Machine learning based early respiratory infection screening system and method

Abstract

The present application relates to the technical field of early infection screening, and particularly relates to a respiratory tract infection early screening system and method based on machine learning, which comprises the following steps: preprocessing collected multi-source time series data to form a standardized time series data unit; based on the standardized time series data unit after a potential infection period, extracting a first group of change characteristics corresponding to respiratory tract infection early stage and non-infectious inflammation in the respiratory rhythm dimension, and a second group of change characteristics corresponding to the autonomic nervous balance dimension, and determining a respiratory tract infection signal purification factor according to the first group of change characteristics and the second group of change characteristics; and determining the probability of respiratory tract infection early stage according to the standardized time series data unit after the potential infection period and the respiratory tract infection signal purification factor. The present application can improve the accuracy of respiratory tract infection early screening.

Description

Technical Field

[0001] This invention relates to the technical field of early infection screening, and specifically to a machine learning-based early screening system and method for respiratory infections. Background Technology

[0002] Respiratory infections have an insidious onset, rapid progression, and strong nonspecific early symptoms, such as mild cough, low-grade fever, changes in respiratory rhythm, and fatigue, which are difficult to identify effectively with traditional physical signs. Conventional diagnosis relies on laboratory and imaging examinations, which have drawbacks such as time lag and high risk of cross-infection. Early detection can significantly reduce the rate of severe illness and the probability of transmission.

[0003] In the early stages of infection, only the respiratory mucosa and local immunity are affected, without triggering significant systemic inflammation. Acoustic and physiological indicators show only slight changes, highly overlapping with the physiological and acoustic characteristics of non-infectious inflammations such as allergic rhinitis. Current technologies mostly rely on single indicators or simple statistical features, which cannot effectively distinguish between the two, resulting in insufficient accuracy and stability in early screening for respiratory infections.

[0004] In other words, the accuracy of existing technologies for early screening of respiratory infections is relatively low. Summary of the Invention

[0005] To address the low accuracy of early respiratory infection screening in existing technologies, the present invention aims to provide a machine learning-based early respiratory infection screening system and method, the specific technical solution of which is as follows: In a first aspect, one embodiment of the present invention provides a method for early screening of respiratory infections based on machine learning, the method comprising: Multi-source time-series data are collected and preprocessed to form standardized time-series data units; the multi-source time-series data includes acoustic data, physiological data, and behavioral state data. Acoustic features are extracted from standardized time-series data units, potential respiratory infection signal features are identified based on the acoustic features, and potential infection time periods are determined based on the potential respiratory infection signal features; the acoustic features include breath sound features and cough sound features; Based on standardized time-series data units following the potential infection period, and combined with the physiological and pathological evolution logic of early respiratory infection and non-infectious inflammation, the first set of change features corresponding to the respiratory rhythm dimension and the second set of change features corresponding to the autonomic balance dimension are extracted for early respiratory infection and non-infectious inflammation, respectively. Based on the first and second set of change features, respiratory infection signal purification factors are determined. The physiological and pathological evolution logic is used to characterize the different temporal change patterns of early respiratory infection and non-infectious inflammation in the respiratory rhythm dimension and autonomic balance dimension. The probability of early respiratory infection was determined based on standardized time-series data units following the potential infection period and respiratory infection signal purification factors.

[0006] In one embodiment, the first set of variation features includes respiratory rhythm feature indices of respiratory tract infection and respiratory rhythm feature indices of non-infectious inflammation; the second set of variation features includes autonomic balance respiratory infection features and autonomic balance non-infectious inflammation features; and the determination of respiratory infection signal purification factors based on the first and second sets of variation features includes: Based on respiratory rhythm characteristic indices of respiratory tract infection and respiratory rhythm characteristic indices of non-infectious inflammation, the target respiratory rhythm infection purification index was determined; Based on the characteristics of autonomic nervous system-balanced respiratory infection and the characteristics of autonomic nervous system-balanced non-infectious inflammation, the purification coefficient of autonomic nervous system-balanced respiratory infection was determined. The respiratory infection signal purification factor was obtained by fusing the target respiratory rhythm infection purification index and the autonomic nervous system balance respiratory infection purification coefficient.

[0007] In one embodiment, the fusion processing of the target respiratory rhythm infection purification index and the autonomic nervous system balance respiratory infection purification coefficient to obtain a respiratory infection signal purification factor includes: Based on the mean values ​​of the target respiratory rhythm infection purification index and the autonomic nervous system balance respiratory infection purification coefficient within each preset time window after the potential infection period, the initial respiratory infection signal purification factor corresponding to the preset time window is determined. The respiratory infection signal purification factor is determined based on the mean value of the initial respiratory infection signal purification factor corresponding to each preset time window.

[0008] In one embodiment, the step of identifying potential respiratory infection signal features based on acoustic characteristics and determining the potential infection time period based on the potential respiratory infection signal features includes: Frequency domain analysis was performed on the breath sound characteristics to determine the proportion of high-frequency energy and the rate of change of the proportion of high-frequency energy corresponding to the breath sound characteristics. The peak intensity sequence, duration sequence, and adjacent interval sequence corresponding to the cough sound features are extracted, and the infection characteristic factors of the cough signal are determined based on the peak intensity sequence, duration sequence, and adjacent interval sequence. Based on the rate of change of high-frequency energy proportion and the infection characteristic factors of cough signals, early infection acoustic cue factors are identified; these early infection acoustic cue factors are used to characterize the signal features of potential respiratory tract infections. Based on early infection acoustic cue factors and preset thresholds, the potential infection time period is determined.

[0009] In one embodiment, the behavioral state data includes resting state, sleep state, and movement state. Before determining the target respiratory rhythm infection purification index based on respiratory rhythm characteristic indices of respiratory tract infection and respiratory rhythm characteristic indices of non-infectious inflammation, the method further includes: Extract the rate of change in respiratory rate and the rate of change in respiratory interval after the potential infection period; Within the current time period, behavioral states are segmented based on their status. For each behavioral state segment, respiratory rhythm characteristic indices for respiratory infections and respiratory rhythm characteristic indices for non-infectious inflammation are determined based on the rate of change in respiratory rate and the rate of change in respiratory interval.

[0010] In one embodiment, determining the target respiratory rhythm infection purification index based on respiratory rhythm characteristic indices of respiratory tract infection and respiratory rhythm characteristic indices of non-infectious inflammation includes: The respiratory rhythm infection purification index for single-behavioral state segments is determined based on the ratio of the respiratory rhythm characteristic index of respiratory tract infection to that of non-infectious inflammation. Using the duration percentage of each behavioral state segment as a weight, the weighted average of the respiratory rhythm infection purification index for all behavioral state segments within the current time period is calculated to obtain the target respiratory rhythm infection purification index.

[0011] In one embodiment, determining the autonomic nervous system-balanced respiratory infection purification coefficient based on autonomic nervous system-balanced respiratory infection characteristics and autonomic nervous system-balanced non-infectious inflammatory characteristics includes: Extract resting heart rate sequences and resting heart rate variability sequences after the potential infection period; The infection index of heart rate characteristics was determined based on the resting heart rate sequence, and the infection index of heart rate variability characteristics was determined based on the resting heart rate variability sequence. The infection characteristics of autonomic nervous system balance breathing were determined based on the infection index of heart rate characteristics and the infection index of heart rate variability characteristics. Based on the contribution of mean heart rate and heart rate recovery coefficient corresponding to cough events, the non-infectious inflammatory characteristics of autonomic nervous system balance were determined; The autonomic nervous system-balanced respiratory infection purification coefficient is determined based on the mean ratio of autonomic nervous system-balanced respiratory infection characteristics to autonomic nervous system-balanced non-infectious inflammatory characteristics in each resting behavior segment within the current time period following the potential infection period.

