AI-based chronic obstructive pulmonary disease stable phase acute exacerbation prediction system

Abstract

The present application relates to the technical field of personal health risk assessment, and particularly relates to an AI-based acute exacerbation prediction system for stable chronic obstructive pulmonary disease, comprising a data acquisition module and a prediction module, the data acquisition module acquires physiological data of each resting moment in a preset time period after medication of a patient, including heart rate and blood oxygen saturation; the prediction module is used for obtaining a cardiorespiratory response efficiency ratio feature according to the increment of the physiological data compared with that before medication; obtaining a matching cost of a target wave peak of heart rate and a target wave peak of blood oxygen saturation according to the difference between the time interval of the target wave peak of heart rate and the target wave peak of blood oxygen saturation and a preset physiological delay time, so as to determine an optimal matching cost and value, thereby obtaining a matching deviation feature of heart rate and blood oxygen saturation; and performing acute exacerbation prediction for stable chronic obstructive pulmonary disease according to the cardiorespiratory response efficiency ratio feature and the matching deviation feature, so that multi-modal feature fusion is realized, and the accuracy of early warning is significantly improved.

Description

Technical Field

[0001] This invention relates to the field of personal health risk assessment technology, specifically to an AI-based system for predicting acute exacerbations of COPD during the stable phase. Background Technology

[0002] For patients with chronic obstructive pulmonary disease (COPD) managing their condition at home, daily monitoring of their condition is crucial. Currently, the mainstream methods for home monitoring primarily rely on finger-clip pulse oximeters or smart wearable devices to collect blood oxygen saturation and heart rate data. However, this conventional monitoring method has significant technical limitations in practical application.

[0003] In the early stages of worsening airway obstruction, the body compensates by increasing heart rate and respiratory rate. At this time, blood oxygen saturation can still be maintained within the normal range, exhibiting a phenomenon of "physiological efficacy decay under normal values." This causes systems that rely solely on threshold alarms to miss warnings and fail to detect potential declines in physiological efficacy in a timely manner. Existing technologies propose collecting blood oxygen saturation and heart rate data over a period of time after medication administration and predicting acute exacerbations by observing changes in these data. However, such methods only focus on calculating the macroscopic slope or mean of a single modality of physiological indicators and fail to conduct in-depth analysis of these physiological indicators, thus affecting the accuracy of predicting acute exacerbations during the stable phase of COPD. Summary of the Invention

[0004] To address the low accuracy of existing methods for predicting acute exacerbations in the stable phase of COPD, this invention aims to provide an AI-based system for predicting acute exacerbations in the stable phase of COPD. The specific technical solution adopted is as follows:

[0005] This invention provides an AI-based system for predicting acute exacerbations in the stable phase of COPD, comprising:

[0006] The data acquisition module is used to collect physiological data of the patient at various resting times within a preset time period after medication; the physiological data includes heart rate and blood oxygen saturation.

[0007] The prediction module, electrically connected to the data acquisition module, is configured as follows:

[0008] Based on the increase in physiological data at each resting time compared to the physiological data before medication, the cumulative increase in physiological data over a preset time period is obtained.

[0009] Based on the cumulative increment of the physiological data, the cardiopulmonary response efficiency ratio characteristic is obtained;

[0010] Based on the difference between the time interval between each target heart rate peak and each target blood oxygen saturation peak and the preset physiological delay time, the matching cost between each target heart rate peak and each target blood oxygen saturation peak is obtained, forming a matching cost matrix; the target heart rate peak and the target blood oxygen saturation peak are obtained from the physiological data increments at each resting time.

[0011] The target peak of heart rate, the target peak of blood oxygen saturation, and the matching cost matrix are input into the bipartite graph model to determine the optimal matching cost and value.

[0012] Based on the optimal matching cost and value, the matching deviation characteristics between heart rate and blood oxygen saturation are obtained;

[0013] Based on the cardiopulmonary response efficiency ratio characteristics and matching bias characteristics, acute exacerbations of COPD during the stable phase are predicted.

[0014] In an exemplary embodiment, the process of obtaining the resting time includes:

[0015] The preset time period is divided into multiple initial time periods;

[0016] Determine the patient's exercise intensity at each initial time point;

[0017] Based on the exercise intensity, a resting period is selected from the initial period; the time within the resting period is taken as the resting moment.

[0018] In an exemplary embodiment, the process of acquiring the cumulative increment of physiological data includes:

[0019] Set the initial cumulative increment of physiological data to 0;

[0020] Determine whether the time interval between two adjacent resting moments is greater than a preset time interval threshold. If it is less than or equal to the preset time interval threshold, update the cumulative increment of physiological data based on the physiological data increment of the two adjacent resting moments and the time interval between them. If it is greater than the preset time interval threshold, do not update the cumulative increment of physiological data.

[0021] Iterate through all resting moments within a preset time period to obtain the final cumulative increase in physiological data.

[0022] In one exemplary embodiment, the cumulative increment of the updated physiological data includes:

[0023] Calculate the average increment of physiological data between two adjacent resting moments;

[0024] The product of the average value of the physiological data increment and the time interval between them is used to obtain the update amount of the cumulative physiological data increment;

[0025] The cumulative increment of physiological data is updated based on the update amount.

[0026] In one exemplary embodiment, the cumulative increment of physiological data includes the cumulative increment of heart rate and the cumulative increment of blood oxygen saturation;

[0027] The process of obtaining the cardiopulmonary response efficiency ratio feature includes: calculating the ratio of the cumulative increase in heart rate to the cumulative increase in blood oxygen saturation to obtain the cardiopulmonary response efficiency ratio feature.

[0028] In one exemplary embodiment, the physiological data increments include heart rate increments and blood oxygen saturation increments;

[0029] The process of obtaining the target heart rate peak includes:

[0030] A sliding window operation is performed over a preset time period to obtain the maximum heart rate increment within each window;

[0031] The maximum heart rate increment that is greater than the preset heart rate increment peak value is selected from the maximum heart rate increments in each window and used as the target heart rate peak.

[0032] The process of obtaining the target peak of blood oxygen saturation includes:

[0033] Obtain the maximum increase in blood oxygen saturation within each window;

[0034] The maximum blood oxygen saturation increment that is greater than the preset peak value of blood oxygen saturation increment is selected from the maximum blood oxygen saturation increments in each window and used as the target peak value of blood oxygen saturation.

[0035] In an exemplary embodiment, the step of inputting the target heart rate peak, the target blood oxygen saturation peak, and the matching cost matrix into the bipartite graph model to determine the optimal matching cost and value includes:

[0036] The target peak of heart rate is taken as the left node of the bipartite graph, the target peak of blood oxygen saturation is taken as the right node of the bipartite graph, and the matching cost matrix is ​​taken as the edge weight.

