A method and system for comprehensive monitoring of multiple parameters of cardiopulmonary function
By using multi-parameter coupling analysis and cross-signal cross-validation, an individualized cardiopulmonary function reference baseline was established, which solved the problems of false alarms and insufficient assessment of spontaneous recovery ability in existing central lung function monitoring technologies, and achieved accurate monitoring and risk assessment of cardiopulmonary function.
Patent Information
- Application Number
- CN202610538664.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2026-04-22
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2046-04-22
AI Technical Summary
Existing cardiopulmonary function monitoring methods cannot effectively capture the dynamic correlation between heart rate, respiration and blood oxygenation signals, leading to frequent false alarms and a lack of individualized assessment of self-recovery ability, resulting in insufficient targeting of clinical intervention decisions.
A personalized cardiopulmonary function reference baseline is established through multi-parameter coupling analysis, cardiopulmonary abnormal features are extracted and cross-signal cross-validation is performed, cardiopulmonary coupling frequency bands are identified, a cardiopulmonary status zoning map is generated, the degree of deviation is quantified, the ability to recover spontaneously is assessed, a comprehensive cardiopulmonary monitoring index is generated, and a monitoring report reflecting the direction of cardiopulmonary function evolution is output.
It achieves adaptive zoning of cardiopulmonary function, distinguishes between true recovery trends and pseudo-recovery patterns, improves the sensitivity and specificity of monitoring, and the output report can accurately reflect the dynamic changes and potential risks of cardiopulmonary function.
Smart Images

Figure CN122376057B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of biomedical signal processing technology, and in particular to a method and system for comprehensive monitoring of multiple parameters of cardiopulmonary function. Background Technology
[0002] Heart rate, respiration, and blood oxygenation are core physiological parameters reflecting cardiopulmonary function. These three parameters are interconnected at the physiological regulation level, and a sustained deviation in any one parameter often leads to secondary responses in the other two. Current continuous monitoring schemes primarily rely on single-parameter threshold-triggered alarms, lacking a systematic description of the dynamic correlation between the three signals. This results in numerous false alarms triggered by brief deviations in a single parameter, while early warning signals from coordinated anomalies in multiple parameters are masked by threshold settings, making it difficult to balance overall monitoring sensitivity and specificity.
[0003] Existing methods have further limitations in state zoning and anomaly identification. On the one hand, for the dynamic transformation of functional states as the disease progresses, fixed zoning strategies cannot adaptively capture the boundary positions between gradual and abrupt state transitions, resulting in a disconnect between the zoning results and the actual granularity of functional state evolution. On the other hand, the assessment of the body's autonomous recovery capacity lacks individualized methods. Current methods struggle to distinguish between genuine recovery trends and pseudo-recovery patterns characterized by temporary compensatory rebounds followed by relapse. The peak characteristics and persistent trends of compensatory mobilization capacity are not effectively integrated into the monitoring conclusions, ultimately leading to monitoring reports that are insufficiently targeted for clinical intervention decisions. Summary of the Invention
[0004] This invention discloses a multi-parameter integrated monitoring method and system for cardiopulmonary function. It establishes an individualized cardiopulmonary function reference baseline by performing multi-parameter coupled analysis of heart rate, respiration and blood oxygenation signals. Based on this, it extracts abnormal cardiopulmonary features and performs cross-signal cross-validation, identifies cardiopulmonary coupling frequency bands and calibrates cardiopulmonary status identifiers to generate a cardiopulmonary status zoning map, and then quantifies the degree of deviation of each zone relative to the baseline, assesses the body's self-recovery ability and identifies the risk of persistent abnormalities, generates a comprehensive cardiopulmonary monitoring index in conjunction with cardiopulmonary status identifiers, and outputs a monitoring report reflecting the overall evolution direction of cardiopulmonary function through compensation peak identification and trend weighted assessment.
[0005] The first aspect of this invention proposes a method for comprehensive monitoring of multiple parameters of cardiopulmonary function, comprising the following steps: Collect the patient's physiological signal set, which includes heart rate signal, respiratory signal and blood oxygen signal, and perform multi-parameter coupling analysis on the physiological signal set to generate a cardiopulmonary function reference baseline; The cardiopulmonary function reference baseline is used to perform fluctuation localization to extract cardiopulmonary abnormal features, and cross-signal cross-validation is performed on the cardiopulmonary abnormal features to generate a coupled validation set. The coupled validation set is used to perform frequency band parsing to extract the cardiopulmonary coupled frequency band. Cardiopulmonary state identifiers are calibrated according to the cardiopulmonary coupling frequency bands, and a cardiopulmonary state partition map is generated by performing pattern clustering partitioning on the physiological signal set based on the cardiopulmonary state identifiers. Based on the cardiopulmonary status zoning map and the cardiopulmonary function reference baseline, offset quantification is performed to generate offset amplitude, self-recovery capability assessment is performed on the offset amplitude to generate recovery capability coefficient, and continuous abnormal time points are identified and an abnormal risk set is generated according to the recovery capability coefficient. The abnormal risk set and the cardiopulmonary status identifier are used to perform joint hierarchical matching to generate a comprehensive cardiopulmonary monitoring index. Pattern recognition analysis is performed on the comprehensive cardiopulmonary monitoring index to generate cardiopulmonary abnormality patterns. Compensation peaks are identified in the cardiopulmonary abnormality patterns to generate compensation status identifiers. Trend weighted evaluation is performed based on the compensation status identifiers to generate a cardiopulmonary function monitoring report.
[0006] A second aspect of this invention provides a multi-parameter integrated monitoring system for cardiopulmonary function, comprising: The signal acquisition module is used to acquire the patient's physiological signal set, which includes heart rate signal, respiratory signal and blood oxygen signal, and to perform multi-parameter coupling analysis on the physiological signal set to generate a cardiopulmonary function reference baseline; The feature extraction module is used to perform fluctuation localization on the cardiopulmonary function reference baseline to extract cardiopulmonary abnormal features, perform cross-signal cross-validation on the cardiopulmonary abnormal features to generate a coupled validation set, and perform frequency band parsing to extract the cardiopulmonary coupled frequency band according to the coupled validation set. The state partitioning module is used to calibrate the cardiopulmonary state identifier according to the cardiopulmonary coupling frequency band, and perform pattern clustering partitioning on the physiological signal set based on the cardiopulmonary state identifier to generate a cardiopulmonary state partitioning map. The risk assessment module is used to generate an offset amplitude by performing offset quantification based on the cardiopulmonary status zoning map and the cardiopulmonary function reference baseline, perform self-recovery capability assessment on the offset amplitude to generate a recovery capability coefficient, and identify continuous abnormal time points and generate an abnormal risk set according to the recovery capability coefficient. The report output module is used to generate a comprehensive cardiopulmonary monitoring index by performing joint hierarchical matching with the abnormal risk set and the cardiopulmonary status identifier, perform pattern recognition analysis on the comprehensive cardiopulmonary monitoring index to generate cardiopulmonary abnormality patterns, identify the compensation peak of the cardiopulmonary abnormality patterns to generate compensation status identifiers, and perform trend weighted evaluation based on the compensation status identifiers to generate a cardiopulmonary function monitoring report.
[0007] The beneficial effects of this invention are reflected in the following points: 1. By simultaneously acquiring heart rate, respiration, and blood oxygen signals, a reference baseline for individual cardiopulmonary function is established based on the multi-parameter coupling characteristics during the initial stable period of acquisition. This baseline covers the coupling strength, phase relationship, and conduction delay of each signal. Cardiopulmonary abnormal features are extracted using this baseline. Effective cross-signal co-occurrence abnormalities are screened through phase lag detection and coupling strength quantification. Cardiopulmonary coupling characteristic frequencies are identified, and three states—functional compensation, functional impairment, and coupling differentiation—are categorized accordingly. Combined with a gradual dense segment detection and mutation supplementation candidate mechanism, adaptive partitioning of the monitoring time series is implemented, overcoming the problem of insufficient perception of multi-parameter coordinated abnormalities and gradual state transitions by single-parameter fixed thresholds. 2. The feature vectors of each partition are compared with the individualized baseline dimension by dimension to generate a comprehensive offset amplitude. The peak offset amplitude and recovery rate are used to jointly generate a recovery capacity coefficient. Based on offset quantification, the true recovery trend and the pseudo-recovery pattern of temporary compensatory rebound followed by relapse are further distinguished. An abnormal risk score is generated through duration statistics and density analysis, realizing a quantitative description of the entire process from signal offset to individual compensatory reserve depletion. 3. Extract the causal driving direction of heart rate and respiratory abnormalities and implement directional weight allocation accordingly. After enhancing the reliability of functional status, the weighted integration generates a comprehensive cardiopulmonary monitoring index. By tracking the compensation intensity and verifying the peak stability, it distinguishes between two states: high-burden compensation threshold and premature compensation threshold. According to the frequency evolution and intensity trend of the compensation status identifier, the monitoring report is output so that the report conclusion can directly reflect the dynamic change direction of cardiopulmonary compensation capacity and the degree of decompensation risk. Attached Figure Description
[0008] Figure 1 This is a flowchart illustrating a multi-parameter comprehensive monitoring method for cardiopulmonary function according to the present invention.
[0009] Figure 2 This is a schematic diagram of the layout of the physiological signal acquisition sensor of the present invention.
[0010] Figure 3 This is a schematic diagram of the cardiopulmonary status zoning diagram of the present invention.
[0011] Figure 4 This is a structural block diagram of a multi-parameter integrated monitoring system for cardiopulmonary function according to the present invention.