[0012] In one embodiment, determining the non-infectious inflammatory characteristics of autonomic balance based on the mean heart rate contribution and heart rate recovery coefficient corresponding to the cough event includes: The coughing time period during the current resting behavior period after the potential infection time period; Determine the first difference between the mean heart rate during the first preset resting behavior segment after the cough event ends and the mean resting heart rate, and determine the contribution of the mean heart rate corresponding to the cough event based on the first difference and the mean heart rate during the first preset resting behavior segment after the cough event ends. Determine the second difference between the heart rate of the last sampling point and the heart rate of the next adjacent sampling point within the first preset resting behavior segment after the cough event ends, and determine the heart rate recovery coefficient based on the second difference and the first difference; The mean contribution of heart rate to cough events and the mean heart rate recovery coefficient were averaged to obtain the non-infectious inflammatory characteristics of autonomic nervous system balance.

[0013] Secondly, another embodiment of the present invention provides a machine learning-based early screening system for respiratory infections, the system comprising: The data acquisition module is used to collect multi-source time-series data, including acoustic data, physiological data, environmental data, and behavioral state data. The data preprocessing module is used to preprocess multi-source time-series data to form standardized time-series data units; The potential infection identification module is used to extract acoustic features from standardized time-series data units, identify potential respiratory infection signal features based on the acoustic features, and determine the potential infection time period based on the potential respiratory infection signal features; the acoustic features include breath sound features and cough sound features; The purification factor construction module is used to extract the first set of change features corresponding to the respiratory rhythm dimension and the second set of change features corresponding to the autonomic balance dimension for the early stage of respiratory infection and the non-infectious inflammation based on the standardized time-series data units after the potential infection period, combined with the physiological and pathological evolution logic of the early stage of respiratory infection and non-infectious inflammation. Based on the first set of change features and the second set of change features, the purification factor of respiratory infection signal is determined. The infection probability determination module is used to determine the probability of early respiratory infection based on standardized time-series data units after the potential infection period and respiratory infection signal purification factors.

[0014] Thirdly, in another embodiment of the present invention, an electronic device is provided, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the method described in the first aspect.

[0015] Fourthly, in another embodiment of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0016] The present invention has the following beneficial effects: This invention provides a unified and stable data foundation for early respiratory infection analysis by collecting and standardizing multi-source time-series data. Based on the acoustic characteristics of breath and cough sounds, it identifies potential infection signals and locates time periods, enabling the early capture of abnormal respiratory signals. Combining the physiological and pathological evolution logic of infectious and non-infectious inflammation, it extracts differentiated features from respiratory rhythm and autonomic nervous system balance dimensions to obtain infection signal purification factors, effectively eliminating interference from non-infectious inflammation and distinguishing between early infection and non-infectious states. Finally, by fusing the purification factors with standardized time-series data, it determines the probability of early infection, improving the accuracy and stability of early respiratory infection screening. Attached Figure Description

[0017] Figure 1 A schematic flowchart illustrating an early screening method for respiratory infections based on machine learning, provided as an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of an early screening system for respiratory infections based on machine learning, provided as an embodiment of the present invention. Detailed Implementation

[0018] The following description, in conjunction with the accompanying drawings, details the specific implementation of a machine learning-based early screening system and method for respiratory infections provided by this invention.

[0019] This invention proposes a machine learning-based method for early screening of respiratory infections. Please refer to [link / reference]. Figure 1 The diagram illustrates a schematic flowchart of a machine learning-based early screening method for respiratory infections according to an embodiment of the present invention. The method includes the following steps: Step S1: Collect multi-source time-series data and preprocess the multi-source time-series data to form standardized time-series data units; the multi-source time-series data includes acoustic data, physiological data and behavioral state data.

[0020] Multi-source time-series data refers to physiological, acoustic, and behavioral data collected continuously in chronological order from different collection devices / types, possessing timestamps and continuity, and serving as the foundational data for early screening of respiratory infections.

[0021] Preprocessing refers to operations performed on the raw data before feature calculation and model input, such as noise reduction, alignment, standardization, and handling of outliers, to eliminate noise, unify the format, and improve data usability.

[0022] Standardized time-series data units refer to standardized structured data formed after uniform processing within a fixed time window (e.g., 15 minutes). These data include timestamps, acoustic signal segments, physiological data, and behavioral / environmental markers, and can be directly used for subsequent feature extraction and calculation.

[0023] Acoustic data refers to cough sounds and nasal breathing sounds collected by microphones and wearable microphones, used to identify respiratory abnormalities and potential infection time periods.

[0024] Physiological data refers to human physiological indicators such as respiratory rate, respiratory interval, resting heart rate, heart rate variability, body surface temperature, and blood oxygen saturation collected by wearable devices.

[0025] Behavioral state data refers to monitoring data used to distinguish human activity states, including three categories: resting, sleep, and exercise. It is used to screen resting behavior segments and ensure the accuracy of feature calculations.

[0026] For example, behavioral state data, which is classified data collected by wearable sensing devices to label the user's current activity type, is an important environmental labeling information for physiological / acoustic data in the early screening system for respiratory infections, and is used to exclude the interference of activities on physiological indicators.

[0027] It should be noted that, in the embodiments of the present invention, firstly, multiple types of time-series data, including acoustic signals of breathing and coughing, multiple physiological indicators, and human behavioral states, are continuously collected through acoustic acquisition devices and wearable sensing devices; then, preprocessing operations such as signal denoising, timestamp alignment, unified time granularity, and behavior segmentation are performed on these raw data, and finally, standardized time-series data units with unified structure that can be directly used for subsequent calculations are formed, providing a standard data foundation for acoustic feature recognition, potential infection period determination, and construction of infection signal purification factors.

[0028] Step S2: Extract acoustic features from standardized time-series data units, identify potential respiratory infection signal features based on acoustic features, and determine the potential infection time period based on the potential respiratory infection signal features; acoustic features include breath sound features and cough sound features.

[0029] Acoustic features refer to quantifiable signal features extracted from cough and breath sounds. They are divided into two categories: breath sound features and cough sound features, and serve as the basis for judging early respiratory abnormalities.

[0030] Breath sound characteristics refer to the features obtained after performing frequency domain analysis on breath sounds, including: low-frequency / high-frequency energy values, high-frequency energy ratio, and high-frequency ratio change rate.

[0031] Cough sound features refer to the temporal variation features extracted from cough signals, including cough peak intensity, duration, cough interval, and their fitting slope, which are used to determine whether the cough is evolving towards short, low intensity, and high frequency repetition.

[0032] Potential respiratory infection signal features refer to the acoustic cue factors for early infection calculated by fusing breath sound features and cough sound features. They are quantitative signals that indicate the possible presence of early infection.

[0033] For example, the potential respiratory infection signal characteristics refer to the specific acoustic changes in cough and breath sounds in the early stages of respiratory infection, which serve as the basis for initially marking the risk period.

[0034] For example, the potential infection time period refers to the time window (e.g., 15 minutes) marked when the acoustic cue factor for early infection is greater than or equal to a preset threshold (e.g., 0.55). This indicates that there are highly suspected acoustic abnormalities for early infection during this period, and the next step of physiological feature identification is required.

[0035] It should be noted that, in the embodiments of the present invention, the preset threshold is obtained based on the statistical distribution of early acoustic features of confirmed cases in the historical sample database.

[0036] It should be noted that, in the embodiments of the present invention, the features of breathing sounds and cough sounds are first extracted from the standardized time-series data units that have been unified in time and aligned for noise reduction; the suspected infection signal is calculated using these two types of acoustic features; and then, based on the magnitude of this signal, the highly suspicious time period is marked as the potential infection time period for subsequent differentiation between infectious / allergic rhinitis.