[0037] The goal of solving the matching cost matrix is ​​to find the target solution that minimizes the sum of the matching costs of all selected pairs, thus obtaining the optimal matching cost sum.

[0038] In an exemplary embodiment, the process of obtaining the matching deviation features includes:

[0039] Determine the maximum number of target heart rate peaks and target blood oxygen saturation peaks;

[0040] The matching deviation feature is obtained by calculating the ratio of the optimal matching cost to the maximum number.

[0041] In an exemplary embodiment, the step of predicting acute exacerbations of COPD during the stable phase based on cardiopulmonary response efficiency ratio characteristics and matching bias characteristics includes:

[0042] When the cardiopulmonary response efficiency ratio feature is less than or equal to the first preset efficiency ratio threshold, and the matching deviation feature is less than or equal to the first preset matching deviation threshold, the prediction result of acute exacerbation of COPD in the stable period is determined to be normal physiological state.

[0043] When the cardiopulmonary response efficiency ratio feature is greater than the first preset efficiency ratio threshold and the matching deviation feature is less than the second preset matching deviation threshold, or when the cardiopulmonary response efficiency ratio feature is less than or equal to the first preset efficiency ratio threshold and the matching deviation feature is greater than the first preset matching deviation threshold and less than the second preset matching deviation threshold, the prediction result of acute exacerbation in the stable phase of COPD is determined to be early stage of airway obstruction exacerbation, and a secondary prompt signal is output; the second preset matching deviation threshold is greater than the first preset matching deviation threshold.

[0044] When the matching deviation feature is greater than or equal to the second preset matching deviation threshold, the prediction result of acute exacerbation in the stable phase of COPD is determined to be a severe increase in airway resistance, and a first-level warning signal is output.

[0045] In an exemplary embodiment, before predicting acute exacerbations of COPD during the stable phase based on the cardiopulmonary response efficiency ratio characteristics and the matching bias characteristics, the cardiopulmonary response efficiency ratio characteristics and the matching bias characteristics are standardized respectively.

[0046] This invention has the following beneficial effects: By using AI to process relevant physiological data of patients, it accurately predicts acute exacerbations of COPD during the stable phase. Specifically, the physiological data at resting times within a preset time period after medication administration can filter out motion noise, ensuring that the data reflects a pure pathophysiological response, thus improving the accuracy and reliability of subsequent data processing. The cumulative increase in physiological data over the preset time period highlights the dynamic changes in physiological data under drug intervention. The cardiopulmonary response efficiency ratio feature enables in-depth mining of cardiopulmonary synergistic response efficiency, improving prediction accuracy. The matching cost matrix characterizes the target peak of heart rate and the target peak of blood oxygen saturation. The study clarifies the degree to which the peak time interval conforms to normal physiological patterns, defines the asynchronicity and matching relationship between heart rate and blood oxygen saturation responses, and quantifies the temporal coupling deviation of cardiopulmonary responses. A graph theory optimization algorithm is introduced to process physiological signal matching, avoiding mismatches caused by simple threshold comparisons. Global optimal matching replaces local point-to-point comparisons, resulting in higher robustness. Finally, based on the cardiopulmonary response efficiency ratio characteristics and matching deviation characteristics, the study predicts acute exacerbations during the stable phase of COPD, achieving multimodal feature fusion. This characterizes cardiopulmonary function status from both efficiency and synchronicity dimensions, significantly improving the accuracy of early warning compared to coarse predictions relying solely on the trends of blood oxygen saturation and heart rate changes. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the module composition of an AI-based COPD stable-phase acute exacerbation prediction system provided in one embodiment of the present invention;

[0048] Figure 2 This is a data processing flowchart of a prediction module provided in one embodiment of the present invention. Detailed Implementation

[0049] This embodiment provides an AI-based system for predicting acute exacerbations of stable COPD (hereinafter referred to as the system), used to predict the risk of acute exacerbations of stable COPD in patients with chronic obstructive pulmonary disease (hereinafter referred to as patients). Figure 1 As shown, the system includes a data acquisition module and a prediction module. The data acquisition module and the prediction module are electrically connected. The connection between the data acquisition module and the prediction module can be wired or wireless (such as, but not limited to, Bluetooth, WiFi).

[0050] To capture subtle physiological changes caused solely by drug action, a unified time coordinate system must be established. In this embodiment, the time point of each patient's medication administration is defined as the drug trigger moment. Recording each medication trigger moment can be done manually, for example, by the patient manually triggering the relevant feedback while taking the medication, such as by manually clicking the medication trigger button set in the system; or, each medication trigger moment can also be recorded automatically by relevant devices. Taking a medication event specifically referring to a maintenance medication (such as a long-acting bronchodilator) that the patient uses regularly every day as an example, the system communicates with the smart inhaler used by the patient. When the electronic contact closure signal of the smart inhaler is detected, it indicates that the patient has completed the release of the inhaler (i.e., completed the medication), thus obtaining each medication trigger moment.

[0051] Based on each drug trigger time, a preset time period is determined after each medication administration. It should be understood that the start time of the preset time period is later than the drug trigger time to avoid the physiological delay period before the drug takes effect. Additionally, for smart inhalers, this also filters out non-pharmacological signal fluctuations caused by limb movements and breathing adjustments during inhalation, ensuring that subsequent data analysis accurately reflects the physiological response after drug action. The time interval between the start time of the preset time period and the drug trigger time is set according to actual conditions; this embodiment uses 180s (where s represents seconds) as an example. The duration of the preset time period is set according to actual conditions; this embodiment uses 1620s as an example. Therefore, the preset time period can be expressed as: .

[0052] The data acquisition module collects the patient's physiological data, including heart rate and blood oxygen saturation. Accordingly, the data acquisition module includes a smart wearable device, such as a smartwatch with real-time heart rate and blood oxygen saturation monitoring capabilities. Heart rate and blood oxygen saturation are sampled at the same frequency and simultaneously; the sampling frequency is set according to the actual testing needs, such as 2Hz.

[0053] Because patients may engage in unexpected physical activities during the preset time period, such as walking or housework, these activities can interfere with physiological data, such as introducing non-pathological increases in heart rate. To eliminate this noise, this embodiment requires processing the physiological data within the preset time period to remove interference caused by movement.

[0054] First, it is necessary to obtain the patient's resting moments within a preset time period after medication administration. Resting moments represent times when the patient is not moving. In an exemplary embodiment, the data acquisition module further includes a triaxial accelerometer mounted on the patient, used to collect triaxial acceleration data at preset sampling frequencies at various moments, thus representing movement. The sampling frequency of the triaxial accelerometer can be the same as the sampling frequency of heart rate and blood oxygen saturation, and they can be sampled synchronously.

