characteristics to monitor the system for fault warning to solve the sudden signal distortion, the application realizes it in the following way: first, in the analysis of data distribution characteristics stage, the system uses a machine learning model (such as a support vector machine) to recognize the comprehensive lung ventilation impedance data, quickly identify abnormal distribution (such as sudden drop in signal amplitude or frequency drift), these abnormal features are directly related to sudden signal distortion, such as sensor failure or environmental interference. Secondly, the warning system is designed to trigger based on threshold or adaptive rules: once the abnormal features are detected, the system immediately activates the fault diagnosis module to perform signal resampling or sensor calibration to eliminate the distortion source; at the same time, through the real-time feedback mechanism (such as sending a short message alarm or visual interface prompt), the user can manually intervene or the system automatically adjusts to ensure the continuous accuracy of lung ventilation impedance monitoring. This method effectively avoids misdiagnosis risk and improves the robustness and clinical practicability of the system. Overall, the application realizes the rapid response and solution to sudden signal distortion through multi-modal fusion and intelligent warning, ensuring the reliability and efficiency of real-time monitoring. (Word count: 365) The methods, programs, systems, apparatuses, etc. of embodiments of the present application can be implemented in a single or multiple computing devices or processing units, or portions thereof, in a distributed computing environment. In embodiments of the present application, tasks performed in these distributed computing environments can be performed by remote processing devices that are connected through a communication network.

[0053] Those skilled in the art will appreciate that embodiments of the present application can be provided as methods, systems or computer program products. Accordingly, the means for implementing the functions / acts of the embodiments illustrated in the above-described embodiments, or the related functions / acts, can be implemented in software, hardware, or a combination thereof.

[0054] Unless specifically stated otherwise, the acts or steps of the methods recited in the embodiments of the present application need not be performed in the order in which they are recited, and / or need not be performed in the order in which they are recited.

[0055] In this document, the terms "embodiment," "embodiments," "example," "exemplary," "specific example," or "some examples" mean serving as an example, instance, or illustration. Unless expressly indicated, the terms "embodiment" or "some embodiments" do not foreclose additional or

[0056] The exemplary systems and methods of this application have been described with reference to the specific embodiments and implementations thereof. It should be understood, however, that the systems and methods described herein are not limited to the embodiments described and / or illustrated, but can be practiced with modification and alteration within the scope of the appended claims.

Claims

1. A real-time lung ventilation impedance monitoring method and system based on multi-modal data fusion, characterized in that, Comprising: acquiring real-time multi-modal data from multiple sensors, the multi-modal data including sound signals, airflow signals, and pressure signals; performing fusion processing on the multi-modal data to generate comprehensive lung ventilation impedance data; analyzing data distribution characteristics based on the comprehensive lung ventilation impedance data to detect abnormalities; when detecting abnormal data distribution characteristics, triggering a fault warning system to address sudden signal distortion.

2. The real-time lung ventilation impedance monitoring method and system based on multi-modal data fusion according to claim 1, characterized in that, The analyzing data distribution characteristics to detect abnormalities further comprises: calculating the mean μ and standard deviation σ of the comprehensive lung ventilation impedance data; calculating the abnormality score of each data point based on the following formula: S = |Z μ| / σ, where Z is the value of the current data point, μ is the data mean, and σ is the standard deviation; if S is greater than a preset abnormality threshold Th, marking the data point as abnormal; when the proportion of abnormal data points exceeds a preset proportion P%, triggering the fault warning system to address sudden signal distortion.

3. The real-time lung ventilation impedance monitoring method and system based on multi-modal data fusion according to claim 2, characterized in that, The triggering the fault warning system to address sudden signal distortion further comprises: dividing the real-time data stream into fixed time windows T_window; within each time window, calculating the proportion of abnormal points R_anomaly of the comprehensive lung ventilation impedance data; based on the following formula: if R_anomaly > Q, detecting a window abnormality, where Q is a preset proportion threshold; when detecting a window abnormality, activating a warning signal and starting a sensor calibration module to address sudden signal distortion.