[0037] Furthermore, the step of identifying potential respiratory infection signal features based on acoustic characteristics and determining the potential infection time period based on the potential respiratory infection signal features includes: Frequency domain analysis was performed on the breath sound characteristics to determine the proportion of high-frequency energy and the rate of change of the high-frequency energy proportion corresponding to the breath sound characteristics.

[0038] Frequency domain analysis refers to converting a continuous time-domain breath sound signal into a frequency dimension through mathematical transformations (such as Fourier transform) for analysis, breaking down the mixed breath sounds into different frequency components to facilitate the identification of signal energy distribution in different frequency bands.

[0039] The high-frequency energy ratio refers to dividing the respiratory sound frequency into low-frequency bands (200Hz~1000Hz) and high-frequency bands (1000Hz~4000Hz), and calculating the proportion of high-frequency signal energy to the total respiratory sound energy. This is used to determine whether early pathological changes such as edema of the respiratory mucosa and increased viscosity of secretions have occurred.

[0040] The rate of change of high-frequency energy ratio refers to the rate of change of the high-frequency energy ratio of respiratory sounds in the current time period compared with the previous time period. The larger the value, the more significant the increase in the edema of the user's respiratory mucosa or the viscosity of secretions, reflecting the dynamic change of the high-frequency energy ratio of respiratory sounds over time.

[0041] It should be noted that, in the embodiments of the present invention, the collected human respiratory sound acoustic signals are separated into different frequency components using frequency domain analysis; the ratio of high-frequency signal energy to the total respiratory sound energy (i.e., high-frequency energy ratio) is calculated, and the change in this ratio between the current time period and the previous time period is calculated. That is, by the ratio and change of high-frequency energy of respiratory sounds, early abnormalities of the respiratory mucosa can be preliminarily identified, which is one of the bases for determining the potential infection time period. The larger the high-frequency energy ratio and its rate of change, the higher the possibility of respiratory mucosal edema and increased secretion viscosity, which is more consistent with the acoustic characteristics of the early stage of respiratory infection.

[0042] The peak intensity sequence, duration sequence, and adjacent interval sequence corresponding to the cough sound features are extracted, and the infection characteristic factors of the cough signal are determined based on the peak intensity sequence, duration sequence, and adjacent interval sequence.

[0043] Among them, the peak intensity sequence refers to the peak intensity value of each cough sound signal arranged in chronological order, reflecting the continuous trend of cough loudness.

[0044] Duration sequence refers to the numerical value of the duration of each cough from start to finish, arranged in chronological order, reflecting the variation in the duration of a single cough.

[0045] The adjacent interval sequence refers to the time interval between two consecutive coughs arranged in chronological order, reflecting the frequency and intensity of coughs.

[0046] The infection characteristic factor of cough signal refers to the mean value obtained by inverting and normalizing the slope of cough intensity, the slope of cough duration, and the slope of cough interval, and is used to quantify whether the cough pattern conforms to the characteristics of early infection.

[0047] It should be noted that, in the embodiments of the present invention, three types of time-series data are extracted from the collected cough sound signals: cough peak intensity sequence, cough duration sequence, and cough adjacent interval sequence; a straight line is fitted to each of the three types of sequences, the slope is calculated and the opposite number is taken, and normalization is performed to obtain three feature values; the average value of the three feature values ​​is taken to calculate the cough signal infection feature factor, which is used to preliminarily determine whether the current cough has a risk of respiratory infection.

[0048] Based on the rate of change of high-frequency energy ratio and the infection characteristic factors of cough signal, early infection acoustic cue factors are determined; early infection acoustic cue factors are used to characterize the signal features of potential respiratory tract infection.

[0049] Among them, the early infection acoustic indication factor refers to the quantitative index of early infection risk calculated based on acoustic signals such as cough and breath sounds. It is obtained by taking the average value of the change rate of high frequency energy proportion of breath sounds and the infection characteristic factor of cough signal. The higher the value, the more obvious the local abnormality of the respiratory tract.

[0050] It should be noted that, in the embodiments of the present invention, the change amplitude of the high-frequency energy ratio of breath sounds is first calculated, and then the infection-related characteristic value of the cough pattern is calculated. The two are averaged to obtain the early infection acoustic indication factor. This factor is specifically used to identify and characterize whether the user has acoustic abnormal signals of early respiratory infection, providing a basis for the system to mark potential infection time periods.

[0051] Based on early infection acoustic cue factors and preset thresholds, the potential infection time period is determined.

[0052] The preset threshold refers to a manually set risk assessment threshold, which can be set to 0.55, and is used to distinguish between normal and potential infection states.

[0053] It should be noted that, in the embodiments of the present invention, the acoustic features of the user's cough and breathing sounds in the current time period are extracted to calculate the acoustic indication factor of early infection; then the acoustic indication factor of early infection is compared with a preset threshold; if the acoustic indication factor of early infection is greater than or equal to the preset threshold, the time period is marked as a potential infection time period.

[0054] Step S3: Based on the standardized time-series data units after the potential infection period, and combined with the physiological and pathological evolution logic of early respiratory infection and non-infectious inflammation, extract the first set of change features corresponding to the respiratory rhythm dimension and the second set of change features corresponding to the autonomic nervous balance dimension for early respiratory infection and non-infectious inflammation, respectively. Based on the first set of change features and the second set of change features, determine the respiratory infection signal purification factor.

[0055] Among them, the physiological and pathological evolution logic is used to characterize the different temporal changes in the early stage of respiratory infection and non-infectious inflammation in the dimensions of respiratory rhythm and autonomic nervous balance.

[0056] The physiological and pathological evolutionary logic refers to the different temporal changes in respiratory rhythm and autonomic nervous system balance in the early stage of respiratory tract infection and non-infectious inflammation (such as allergic rhinitis), which is the basis for distinguishing between the two.

[0057] The temporal change pattern refers to the characteristic change trend that appears over time, including differences such as continuous accumulation / instantaneous fluctuation, slow change / rapid recovery, and whether there are clear time boundaries.

[0058] The respiratory rhythm dimension refers to the respiratory pattern characteristic dimension with respiratory frequency, respiratory interval and their rate of change as the core.

[0059] It should be noted that changes in respiratory rhythm in non-infectious inflammation have clear time boundaries and do not exhibit a trend of continuous accumulation.

[0060] The autonomic nervous system balance dimension refers to the characteristic dimension that reflects the body's autonomic nervous system regulation state, with resting heart rate and heart rate variability as its core.

[0061] The first set of changes refers to the respiratory rhythm changes in the early stages of respiratory infection and non-infectious inflammation, respectively.

[0062] The second set of changes refers to the changes in autonomic nervous system regulation in the early stages of respiratory infections and non-infectious inflammation, respectively, on the autonomic nervous system balance dimension.

[0063] Respiratory infection signal purification factors are used to weaken non-infectious inflammatory interference and purify the true signals in the early stages of infection, and are key prior features of machine learning models.

[0064] It should be noted that, in the embodiments of the present invention, standardized time-series data after the potential infection period are used as the basis for analysis. Based on the different time-series changes in respiratory rhythm and autonomic balance between early respiratory infection and non-infectious inflammation, the first set of change features in the respiratory rhythm dimension and the second set of change features in the autonomic balance dimension of the two types of pathological states are extracted respectively. Then, the above two sets of features are calculated and fused to obtain the respiratory infection signal purification factor, thereby eliminating the interference caused by non-infectious inflammation and retaining the true physiological signals of early infection.

[0065] Further, the determination of respiratory infection signal purification factors based on the first set of change characteristics and the second set of change characteristics includes: Based on respiratory rhythm characteristic indices of respiratory tract infection and respiratory rhythm characteristic indices of non-infectious inflammation, the target respiratory rhythm infection purification index was determined.