[0055] The preset time period is divided into multiple initial time periods. It should be understood that the initial time period serves as the unit time period for resting time analysis. To improve the reliability of data analysis, its length does not need to be large; this embodiment uses 10 seconds as an example. There is no overlap between the initial time periods. If the preset time period cannot be evenly divided into multiple initial time periods, the last remaining time period is used as the last initial time period.

[0056] For any given initial time period, the patient's exercise intensity during that initial time period is determined. In an exemplary embodiment, the standard deviation of the triaxial acceleration during that initial time period is calculated as the exercise intensity. Since triaxial acceleration includes acceleration data in the X, Y, and Z axes, a specific formula for calculating exercise intensity is given below:

[0057] ;

[0058] in, This represents the motion intensity during the initial time period, where N represents the number of moments within that initial time period. This represents the acceleration in the X-axis direction at time i within the initial time period. This represents the average acceleration in the X-axis direction during this initial time period. This represents the acceleration in the Y-axis direction at time i within the initial time period. This represents the average acceleration in the Y-axis direction during this initial time period. This represents the acceleration in the Z-axis direction at time i within the initial time period. This represents the average acceleration along the Z-axis during the initial time period. It should be understood that in this embodiment, the units of acceleration in all three directions are the units of gravitational acceleration g.

[0059] Using the above method, the motion intensity (in units of gravitational acceleration g) for each initial time period is obtained. Motion intensity characterizes the degree of acceleration fluctuation. The greater the motion intensity, the more intense the patient's movement within the corresponding initial time period, and the less likely the included moments are resting moments. In an exemplary embodiment, this embodiment presets an acceleration fluctuation threshold, the specific value of which is obtained empirically; here, 0.02g is used as an example. The motion intensity of each initial time period is compared with the acceleration fluctuation threshold: if it is greater than or equal to the acceleration fluctuation threshold, the patient is determined to be in motion within the corresponding initial time period; if it is less than the acceleration fluctuation threshold, the patient is determined to be in resting state within the corresponding initial time period. Thus, each initial time period corresponding to motion intensity less than the acceleration fluctuation threshold is obtained, and each initial time period corresponding to motion intensity less than the acceleration fluctuation threshold is determined as a resting time period. Each moment within a resting time period is considered a resting moment. In addition, heart rate and blood oxygen saturation in each initial time period corresponding to exercise intensity below the acceleration fluctuation threshold are marked as valid data; heart rate and blood oxygen saturation in each initial time period corresponding to exercise intensity greater than or equal to the acceleration fluctuation threshold are marked as invalid and discarded.

[0060] Through the above screening, physiological data of patients at resting moments within a preset time period after medication are obtained. Specifically, the physiological data may be discontinuous on the time axis (interrupted by the initial period corresponding to the movement state), but are uniformly in the resting state in terms of physiological state.

[0061] The prediction module receives patient physiological data collected by the data acquisition module and then processes the received data signals. The prediction module can be a conventional data processing chip, such as a CPU (Central Processing Unit), or a System on Chip (SoC). The SoC can be configured in a medical monitoring platform or in the patient's smart terminal.

[0062] Figure 2 The implementation steps of the data processing process for the prediction module are as follows: Figure 2 As shown, the data processing steps of the prediction module include:

[0063] Step S1: Based on the increase in physiological data at each resting time compared to the physiological data before medication, obtain the cumulative increase in physiological data over a preset time period;

[0064] Step S2: Based on the cumulative increment of physiological data, obtain the cardiopulmonary response efficiency ratio characteristics;

[0065] Step S3: Based on the difference between the time interval between each heart rate target peak and each blood oxygen saturation target peak and the preset physiological delay time, obtain the matching cost between each heart rate target peak and each blood oxygen saturation target peak, and construct a matching cost matrix.

[0066] Step S4: Input the target heart rate peak, target blood oxygen saturation peak, and matching cost matrix into the bipartite graph model to determine the optimal matching cost and value;

[0067] Step S5: Based on the optimal matching cost and value, obtain the matching deviation characteristics between heart rate and blood oxygen saturation;

[0068] Step S6: Based on the cardiopulmonary response efficiency ratio characteristics and matching bias characteristics, predict acute exacerbations during the stable phase of COPD.

[0069] The specific implementation process of each step is explained below.

[0070] Step S1: Based on the increase in physiological data at each resting time compared to the physiological data before medication, obtain the cumulative increase in physiological data over a preset time period.

[0071] This embodiment uses the most recent medication administration as an example. After medication administration, the patient's physiological data usually increases compared to before administration. Therefore, the increase in physiological data at each resting time compared to before medication is obtained as the relative change caused by the drug. To obtain the increase in physiological data before and after medication, it is necessary to first obtain the physiological data before medication as the pre-medication baseline value. The pre-medication baseline value can be a preset physiological standard value (e.g., heart rate of 75 bpm, blood oxygen saturation of 95%). To ensure that the pre-medication baseline value is more relevant to the patient, in an exemplary embodiment, a baseline reference time interval before medication is obtained, such as the time of drug triggering. The preceding moment is used as the set duration of the time interval for the cutoff moment. In this embodiment, the numerical range of the baseline reference time interval is 600s-1800s. The specific value is set according to actual needs. In this embodiment, the longest time interval (i.e., 1800s) is selected. At the same time, the above process is used to obtain each resting moment within the baseline reference time interval.

[0072] Heart rate is obtained at each resting moment within the baseline reference time interval, and the average heart rate is calculated as the pre-medication baseline heart rate value. Similarly, blood oxygen saturation is obtained at each resting moment within the baseline reference time interval, and the average blood oxygen saturation is calculated as the pre-medication baseline blood oxygen saturation value. It should be understood that if the number of resting moments within the baseline reference time interval is too small, such as less than 12, it indicates that the patient was in a state of continuous activity for a period of time before medication. In this case, the data is insufficient to represent the resting state, and the aforementioned preset physiological standard value can be directly used as the pre-medication baseline value.

[0073] The physiological data increments include heart rate increments and blood oxygen saturation increments. For any resting moment within a preset time period, the heart rate at that resting moment is subtracted from the baseline heart rate before medication, and the difference is taken as the heart rate increment at that resting moment. It should be understood that if the difference is negative, it indicates that the heart rate after medication is lower than the baseline heart rate before medication, which may be caused by measurement noise. Moreover, this embodiment requires a positive increment, so when the difference is negative, the heart rate increment is limited to 0. Similarly, the blood oxygen saturation at that resting moment is subtracted from the baseline blood oxygen saturation before medication, and the difference is taken as the blood oxygen saturation increment at that resting moment. It should be understood that if the difference is negative, it indicates that the blood oxygen saturation after medication is lower than the baseline blood oxygen saturation before medication, which may be caused by measurement noise. Moreover, this embodiment requires a positive increment, so when the difference is negative, the blood oxygen saturation increment is limited to 0. Thus, the heart rate increments and blood oxygen saturation increments at each resting moment within the preset time period are obtained.