[0012] The components are as follows: 1- Chest ECG acquisition electrode; 2- Earlobe reference electrode; 3- Chest impedance sensor; 4- Abdominal impedance sensor; 5- Earlobe photoplethysmography (PPG) sensor; 6- Signal acquisition host; 7- Wireless data transmission module; 8- R-wave peak detection point; 9- Functional compensation state partition; 10- Functional impairment state partition; 11- Coupling differentiation state partition; 12- State boundary point; 13- Comprehensive offset amplitude curve; 14- Recovery start marker; 15- Pseudo-recovery marker; 16- Oscillation segment label. Detailed Implementation
[0013] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.
[0014] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.
[0015] The technical solutions of the embodiments of this application will be described below.
[0016] like Figure 1 As shown, this embodiment of the invention provides a method for comprehensive monitoring of multiple parameters of cardiopulmonary function, including the following steps S110-S150: Step S110: Collect the patient's physiological signal set, which includes heart rate signal, respiratory signal and blood oxygen signal, and perform multi-parameter coupling analysis on the physiological signal set to generate a cardiopulmonary function reference baseline.
[0017] Specifically, it involves collecting a set of physiological signals from the patient. For example... Figure 2As shown, the heart rate signal continuously records cardiac electrical activity via chest ECG acquisition electrode 1 and earlobe reference electrode 2, and extracts the time sequence of R-wave peak detection point 8 as the basis for time-domain analysis. An adaptive threshold strategy is introduced for R-wave detection, with the threshold dynamically updated according to local ECG amplitude to avoid missing low-amplitude R-waves caused by baseline drift due to body position adjustments. The respiratory signal obtains chest wall fluctuation information using chest impedance sensor 3 and abdominal impedance sensor 4. Since the respiratory signal sampling rate is lower than the heart rate signal, the two data streams must be aligned using a unified time reference before being synchronously included in the physiological signal set. Resampling correction is initiated when the alignment error exceeds the allowable range. The blood oxygen signal originates from earlobe photoplethysmography (PPG) sensor 5. When insufficient peripheral perfusion causes the signal-to-noise ratio to remain below the acceptable threshold, a low-confidence label is added to the quality score for the corresponding time period. The contribution weight of the low-confidence labeled segment is reduced proportionally to the difference between the quality score and the acceptable threshold. The heart rate, respiratory, and blood oxygen signals together constitute the physiological signal set. After being aggregated and processed by the signal acquisition host 6, the physiological signal set is uploaded to the monitoring system in real time via the wireless data transmission module 7. The quality metadata of each signal in the physiological signal set is updated and written in real time. The content includes the signal-to-noise ratio estimate and the acquisition continuity score. These two indicators serve as the quantitative basis for the dynamic allocation of the weights of each signal in the multi-parameter coupling analysis stage.
[0018] Multi-parameter coupling analysis was performed on the physiological signal set to generate a cardiopulmonary function reference baseline. The multi-parameter coupling analysis received synchronous time-series data from three signals within the physiological signal set. The coupling relationship between heart rate and respiratory signals manifested as rhythmic modulation of the RR interval by the respiratory phase—in individuals with normal cardiopulmonary function, the RR interval shortened during inspiration and prolonged during expiration, forming a periodic alternation of long and short intervals. The amplitude of this physiological modulation was statistically analyzed during the initial acquisition phase of the physiological signal set, selecting stable signal segments and measuring it as the ratio of the difference between the mean RR intervals of the expiratory and inspiratory phases to the mean of the entire segment. The resulting value was written into the cardiopulmonary function reference baseline. The coupling between blood oxygenation and heart rate signals presented another form: when the heart rate increased, peripheral blood oxygen saturation did not change immediately but exhibited a measurable lag response. This lag essentially reflected the conduction delay of the peripheral circulatory system in response to changes in heart rate. Clinically, the conduction delay was significantly longer in patients with chronic heart failure than in healthy individuals. When processing low-confidence labeled segments, coupling analysis applies weighted reduction to the estimation results for the corresponding time windows. The reduced coupling estimates still participate in the statistical summarization of the cardiopulmonary function reference baseline, but their contribution ratio is constrained. The final generated cardiopulmonary function reference baseline is based on the multi-parameter coupling features extracted during the initial stable period of acquisition, specifically covering six types of parameters: mean heart rate and its variation range, mean respiratory rate and amplitude, mean blood oxygen saturation and its fluctuation range, and the coupling strength and phase relationship between the three signals. These parameters together constitute the individual-oriented cardiopulmonary function reference baseline.
[0019] Step S120: Perform fluctuation localization on the cardiopulmonary function reference baseline to extract cardiopulmonary abnormal features, perform cross-signal cross-validation on the cardiopulmonary abnormal features to generate a coupled validation set, and perform frequency band analysis to extract the cardiopulmonary coupled frequency band according to the coupled validation set.
[0020] Specifically, fluctuation localization is performed on the cardiopulmonary function reference baseline to extract abnormal cardiopulmonary features. The determination of fluctuation localization is based on the mean and fluctuation range of each parameter in the cardiopulmonary function reference baseline. At each sampling moment of continuous acquisition of physiological signals, the measured value of each parameter is compared with the baseline range one by one. When the measured value of a parameter deviates from the baseline fluctuation range for more than two sampling windows, the position is determined to be an abnormal fluctuation position; if the deviation time is less than two sampling windows, it is considered a temporary deviation caused by physiological noise and is not marked. Specifically, when the mean or variation range of heart rate exceeds the baseline limit, it is marked as an abnormal heart rate; when the mean respiratory rate or amplitude exceeds the baseline range, it is marked as an abnormal respiratory rate; when blood oxygen saturation drops below the difference between the baseline mean and the lower limit of fluctuation, it is marked as an abnormal blood oxygen. At the same time, the coupling strength and phase relationship between any two of the three signals are also included in the fluctuation localization detection scope—once the coupling strength deviates from the baseline range, or the phase relationship shows a identifiable hysteresis drift, coupling offset marking is triggered. In clinical settings, patients with both chronic obstructive pulmonary disease (COPD) and heart failure often experience simultaneous and concentrated respiratory and heart rate abnormalities at night. The co-occurrence of heart rate, respiration, and blood oxygenation data precisely reflects the typical pathological pattern of multi-parameter combined abnormalities in these patients. Each cardiopulmonary abnormality record includes four basic attributes: the time of abnormality, the type of parameter involved, the magnitude of deviation, and the duration. The magnitude of deviation is calculated by dividing the difference between the measured value and the mean of the cardiopulmonary function baseline by the baseline standard deviation. Entries with larger deviations are assigned higher priority weights in the cross-validation process.
[0021] In some embodiments, performing cross-signal cross-validation on the cardiopulmonary abnormal features to generate a coupled validation set includes: decomposing the cardiopulmonary abnormal features according to signal type to generate a single-parameter abnormal sequence; performing phase lag detection on the single-parameter abnormal sequence to generate a lag time series distribution; performing coupling strength quantization on the lag time series distribution to generate a coupling strength distribution; and filtering phase lag co-occurrence anomalies according to the coupling strength distribution to generate a coupled validation set.
[0022] Cardiopulmonary abnormalities were decomposed into single-parameter abnormality sequences based on signal type. Entries categorized into heart rate abnormalities, respiratory abnormalities, and blood oxygen abnormalities were extracted separately. Within each category, entries were arranged in ascending order of time based on time and deviation magnitude, thus forming three independent single-parameter abnormality sequences. Taking a patient with both chronic obstructive pulmonary disease and heart failure as an example, entries in the heart rate single-parameter abnormality sequence typically showed dense distribution at night, while the respiratory single-parameter abnormality sequence also exhibited high-density clustering during the same time period. The temporal distributions of the two sequences highly overlapped, providing a representative cross-signal comparison basis for subsequent phase lag detection. Coupled offset entries in the cardiopulmonary abnormality features were independently assigned to the coupled offset sequence during decomposition, without being mixed into the three single-signal sequences mentioned above. The coupled offset sequence provides an auxiliary reference for the phase relationship between single-signal abnormal events during the phase lag detection stage, verifying whether it is consistent with the baseline coupling characteristics. Furthermore, entries in the single-parameter abnormality sequence with a deviation magnitude less than 50% of the overall sequence mean were marked as weak abnormalities. These weakly abnormal entries remained in the sequence for lag detection, but their contribution coefficient in the density statistics of the lag temporal distribution decreased to 0.5. Taking patients with arrhythmia as an example, the proportion of weak abnormal entries in their single-parameter abnormal heart rate sequences is often high. If they are directly counted without distinction, it will mask the true temporal pattern of strong heart rate abnormal events. Once the proportion of weak abnormalities exceeds 60%, the single-parameter abnormal sequence will be marked with high noise, so that the phase lag detection stage will apply an overall reduction to its lag estimation results.
[0023] Phase lag detection is performed on single-parameter abnormal sequences to generate a lag time series distribution. Phase lag detection uses the heart rate single-parameter abnormal sequence as the driving sequence and the respiration and blood oxygenation single-parameter abnormal sequences as the response sequences. The aim is to capture the delays in the appearance of respiratory and blood oxygenation abnormalities after a heart rate abnormality event. The lag duration is calculated as the difference between the driving event time and the nearest neighbor response event time on the time axis: a positive difference indicates that the response event lags behind the driving event, while a negative difference indicates that the respiratory or blood oxygenation abnormality is triggered before the heart rate abnormality, suggesting that the causal conduction direction in the heart and lungs of this event group is opposite to the preset driving direction. When writing the lag duration into the lag time series distribution, the estimation results for weak anomaly entries are multiplied by a reduction factor of 0.5, all estimation results for high-noise labeled sequences are multiplied by 0.4, and strong anomaly entries are written in full with a factor of 1.0. The entire lag time series distribution is organized with each driving event as a node and its corresponding lag duration as an attribute. The distribution pattern of all nodes arranged on the time axis has clinical interpretation value: if the lag duration is concentrated in a certain fixed interval, it suggests a stable pattern in cardiopulmonary conduction delay; if the duration is highly dispersed, it indicates that the conduction relationship is in an unstable state. In practical applications, the lag duration of blood oxygen response to abnormal heart rate in patients with chronic heart failure often clusters in a relatively fixed interval, while the lag duration distribution is significantly diffuse during acute myocardial ischemia. The coupling offset sequence is synchronously compared with the nodes of the lag time series distribution at this stage. When the time of the coupling offset falls exactly into the concentrated interval of lag duration, the weight of the corresponding node is increased, indicating that the single signal lag relationship at that time has been independently corroborated by the coupling offset record. The lag time series distribution also synchronously records the deviation magnitude of the driving event corresponding to each node. This information will be used as input for weighted calculation in the coupling strength quantification stage.