[0066] The respiratory rhythm characteristic index of respiratory tract infection refers to the quantitative value obtained by normalizing the ratio of the mean rate of change of respiratory interval to the mean rate of change of respiratory frequency after taking the mean of the rate of change of respiratory frequency and the rate of change of respiratory interval respectively within the corresponding behavioral state segment after the potential infection period. The larger the value, the more it indicates that the respiratory frequency shows a continuous and slow upward trend and the respiratory interval fluctuates gradually, which is more consistent with the respiratory rhythm change characteristics in the early stage of respiratory tract infection.

[0067] The respiratory rhythm characteristic index of non-infectious inflammation refers to the mean value obtained by calculating the weighted standard deviation of the respiratory frequency and respiratory interval change rate sequence. The larger the value, the more it conforms to the respiratory rhythm characteristics of non-infectious inflammation, reflecting the respiratory characteristics of non-infectious inflammation such as allergic rhinitis, with sudden fluctuations, discrete initial segments, and no continuous accumulation trend.

[0068] The target respiratory rhythm infection purification index is determined by the ratio of the respiratory rhythm characteristic index of respiratory infection to the respiratory rhythm characteristic index of non-infectious inflammation within the same behavioral phase, and obtained by weighted averaging of the ratios across all behavioral phases within the current time period.

[0069] It should be noted that, in the embodiments of the present invention, a purity index is obtained by calculating the ratio of respiratory rhythm characteristics representing infection to respiratory rhythm characteristics representing non-infectious inflammation, which can distinguish between infection and allergy and eliminate interference, and is used to subsequently determine whether it is an early respiratory infection.

[0070] Based on the characteristics of autonomic nervous system-balanced respiratory infection and the characteristics of autonomic nervous system-balanced non-infectious inflammation, the purification coefficient of autonomic nervous system-balanced respiratory infection was determined.

[0071] Among them, autonomic nervous system balance refers to the stable state of regulation between the sympathetic and parasympathetic nervous systems in the human body; early respiratory infections will continuously and slowly affect the autonomic nervous system, while non-infectious inflammations such as allergic rhinitis are mostly short-term and stimulus-driven, and the two have different patterns.

[0072] The autonomic nervous system balance respiratory infection characteristics refer to the characteristic values ​​that reflect early autonomic nervous system changes in respiratory tract infection. It is a comprehensive characteristic obtained by averaging the heart rate characteristic infection index and the heart rate variability characteristic infection index, which reflects the infection characteristics of a slow increase in resting heart rate and a continuous decrease in the high-frequency component of heart rate variability.

[0073] The autonomic nervous system balance non-infectious inflammatory characteristics refer to the autonomic nervous system characteristics that reflect non-infectious inflammations such as allergic rhinitis. It is obtained by averaging the contribution of the mean heart rate to short-term stimulation and the heart rate recovery coefficient, reflecting the non-infectious characteristics of heart rate fluctuations caused by short-term stimulations such as coughing and rapid recovery after the stimulation ends.

[0074] The purification coefficient of autonomic nervous system-balanced respiratory infection was determined by the ratio of autonomic nervous system-balanced respiratory infection characteristics to autonomic nervous system-balanced non-infectious inflammatory characteristics. It was used to distinguish between persistent heart rate drift in the early stage of infection and heart rate fluctuations caused by short-term allergic stimulation.

[0075] It should be noted that, in the embodiments of the present invention, the autonomic nerve characteristics representing respiratory tract infection and the autonomic nerve characteristics representing non-infectious inflammation such as allergic rhinitis are calculated first, and then the ratio of the two is used to calculate the autonomic nerve balance respiratory infection purification coefficient.

[0076] For example, in situations where acoustic signals cannot distinguish between infected and non-infected, the difference between continuous changes in the autonomic nervous system and short-term stimulation changes can be used to purify the infection signal from the interference, reduce false positives, and improve the accuracy of early screening.

[0077] The respiratory infection signal purification factor was obtained by fusing the target respiratory rhythm infection purification index and the autonomic nervous system balance respiratory infection purification coefficient.

[0078] It should be noted that, in the embodiments of the present invention, the target respiratory rhythm infection purification index and the autonomic nervous balance respiratory infection purification coefficient used to filter non-infectious interference are first calculated separately, and then the average value of these two indicators is taken to complete the fusion, and finally the respiratory infection signal purification factor is obtained. This respiratory infection signal purification factor can help machine learning models distinguish between early respiratory infection and non-infectious inflammation such as allergic rhinitis, thereby improving the accuracy and stability of screening.

[0079] The first group of changes includes respiratory rhythm characteristic index of respiratory tract infection and respiratory rhythm characteristic index of non-infectious inflammation. The second group of changes includes autonomic nervous system balance respiratory infection characteristics and autonomic nervous system balance non-infectious inflammation characteristics.

[0080] It should be noted that, in order to distinguish between early respiratory infections and non-infectious inflammations such as allergic rhinitis, two sets of differential change features were extracted from physiological data. Two types of indicators, respiratory rhythm and autonomic nervous system response, were used to create a pair of infection / non-infection features to distinguish between early infection and allergic interference.

[0081] Furthermore, the fusion processing of the target respiratory rhythm infection purification index and the autonomic nervous system balance respiratory infection purification coefficient yields respiratory infection signal purification factors, including: The initial respiratory infection signal purification factor corresponding to the preset time window is determined based on the mean of the target respiratory rhythm infection purification index and the autonomic nervous system balanced respiratory infection purification coefficient within each preset time window after the potential infection period.

[0082] The preset time window refers to a fixed analytical time window used to extract physiological characteristics and calculate purification factors after the potential infection period. It is a standard statistical interval for continuous time series data (e.g., 15 minutes).

[0083] It should be noted that after the preset analysis period following the locking of the potential infection time period, the target respiratory rhythm infection purification index and the autonomic nervous system balance respiratory infection purification coefficient are calculated within this period. The arithmetic mean of the two indicators is used as the initial respiratory infection signal purification factor for this preset time window. This factor is used as the input for subsequent models to filter allergy interference and improve the accuracy of early infection screening.

[0084] The respiratory infection signal purification factor is determined based on the mean value of the initial respiratory infection signal purification factor corresponding to each preset time window.

[0085] For example, the initial respiratory infection signal purification factor refers to the single-time purification feature obtained by averaging the respiratory rhythm infection purification index and the autonomic nervous system balance respiratory infection purification coefficient within a single preset time window, which is used to characterize the purity of infection-related signals during that time period.

[0086] For example, the final respiratory infection signal purification factor is obtained by averaging the initial respiratory infection signal purification factors of all preset time windows after the potential infection period. This factor is used to weaken the fluctuations in a single time period and stably represent the overall respiratory infection characteristics of the user. It is one of the core inputs of the machine learning model and is used to remove interference from non-infectious inflammation such as allergic rhinitis.

[0087] It should be noted that after determining the potential infection period, the initial respiratory infection signal purification factor is calculated for each time period, using a preset time window as the unit. Then, the initial purification factor of all preset time windows after the potential infection period is averaged, and the overall average value is used to eliminate the data fluctuation of a single time period. Finally, the final respiratory infection signal purification factor is obtained, which is used as input for subsequent machine learning models and early screening of respiratory infections.

[0088] Furthermore, before determining the target respiratory rhythm infection purification index based on respiratory rhythm characteristic indices of respiratory tract infection and respiratory rhythm characteristic indices of non-infectious inflammation, the method further includes: Extract the rate of change in respiratory rate and the rate of change in respiratory interval after the potential infection period.

[0089] Among them, the rate of change of respiratory rate refers to the magnitude and trend of the change of the user's respiratory rate between different sampling points after the potential infection period, reflecting the continuous change in the speed of breathing.

[0090] The rate of change in respiratory interval refers to the magnitude and trend of the variation in the time interval between each breath of a user at different sampling points after the potential infection period, reflecting the stability changes in respiratory rhythm.