[0074] Based on the heart rate and blood oxygen saturation increments at various resting moments within a preset time period, the data are encapsulated into a physiological data increment sequence in chronological order. :

[0075] ;

[0076] in, This indicates the first resting moment within a preset time period. This indicates the increase in heart rate at the first resting moment. This represents the increase in blood oxygen saturation at the first resting moment. This indicates the second resting moment within a preset time period. This indicates the increase in heart rate at the second resting moment. This indicates the increase in blood oxygen saturation at the second resting moment. This indicates the third resting moment within a preset time period. This indicates the heart rate increment at the third resting moment. This indicates the increase in blood oxygen saturation at the third resting moment. This represents the k-th resting moment within a preset time period. This represents the heart rate increment at the k-th resting moment. This represents the increase in blood oxygen saturation at the k-th resting moment. This represents the Mth resting moment within a preset time period. This represents the heart rate increment at the Mth resting moment. This represents the increase in blood oxygen saturation at the Mth resting moment, where M represents the number of resting moments within the preset time period.

[0077] Incremental sequences of physiological data It is a time-discrete sequence containing only physiological data increments during the period of drug action and at rest. Therefore, based on the physiological data increment sequence... The cumulative increment of physiological data over a preset time period is obtained. The cumulative increment of physiological data represents the accumulation of physiological data increment within the preset time period. This yields the cardiopulmonary response efficiency ratio characteristic, which reflects the patient's physiological response pattern from a macroscopic perspective.

[0078] To quantify the cumulative physiological cost borne by patients during the drug's effect (i.e., the preset time period) to maintain oxygenation levels, this embodiment incrementally sequences physiological data. Perform integral calculations. Due to the incremental sequence of physiological data... Since the time sequence is discontinuous and there are breakpoints on the time axis, directly using conventional integration will lead to errors. Therefore, this embodiment performs a piecewise trapezoidal integration operation with time interval constraints, as follows:

[0079] First, the initial cumulative increment of physiological data is set to 0, that is, the initial cumulative increment of heart rate and cumulative increment of blood oxygen saturation are both 0.

[0080] Then, according to the physiological data increment sequence In chronological order, traverse the incremental sequence of physiological data. Each two adjacent resting moments are considered. In this embodiment, taking the k-th resting moment and the (k+1)-th resting moment as examples, the time interval between the k-th resting moment and the (k+1)-th resting moment is determined. Determine the time interval Is the time interval greater than a preset time interval threshold? The preset time interval threshold is an introduced preset integral interpolation allowable interval, which is set to 15 seconds in this embodiment. If the time interval... If the time interval is less than or equal to a preset time interval threshold, it is determined that the k-th resting moment and the (k+1)-th resting moment are physiologically continuous and can be linearly interpolated. Then, based on the physiological data increments and time intervals at the k-th and (k+1)-th resting moments, the... Update the cumulative increment of physiological data to obtain the cumulative increment of physiological data up to the (k+1)th resting time; if the time interval... If the time interval exceeds a preset threshold, it indicates a prolonged period of motion interference or data loss between the k-th and (k+1)-th resting moments. In this case, interpolation would introduce significant errors. Therefore, the accumulated physiological data increment is not updated, and the process proceeds directly to the next two adjacent resting moments. This process iterates through all resting moments within the preset time period to obtain the final accumulated physiological data increment.

[0081] The process of updating the cumulative increment of physiological data is as follows: Calculate the average increment of physiological data at the k-th resting time and the (k+1)-th resting time; then calculate the average increment of physiological data and the time interval. The product of the products is used as the update amount of the cumulative physiological data increment corresponding to the (k+1)th resting time. Finally, the cumulative physiological data increment is updated based on the update amount of the cumulative physiological data increment corresponding to the (k+1)th resting time.

[0082] The cumulative increment of physiological data includes the cumulative increment of heart rate and the cumulative increment of blood oxygen saturation. Based on the above process, the specific loop implementation process is given below:

[0083] The cumulative increment of physiological data corresponding to the first resting moment is set as the initial cumulative increment of physiological data, that is, the cumulative increment of heart rate and cumulative increment of blood oxygen saturation corresponding to the first resting moment are both 0.

[0084] For the cumulative increase in physiological data corresponding to the second resting moment, determine the time interval between the first and second resting moments. Is the time interval greater than the preset threshold? If the data is less than or equal to a preset time interval threshold, then the physiological data increment and time interval at the first and second resting moments will be used as the basis for the calculation. Update the cumulative physiological data increment to obtain the cumulative physiological data increment up to the second resting time. The calculation method is as follows:

[0085] ;

[0086] ;

[0087] in, This indicates the cumulative increase in heart rate up to the second resting moment. This represents the initial cumulative heart rate increment (which is 0). This represents the update amount of the cumulative increase in heart rate corresponding to the second resting moment. This indicates the cumulative increase in blood oxygen saturation up to the second resting time. This represents the initial cumulative increase in blood oxygen saturation (which is 0). This represents the update amount of the cumulative increase in blood oxygen saturation corresponding to the second resting time.

[0088] If time interval If the cumulative increase in physiological data exceeds the preset time interval threshold, the cumulative increase in heart rate up to the second resting time will not be updated. The cumulative increase in heart rate up to the second resting time will be the initial cumulative increase in heart rate, and the cumulative increase in blood oxygen saturation up to the second resting time will be the initial cumulative increase in blood oxygen saturation.

[0089] For the cumulative increase in physiological data corresponding to the third resting moment, determine the time interval between the second and third resting moments. Is the time interval greater than the preset threshold? If the data is less than or equal to a preset time interval threshold, then the physiological data increments and time intervals at the second and third resting moments will be used as the basis for the calculation. Update the cumulative physiological data increment to obtain the cumulative physiological data increment up to the 3rd resting time. The calculation method is as follows:

[0090] ;

[0091] ;

[0092] in, This indicates the cumulative increase in heart rate up to the third resting moment. This represents the update amount of the cumulative heart rate increment corresponding to the 3rd resting moment. This indicates the cumulative increase in blood oxygen saturation up to the third resting time. This represents the update amount of the cumulative increase in blood oxygen saturation corresponding to the 3rd resting time.

[0093] If time interval If the cumulative increase in physiological data exceeds the preset time interval threshold, the cumulative increase in heart rate up to the third resting time is equal to the cumulative increase in heart rate up to the second resting time, and the cumulative increase in blood oxygen saturation up to the third resting time is equal to the cumulative increase in blood oxygen saturation up to the second resting time.