[0024] The coupling strength distribution is generated by quantifying the coupling strength of the lag time series distribution. The coupling strength quantification comprehensively considers the concentration of lag durations and the deviation of driving events at each node in the lag time series distribution. Nodes whose lag durations are closer to the mean of the entire distribution contribute more to the coupling strength. The coupling strength C_k of each node is determined according to the formula C_k=w_d×A_k×(1-|τ_k-τ_mean| / τ_max), where A_k is the normalized value of the deviation of the driving event at the k-th node, τ_k is the lag duration of that node, τ_mean is the mean lag duration of the entire distribution, τ_max is the maximum lag duration within the distribution, and w_d is the driving direction weight. For heart rate driving the respiratory direction, w_d is set to 1.2 to highlight the contribution of the cardiogenic conduction pathway; for respiratory driving the heart rate direction, w_d is set to 0.8. The value range of C_k is limited to 0 to 1. Values below 0 are truncated to 0, and values above 1.0 are truncated to 1.0. For nodes whose C_k is supported by the coupling offset sequence, an additional 0.1 is added after the formula calculation. If the value exceeds 1.0 after the addition, it is truncated to 1.0. Double-validation nodes tend to be concentrated during acute fluctuations in cardiac function, causing the coupling strength distribution to exhibit local peaks in the corresponding regions. The C_k of all nodes arranged in chronological order constitutes the coupling strength distribution. Periods with consistently high C_k correspond to stable and considerable cardiopulmonary coupling conduction, while periods with drastic fluctuations in C_k reflect an unstable coupling relationship. For nodes with negative differences in the lag time series distribution, their C_k is marked with a reverse label in the coupling strength distribution. When reverse-labeled nodes appear densely, it indicates a systematic reversal of the causal direction of cardiopulmonary drive during that period.
[0025] A coupling validation set is generated by filtering co-occurring anomalies with phase lag based on the coupling strength distribution. The selection range of the coupling validation set is limited to strongly coupled nodes in the coupling strength distribution where C_k is not less than 0.3. The heart rate abnormality event corresponding to each strongly coupled node and the respiratory or blood oxygenation response event it triggers together constitute a pair of co-occurring anomalies. The prerequisite for determining a valid co-occurrence is that the two types of anomalies appear sequentially within a specified allowable lag time range. The allowable range is set as the mean of the lag time of the entire distribution plus or minus one standard deviation. If the response event exceeds this allowable window, it is not included in the co-occurrence statistics of the driving node. Each member of the coupling validation set contains three attributes: lag time, coupling strength C_k, and driving direction label. These three attributes jointly describe a complete cross-signal coupling anomaly event. For nodes with reverse labels in the coupling strength distribution, their corresponding co-occurring anomalies are written into the coupling validation set in the reverse driving direction and stored alongside the forward co-occurring anomalies. They are processed separately according to the direction label during the frequency band resolution stage. All members are arranged in descending order of C_k. The higher C_k members at the top of the list represent the most stable and reliable co-occurring abnormal events in cardiopulmonary coupling, forming the core input for frequency band analysis. The lower C_k members at the bottom of the list serve as auxiliary references, participating in the fine-tuning and correction of frequency band boundaries. During the arrangement process, when C_k values are the same, members contributed by dual-verification nodes take priority over members that only receive single verification, reflecting the advantage of dual corroboration in determining the reliability of events.
[0026] Frequency band analysis was performed to extract the cardiopulmonary coupling frequency band based on the coupling validation set. Frequency band analysis was performed using the lag duration sequences of each member in the coupling validation set as time-domain input. Power spectrum estimation was performed on this sequence using C_k weighting, with higher C_k members contributing more significantly to the spectral shape. The peak position of the power spectrum corresponds to the repetition period of the coupled event in the time dimension. Taking the reciprocal of the repetition period yields the characteristic frequency of cardiopulmonary coupling, and the frequency range of this characteristic frequency is the core interval of the cardiopulmonary coupling frequency band. The coupling frequencies of patients with chronic cardiopulmonary insufficiency tend to fall into the low-frequency band, while the coupling frequencies of healthy individuals are mainly concentrated in the high-frequency band corresponding to respiratory rate. The lag duration of the reverse-labeled members in the coupling validation set is processed separately during power spectrum estimation. The peak frequency of the power spectrum of the obtained reverse members is compared with the corresponding result of the forward members. When the difference between the two exceeds 20% of the peak frequency of the forward members, it indicates that the conduction period of the forward and reverse coupling paths has diverged. The cardiopulmonary coupling frequency band records the core frequency ranges of forward and reverse coupling respectively. The reverse coupling frequency range is used as an independent discrimination criterion during the cardiopulmonary status identification and calibration stage and is not merged with the forward coupling frequency range. During the frequency band resolution process, for periods when the number of effective members in the coupling validation set is insufficient to support reliable power spectrum estimation, the benchmark coupling frequency recorded in the cardiopulmonary function reference baseline is substituted as a substitute estimate, and the corresponding period is marked as the baseline substitute in the cardiopulmonary coupling frequency band.
[0027] Step S130: Calibrate the cardiopulmonary state identifier according to the cardiopulmonary coupling frequency band, and perform pattern clustering partitioning on the physiological signal set based on the cardiopulmonary state identifier to generate a cardiopulmonary state partition map.
[0028] Specifically, cardiopulmonary status markers are calibrated according to the cardiopulmonary coupling frequency band. The core frequency intervals of forward coupling and reverse coupling are respectively assigned to the corresponding status markers. When the forward coupling frequency interval falls into the high-frequency band, it indicates relatively intact autonomic nervous system regulation, and is calibrated as a functional compensation status marker; falling into the low-frequency band indicates a decline in autonomic nervous system regulation, and is calibrated as a functional impairment status marker. These two types of cardiopulmonary status markers cover the entire time period of the forward coupling portion of the cardiopulmonary coupling frequency band. The reverse coupling frequency interval is calibrated independently: once the difference between the reverse coupling core frequency and the forward coupling core frequency exceeds 20%, it is calibrated as a coupling differentiation status marker, indicating a systematic reversal of the causal direction of cardiopulmonary drive during that time period. The coupling differentiation status marker is processed as an independent category during pattern clustering and is not merged with functional compensation or functional impairment states. For time periods marked as baseline replacement in the cardiopulmonary coupling frequency band, the calibration range of cardiopulmonary status markers needs to be expanded—the allowable range of the frequency band boundary is widened, the judgment thresholds for both functional compensation and functional impairment cardiopulmonary status markers are simultaneously extended, and the cardiopulmonary status markers for the corresponding time periods are additionally labeled with insufficient estimation. During the pattern clustering partitioning phase, the clustering weight for that time period is reduced to 70% of that for normal time periods to avoid underestimating the dominant partition boundaries of under-estimated segments. The cardiopulmonary status identifier for each time period carries two attributes: status category and the mean C_k of the coupling strength. A higher mean C_k indicates stronger reliability in the determination.
[0029] In some embodiments, the step of generating a cardiopulmonary state partition map by performing pattern clustering partitioning on the physiological signal set based on the cardiopulmonary state identifier includes: performing state sequence merging on the cardiopulmonary state identifier to generate a state category set; performing temporal mapping between the physiological signal set and the state category set to generate a mapping sequence; identifying dense state transition segments within the mapping sequence and performing priority boundary marking to generate a state boundary point set; and constructing a cardiopulmonary state partition map using the state boundary point set.
[0030] The cardiopulmonary status identifiers are merged into a status category set. After the cardiopulmonary status identifiers are arranged chronologically, if adjacent time periods have the same status category and the difference in the mean C_k does not exceed 0.15, the corresponding cardiopulmonary status identifiers are merged into the same status segment. The merged mean C_k is the weighted average of the mean C_k of each sub-segment, with the weights allocated according to the duration of each sub-segment—the longer the duration, the greater the contribution; short-lived sub-segments of the same type of status have a lower contribution weight due to their limited duration, thus avoiding excessive skewing of the merged mean by instantaneous similar identifiers. Adjacent time periods with different status categories remain independent and are not merged; the boundary between functional compensation and functional impairment must be accurately recorded. The greater the difference in the mean C_k within a sampling window on both sides of the boundary, the more drastic the state transition. A boundary with a high degree of drastic transition reflects a sudden rather than gradual change in functional status. Patients with multiple system diseases frequently alternate between status categories during monitoring, and the number of status segments in the status category set is significantly greater than that of patients with a single disease, and the duration of each status segment is also generally shorter. After merging, the state categories are concentrated, and the proportions of functional compensation, functional impairment, and coupling differentiation outline the overall distribution pattern of cardiopulmonary function in the current monitoring phase. When the proportion of coupling differentiation exceeds 30%, it indicates a systemic disorder in the cardiopulmonary causal relationship. The combined distribution of the duration proportions of the three states and the intensity of transitions constitutes the core input basis for the time-series mapping process.