[0091] Within the current time period, behavioral states are segmented based on their status. For each behavioral state segment, respiratory rhythm characteristic indices for respiratory infections and respiratory rhythm characteristic indices for non-infectious inflammation are determined based on the rate of change in respiratory rate and the rate of change in respiratory interval.

[0092] Among them, behavioral state refers to the user's current activity state, including rest, sleep, exercise, etc., which is collected and identified by wearable devices to eliminate the interference of behavior on physiological signals.

[0093] Behavioral state segmentation refers to dividing a time period into continuous segments of the same behavioral state based on changes in user behavior, in order to avoid the distortion of respiratory data caused by mixing different behaviors (such as exercise -> rest).

[0094] It should be noted that, in the embodiments of the present invention, after dividing the user behavior state into segments, the respiratory rhythm infection characteristic index and the non-infectious inflammatory respiratory rhythm characteristic index are calculated in each segment to distinguish between infection signals that are continuously progressing and non-infectious signals that do not have a continuous accumulation trend, thus providing a basis for calculating the respiratory rhythm infection purification index.

[0095] It should be noted that within the current time period, behavior state segments are divided according to behavior state, and behavior state segments with a duration less than a preset time threshold (e.g., 30 seconds) are removed or merged into adjacent segments.

[0096] Behavioral state data includes resting state, sleep state, and movement state.

[0097] The resting state refers to a state in which the user sits quietly without limb movement or obvious activity, and is an effective state for calculating infection-related characteristics such as baseline heart rate, heart rate variability, and respiratory rhythm.

[0098] Sleep state refers to the physiological state in which a user is asleep. In this state, physiological signals are stable and there is less interference, which is used to improve the accuracy and reliability of data such as breathing and heart rate.

[0099] The motion state refers to the state in which a user engages in limb activities and movements. This state can significantly change heart rate and respiratory rate, and should be excluded or segmented separately during feature calculation to avoid interfering with the judgment of infection signals.

[0100] Furthermore, the determination of the target respiratory rhythm infection purification index based on respiratory rhythm characteristic indices of respiratory tract infection and respiratory rhythm characteristic indices of non-infectious inflammation includes: The respiratory rhythm infection purification index for single-behavioral state segments is determined based on the ratio of the respiratory rhythm characteristic index of respiratory tract infection to that of non-infectious inflammation.

[0101] Among them, single behavior state segmentation refers to a single behavior segment after dividing the time axis according to user behavior (rest / sleep / exercise, etc.), which is the smallest time unit for calculating features.

[0102] The respiratory rhythm infection purification index is a quantitative indicator that measures whether a respiratory rhythm is more like an early infection than allergic rhinitis; the higher the ratio, the more it is biased towards infection.

[0103] It should be noted that, in the embodiments of the present invention, in the process of determining the respiratory rhythm infection purification index of a single-behavior state segment based on the ratio of the respiratory rhythm characteristic index of respiratory tract infection to the respiratory rhythm characteristic index of non-infectious inflammation, a very small positive number is first introduced into the denominator to avoid the denominator being zero, and then the ratio calculation result is normalized to avoid the ratio from generating extreme values ​​due to numerical fluctuations, thus ensuring the stability of the subsequent machine learning model input data distribution.

[0104] For example, the infection purification index of respiratory rhythm segmentation for single-behavior state can be expressed as: in, Indicators representing respiratory rhythm characteristics of respiratory tract infections Indicators representing non-infectious inflammatory respiratory rhythm characteristics. It is a very small positive number, used to prevent the denominator from being zero, and can take the value 0.0001. This represents the normalization function.

[0105] For example, It can be a min-max normalization function, used to map feature values ​​with different physical meanings and numerical scales to the [0,1] interval, making feature parameters that were originally not directly comparable or fused comparable and computable.

[0106] For example, the respiratory rhythm characteristic index of respiratory tract infection can be expressed as: in, This represents the mean rate of change in respiratory intervals. express The mean of the rate of change of respiratory rate after taking the absolute value. Represents the normalization function. It is a very small positive number, used to prevent the denominator from being zero, and can take the value 0.0001. This represents the rate of change of respiratory rate at the i-th time point.

[0107] It should be noted that, in the embodiments of the present invention, when the value at a single time point is negative, the absolute value (penalty term) must be taken before calculating the mean.

[0108] For example, the non-infectious inflammatory respiratory rhythm characteristic index can be expressed as: in, The weighted standard deviation representing the rate of change in respiratory rate. The weighted standard deviation of the rate of change in respiratory intervals. This represents the mean function.

[0109] Using the duration percentage of each behavioral state segment as a weight, the weighted average of the respiratory rhythm infection purification index for all behavioral state segments within the current time period is calculated to obtain the target respiratory rhythm infection purification index.

[0110] The duration percentage refers to the ratio of the duration of a certain behavior state segment to the total duration of the current time period, which is used to indicate the weight of that segment in the whole.

[0111] For example, the target respiratory rhythm infection purification index, obtained by weighted averaging of the respiratory rhythm infection purification indices of all behavioral segments within the current time period, is the input for subsequent calculation of infection signal purification factors.

[0112] It should be noted that, in the embodiments of the present invention, within the same behavioral segment, the respiratory rhythm characteristic index of respiratory infection is compared with the respiratory rhythm characteristic index of non-infectious inflammation to obtain the respiratory rhythm infection purification index of that behavioral segment. Then, the index of all behavioral segments within the current time period is weighted by time length to obtain the final respiratory rhythm infection purification index, which is used to quantify the purity of the infection characteristics of the respiratory rhythm in the current time period. The larger the ratio, the more significant the infection characteristics and the weaker the interference of non-infectious inflammation.

[0113] Furthermore, the determination of the autonomic nervous system-balanced respiratory infection purification coefficient based on the characteristics of autonomic nervous system-balanced respiratory infection and the characteristics of autonomic nervous system-balanced non-infectious inflammation includes: Extract resting heart rate sequences and resting heart rate variability sequences after the potential infection period.

[0114] Among them, the resting heart rate sequence refers to the sequence of heart rate averages collected continuously in chronological order within a resting behavior period, which is used to reflect the trend of heart rate changes over time in a resting state.

[0115] The resting heart rate variability sequence refers to the sequence of high-frequency power values ​​obtained by analyzing the heart rate intervals within a resting behavior segment, arranged in chronological order. It is used to reflect the changes in the regulatory state of the autonomic nervous system under resting conditions.

[0116] The resting behavior segment refers to the time segment in which the user is in a resting state (not sleeping, not exercising) based on behavioral state data. It is an effective time period for calculating heart rate and heart rate variability characteristics.

[0117] It should be noted that, in the embodiments of the invention, after determining the potential infection time period, the corresponding continuous resting heart rate data sequence and continuous resting heart rate variability data sequence are screened and extracted from the standardized time-series data units during the period after the potential infection time period and when the user is in a resting state. This provides basic data for subsequent calculation of the heart rate characteristic infection index and the heart rate variability characteristic infection index, thereby determining the autonomic nervous system balance respiratory infection characteristics.

[0118] The infection index of heart rate characteristics was determined based on the resting heart rate sequence, and the infection index of heart rate variability characteristics was determined based on the resting heart rate variability sequence.

[0119] Among them, the heart rate characteristic infection index refers to an index calculated based on the time-series change characteristics of a slow and continuous increase in resting heart rate without obvious fluctuations or reversals. It is used to reflect the continuous impact of early respiratory infection on heart rate. The larger the value, the more it matches the heart rate characteristics of early respiratory infection (slow and continuous increase without violent fluctuations).

[0120] The Heart Rate Variability Infection Index is an index calculated based on the temporal variation characteristics of high frequency power (HF) (0.15–0.4 Hz) in heart rate variability. It is obtained by a series of operations such as fitting, differentiating, and calculating variance of the temporal sequence of HF values ​​after the potential infection period. Since this index corresponds to the autonomic nervous system change pattern of the continuous decline of HF values ​​in the early stage of infection, the larger the index value, the more consistent it is with the autonomic nervous system change characteristics in the early stage of respiratory infection.