4. The real-time lung ventilation impedance monitoring method and system based on multi-modal data fusion according to claim 3, characterized in that, The activating a warning signal and starting a sensor calibration module further comprises: training a classifier model using historical data, with the model input being data distribution characteristics; inputting the data distribution characteristics of the current time window into the classifier model to obtain an abnormality probability P_anomaly; based on the following formula: if P_anomaly > P_th, confirming an abnormality and triggering an advanced warning, where P_th is a preset probability threshold; after triggering an advanced warning, performing sensor automatic calibration to correct signal distortion.

5. The real-time lung ventilation impedance monitoring method and system based on multi-modal data fusion according to claim 4, characterized in that, The performing sensor automatic calibration further comprises: when triggering an advanced warning, analyzing the signal quality scores Q_i of each sensor; calculating the drift D of the current comprehensive lung ventilation impedance data; based on the following formula: if D > D_th, adjusting the multi-modal data fusion weights, where D = | μ_current μ_reference|, μ_current is the current mean, μ_reference is the reference mean, and D_th is a preset drift threshold; after adjustment, re-fusing the data and updating the comprehensive lung ventilation impedance data to address sudden signal distortion.

6. The real-time lung ventilation impedance monitoring method and system based on multi-modal data fusion according to claim 5, characterized in that, The adjusting the multi-modal data fusion weights further comprises: monitoring environmental parameter changes ΔE, where ΔE includes temperature or humidity changes; updating the abnormality threshold Th based on the environmental parameters; adjusting the fusion weights based on the following formula: W_i = Q_i / Σ Q_i, where W_i is the weight of the i-th sensor, and Q_i is the signal quality score; if the weight change rate after adjustment is greater than a preset threshold, re-triggering data fusion to eliminate signal distortion.

7. The real-time lung ventilation impedance monitoring method and system based on multi-modal data fusion according to claim 6, characterized in that, The re-triggering data fusion further comprises: Defining a multi-level early warning system, where levels are based on anomaly severity; Calculating a severity score S_severity = R_anomaly * avg(S), where avg(S) is the average anomaly score; Condition checking based on the following formula: if S_severity > S1, trigger low-level early warning; if S_severity > S2, trigger medium-level early warning; if S_severity > S3, trigger high-level early warning, where S1, S2, S3 are preset severity thresholds; Taking corresponding measures for different early warning levels, including alarm notification or system shutdown, to address sudden signal distortion.

8. The real-time lung ventilation impedance monitoring method and system based on multi-modal data fusion according to claim 7, characterized in that, The execution of corresponding measures further includes: Recording the early warning response time T_response; Evaluating system performance parameters based on response effects; Adjusting anomaly thresholds based on the following formula: if T_response > T_max, Th = Th * 95, where T_max is the preset maximum response time threshold; Reanalyzing data distribution characteristics using adjusted thresholds to optimize fault early warning.

9. The real-time lung ventilation impedance monitoring method and system based on multi-modal data fusion according to claim 8, characterized in that, The reanalysis of data distribution characteristics further includes: Applying a moving average filter to smooth the comprehensive lung ventilation impedance data; Calculating the mean μ_smooth and standard deviation σ_smooth of the smoothed data; Detecting anomalies based on the following formula: if |ZMA| > M * σ_smooth, mark as anomaly, where MA is the moving average, M is the preset multiplier; When an anomaly is detected, triggering a feedback control loop to dynamically adjust the fusion parameters to address sudden signal distortion.

10. The real-time lung ventilation impedance monitoring method and system based on multi-modal data fusion according to claim 9, characterized in that, The triggering of the feedback control loop further includes: When the anomaly is confirmed, checking the normal state of individual sensor data; Calculating the number of normal sensors N_normal; Condition checking based on the following formula: if N_normal < N_min, confirm system-level anomaly, where N_min is the preset minimum normal sensor threshold; After confirming the system-level anomaly, performing multi-sensor redundancy switching to address sudden signal distortion.

Images