[0031] A mapping sequence is generated by temporally mapping the physiological signal set to the state category set. The mapping process indexes each unit in the state category set according to its start and end times. The measured value sequences of the three signals in the physiological signal set corresponding to different time periods are associated with the state category attributes time-by-time. The measured values of each signal in the physiological signal set retain their original amplitude information at the mapping nodes. The association result forms a mapping sequence: each time node carries a state category label and the measured values of the three signals. At the same time, the combined measured values of heart rate, respiration, and blood oxygenation constitute the multi-parameter feature vector of that node. Each dimension of the feature vector is baseline normalized: the measured value is subtracted from the mean of the corresponding parameter of the cardiopulmonary function reference baseline, and then divided by the baseline standard deviation. After normalization, each dimension directly represents the degree of deviation from the baseline. Taking a patient with both heart failure and chronic obstructive pulmonary disease as an example, a normalized value of 2.5 for the heart rate dimension at a certain moment in the physiological signal set means that the heart rate deviates from the baseline mean by 2.5 standard deviations. At the same moment, a normalized value of -1.2 for the blood oxygen dimension indicates that the blood oxygen is slightly lower than the baseline. The state category set labels this period as a functional impairment state. The two types of information are completely correlated in the mapping sequence, intuitively presenting a cross-dimensional abnormal pattern of compensatory increase in heart rate and synchronous decrease in blood oxygen. Differentiation labels are added to the mapping sequence nodes corresponding to the coupling differentiation state periods. The differentiation label nodes have their transition intensity parameter increased by 0.2 in the identification of dense state transition sections, so that the coupling causal reversal moment gets higher weight in the dense section judgment. The contribution coefficient of the feature vector of nodes with insufficient estimation of the labeled period in the mapping sequence is reduced to prevent the feature vector noise of low confidence segments from masking the true state transition signal.
[0032] For example, the step of identifying densely transitioned state segments within the mapping sequence and implementing priority boundary marking to generate a set of state boundary points includes: extracting state abrupt change locations within the mapping sequence to generate a set of abrupt change points; performing inverse weighted mapping of abrupt change magnitude on the set of abrupt change points to generate a set of weak abrupt change weights; identifying high-weight segments within the set of weak abrupt change weights and implementing boundary expansion to generate an extended boundary set; and determining the set of state boundary points according to the extended boundary set and the set of abrupt change points.
[0033] A mutation point set is generated by extracting the locations of state mutations within the mapping sequence. The criterion is the moment when the Euclidean distance between the feature vectors of adjacent nodes in the mapping sequence exceeds a threshold, which is the mean of the distances between pairs of global feature vectors in the mapping sequence plus one standard deviation. Before calculating the Euclidean distance, the differences in each dimension are weighted according to the acquisition quality score of the corresponding signal in the physiological signal set quality metadata. The weight of dimensions with low quality scores is reduced accordingly to ensure a more balanced contribution of the three signals to the mutation amplitude. The combination of the state categories of the predecessor and successor nodes determines the classification of the transformation direction of each member in the mutation point set: members of the mutation point set whose function shifts from functional compensation to functional impairment are labeled as deterioration mutations, members in the opposite direction are labeled as recovery mutations, and members of the mutation point set whose transformation involves coupling differentiation state are labeled as differentiation mutations. This situation is more common in patients with severe arrhythmias—during the arrhythmia attack, the cardiopulmonary causal relationship frequently reverses, differentiation mutations appear densely in a short period of time, and the number of deterioration and recovery mutations is similar, indicating that the functional state is repeatedly oscillating in a highly unstable state rather than unidirectional deterioration. Uncertain labels are added to the mutation locations within the estimated insufficient labeling period, and the magnitudes of the corresponding mutation point set members are treated with a reduction factor of 0.7 during the inverse weighting mapping stage.
[0034] A weak mutation weight set is generated by performing an inverse weighted mapping of mutation amplitude on the mutation point set. This mapping transforms mutation points with large amplitudes into low weights and mutation points with small amplitudes into high weights. The specific mapping rule is W_i = 1 - A_i / A_max, where A_i is the normalized value of the mutation amplitude of the i-th mutation point, A_max is the maximum value of the mutation amplitude within the mutation point set, and W_i ranges from 0 to 1. The reason for using inverse mapping is that large mutation locations already have significant boundary marker value, while small mutation locations are easily overlooked in conventional detection, but may mark key turning points in the gradual state transition process. State transitions in patients with multiple comorbidities are often gradual rather than abrupt. For example, when a patient's heart rate and blood oxygenation slowly deviate from the baseline, a single mutation amplitude is insufficient to trigger a highly significant label, but multiple consecutive small mutation locations are densely clustered with high weights in the weak mutation weight set, revealing a systematic gradual trend in this segment. The A_i of deterioration-type mutations is usually higher than that of recovery-type mutations. This is because during functional deterioration, multiple signals synchronously deviate from the baseline, causing the Euclidean distance to be contributed by multiple dimensions. In contrast, recovery-type mutations are often driven by a single-dimensional improvement, with other dimensions lagging behind, resulting in a smaller overall Euclidean distance. Therefore, inverse weight mapping assigns a higher W_i to recovery-type mutations, ensuring that small improvement nodes in the gradual recovery process receive sufficient attention. This forms a weak mutation weight set. Higher W_i values for members in the weak mutation weight set indicate a higher priority for the mutation point in boundary labeling. For uncertainly labeled mutation points, the W_i is multiplied by a reduction factor of 0.7 based on the original mapping result. The members of the reduced weak mutation weight set still participate in the identification of high-weight segments, but their contributions are clearly limited.
[0035] Within the weak mutation weight set, high-weight segments are identified and their boundaries are expanded to generate an extended boundary set. The mean W_i is calculated for each time window of the weak mutation weight set. Window periods where the mean W_i exceeds the global mean W_i are identified as high-weight segments, with their start and end boundaries determined by consecutive time windows where the mean W_i consistently exceeds the global mean. When the mean W_i within a segment is consistently higher than the global mean and the mean curve shows an upward trend, it corresponds to a gradual deterioration of the condition rather than random fluctuations, and the boundary range expands synchronously with the continuous rise of the mean curve. The sliding window length is consistent with the basic sampling window length during physiological signal acquisition. The window is stepped by half a sampling window to ensure 50% overlap between adjacent statistical intervals, resulting in a smooth transition of the mean W_i curve and preventing high-weight segments from being artificially truncated due to window alignment errors. The boundary expansion operation extends the start and end boundaries of each high-weight segment by one sampling window length to both sides, covering mutation points that may be truncated by the window at the edges. When the boundaries of adjacent high-weight segments overlap after expansion, they are merged into the same extended boundary set member, and the union of the two extended boundaries is taken. In patients with chronic cardiopulmonary dysfunction, adjacent high-weight segments frequently trigger merging during the continuous progression of the disease, eventually forming a large-scale extended boundary set member spanning several hours, which is in stark contrast to the generally shorter duration of extended boundary set members during occasional and brief fluctuations in healthy individuals.
[0036] The state boundary point set is determined based on the extended boundary set and the mutation point set. Within each member of the extended boundary set, the mutation point with the highest W_i is selected as the representative boundary candidate for that segment; the candidate's time corresponds to the precise boundary location. If a differentiated mutation point exists within half a sampling window at that time, the precise boundary location is corrected to the time of the differentiated mutation point. Priority is given to including the time of causal inversion in the precise boundary, ensuring that the boundary location in the state boundary point set is highly aligned with the time of causal direction change. Simultaneously, deterioration and recovery mutation points corresponding to significant mutation locations in the mutation point set are introduced as supplementary boundary candidates to ensure that the clear boundary generated by a single drastic state jump is not obscured by low weights in the inverse weighted mapping due to its large amplitude. For example, if a patient experiences an acute arrhythmia and the Euclidean distance of the heart rate feature vector suddenly increases to more than three times the global mean, the corresponding significant deterioration mutation point would not be able to obtain boundary labeling in the gradually changing extended boundary set without a supplementary candidate mechanism. The supplementary mechanism is designed for the accurate location of boundaries in such acute events. After merging the two candidate paths and removing duplicates, candidates with a time difference within one sampling window are retained based on the larger mutation magnitude. Boundary points from the high W_i mean extension segment are given higher weights than candidates with large mutations. The merged and deduplicated candidate set is the state boundary point set. Each boundary point in the state boundary point set carries time information for subsequent partitioning of the partition map.
[0037] Construct a cardiopulmonary state partitioning map using a set of state boundary points. For example... Figure 3As shown, using the time of state boundary point 12 as the dividing line, the time axis of the mapping sequence is divided into functional compensation state partition 9, functional impairment state partition 10, and coupled differentiation state partition 11. Each partition corresponds to a continuous and stable time course of state category. The partition attributes of the cardiopulmonary state partition map are jointly constructed by the representative category and the mean of the feature vectors of the three signals in the corresponding time period. The mean of the feature vectors is compared with the cardiopulmonary function reference baseline and then written into the offset amplitude parameter of the corresponding partition of the comprehensive offset amplitude curve 13. The offset amplitude in functional impairment state partition 10 is usually higher than that in functional compensation state partition 9. However, the cross-dimensional difference in offset amplitude in functional impairment state partition 10 is greater in patients with multiple comorbidities—the offset directions of the heart rate dimension and blood oxygen dimension are sometimes opposite, suggesting that compensatory increase in heart rate and decrease in peripheral oxygenation coexist in the same partition. This is fundamentally different from the typical pattern of consistent offset directions of heart rate and blood oxygen in patients with simple heart failure. The coupled differentiation state partition 11 is supplemented with a differentiation label. The offset amplitude parameter of the differentiation label partition is processed separately in the self-recovery ability assessment stage and is not combined with the adjacent undifferentiated partitions for statistics. When adjacent partitions represent the same category and the difference in the mean of feature vectors is less than 10% of the global mean, partition merging is triggered to compress the number of fragmented partitions, so that the partition structure of the cardiopulmonary status partition map matches the actual evolution granularity of cardiopulmonary function status. The final generated cardiopulmonary status partition map completely records the overall evolution of cardiopulmonary function during the monitoring period.