[0094] Similarly, for the cumulative increment of physiological data corresponding to the (k+1)th resting moment, if the time interval... If the data is less than or equal to a preset time interval threshold, then the physiological data increments and time intervals at the k-th and (k+1)-th resting times are used as the basis. Update the cumulative increment of physiological data to obtain the cumulative increment of physiological data up to the (k+1)th resting time. The calculation method is as follows:

[0095] ;

[0096] ;

[0097] in, This represents the cumulative increase in heart rate up to the (k+1)th resting time. This represents the cumulative increase in heart rate up to the k-th resting time. This represents the increase in heart rate at the (k+1)th resting moment. This represents the update amount of the cumulative heart rate increment corresponding to the (k+1)th resting time. This represents the increase in blood oxygen saturation at the (k+1)th resting time. This represents the cumulative increase in blood oxygen saturation up to the (k+1)th resting time. This represents the cumulative increase in blood oxygen saturation up to the k-th resting time. This represents the update amount of the cumulative increase in blood oxygen saturation at the (k+1)th resting time.

[0098] If time interval If the time interval exceeds the preset threshold, the cumulative increase in physiological data will not be updated. The cumulative increase in heart rate up to the (k+1)th resting time is equal to the cumulative increase in heart rate up to the kth resting time, and the cumulative increase in blood oxygen saturation up to the (k+1)th resting time is equal to the cumulative increase in blood oxygen saturation up to the kth resting time.

[0099] Until the last resting moment, i.e., the Mth resting moment, the time interval between the Mth resting moment and the (M-1)th resting moment is determined by the cumulative increment of physiological data corresponding to the Mth resting moment. If the time interval If the value is less than or equal to a preset time interval threshold, then the physiological data increment and time interval between the Mth resting moment and the (M-1)th resting moment are used as the basis. The cumulative physiological data increment is updated to obtain the cumulative physiological data increment up to the Mth resting time. The cumulative physiological data increment up to the Mth resting time is used as the final calculated cumulative physiological data increment. The calculation method is as follows:

[0100] ;

[0101] ;

[0102] in, This represents the cumulative increase in heart rate up to the Mth resting time. This represents the cumulative increase in heart rate up to the (M-1)th resting time. This represents the heart rate increment at rest time M-1. This represents the update amount of the cumulative heart rate increment corresponding to the Mth resting moment. This represents the cumulative increase in blood oxygen saturation up to the Mth resting time. This represents the cumulative increase in blood oxygen saturation up to the (M-1)th resting time. This represents the increase in blood oxygen saturation at the (M-1)th resting time. This represents the update amount of the cumulative increase in blood oxygen saturation at the Mth resting time.

[0103] If time interval If the cumulative increase in physiological data exceeds the preset time interval threshold, the cumulative increase in heart rate up to the Mth resting time is equal to the cumulative increase in heart rate up to the (M-1)th resting time, and the cumulative increase in blood oxygen saturation up to the Mth resting time is equal to the cumulative increase in blood oxygen saturation up to the (M-1)th resting time.

[0104] Step S2: Based on the cumulative increment of physiological data, obtain the cardiopulmonary response efficiency ratio characteristics.

[0105] Based on step S1, the final cumulative increase in heart rate and cumulative increase in blood oxygen saturation are obtained, and the cardiopulmonary response efficiency ratio is calculated accordingly. In an exemplary embodiment, the calculation method is given below:

[0106] ;

[0107] in, This indicates the cardiopulmonary response efficiency ratio characteristic. This indicates the final cumulative increase in heart rate. This represents the final cumulative increase in blood oxygen saturation. This represents the preset integral smoothing constant (in this embodiment, it is denoted as...). For example, its dimension is the same as the cumulative increase in blood oxygen saturation. Its function is to prevent when When the value is 0 (i.e. the drug does not produce any improvement in blood oxygenation), the denominator of the formula is zero, causing the calculation to crash. At the same time, it also smooths out the numerical fluctuations in the low-yield range.

[0108] Cardiopulmonary response efficiency ratio characteristics The higher the value, the greater the increase in heart rate required to produce one unit of effective improvement in blood oxygenation (the denominator), which is the numerator. In the compensated phase of COPD, patients often need to significantly increase their heart rate to maintain normal blood oxygenation; therefore, the cardiopulmonary response efficiency ratio is a key characteristic. The value will increase significantly, thus enabling the quantification of early inefficient compensatory states.

[0109] Step S3: Based on the difference between the time interval between each heart rate target peak and each blood oxygen saturation target peak and the preset physiological delay time, obtain the matching cost between each heart rate target peak and each blood oxygen saturation target peak, and construct a matching cost matrix.

[0110] First, the target peak of heart rate within the preset time period is obtained based on the heart rate increment at each resting moment within the preset time period, and the target peak of blood oxygen saturation within the preset time period is obtained based on the blood oxygen saturation increment at each resting moment within the preset time period.

[0111] This embodiment uses a local maximum detection algorithm to obtain the target peaks of heart rate and blood oxygen saturation. Specifically, it employs a sliding window method to incrementally sequence physiological data over a preset time period. Perform a sliding window operation, where the window width is set according to actual needs. In this embodiment, 30 seconds is taken as an example, and the sliding step is the window width.

[0112] During each window slide, the maximum heart rate increment within the window is acquired, thus obtaining the maximum heart rate increment for each window. To filter out noise fluctuations, this embodiment presets a peak heart rate increment value, which is set according to actual judgment needs, such as 2 bpm. The maximum heart rate increment within each window is compared with the magnitude of this peak heart rate increment. The maximum heart rate increment greater than this peak heart rate increment is selected from the maximum heart rate increments within each window and taken as the target heart rate peak.

[0113] Similarly, during each window sliding operation, the maximum blood oxygen saturation increment within the window is obtained, thus yielding the maximum blood oxygen saturation increment for each window. To filter out noise fluctuations, this embodiment presets a peak value for the blood oxygen saturation increment. The value of this peak value is set according to actual judgment needs, such as 1%. The maximum blood oxygen saturation increment within each window is compared with the magnitude of this peak value. The maximum blood oxygen saturation increment greater than this peak value is selected from the maximum blood oxygen saturation increments within each window and taken as the target blood oxygen saturation peak.

[0114] The target heart rate peaks within a preset time period are sorted chronologically to obtain a set of local heart rate peak events, and the target blood oxygen saturation peaks within the same preset time period are sorted chronologically to obtain a set of local blood oxygen saturation peak events. It should be understood that if the number of elements in both the local heart rate peak event set and the local blood oxygen saturation peak event set is greater than 0, indicating the existence of a certain number of target heart rate and blood oxygen saturation peaks, subsequent data processing can proceed normally. If at least one set in either the local heart rate peak event set or the local blood oxygen saturation peak event set has a number of elements equal to 0, it indicates that the physiological data within the preset time period is too stable and lacks micro-fluctuation characteristics. In this case, the subsequent matching deviation characteristics can be directly assigned a preset default value (set to 0 in this embodiment, representing no significant decoupling risk) to ensure the robustness of the data analysis.