[0121] The infection characteristics of autonomic nervous system balance respiration were determined based on the infection index of heart rate characteristics and the infection index of heart rate variability characteristics.

[0122] It should be noted that the heart rate characteristic infection index and the heart rate variability characteristic infection index are averaged to obtain a comprehensive characteristic value, namely the autonomic nervous system balanced respiratory infection characteristic, which is used to uniformly characterize the typical changes of the autonomic nervous system in the early stage of infection.

[0123] Based on the contribution of mean heart rate and heart rate recovery coefficient corresponding to cough events, the characteristics of non-infectious inflammation of autonomic nervous system balance were determined.

[0124] Among them, the cough event refers to the coughing action that occurs when a user is at rest, which is a stimulus event that triggers short-term heart rate fluctuations.

[0125] The contribution of mean heart rate refers to the difference between the mean heart rate during the coughing period and the mean heart rate during the resting period, which is obtained by comparing it with the mean heart rate during the coughing period. It is used to quantify the short-term increase in heart rate caused by coughing.

[0126] For example, the contribution of mean heart rate can be expressed as: in, This represents the mean heart rate during the resting phase. This represents the mean heart rate during the coughing period. It is a very small positive number used to prevent the denominator from being zero, and can take the value 0.0001.

[0127] The heart rate recovery coefficient is a parameter obtained by comparing the difference between the heart rate during the coughing period and the heart rate at the next sampling point with the difference between the mean heart rate during the coughing period and the mean heart rate during the resting period. It is used to measure the speed at which the heart rate recovers rapidly after a coughing stimulus.

[0128] For example, the heart rate recovery coefficient can be expressed as: in, Indicates the heart rate of the next sample. This indicates the heart rate during the coughing episode. This represents the mean heart rate during the resting phase. This represents the mean heart rate during the coughing period. It is a very small positive number used to prevent the denominator from being zero, and can take the value 0.0001.

[0129] It should be noted that, in the embodiments of the present invention, the mean contribution of heart rate caused by a cough event (the magnitude of heart rate increase) and the heart rate recovery coefficient (the speed of heart rate decrease) are first calculated, and then the two values ​​are averaged to obtain the non-infectious inflammatory feature of autonomic nervous system balance. This feature is used to identify short-term heart rate fluctuations caused by non-infectious stimuli such as allergic rhinitis, avoiding misjudgment by machine learning models as respiratory infections.

[0130] The autonomic nervous system-balanced respiratory infection purification coefficient is determined based on the mean ratio of autonomic nervous system-balanced respiratory infection characteristics to autonomic nervous system-balanced non-infectious inflammatory characteristics in each resting behavior segment within the current time period following the potential infection period.

[0131] The current time period refers to the target analysis period used to calculate infection / non-infection characteristics, which is after the potential infection period.

[0132] For example, the autonomic nervous system balance respiratory infection purification coefficient It can be represented as: in, This represents the autonomic nervous system's balance and respiratory infection characteristics in the i-th resting behavioral segment. This represents the non-infectious inflammatory characteristics of the autonomic balance in the i-th resting behavioral segment. This is a very small positive number used to prevent the denominator from being zero; it can take the value 0.0001. N represents the number of valid resting behavior segments participating in the calculation within the current time period (excluding fragmented segments with a duration of <30 seconds), and... Minimum normalization is performed to map the ratio to the [0,1] interval to avoid extreme values ​​due to numerical fluctuations and to ensure the stability of the input data distribution for subsequent machine learning models.

[0133] It should be noted that, within the target analysis period following the potential infection period, all resting behavioral segments within that period are traversed, and the ratio of autonomic nervous system-balanced respiratory infection characteristics to autonomic nervous system-balanced non-infectious inflammatory characteristics is calculated for each resting segment. The average of these ratios for all resting segments is then used to obtain the autonomic nervous system-balanced respiratory infection purification coefficient.

[0134] Furthermore, the determination of non-infectious inflammatory characteristics of autonomic balance based on the mean heart rate contribution and heart rate recovery coefficient corresponding to cough events includes: The coughing period during the current resting behavior period following the period of potential infection.

[0135] The current resting behavior segment refers to the continuous data period after the potential infection period when the user is in a resting state (not sleeping, not exercising), which is a fixed behavioral scenario for calculating heart rate-related features.

[0136] The cough time interval refers to the specific time range from the onset to the end of a cough, and it is an analytical window used to extract short-term heart rate stimulation features.

[0137] Determine the first difference between the mean heart rate during the first preset resting behavior segment after the cough event ends and the mean resting heart rate. Based on the first difference and the mean heart rate during the first preset resting behavior segment after the cough event ends, determine the contribution of the mean heart rate corresponding to the cough event.

[0138] The mean heart rate refers to the average value of the heart rate data collected within the corresponding time segment, which is used to reflect the average heart rate level during that period.

[0139] The mean heart rate during the first preset resting behavior period after the cough event ends refers to the average value of the user's heart rate data during the resting behavior period after the cough event ends.

[0140] Mean resting heart rate refers to the average baseline heart rate of a user during a resting period without external stimuli such as coughing.

[0141] It should be noted that, in the embodiments of the present invention, the average resting heart rate over a long period of 24 hours prior to the current time period is extracted as the baseline of the average resting heart rate.

[0142] The first difference refers to the difference between the mean heart rate during the first preset resting behavior segment after the cough event ends and the mean resting heart rate, which is used to characterize the instantaneous increase in heart rate caused by the cough event.

[0143] It should be noted that, in the embodiments of the present invention, the difference between the average heart rate during the coughing period and the baseline resting average heart rate is first calculated to obtain a first difference; then, the first difference is compared with the average heart rate during the first preset resting behavior segment after the coughing event ends to obtain the contribution of the short-term stimulation heart rate average to the coughing event; this parameter is used to quantify the degree of instantaneous stimulation of heart rate by coughing and is the basis for constructing non-infectious inflammatory autonomic nerve characteristics.

[0144] Determine the second difference between the heart rate of the last sampling point and the heart rate of the next adjacent sampling point within the first preset resting behavior segment after the cough event ends, and determine the heart rate recovery coefficient based on the second difference and the first difference.

[0145] The heart rate of the last sample point in the first preset resting behavior segment after the cough event ends refers to the last instantaneous heart rate sample with a timestamp in the first preset resting behavior segment that meets the preset time threshold (no fragmentation, such as ≥30 seconds) after the cough event ends.

[0146] For example, the heart rate of the next adjacent sampling point refers to the next instantaneous heart rate sample value that is temporally adjacent to and immediately following the last sampling point within the first preset resting behavior segment after the cough event ends.

[0147] The second difference refers to the difference between the heart rate of the last sampling point and the heart rate of the next adjacent sampling point within the first preset resting behavior segment after the cough event ends.

[0148] It should be noted that, in the embodiments of the present invention, the decrease in heart rate immediately after coughing is calculated (second difference), and then this decrease is divided by the increase in heart rate caused by coughing (first difference) to finally obtain the heart rate recovery coefficient.

[0149] The mean contribution of heart rate to cough events and the mean heart rate recovery coefficient were averaged to obtain the non-infectious inflammatory characteristics of autonomic nervous system balance.