[0038] Step S140: Based on the cardiopulmonary status zoning map and the cardiopulmonary function reference baseline, offset quantification is performed to generate offset amplitude, self-recovery capability assessment is performed on the offset amplitude to generate recovery capability coefficient, and the abnormal risk set is generated by identifying the time points of continuous abnormality according to the recovery capability coefficient.
[0039] Specifically, offset quantization is performed to generate offset amplitude based on the cardiopulmonary status zoning map and the cardiopulmonary function reference baseline. The mean of the feature vectors already written in the attributes of each zone of the cardiopulmonary status zoning map constitutes the starting point for offset quantization calculation. The mean of the feature vectors of the three signals is subtracted from the mean of the corresponding parameter of the cardiopulmonary function reference baseline dimension by dimension. The difference is divided by the baseline standard deviation to obtain the standardized offset of each dimension. The vector magnitude of the standardized offset of the three dimensions is the comprehensive offset amplitude of the zone. The larger the magnitude, the more significant the overall deviation of multiple parameters within the zone from the baseline. The offset amplitude of the functional impairment zone is usually higher than that of the functional compensation zone, but the difference in offset amplitude between the two types of zones is significantly smaller in patients with multiple comorbidities than in patients with a single disease. The reason for this phenomenon is that the cardiopulmonary function reference baseline of patients with multiple comorbidities already reflects the systemic offset under chronic pathological conditions. After the overall baseline is raised, the relative difference in the offset amplitude between the sub-zones is compressed. The offset amplitude of the functional compensation zone of such patients may still be higher than that of the functional impairment zone of patients with a single disease, suggesting that baseline individualization is a key prerequisite for the accuracy of offset quantization. Due to the addition of differentiation markers, the offset magnitude calculation for coupled differentiation state partitions incorporates the coupling phase offset as a fourth dimension in addition to the three-channel signal calculation. This offset equals the mean of the lag time series distribution within the partition minus the baseline coupling phase recorded in the cardiopulmonary function reference baseline. The coupling phase offset is normalized by dividing by the baseline coupling phase recorded in the cardiopulmonary function reference baseline and then used as the fourth dimension. The magnitude of this four-dimensional vector constitutes the comprehensive offset magnitude of the coupled differentiation partition, which is typically higher than that of undifferentiated partitions with the same degree of functional impairment. The introduction of the coupling phase offset enables the differentiation partitions to capture cardiopulmonary temporal decoupling information that cannot be reflected by simple signal amplitude changes.
[0040] In some embodiments, the step of performing an autonomous recovery capability assessment on the offset amplitude to generate a recovery capability coefficient includes: performing time-series tracking on the offset amplitude to generate an offset time-series curve; identifying the recovery start point of the offset time-series curve to generate a recovery start marker; calculating the abnormal recovery rate according to the recovery start marker to generate a recovery rate value; and generating a recovery capability coefficient based on a comprehensive quantization of the recovery rate value and the offset amplitude.
[0041] The offset amplitude is time-series tracked to generate offset time-series curves. The comprehensive offset amplitude of each zone is arranged in chronological order according to the cardiopulmonary status zoning map. The midpoint of the start and end times of each zone is taken as the representative point. Adjacent representative points are connected by linear interpolation to form a continuous offset time-series curve. The slope of the interpolation segment is equal to the difference in offset amplitude between adjacent representative points divided by the time interval. A positive slope indicates that the offset amplitude between two zones is continuously increasing, while a negative slope indicates that the offset amplitude is converging towards the baseline. The position of positive-negative transition marks the node of functional status change. Taking a patient with acute exacerbation of chronic obstructive pulmonary disease as an example, the functional impairment state zone appears continuously during the acute exacerbation period. The slope of the corresponding offset time-series curve segment is continuously positive and gradually increases. The offset amplitude continuously climbs from the upper limit of the baseline tolerance to more than twice within a few hours. After treatment intervention, the functional compensation state zone appears. The slope of the corresponding segment turns negative, and the offset amplitude begins to converge towards the baseline. The offset time-series curve thus shows a typical arc-shaped trajectory of first rising and then falling. In patients with concurrent acute infection-induced cardiopulmonary dysfunction, the slope of the offset time-series curve abruptly rises to a high positive value during the peak of infection. The extreme slope of the former is usually more than three times that of patients with chronic cardiopulmonary insufficiency, reflecting the essential difference in the temporal dimension between the two types of deterioration mechanisms: acute-driven and chronic-accumulated. Representative points corresponding to the coupling differentiation zones are distinguished by hollow markers. During the recovery starting point identification stage, the hollow marker points use four-dimensional vector magnitudes instead of three-dimensional magnitudes in slope estimation to ensure that the offset change trend during the coupling differentiation period is not underestimated due to dimensional inconsistencies.
[0042] The recovery start point of the offset time-series curve is identified to generate a recovery start marker. The recovery start point is the position where the slope sign of the offset time-series curve reverses from rising to falling, corresponding to the moment when the body's self-recovery mechanism begins to take effect and the deviation from the baseline begins to narrow. The reversal is determined by the slope estimate of the representative point of the adjacent partition changing from positive to negative. A single sign reversal lasting less than two partitions is considered a brief fluctuation and does not trigger a marker. A reversal lasting two or more partitions is confirmed as a valid reversal. The valid reversal position is written as recovery start marker 14 in the corresponding partition of the cardiopulmonary status partition map. When the slope of the offset time-series curve switches rapidly between positive and negative multiple times within the same time period, it indicates that the offset amplitude is oscillating repeatedly rather than trending downward. This oscillation pattern is often seen in patients with arrhythmias: alternating arrhythmia attacks and remissions cause the offset time-series curve to be sawtooth-shaped, with each remission causing a brief negative slope, and the slope quickly returning to positive when the attack resumes. When the interval between adjacent positive and negative switching is less than the duration of one partition, no recovery start marker is generated at the corresponding position. When there are more than 3 consecutive oscillations and no single sustained effective decline in more than two partitions, the oscillation segment marker 16 is written into the corresponding partition of the cardiopulmonary status partition map, indicating that although the autonomous recovery mechanism is intermittently activated, its sustainability is insufficient. Each recovery start marker records the offset value between the corresponding time and that time, and this offset value is the recovery start reference amplitude. The higher the reference amplitude, the more severe the deviation state needs to be from in the recovery process, and the greater the recovery difficulty. Under the same recovery rate, a marker with a higher reference amplitude corresponds to a longer expected recovery time.
[0043] Recovery rate values are generated by calculating abnormal recovery rates based on recovery initiation markers. The calculation of the recovery rate starts from the time corresponding to the recovery initiation marker and tracks the subsequent decline of the offset time-series curve. The decline rate is taken as the average absolute value of the curve slope, and the calculation range covers the complete decline from the recovery initiation marker time to the offset time-series curve re-reaching the baseline tolerance range. The baseline tolerance range is defined as the upper limit of the comprehensive offset amplitude corresponding to the fluctuation range of the corresponding parameters of the cardiopulmonary function reference baseline. If a local rebound occurs during the decline and the rebound amplitude exceeds the recovery initiation baseline amplitude by 10%, the rebound segment is excluded from the slope average calculation, and the segment that continues to decline after the rebound is included again to ensure that the recovery rate value only reflects the true recovery trend. An unstable recovery marker is added when the local rebound amplitude exceeds 30%. This marker is triggered more frequently in patients with unstable volume overload—diuretic therapy causes rapid fluctuations in blood volume, and the offset time-series curve rebounds multiple times during its decline, each rebound corresponding to a temporary functional regression caused by volume redistribution. Recovery rate values are calculated separately for multiple effective recovery initiation markers, and the resulting recovery rate value sequence reflects the dynamic changes in the body's spontaneous recovery ability over different time periods. A continuously declining recovery rate indicates that the ability to recover spontaneously gradually declines as the monitoring period progresses. This pattern is commonly observed in patients with chronic cardiopulmonary insufficiency and recurrent acute exacerbations.
[0044] The recovery capacity coefficient is generated by comprehensively quantifying the recovery rate value and the offset amplitude. The recovery rate value reflects the recovery momentum, while the peak offset amplitude reflects the degree of functional deviation. Together, they determine the body's autonomous recovery capacity level under the current functional state. Two typical patient types illustrate the combined effect: A patient with acute heart failure experiences a rapid decrease in the offset time-series curve after diuretic treatment, resulting in a high recovery rate value. However, the patient's peak offset amplitude is also high, indicating that the burden remains heavy despite strong compensatory mobilization. The corresponding recovery capacity coefficient, suppressed by the denominator, is at a moderate level. Conversely, a patient in a chronic stable phase experiences a slow and prolonged convergence of the offset time-series curve, resulting in a low recovery rate value but also a low peak offset amplitude. This reflects a chronic stable state with low burden and low mobilization, and the corresponding recovery capacity coefficient is also at a moderate level. The clinical intervention priorities for these two types of patients differ fundamentally. The recovery capability coefficient R_c is calculated using the formula R_c = V_rec / (1 + λ × P_peak), where R_c is the recovery capability coefficient, V_rec is the recovery rate value normalized to 0-1, P_peak is the normalized value of the peak offset amplitude before the corresponding recovery initiation mark, and λ is the burden adjustment coefficient, set to 0.5. The purpose of introducing λ × P_peak in the denominator is to indicate that the higher the peak offset amplitude, the greater the suppression of R_c, reflecting the actual loss of recovery capability under high burden conditions. The value of R_c ranges from 0 to 1; a value above 0.7 indicates good autonomous recovery capability, while a value below 0.3 indicates severely impaired recovery capability. For recovery rate values with unstable recovery labels, V_rec is multiplied by a reduction factor in the formula. The reduction factor is determined by the number of times the unstable recovery label is triggered: 0.8 for one trigger, 0.6 for two triggers, and 0.5 for three or more triggers, reflecting that the more local rebounds during the recovery process, the more significant the weakening of the actual recovery capability. When the oscillating segment in the offset time series curve does not generate an effective recovery rate value, the recovery capability coefficient of the corresponding time period is assigned a value of 0 and directly included in the low recovery interval set.