[0115] Based on the difference between the time interval between each target heart rate peak and each target blood oxygen saturation peak and the preset physiological delay time, the matching cost of each target heart rate peak and each target blood oxygen saturation peak is obtained, forming a matching cost matrix. The matching cost matrix is ​​used as input to the subsequently constructed bipartite graph model and solved to quantify the temporal synchronicity of heart rate fluctuations and blood oxygen saturation fluctuations.

[0116] To assess the potential causal relationship between target heart rate peaks and target blood oxygen saturation peaks, it is necessary to quantify the degree of irrationality (i.e., matching cost) of any pair of peaks (i.e., any target heart rate peak and any target blood oxygen saturation peak). This embodiment constructs a matching cost matrix W of dimension H×J, where H represents the number of target heart rate peaks within the local heart rate peak event set, and J represents the number of target blood oxygen saturation peaks within the local blood oxygen saturation peak event set. Elements within the matching cost matrix W... This represents the matching cost corresponding to pairing the h-th heart rate target peak with the j-th blood oxygen saturation target peak.

[0117] Determine the time interval between the h-th target heart rate peak and the j-th target blood oxygen saturation peak. Specifically, this time interval is the length of time by which the h-th target heart rate peak precedes the j-th target blood oxygen saturation peak. If the time interval is greater than 0, it means that the h-th target heart rate peak precedes the j-th target blood oxygen saturation peak. If the time interval is less than 0, it means that the h-th target heart rate peak lags behind the j-th target blood oxygen saturation peak. If the time interval is equal to 0, it means that the h-th target heart rate peak and the j-th target blood oxygen saturation peak are synchronized.

[0118] This embodiment presets a physiological delay time, which characterizes the normal physiological transmission delay between heart rate and blood oxygen saturation in the cardiovascular system. In this embodiment, it is set to 20 seconds. The smaller the difference between the time interval between the h-th target heart rate peak and the j-th target blood oxygen saturation peak and this preset physiological delay time, the more the time interval between the h-th target heart rate peak and the j-th target blood oxygen saturation peak conforms to the normal physiological transmission pattern, and the lower the matching cost.

[0119] The formula for calculating the matching cost between the h-th target heart rate peak and the j-th target blood oxygen saturation peak is as follows:

[0120] ;

[0121] in, This represents the matching cost between the h-th target heart rate peak and the j-th target blood oxygen saturation peak. This represents the time interval between the h-th target heart rate peak and the j-th target blood oxygen saturation peak. This represents the preset causality penalty value, set as a very large constant (1000 in this embodiment, dimensionless). This means that the target peak of blood oxygen saturation appears before the target peak of heart rate, which violates the physiological causal law that heart rate drives changes in blood oxygen saturation. The purpose of assigning a large penalty value is to avoid selecting such a causal pairing as much as possible in subsequent matching algorithms. This indicates the preset physiological delay time. This represents the preset time weighting coefficient, which is set to [value] in this embodiment. , The unit of measurement, the reciprocal of a second, is used to represent differences in the time dimension. Convert to dimensionless matching value.

[0122] Using the above process, the matching cost between each target heart rate peak and each target blood oxygen saturation peak is obtained, thus forming a matching cost matrix W. Each matching cost in the matching cost matrix W intuitively reflects the temporal and morphological matching difficulty between the corresponding target heart rate peak and target blood oxygen saturation peak. The smaller the matching cost value, the more reasonable the matching.

[0123] Step S4: Input the target heart rate peak, target blood oxygen saturation peak, and matching cost matrix into the bipartite graph model to determine the optimal matching cost and value.

[0124] After constructing the matching cost matrix W, a graph theory algorithm is used to find the globally optimal pairing scheme by combining the local peak event sets of heart rate and blood oxygen saturation to evaluate the overall signal synchronization. Specifically, the local peak event sets of heart rate and blood oxygen saturation, along with the matching cost matrix, are input into a bipartite graph model to determine the optimal matching cost and value, as follows:

[0125] Using the local peak heart rate event set as the left node of the bipartite graph and the local peak blood oxygen saturation event set as the right node, and the matching cost matrix W as the edge weight, this is a typical bipartite graph minimum weight matching problem (or assignment problem).

[0126] The matching cost matrix W is solved using either the Hungarian Algorithm or the Kuhn-Munkres Algorithm in an exemplary embodiment. The goal of the algorithm is to find a one-to-one mapping scheme (i.e., a set of non-conflicting pairing indices in the matching cost matrix W), specifically, to find a target scheme that minimizes the sum of the matching costs of all selected pairs. The minimum sum of matching costs is taken as the optimal matching cost sum.

[0127] It should be understood that if H and J are not equal, the algorithm allows some nodes not to participate in the matching. For unmatched nodes, the matching cost matrix W can be completed into a square matrix by adding virtual nodes. The matching cost of unmatched nodes in the square matrix can be set to a sufficiently large constant (e.g., greater than the sum of all matching costs) to ensure that the algorithm prioritizes matching real nodes.

[0128] Step S5: Based on the optimal matching cost and value, obtain the matching deviation characteristics between heart rate and blood oxygen saturation.

[0129] After obtaining the optimal matching cost and value, in order to eliminate the influence of the number of target heart rate peaks and target blood oxygen saturation peaks on the magnitude of the optimal matching cost and value, the result is processed. The processed result serves as a matching deviation feature between heart rate and blood oxygen saturation. The processing procedure is as follows:

[0130] ;

[0131] in, This indicates the characteristics of the mismatch between heart rate and blood oxygen saturation. This represents the optimal matching cost and value. This indicates that the maximum value among H, J, and 3 is taken. The reason for including the value 3 when taking the maximum number of target heart rate and target oxygen saturation peaks is to smooth out cases where the number of target heart rate and target oxygen saturation peaks is too small, preventing a single peak from determining the overall score. If both H and J are greater than 3, then... Essentially, it takes the maximum value between H and J.

[0132] Heart rate and blood oxygen saturation matching deviation characteristics The magnitude of the value directly reflects the state of cardiopulmonary coupling: a small value indicates that the target peaks of heart rate and blood oxygen saturation correspond well in time (consistent with physiological delay), suggesting tight cardiopulmonary coupling, typically corresponding to normal recovery or a healthy state after exercise; a large value indicates that regardless of how the target peaks of heart rate and blood oxygen saturation are paired, a large number of target peaks will violate the time sequence or fail to match (high matching cost), indicating that heart rate fluctuations have failed to effectively drive changes in blood oxygen saturation, i.e., cardiopulmonary decoupling has occurred. This decoupling characteristic is a specific manifestation of worsening airway obstruction in COPD. Through the above data processing, the complex time-series waveform relationship is transformed into a single quantitative indicator. .