[0150] It should be noted that, in the embodiments of the present invention, the construction of non-infectious inflammatory features of autonomic nervous balance through cough event-related heart rate features does not directly use a single heart rate fluctuation ratio as the basis for distinguishing between respiratory infection and non-infectious inflammation, nor does it define this ratio as a non-infectious inflammation-related feature alone. Instead, it is based on the differences in the clinical and pathological evolution of heart rate fluctuations and recovery processes in the early stage of respiratory infection and non-infectious inflammation, and integrates the contribution of mean heart rate and heart rate recovery coefficient into a composite time-series feature to characterize the non-infectious inflammatory features. Moreover, all heart rate features are extracted from multi-source time-series data collected by wearable devices. Through individual resting heart rate historical baseline calibration and multi-dimensional feature fusion, the physiological fluctuations collected by the hardware are transformed into quantitative features with the ability to distinguish between infectious and non-infectious inflammation.

[0151] Step S4: Determine the probability of early respiratory infection based on standardized time-series data units and respiratory infection signal purification factors after the potential infection period.

[0152] Among them, the probability of early respiratory infection refers to the probability value between 0 and 1 output by the machine learning model, which represents the possibility that the user is currently in the early stage of respiratory infection and is the basis for the system to issue risk warnings.

[0153] It should be noted that, in the embodiments of the present invention, the pre-processed standardized time-series data units after the potential infection period are used as the base input, and then combined with a respiratory infection signal purification factor that is specially used to remove non-infectious interference, and input together into the trained machine learning model. The model calculates and outputs the probability value that the user is in the early stage of respiratory infection, thus completing the early screening.

[0154] It should be noted that, in the embodiments of the present invention, multi-source time-series data of the infected group and the non-infected group are acquired, and continuous non-invasive time-series monitoring is performed on the infected group and the non-infected group. The monitoring period covers at least the common incubation period of respiratory infection and the clinical symptom manifestation period (e.g., 7-14 days), ensuring that complete incubation period data before the diagnosis time point can be extracted; the non-infected group is monitored in the same period simultaneously to ensure that the data dimension and time granularity are consistent with the infected group.

[0155] For example, the time point of diagnosis with a positive clinical nucleic acid / antigen test is used as the anchor point as the time point when clinical symptoms / signs of respiratory infection clearly appear. Based on the common incubation period range of respiratory infection pathogens, a fixed incubation period is extrapolated backward from the time point of diagnosis. The standardized time series data within this period is marked as early infection positive samples (label=1), and all monitoring data of the non-infected group and the non-incubation period / post-diagnosis data of the infected group are marked as non-infected negative samples (label=0). The negative samples include non-infectious inflammatory data such as allergic rhinitis and data of healthy people.

[0156] For example, standardized time-series data units following the potential infection period are integrated with respiratory infection signal purification factors to form a modeling feature set. Samples are labeled as [1 = early infection, 0 = non-infection], and the dataset is cleaned. The training / test / validation sets are divided in a 7:2:1 ratio. The training set is balanced using the Synthetic Minority Over-sampling Technique (SMOTE). The respiratory infection signal purification factors and high-discrimination original features are retained (by aligning and denoising multi-source time-series data, original features with temporal evolution differences are extracted from acoustics, respiratory rhythm, and heart rate; combined with infection signal purification factors to filter out non-infection interference; and then filtered by model feature importance to finally obtain high-discrimination original features). The respiratory infection signal purification factors are used as key features input to the model, or the respiratory infection signal purification factors are forced as the front-layer splitting node when constructing the tree model. Extreme gradient boosting is selected. The Boosting (XGBoost) model is trained using a training set, prioritizing an infection recall rate greater than or equal to 95%. After parameter tuning and optimization, false positives due to allergies are reduced. The accuracy is verified using a validation set to ensure that the false positive rate due to allergies is less than or equal to 10%. The prediction instruction is set as the output probability value, and the screening model is output.

[0157] The patient monitoring data and corresponding respiratory infection signal purification factors are input into the screening model, which outputs the probability of early respiratory infection for the current user. A risk threshold is set according to clinical needs (e.g., N=0.7). When the probability of early infection is greater than or equal to the risk threshold, an early respiratory infection risk warning is issued to the user, thus achieving early screening of respiratory infection.

[0158] It should be noted that in the embodiments of the present invention, all preset thresholds are empirical values ​​based on specific acquisition devices and sample sets. In practical applications, parameters can be optimized or adaptively adjusted according to the signal-to-noise ratio of the device and the physiological statistical distribution of the target population.

[0159] It is important to emphasize that the above-mentioned probability results for early respiratory infection are only used to indicate the user's current potential early risk status of respiratory infection (usually the probability results output by the model, used to reduce interference with subsequent clinical diagnosis while ensuring the effectiveness of early screening), in order to assist medical staff in completing the preliminary assessment of the user's early respiratory infection status more quickly; however, the definitive diagnosis of the user's respiratory infection and the formulation and implementation of subsequent clinical treatment plans must be determined by medical staff based on professional medical knowledge, clinical experience and the user's actual examination results.

[0160] In summary, this invention provides a unified and stable data foundation for early respiratory infection analysis by collecting and standardizing multi-source time-series data; it identifies potential infection signals and locates time periods based on the acoustic characteristics of breath sounds and cough sounds, enabling the early capture of abnormal respiratory signals; by combining the physiological and pathological evolution logic of infectious and non-infectious inflammation, it extracts differentiated features from the dimensions of respiratory rhythm and autonomic nervous system balance to obtain infection signal purification factors, which can effectively eliminate interference from non-infectious inflammation and distinguish between early infection and non-infectious states; finally, by fusing the purification factors and standardized time-series data to determine the probability of early infection, it can improve the accuracy and stability of early respiratory infection screening.

[0161] This invention proposes a machine learning-based early screening system for respiratory infections. Please refer to [link / reference]. Figure 2 The diagram illustrates a structural schematic of a machine learning-based early screening system 200 for respiratory infections according to an embodiment of the present invention. The system includes: The data acquisition module 201 is used to acquire multi-source time-series data, including acoustic data, physiological data, environmental data, and behavioral state data. Data preprocessing module 202 is used to preprocess multi-source time series data to form standardized time series data units; The potential infection identification module 203 is used to extract acoustic features from standardized time-series data units, identify potential respiratory infection signal features based on the acoustic features, and determine the potential infection time period based on the potential respiratory infection signal features; the acoustic features include breath sound features and cough sound features; The purification factor construction module 204 is used to extract the first set of change features corresponding to the respiratory rhythm dimension and the second set of change features corresponding to the autonomic balance dimension of the early respiratory infection and non-infectious inflammation based on the standardized time-series data units after the potential infection period, combined with the physiological and pathological evolution logic of the early respiratory infection and non-infectious inflammation. Based on the first set of change features and the second set of change features, the purification factor of the respiratory infection signal is determined. The infection probability determination module 205 is used to determine the probability of early respiratory infection based on standardized time-series data units after the potential infection period and respiratory infection signal purification factors.

[0162] It should be noted that the system provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above. In addition, the machine learning-based early screening system for respiratory infections and the machine learning-based early screening method for respiratory infections provided in the above embodiments belong to the same concept, and their specific implementation process is detailed in the method embodiment, which will not be repeated here.

Claims

1. A method for early screening of respiratory infections based on machine learning, characterized in that, The method includes: Multi-source time-series data are collected and preprocessed to form standardized time-series data units; the multi-source time-series data includes acoustic data, physiological data, and behavioral state data. Acoustic features are extracted from the standardized time-series data units, potential respiratory infection signal features are identified based on the acoustic features, and potential infection time periods are determined according to the potential respiratory infection signal features; the acoustic features include breath sound features and cough sound features; Based on standardized time-series data units following the potential infection period, and combined with the physiological and pathological evolution logic of early respiratory infection and non-infectious inflammation, the first set of change features corresponding to the early respiratory infection and non-infectious inflammation in the respiratory rhythm dimension, and the second set of change features corresponding to the autonomic nervous balance dimension are extracted respectively. Based on the first set of change features and the second set of change features, a respiratory infection signal purification factor is determined. The physiological and pathological evolution logic is used to characterize the different temporal change patterns of early respiratory infection and non-infectious inflammation in the respiratory rhythm dimension and the autonomic nervous balance dimension. The probability of early respiratory infection is determined based on standardized time-series data units following the potential infection period and the respiratory infection signal purification factor.