[0045] In some embodiments, the step of identifying persistent abnormal time points and generating an abnormal risk set based on the recovery capability coefficient includes: locating low-value intervals of the recovery capability coefficient to generate a low-recovery interval set; identifying fluctuation characteristics of a brief rebound followed by a decline in the low-recovery interval set to generate a pseudo-recovery marker set; statistically analyzing the duration of the pseudo-recovery marker set to generate a duration distribution; and integrating the duration distribution with the pseudo-recovery marker set to generate an abnormal risk set.
[0046] Low-recovery interval sets are generated by locating low-value intervals in the recovery ability coefficient. The detection object is the temporal variation of the recovery ability coefficient. Periods where R_c is lower than the mean of the entire sequence minus 0.5 times are identified as low-recovery intervals. The threshold is set at this level to ensure that normal fluctuations in R_c slightly below the mean do not trigger annotation, and only periods of significant decline in recovery ability are captured. The threshold is dynamically updated with the monitoring time: in the early monitoring stage when the sample size is insufficient, the population reference mean is used to replace the individual sequence mean. After more than 10 valid R_c values are accumulated in the individual sequences, the individual mean is switched to. Oscillating segments with a recovery ability coefficient of 0 are directly included in the low-recovery intervals without threshold screening. When the time interval between two adjacent low-recovery intervals does not exceed one partition duration, they are merged into the same member, thus identifying the situation of a brief rise followed by a rapid decline again as a continuous low-recovery process rather than independent and scattered low value points. The recovery ability coefficient sequences of patients in the acute exacerbation of chronic cardiopulmonary failure usually show long-term sustained low values, and the duration of the low-recovery interval set members is significantly longer than that of patients with intermittent acute events. For example, a patient with acute exacerbation of chronic obstructive pulmonary disease (COPD) whose recovery coefficient remains below the threshold for 72 hours during hospitalization forms a single low-recovery interval set spanning the entire pre-hospitalization period. In contrast, the low-recovery interval sets for patients with intermittent arrhythmias typically last for hours and are independent of each other. Each member's start and end times and the mean R_c value within the interval are recorded. Intervals with a mean R_c value below 0.3 are labeled as severely low-recovery intervals. The threshold for determining severely low-recovery intervals is tightened accordingly during the pseudo-recovery label set generation stage.
[0047] A pseudo-recovery marker set is generated to identify fluctuations in the low recovery interval set, characterized by a brief rise followed by a decline. The detection target is the R_c time series within each member of the low recovery interval set. Positions where R_c shows a local rise followed by a decline within the interval are identified as pseudo-recovery candidates. The local rise must exceed 20% of the mean R_c within the interval to trigger candidate labeling; minor fluctuations below this range are considered noise and ignored. For severe low recovery intervals, the candidate trigger threshold is tightened to 10% to ensure that minor pseudo-recovery signals during periods of severe illness are not missed. The complete process of the rise followed by a decline must be completed within two time intervals; rises exceeding this duration are considered genuine recovery trends rather than pseudo-recoveries. In patients with both heart failure and chronic obstructive pulmonary disease, pseudo-recovery candidates frequently appear within the low recovery interval set during nocturnal hypoxic stress periods—a single hypoxic stress triggers a brief drop in heart rate, causing a local rise in R_c, which then declines again due to increased respiratory burden. The direct cause of this pattern is the prior improvement in heart rate signals followed by a lagging deterioration in respiratory signals, resulting in a brief contraction followed by a expansion of the three-dimensional feature vector magnitude. Candidate locations meeting the above conditions are aggregated to form a pseudo-recovery marker set. Each member's location is written with pseudo-recovery marker 15 in the corresponding partition of the cardiopulmonary status zoning map. Each member of the pseudo-recovery marker set records three attributes: pseudo-recovery start time, fall time, and peak R_c. A higher peak R_c in the pseudo-recovery marker set indicates a greater increase in R_c during the pseudo-recovery process, corresponding to a stronger degree of compensatory capacity mobilization but insufficient sustainability. Pseudo-recovery marker set members with a peak R_c exceeding 0.5 indicate that the body has approached the normal compensatory level during the short recovery period, and the subsequent fall reflects that the external stress continues to exceed the upper limit of the compensatory maintenance capacity.
[0048] The duration of the pseudo-recovery marker set was statistically analyzed to generate a duration distribution. The statistical object was the time span from the start time to the fall time for each member of the pseudo-recovery marker set. The durations of each member were arranged in ascending order to form a duration distribution sequence. The overall shape of the duration distribution reflects the regularity of the pseudo-recovery process in the time dimension. If the duration distribution is concentrated in a short range, it suggests that the pseudo-recovery process is periodic, with the body repeatedly triggering ineffective compensation at a fixed rhythm. If the duration distribution is dispersed, it suggests that pseudo-recovery is driven by multiple different mechanisms, with differences in the triggering conditions and decay rates of each mechanism. Patients with combined sympathetic hyperactivation and parasympathetic inhibition typically have uniform and short durations of pseudo-recovery, contrasting with the highly dispersed duration distribution pattern in patients with multiple target organ dysfunction. The mean and standard deviation of the duration distribution jointly reflect the temporal regularity of the pseudo-recovery process: a short mean and a small standard deviation indicate a rapidly failing, periodic pseudo-recovery, while a long mean and a large standard deviation indicate an irregular pseudo-recovery timeline. The duration of pseudo-recovery markers within the severely low-recovery zone is typically shorter than that within the mildly low-recovery zone, reflecting a faster decline in compensatory mobilization capacity under severe impairment. Pseudo-recovery markers with a duration exceeding two partition durations are separately labeled as long-term pseudo-recovery. The peak R_c of these markers is usually higher than that of short-term pseudo-recovery markers, suggesting a positive correlation between compensatory mobilization intensity and duration. Periods with dense occurrences of long-term pseudo-recovery correspond to high-density tail intervals in the duration distribution, and are typical precursors to progressive functional decline.
[0049] An anomaly risk set is generated by integrating the duration distribution and the pseudo-recovery marker set. The risk weights are based on the pseudo-recovery member density corresponding to each duration interval in the duration distribution: intervals with short duration and high density correspond to dense pseudo-recoveries with rapid failure, and are assigned high risk weights; intervals with long duration but low density correspond to occasional long-term pseudo-recoveries, and are assigned median risk weights. The difference in risk weights between the two types of patterns reflects that dense short-cycle ineffective compensation is a stronger indicator of persistent anomalies than occasional long-cycle pseudo-recoveries. For each member in the pseudo-recovery marker set, an anomaly risk score is calculated by weighting three parameters: the pseudo-recovery member density in its duration distribution interval, the mean of the corresponding risk weight, and the mean R_c of its low-recovery interval. The weights are 0.4, 0.3, and 0.3, respectively. A higher score indicates a more severe persistent anomaly in the corresponding time period and a lower probability of successful autonomous recovery. The anomaly risk score of members in the pseudo-recovery marker set corresponding to the severely low-recovery interval is increased by an additional 0.1 on top of the weighted average of the three parameters, reflecting the additional contribution of the severely damaged state to the risk of persistent anomalies. After calculating the abnormal risk scores of all members of the pseudo-recovery marker set, the abnormal risk set is formed by arranging and summarizing the scores in descending order. The high-scoring members in the abnormal risk set represent the period with the highest risk of continuous deterioration of cardiopulmonary function. The number of high-scoring members in the abnormal risk set of patients with multiple system diseases is usually greater than that of patients with single diseases. The final abnormal risk set is then fed into the subsequent joint grading and matching process.
[0050] Step S150: Use the abnormal risk set and cardiopulmonary status identifier to perform joint hierarchical matching to generate a comprehensive cardiopulmonary monitoring index. Perform pattern recognition analysis on the comprehensive cardiopulmonary monitoring index to generate cardiopulmonary abnormality patterns. Perform compensation peak identification on the cardiopulmonary abnormality patterns to generate compensation status identifiers. Perform trend weighted evaluation based on the compensation status identifiers to generate a cardiopulmonary function monitoring report.
[0051] In some embodiments, the step of generating a comprehensive cardiopulmonary monitoring index by jointly performing hierarchical matching using the abnormal risk set and the cardiopulmonary status identifier includes: performing cross-state association analysis on the abnormal risk set to generate an association strength distribution; performing stable state weight enhancement on the association strength distribution and the cardiopulmonary status identifier to generate a hierarchical weight set; performing weighted integration on the abnormal risk set according to the hierarchical weight set to generate a comprehensive abnormality quantity; and using the comprehensive abnormality quantity for normalization processing to generate a comprehensive cardiopulmonary monitoring index.
[0052] For example, the step of performing cross-state association analysis on the abnormal risk set to generate an association strength distribution includes: performing state attribution labeling on the abnormal risk set to generate a state attribution set; extracting the causal direction of heart rate abnormality and respiratory abnormality from the state attribution set to generate an asymmetric causal matrix; performing directional weight allocation based on the asymmetric causal matrix to generate a weight allocation table; and using the weight allocation table to quantify the degree of cross-state association to generate an association strength distribution.