[0133] Step S6: Based on the cardiopulmonary response efficiency ratio characteristics and matching bias characteristics, predict acute exacerbations during the stable phase of COPD.

[0134] Based on the cardiopulmonary response efficiency ratio characteristics and matching bias characteristics, the feature vector is obtained. According to the feature vector Predicting acute exacerbations during the stable phase of COPD.

[0135] It should be understood that, in order to further improve the reliability of prediction and eliminate the influence of individual basic physiological differences and different feature dimensions, this embodiment standardizes the cardiopulmonary response efficiency ratio characteristics and matching bias characteristics before predicting acute exacerbations of COPD in the stable phase based on the cardiopulmonary response efficiency ratio characteristics and matching bias characteristics. In an exemplary embodiment, a specific implementation process of standardization is given below:

[0136] Using the above process, based on physiological data collected from multiple stable COPD patients (defined as historical patients, e.g., 500 patients), the cardiopulmonary response efficiency ratio characteristics and matching bias characteristics of each historical patient were obtained. Then, the mean of the cardiopulmonary response efficiency ratio characteristics was calculated based on these characteristics. and standard deviation The mean of the matching deviation characteristics was calculated based on the matching deviation characteristics of each historical patient. and standard deviation .

[0137] The Z-Score standardization method was used to standardize the cardiopulmonary response efficiency ratio and matching bias characteristics respectively:

[0138] ;

[0139] ;

[0140] in, This represents the standardized cardiopulmonary response efficiency ratio characteristic. This represents the standardized matching bias characteristics. The minimum standard deviation threshold representing the cardiopulmonary response efficiency ratio characteristic. Its purpose is to prevent the group data from becoming too uniform (i.e., When the denominator is extremely small, the cardiopulmonary response efficiency ratio characteristic becomes overly sensitive to minute fluctuations. Introducing the max function ensures a lower limit for the denominator, thus suppressing false alarms caused by numerical noise. Similarly, The minimum standard deviation threshold representing the matching bias characteristic.

[0141] and This directly reflects the deviation of the patient's current physiological state from that of a similar patient group. A value greater than 0 indicates that the indicator is above average (with a certain risk), and a larger value indicates a more severe deviation. For example, This means that the patient's cardiopulmonary decoupling level is 2 standard deviations higher than the group average, which is considered a significant abnormality.

[0142] Based on the standardized cardiopulmonary response efficiency ratio characteristics and matching deviation features The specific process for predicting acute exacerbations during the stable phase of COPD is as follows:

[0143] Cardiopulmonary response efficiency ratio characteristics Less than or equal to the first preset performance ratio threshold, and matching deviation characteristics When the data is less than or equal to the first preset matching deviation threshold, the patient's physiological data is determined to be within the normal range for the population, indicating good cardiopulmonary coupling. Therefore, the prediction result for acute exacerbation during the stable phase of COPD is considered to be a normal physiological state, and only data is recorded without outputting a prompt signal. The specific value of the first preset efficacy ratio threshold is set according to actual judgment needs; this embodiment uses 1.0 as an example. The judgment condition is as follows: ,and .

[0144] Cardiopulmonary response efficiency ratio characteristics Features greater than the first preset performance ratio threshold and matching deviation When the deviation is less than the second preset matching deviation threshold, or in the cardiopulmonary response efficiency ratio feature Less than or equal to the first preset performance ratio threshold and matching deviation feature When the cardiopulmonary response efficiency ratio is greater than the first preset matching deviation threshold and less than the second preset matching deviation threshold, the characteristic value is... Or matching deviation features The abnormality indicates that the patient is in a highly compensated state (i.e., paying a high cardiac rate cost to maintain blood oxygen saturation), but the cardiopulmonary temporal structure has not yet been decoupled (i.e., Although it may be high, it is still in a non-abnormal state. This usually corresponds to the early stage of airway obstruction aggravation or the recovery period after exercise. Therefore, the prediction result of acute exacerbation in the stable phase of COPD is determined to be the early stage of airway obstruction aggravation, and a secondary prompt signal is output. Among them, the second preset matching deviation threshold is greater than the first preset matching deviation threshold. In this embodiment, the second preset matching deviation threshold is taken as 2.0. Then the above judgment condition is: and ,or, and .

[0145] In matching deviation features When the deviation is greater than or equal to the second preset matching deviation threshold, the cardiopulmonary response efficiency ratio feature is irrelevant. The numerical value depends on the matching bias characteristics. If the value is too high, it indicates a significant abnormality in the temporal matching of heart rate and blood oxygen saturation, suggesting that heart rate fluctuations are no longer able to drive effective changes in blood oxygen saturation. This indicates decoupling of the patient's cardiopulmonary system, a specific signal of severely increased airway resistance. Therefore, the prediction result for acute exacerbation in the stable phase of COPD is determined to be severely increased airway resistance, and a level one warning signal is output. The above judgment conditions are as follows: .

[0146] It should be understood that the severity of a Level 1 warning signal is higher than that of a Level 2 alert signal, and the severity of a Level 2 alert signal is higher than that of a normal physiological state. Furthermore, in this embodiment, the standardization of the aforementioned cardiopulmonary response efficiency ratio characteristics and matching bias characteristics may be omitted, and the result will be based on… To predict acute exacerbations during the stable phase of COPD, the values ​​of the relevant thresholds mentioned above need to be adjusted accordingly.

[0147] In addition to the above-mentioned prediction and determination of acute exacerbations of COPD during the stable period based on a single dose, this embodiment can also perform trend tracking based on the cardiopulmonary response efficiency ratio characteristics and matching deviation characteristics of several doses, thereby achieving trend early warning. Specifically:

[0148] The cardiopulmonary response efficiency ratio and matching bias characteristics of the patient's most recent medications are obtained. In this embodiment, the most recent 7 medications are used as an example to obtain 7 feature vectors. The abnormality index is obtained by fusing the cardiopulmonary response efficiency ratio feature and the matching bias feature using the following synthesis equation. : Here, 0.6 and 0.4 are the weighting coefficients for the two features, respectively. This embodiment is not limited to these weighting coefficient settings, and other feasible values ​​can be selected according to the actual situation. Through fusion, seven anomaly indices are obtained. A two-dimensional coordinate system is constructed with the serial number of each medication administration as the x-axis and the abnormal index as the y-axis, and the seven abnormal indices are plotted. The curve is mapped to a two-dimensional coordinate system and fitted with a straight line to obtain the slope of the fitted line. If the slope is greater than 0 and greater than a preset slope threshold (e.g., 0.1), the abnormal index is judged to be on a continuous upward trend. Even if the latest medication corresponds to a prediction result of acute exacerbation of COPD in the stable period, which is a normal physiological state or an early stage of airway obstruction, a level one warning signal is still output to capture the risk of dynamic deterioration.