2. The early screening method for respiratory infections based on machine learning according to claim 1, characterized in that, The first set of change features includes respiratory rhythm characteristic indices of respiratory tract infection and respiratory rhythm characteristic indices of non-infectious inflammation; the second set of change features includes autonomic balance respiratory infection features and autonomic balance non-infectious inflammation features; the step of determining respiratory infection signal purification factors based on the first set of change features and the second set of change features includes: Based on the respiratory rhythm characteristic index of respiratory tract infection and the respiratory rhythm characteristic index of non-infectious inflammation, the target respiratory rhythm infection purification index is determined; Based on the aforementioned characteristics of autonomic nervous system-balanced respiratory infection and the aforementioned characteristics of autonomic nervous system-balanced non-infectious inflammation, the purification coefficient of autonomic nervous system-balanced respiratory infection was determined. The target respiratory rhythm infection purification index and the autonomic nervous system balance respiratory infection purification coefficient are fused to obtain the respiratory infection signal purification factor.

3. The early screening method for respiratory infections based on machine learning according to claim 2, characterized in that, The process of fusing the target respiratory rhythm infection purification index and the autonomic nervous system balance respiratory infection purification coefficient to obtain the respiratory infection signal purification factor includes: Based on the mean of the target respiratory rhythm infection purification index and the autonomic nervous system balanced respiratory infection purification coefficient within each preset time window after the potential infection period, the initial respiratory infection signal purification factor corresponding to the preset time window is determined. The respiratory infection signal purification factor is determined based on the mean value of the initial respiratory infection signal purification factor corresponding to each preset time window.

4. The early screening method for respiratory infections based on machine learning according to claim 1, characterized in that, The step of identifying potential respiratory infection signal features based on the acoustic features and determining the potential infection time period based on the potential respiratory infection signal features includes: Frequency domain analysis is performed on the respiratory sound features to determine the proportion of high-frequency energy corresponding to the respiratory sound features and the rate of change of the proportion of high-frequency energy. Extract the peak intensity sequence, duration sequence, and adjacent interval sequence corresponding to the cough sound features, and determine the infection feature factors of the cough signal based on the peak intensity sequence, the duration sequence, and the adjacent interval sequence; Based on the rate of change of the high-frequency energy ratio and the infection characteristic factors of the cough signal, an early infection acoustic cue factor is determined; the early infection acoustic cue factor is used to characterize the signal features of potential respiratory tract infection. Based on the early infection acoustic cue factor and preset threshold, the potential infection time period is determined.

5. The early screening method for respiratory infections based on machine learning according to claim 2, characterized in that, The behavioral state data includes resting state, sleep state, and movement state. Before determining the target respiratory rhythm infection purification index based on the respiratory tract infection respiratory rhythm characteristic index and the non-infectious inflammatory respiratory rhythm characteristic index, the method further includes: Extract the rate of change in respiratory rate and the rate of change in respiratory interval after the potential infection period; Within the current time period, behavioral states are divided into segments. For each behavioral state segment, respiratory rhythm characteristic indexes for respiratory tract infection and respiratory rhythm characteristic indexes for non-infectious inflammation are determined based on the rate of change of respiratory frequency and the rate of change of respiratory interval.

6. The early screening method for respiratory infections based on machine learning according to claim 2, characterized in that, The determination of the target respiratory rhythm infection purification index based on the respiratory rhythm characteristic index of respiratory tract infection and the respiratory rhythm characteristic index of non-infectious inflammation includes: The respiratory rhythm infection purification index for single-behavioral state segments is determined based on the ratio of the respiratory rhythm characteristic index of respiratory tract infection to the respiratory rhythm characteristic index of non-infectious inflammation. Using the duration percentage of the aforementioned behavioral state segments as weights, the weighted average of the respiratory rhythm infection purification index for all behavioral state segments within the current time period is calculated to obtain the target respiratory rhythm infection purification index.

7. The early screening method for respiratory infections based on machine learning according to claim 2, characterized in that, The determination of the autonomic nervous system-balanced respiratory infection purification coefficient based on the autonomic nervous system-balanced respiratory infection characteristics and the autonomic nervous system-balanced non-infectious inflammatory characteristics includes: Extract the resting heart rate sequence and resting heart rate variability sequence after the potential infection period; The heart rate characteristic infection index is determined based on the resting heart rate sequence, and the heart rate variability characteristic infection index is determined based on the resting heart rate variability sequence. Based on the heart rate characteristic infection index and the heart rate variability characteristic infection index, the autonomic nervous system balance respiratory infection characteristics are determined; Based on the contribution of mean heart rate and heart rate recovery coefficient corresponding to cough events, the non-infectious inflammatory characteristics of autonomic nervous system balance were determined; The autonomic nervous system balanced respiratory infection purification coefficient is determined based on the average ratio of the autonomic nervous system balanced non-infectious inflammatory characteristics to the autonomic nervous system balanced respiratory infection characteristics in each resting behavior segment within the current time period following the potential infection time period.

8. The method for early screening of respiratory infections based on machine learning according to claim 7, characterized in that, The determination of non-infectious inflammatory characteristics of autonomic nervous system balance based on the contribution of mean heart rate and heart rate recovery coefficient corresponding to cough events includes: Obtain the cough time period during which a cough event occurred within the current resting behavior period after the potential infection time period; Determine the first difference between the mean heart rate during the first preset resting behavior segment after the cough event ends and the mean resting heart rate, and determine the contribution of the mean heart rate corresponding to the cough event based on the first difference and the mean heart rate during the first preset resting behavior segment after the cough event ends. Determine the second difference between the heart rate of the last sampling point and the heart rate of the next adjacent sampling point within the first preset resting behavior segment after the cough event ends, and determine the heart rate recovery coefficient based on the second difference and the first difference; The mean contribution of the heart rate to the cough event and the mean of the heart rate recovery coefficient are averaged to obtain the non-infectious inflammatory characteristics of autonomic nervous system balance.

9. A machine learning-based early screening system for respiratory infections, characterized in that, The system includes: The data acquisition module is used to acquire multi-source time-series data, including acoustic data, physiological data, environmental data, and behavioral state data. The data preprocessing module is used to preprocess the multi-source time-series data to form standardized time-series data units; A potential infection identification module is used to extract acoustic features from the standardized time-series data units, identify potential respiratory infection signal features based on the acoustic features, and determine the potential infection time period based on the potential respiratory infection signal features; the acoustic features include breath sound features and cough sound features; The purification factor construction module is used to extract, based on the standardized time-series data units after the potential infection period, and combined with the physiological and pathological evolution logic of early respiratory infection and non-infectious inflammation, the first set of change features corresponding to the respiratory rhythm dimension and the second set of change features corresponding to the autonomic nervous balance dimension of the early respiratory infection and the non-infectious inflammation, respectively, and determine the respiratory infection signal purification factor according to the first set of change features and the second set of change features. The infection probability determination module is used to determine the probability of early respiratory infection based on the standardized time-series data unit after the potential infection time period and the respiratory infection signal purification factor.

10. The machine learning-based early screening system for respiratory infections according to claim 9, characterized in that, The purification factor construction module includes: The fusion module is used to determine a target respiratory rhythm infection purification index based on the respiratory rhythm characteristic index of respiratory tract infection and the respiratory rhythm characteristic index of non-infectious inflammation; to determine an autonomic nervous system balanced respiratory infection purification coefficient based on the autonomic nervous system balanced non-infectious inflammation; and to fuse the target respiratory rhythm infection purification index and the autonomic nervous system balanced respiratory infection purification coefficient to obtain the respiratory tract infection signal purification factor.

Images