[0053] Anomaly risk sets are generated by performing state attribution labeling. The labeling process indexes each member of the anomaly risk set according to its start and end times, mapping the state category sequence of cardiopulmonary status identifiers for the corresponding time period to each member of the anomaly risk set. This provides each member with a complete temporal context of its state categories. The state category sequence for each member records the functional compensation, functional impairment, or coupling differentiation state at each moment within the member's time period, with the duration percentage of each category written into the member attributes of the state attribution set. Members of the state attribution set with a functional impairment state percentage exceeding 70% are labeled as impairment-dominant; members with a functional compensation state percentage exceeding 50% are labeled as compensation-dominant; and members with both categories below 50% are labeled as mixed states. For mixed-state members, a weighted merging method is used in the causal direction extraction stage, where each state time period is analyzed independently and then weighted. The weights are determined by the duration percentage of each state, preventing the dominant state from masking causal signals in a few time periods. When the mean C_k value is low during the coupling differentiation state period, the reliability of the corresponding state attribution label decreases. The attribution conclusion for this period is marked with a low reliability label, and the weight of the statistical results for the low reliability labeled period is reduced to 0.7 during the causal direction extraction stage. Taking a patient with both heart failure and chronic obstructive pulmonary disease as an example, the members of their abnormal risk concentration during the nighttime period usually span both functional compensation and functional impairment states. The state attribution set labels this type of member as a mixed state. Members of the mixed state are analyzed independently by time period during the subsequent causal direction extraction to ensure that the causal signals driven by heart rate and breathing are completely captured.
[0054] Asymmetric causal matrices are generated by extracting the causal directions of heart rate abnormalities and respiratory abnormalities from state attribution sets. The extraction is based on the temporal sequence of single-parameter heart rate abnormality sequences and single-parameter respiratory abnormality sequences for each member of the state attribution set within the corresponding time period. The causal direction conclusion for each member in the state attribution set is written into the corresponding element of the asymmetric causal matrix. Cases where heart rate abnormalities precede respiratory abnormalities are marked as heart rate-driven directions, and cases where respiratory abnormalities precede heart rate abnormalities are marked as respiratory-driven directions. The time difference between the occurrences must be within one sampling window to determine a causal relationship; events exceeding this range are considered independent events and are not included in the element statistics of the asymmetric causal matrix. The row and column indices of the asymmetric causal matrix are both state categories. The element M_pq represents the ratio of heart rate-driven frequency to respiratory-driven frequency during the transition from state p to state q. A ratio greater than 1 indicates that heart rate abnormalities usually trigger before respiratory abnormalities, with cardiogenic drive being dominant; a ratio less than 1 indicates that respiratory drive is dominant. Taking a patient with acute exacerbation of chronic obstructive pulmonary disease and decompensated heart failure as an example, during the transition from functional compensation to functional impairment, the respiratory abnormality events of the members corresponding to the state affiliation set systematically occurred earlier than the heart rate abnormality events, resulting in the ratio of elements in this direction in the asymmetric causal matrix being much less than 1, intuitively showing the direction of disease evolution dominated by pulmonary drivers; the diagonal elements of the functional impairment state also showed an uneven distribution with respiratory drive frequency significantly higher than heart rate drive frequency. The asymmetric causal matrix fully recorded the differences in drive direction across states and within states, providing a directional basis for subsequent weight allocation.
[0055] A weight allocation table is generated based on directional weight allocation using an asymmetric causal matrix. The ratios of each element in the asymmetric causal matrix form the starting basis for weight calculation: the weight of heart rate abnormality members is increased in the direction of state transition with a higher heart rate-driven ratio, and the weight of respiratory abnormality members is increased in the direction of respiratory-driven ratio. The increase is linearly determined by the deviation of the matrix element ratio from 1.0; the larger the deviation, the higher the increase. When the deviation is less than 0.1, it is considered that the driving direction is balanced and no weight adjustment is made. The row index of the weight allocation table is the direction of state transition, and the column index is the type of abnormality. Each element records the weight coefficient of heart rate abnormality and respiratory abnormality members in the corresponding direction. In the direction dominated by cardiogenic factors, the weight coefficient of heart rate abnormality is usually 1.2 to 1.5, and the weight coefficient of respiratory abnormality is 0.8 to 1.0. The opposite is true when pulmonary factors are dominant. Patients with both acute exacerbations of chronic obstructive pulmonary disease (COPD) and decompensated heart failure exhibit an abnormally high proportion of respiratory drive in the direction of functional compensation shifting towards functional impairment. The corresponding respiratory abnormality weight coefficients in the weighting table are significantly elevated, reflecting the typical disease progression pattern of lung function deterioration preceding cardiac function deterioration in these patients. The directional differences in the weighting table directly affect the subsequent calculation results of cross-state association strength. The difference between the weight coefficients for cardiac and pulmonary origins increases linearly with the deviation of the element ratios in the asymmetric causal matrix.
[0056] A weighted allocation table is used to quantify the degree of cross-state association and generate an association strength distribution. The weight coefficients corresponding to each state transition direction covered by the abnormal risk set member are multiplied by the mean abnormal risk score within that direction, and the sum is obtained. Dividing the sum by the number of covered directions yields the normalized association strength, eliminating the interference of the number of states crossed and making single-state abnormal risk set members comparable to those across multiple states. The normalized association strengths of all abnormal risk set members are arranged chronologically to form the association strength distribution. Members with association strengths higher than the overall distribution mean are marked as strongly associated members. These members receive higher grading weights during the steady-state weight enhancement phase. Taking a patient primarily driven by respiratory function as an example, after weighting the cross-state abnormal events in the abnormal risk set according to the directions with higher respiratory abnormality weight coefficients in the weighted allocation table, the association strength of the corresponding nodes in the association strength distribution increases by 30% to 50% compared to the baseline association strength without directional weighting. This places the member in the strongly associated interval of the association strength distribution, suggesting that the persistent abnormal risk of this abnormal risk set member is mainly driven by lung function. When calculating the association strength, the weight reduction effect of the abnormal risk set members corresponding to the low reliability label period is synchronously transmitted. After the association strength of the corresponding association strength distribution node is reduced, a low reliability label is added. During the stable state weight enhancement stage, the upper limit of the enhancement coefficient of such association strength distribution nodes is compressed to 80% of the normal level, ensuring that abnormal events in the low reliability period do not excessively dominate the overall pattern of association strength distribution.
[0057] A tiered weight set is generated by enhancing the stable-state weights of the association strength distribution and cardiopulmonary status identifiers. The enhancement is based on the mean C_k value of each time period in the cardiopulmonary status identifiers. Time periods with higher mean C_k values correspond to stronger reliability. The association strength of the association strength distribution nodes within that time period is multiplied by an enhancement coefficient, which is equal to the mean C_k divided by the global mean C_k value. Time periods with a mean C_k value higher than the global mean have an enhancement coefficient greater than 1, while those with a mean C_k value lower than the global mean have an enhancement coefficient less than 1. The mean C_k value for functionally compensated time periods in the cardiopulmonary status identifiers is generally higher than that for functionally impaired time periods. The enhanced nodes in the association strength distribution under stable compensated states have higher weights, reflecting higher reliability of the association degree of abnormal events under a stable functional background. When the mean C_k value for functionally impaired time periods is low, the enhancement coefficient is less than 1, corresponding to a reduction in the weight of the association strength distribution nodes. The mean C_k value for functionally impaired time periods in patients with multiple system diseases is generally lower than that for patients with a single disease, with a larger reduction. The overall weight of members in the functionally impaired time periods in the tiered weight set is lower than that of similar time periods in patients with a single disease. This difference reflects the impact of the systematically low reliability of the cardiopulmonary status identifiers in multi-disease comorbidity scenarios on weight allocation. In cardiopulmonary status identification, the mean C_k of the coupled differentiation state period usually falls between functional compensation and functional impairment, with the enhancement coefficient in the middle. However, the distribution nodes of the association strength of the coupled differentiation period have already received additional enhancement in the cross-state association analysis stage. After the weight enhancement of the stable state, the final weight is usually still higher than that of the undifferentiated period with the same C_k mean. The higher the weight of a member in the hierarchical weight set, the greater the proportion of contribution of that member to the comprehensive outlier in the weighted integration.
[0058] A comprehensive anomaly quantity is generated by weighting and integrating the anomaly risk set according to the hierarchical weight set. The weights of each member in the hierarchical weight set are used as coefficients to weight and sum the anomaly risk scores of the corresponding members in the anomaly risk set. The summation result is divided by the sum of the weights of all members in the hierarchical weight set to obtain the normalized comprehensive anomaly quantity. This normalization operation ensures that the comprehensive anomaly quantity is not affected by the number of members in the anomaly risk set, guaranteeing comparability across different monitoring time periods. The anomaly risk scores of high-weight members contribute a larger proportion to the comprehensive anomaly quantity, while the high scores of low-weight members have a suppressed effect on increasing the comprehensive anomaly quantity, making the comprehensive anomaly quantity more focused on reflecting persistent anomalies occurring during periods of reliable status and high correlation strength. The anomaly risk scores of members in the severely low recovery interval have already been additionally adjusted in stage S140. When entering the weighted integration, the high scores and high weights in the hierarchical weight set have a double superposition, making their contribution to the comprehensive anomaly quantity particularly prominent. Taking a patient with a triple comorbidity as an example, the top three contributors to the abnormal risk set belong to the abnormality types driven by heart failure, chronic obstructive pulmonary disease, and chronic kidney disease, respectively. The overall abnormality is driven by the high-scoring members of the three comorbidity types. In contrast, for patients with a single disease, the top three contributors to the abnormal risk set are usually concentrated in the same abnormality type, and the driving source of the overall abnormality is singular. The difference in the distribution of contributing members between the two types of patients is fully preserved in the overall abnormality calculation record. The final generated overall abnormality is passed into the subsequent normalization process in the form of a single numerical value.