Claims

1. An AI-based system for predicting acute exacerbations in stable COPD patients, characterized in that, include: The data acquisition module is used to collect physiological data of patients at various resting times within a preset time period after medication. The physiological data include heart rate and blood oxygen saturation; The prediction module, electrically connected to the data acquisition module, is configured as follows: Based on the increase in physiological data at each resting time compared to the physiological data before medication, the cumulative increase in physiological data over a preset time period is obtained. Based on the cumulative increment of the physiological data, the cardiopulmonary response efficiency ratio characteristic is obtained; Based on the difference between the time interval between each target heart rate peak and each target blood oxygen saturation peak and the preset physiological delay time, the matching cost between each target heart rate peak and each target blood oxygen saturation peak is obtained, forming a matching cost matrix; the target heart rate peak and the target blood oxygen saturation peak are obtained from the physiological data increments at each resting time. The target peak of heart rate, the target peak of blood oxygen saturation, and the matching cost matrix are input into the bipartite graph model to determine the optimal matching cost and value. Based on the optimal matching cost and value, the matching deviation characteristics between heart rate and blood oxygen saturation are obtained; Based on the cardiopulmonary response efficiency ratio characteristics and matching bias characteristics, acute exacerbations of COPD during the stable phase are predicted.

2. The AI-based COPD stable-phase acute exacerbation prediction system as described in claim 1, characterized in that, The process of obtaining the resting time includes: The preset time period is divided into multiple initial time periods; Determine the patient's exercise intensity at each initial time point; Based on the exercise intensity, a resting period is selected from the initial period; the time within the resting period is taken as the resting moment.

3. The AI-based COPD stable-phase acute exacerbation prediction system as described in claim 1, characterized in that, The process of acquiring the cumulative increment of the physiological data includes: Set the initial cumulative increment of physiological data to 0; Determine whether the time interval between two adjacent resting moments is greater than a preset time interval threshold. If it is less than or equal to the preset time interval threshold, update the cumulative increment of physiological data based on the physiological data increment of the two adjacent resting moments and the time interval between them. If it is greater than the preset time interval threshold, do not update the cumulative increment of physiological data. Iterate through all resting moments within a preset time period to obtain the final cumulative increase in physiological data.

4. The AI-based COPD stable-phase acute exacerbation prediction system as described in claim 3, characterized in that, The cumulative increment of the updated physiological data includes: Calculate the average increment of physiological data between two adjacent resting moments; The product of the average value of the physiological data increment and the time interval between them is used to obtain the update amount of the cumulative physiological data increment; The cumulative increment of physiological data is updated based on the update amount.

5. The AI-based COPD stable-phase acute exacerbation prediction system as described in claim 1, characterized in that, The cumulative increment of physiological data includes the cumulative increment of heart rate and the cumulative increment of blood oxygen saturation; The process of obtaining the cardiopulmonary response efficiency ratio feature includes: calculating the ratio of the cumulative increase in heart rate to the cumulative increase in blood oxygen saturation to obtain the cardiopulmonary response efficiency ratio feature.

6. The AI-based COPD stable-phase acute exacerbation prediction system as described in claim 1, characterized in that, The physiological data increments include heart rate increments and blood oxygen saturation increments; The process of obtaining the target heart rate peak includes: A sliding window operation is performed over a preset time period to obtain the maximum heart rate increment within each window; The maximum heart rate increment that is greater than the preset heart rate increment peak value is selected from the maximum heart rate increments in each window and used as the target heart rate peak. The process of obtaining the target peak of blood oxygen saturation includes: Obtain the maximum increase in blood oxygen saturation within each window; The maximum blood oxygen saturation increment that is greater than the preset peak value of blood oxygen saturation increment is selected from the maximum blood oxygen saturation increments in each window and used as the target peak value of blood oxygen saturation.

7. The AI-based COPD stable-phase acute exacerbation prediction system as described in claim 1, characterized in that, The step of inputting the target heart rate peak, target blood oxygen saturation peak, and matching cost matrix into the bipartite graph model to determine the optimal matching cost and value includes: The target peak of heart rate is taken as the left node of the bipartite graph, the target peak of blood oxygen saturation is taken as the right node of the bipartite graph, and the matching cost matrix is ​​taken as the edge weight. The goal of solving the matching cost matrix is ​​to find the target solution that minimizes the sum of the matching costs of all selected pairs, thus obtaining the optimal matching cost sum.

8. The AI-based COPD stable-phase acute exacerbation prediction system as described in claim 1, characterized in that, The process of obtaining the matching deviation features includes: Determine the maximum number of target heart rate peaks and target blood oxygen saturation peaks; The matching deviation feature is obtained by calculating the ratio of the optimal matching cost to the maximum number.

9. The AI-based COPD stable-phase acute exacerbation prediction system as described in claim 1, characterized in that, The method for predicting acute exacerbations during the stable phase of COPD based on cardiopulmonary response efficiency ratio characteristics and matching bias characteristics includes: When the cardiopulmonary response efficiency ratio feature is less than or equal to the first preset efficiency ratio threshold, and the matching deviation feature is less than or equal to the first preset matching deviation threshold, the prediction result of acute exacerbation of COPD in the stable period is determined to be normal physiological state. When the cardiopulmonary response efficiency ratio feature is greater than the first preset efficiency ratio threshold and the matching deviation feature is less than the second preset matching deviation threshold, or when the cardiopulmonary response efficiency ratio feature is less than or equal to the first preset efficiency ratio threshold and the matching deviation feature is greater than the first preset matching deviation threshold and less than the second preset matching deviation threshold, the prediction result of acute exacerbation in the stable phase of COPD is determined to be early stage of airway obstruction exacerbation, and a secondary prompt signal is output; the second preset matching deviation threshold is greater than the first preset matching deviation threshold. When the matching deviation feature is greater than or equal to the second preset matching deviation threshold, the prediction result of acute exacerbation in the stable phase of COPD is determined to be a severe increase in airway resistance, and a first-level warning signal is output.

10. The AI-based COPD stable-phase acute exacerbation prediction system as described in claim 1, characterized in that, Before predicting acute exacerbations of COPD during the stable phase based on cardiopulmonary response efficiency ratio characteristics and matching bias characteristics, the cardiopulmonary response efficiency ratio characteristics and matching bias characteristics are standardized respectively.

Images