[0059] A comprehensive cardiopulmonary monitoring index (CPR) is generated using normalized processing of comprehensive abnormalities. The core of the normalization process is to map individual comprehensive abnormalities to a population reference distribution using percentiles, mapping the comprehensive abnormality to a standardized range of 0 to 1. A higher percentile for an individual's comprehensive abnormality within the population distribution corresponds to a higher CPR. A CPR above the 90th percentile (above 0.9) indicates that the individual's persistent abnormality is at the severe end among patients with similar comorbidities. The population reference distribution is stratified by comorbidity type: the reference distribution for heart failure with chronic obstructive pulmonary disease (COPD) is independent of the reference distribution for heart failure with chronic kidney disease (CHD). Stratification avoids mapping bias caused by differences in baseline levels of comprehensive abnormalities between different comorbidity combinations. Reference distributions for patients with a single disease and those with multiple comorbidities are also maintained independently. When no corresponding stratification exists for a new comorbidity combination, the reference distribution of the nearest neighbor comorbidity combination is used. A CPR exceeding 0.7 indicates a relatively severe overall abnormality in the current monitoring phase, while a value below 0.3 indicates a mild overall abnormality. When the comprehensive cardiopulmonary monitoring index continues to rise for three consecutive monitoring cycles, an overall trend deterioration label is triggered; when it continues to decline, an overall trend improvement label is triggered. Both types of trend labels are transmitted to the pattern recognition analysis stage in sync with the comprehensive index, serving as a global background reference for determining abnormal cardiopulmonary patterns.
[0060] Pattern recognition analysis is performed on the cardiopulmonary monitoring comprehensive index to generate cardiopulmonary abnormality patterns. Pattern recognition segments the cardiopulmonary monitoring comprehensive index sequence into time windows, extracting statistical features window by window. These features include the mean, standard deviation, number of monotonic segments, and slope distribution of the comprehensive index within each window. These four features together constitute the pattern feature vector for that window. Adjacent windows are incremented by half a window length to ensure the temporal continuity of the feature sequence. The pattern feature vector is then matched window by window against a pre-defined abnormality pattern category library, which covers four typical cardiopulmonary abnormality patterns: persistently high (corresponding to a long-term high cardiopulmonary monitoring comprehensive index with a small standard deviation), fluctuating and oscillating (corresponding to repeated rises and falls in the comprehensive index with a large number of monotonic segments), stepwise deterioration (corresponding to a phased jump in the comprehensive index followed by stabilization at a new high), and slowly climbing (corresponding to a long-term sustained rise in the comprehensive index with a low slope). The coexistence of slow-rising and step-like deterioration patterns is relatively common in patients with multi-system diseases. Taking a patient with both heart failure and chronic kidney disease as an example, the decline in chronic kidney function leads to a slow and continuous increase in the cardiopulmonary monitoring composite index. This is compounded by a step-like jump in the composite index during acute fluid overload. These two patterns coexist within the same monitoring timeframe, triggering dual-pattern matching in the pattern category library based on the corresponding feature vectors. The matching results for each time window are composed of pattern categories. Windows with a similarity below 0.6 are considered to have an ambiguous pattern. During the compensation peak identification stage, ambiguous windows in the cardiopulmonary abnormality pattern are processed by substituting the pattern category of the nearest-neighbor matched window.
[0061] In some embodiments, the step of identifying the compensatory peak of the cardiopulmonary abnormality pattern and generating a compensatory state identifier includes: tracking the compensatory intensity of the cardiopulmonary abnormality pattern to generate a compensatory intensity sequence; identifying extreme inflection points in the compensatory intensity sequence to generate a peak candidate set; performing peak stability verification based on the peak candidate set to generate an effective peak point; and calibrating the compensatory critical state through the effective peak point to generate a compensatory state identifier.
[0062] Compensation intensity is tracked to generate a compensation intensity sequence for abnormal cardiopulmonary patterns. The tracking process combines the matching pattern category of each time window of the abnormal cardiopulmonary pattern with the time sequence of the recovery capacity coefficient for the corresponding time period. The current level of compensatory mobilization is estimated window by window. The level of compensatory mobilization reflects the body's ability to spontaneously compensate for functional impairment and maintain physiological homeostasis within the current time period; the more active the compensatory mobilization, the more active the body's autonomous recovery mechanism. The method for estimating compensation intensity is to calculate the recovery capacity coefficient relative to the local minimum value within the window. The larger the recovery magnitude, the more active the body's compensatory mobilization within that window. Taking a patient with acute exacerbation of heart failure as an example, during the period of persistent high-level cardiopulmonary abnormality, the recovery capacity coefficient remained in the low range of 0.2 to 0.3 for a long time, with an increase of less than 0.1 in each window. The corresponding compensation intensity sequence showed a persistently flat low-value pattern, indicating that the body's compensatory mobilization capacity was almost exhausted during this stage. After diuretic intervention, the same patient entered the period of fluctuating cardiopulmonary abnormality, with the recovery capacity coefficient repeatedly rising. The compensation intensity sequence locally increased to 0.4 to 0.6 with each rise, showing high-frequency oscillation characteristics, indicating that the compensatory capacity began to recover but was still unstable. The compensation intensity sequence subsequences corresponding to different cardiopulmonary abnormality patterns need to be processed separately in the peak identification stage. After classification, the peak identification threshold of each subsequence is determined independently based on the historical distribution of the corresponding cardiopulmonary abnormality pattern category. The 75th percentile of the historical distribution of the compensation intensity sequence of the pattern category is taken as the judgment threshold to ensure that peak candidates are screened as relatively high values in each pattern category.
[0063] A peak candidate set is generated to identify extreme inflection points in the compensation intensity sequence. Extreme inflection points correspond to the locations of local maxima in the compensation intensity sequence. The determination method involves comparing the current window's compensation intensity sequence value with the values of the two windows before and after it; a value higher than either of the two windows is considered an extreme inflection point. Using two windows before and after, rather than a single window, as the comparison range filters out false extremes caused by single-window noise, requiring the local advantage of the compensation intensity sequence to remain stable over a certain time span. Extreme inflection points for different mode category subsequences are extracted independently. The determination threshold for each subsequence is the 75th percentile of the historical distribution of the corresponding category's compensation intensity sequence. The 75th percentile for fluctuating and oscillating subsequences is typically higher than that for persistently high-level subsequences, resulting in a correspondingly higher determination threshold. Only inflection points with truly prominent compensation mobilization levels trigger candidate labeling, filtering out low-amplitude false inflection points caused by frequent fluctuations in the oscillating mode. After the extreme inflection points of each pattern category are extracted, they are aggregated into the same peak candidate set. Each candidate carries two attributes: the extreme moment and the compensation intensity value. When the pattern category switches before and after the same moment, the extreme inflection point is assigned to the pattern category period with the higher compensation intensity value. The number of peak candidate set members in fluctuating and oscillating periods is usually much larger than that in sustained high-level periods. In a patient, the number of peak candidate set members on a single day during a period of frequent arrhythmia can be several times that of the stable period.
[0064] Valid peak points are generated through peak stability verification based on a peak candidate set. The verification process uses the compensation intensity value of each candidate in the peak candidate set and the local standard deviation of the compensation intensity sequence within its respective time period. The local standard deviation is the standard deviation of the compensation intensity sequence within three windows before and after the candidate's time. A large local standard deviation indicates that the compensation intensity near the candidate fluctuates violently, and the corresponding extreme inflection point in the peak candidate set may be misjudged as a true compensation peak due to noise superposition. The verification rule requires that the compensation intensity value of each candidate in the peak candidate set exceeds the local mean plus 1.5 times the local standard deviation to pass the verification and be confirmed as a valid peak point; candidates that fail the verification are marked as unstable candidates. Taking a patient with arrhythmia as an example, during the fluctuating oscillation pattern, the compensatory intensity sequence fluctuated frequently, with the local standard deviation consistently high, ranging from 0.15 to 0.25. Consequently, the validation threshold shifted upward. A large number of members in the peak candidate set for the corresponding period were marked as unstable candidates because their compensatory intensity values failed to exceed the shifted validation threshold. Only the candidates corresponding to the actual concentrated outbreak of compensatory mobilization passed the validation and became valid peak points, effectively distinguishing between the oscillating noise caused by frequent arrhythmia episodes and the true compensatory mobilization peak. In contrast, during the step-deterioration pattern of the same patient, the local standard deviation was lower, and the compensatory intensity values of most members in the peak candidate set exceeded the validation threshold. The number of valid peak points was significantly greater than that during the oscillation pattern. The difference in the density of valid peak points between the two pattern periods was fully preserved in the validation results.
[0065] The compensation status identifier is generated by calibrating the critical state of compensation using effective peak points. The calibration criterion is the combined performance of the compensation intensity value of the effective peak point and the comprehensive cardiopulmonary monitoring index of the corresponding time period. When the compensation intensity value is high and the comprehensive cardiopulmonary monitoring index is also high, it indicates that the body maintains strong compensatory mobilization under severe abnormal burden, and is calibrated as a high-burden compensation critical state. When the compensation intensity value is high and the comprehensive cardiopulmonary monitoring index is at a medium to low level, it indicates that the compensatory mobilization is ahead of the actual needs of the abnormal burden, and is calibrated as an advanced compensation critical state. The two types of calibrations differ fundamentally in clinical significance. A high-burden compensation threshold indicates that compensatory resources are nearing depletion, while a pre-compensation threshold indicates that the body has a margin for active regulation. Taking an acute exacerbation of chronic heart failure as an example, the compensatory intensity at the effective peak point during the exacerbation's peak and the cardiopulmonary monitoring comprehensive index are both at a high level, calibrated as a high-burden compensation threshold. Conversely, the same patient's effective peak point compensatory intensity during the stable period is also high, but the cardiopulmonary monitoring comprehensive index is at a low to medium level, calibrated as a pre-compensation threshold. The sequential appearance of these two types of compensation status indicators fully records the evolutionary trajectory from sufficient compensation to compensation depletion. The compensation status indicator is composed of the calibration results of each effective peak point, and each compensation status indicator unit carries three attributes: indicator category, corresponding time, and compensatory intensity value. If adjacent valid peak points correspond to the same compensation status identifier category and the time interval is less than two windows, they are merged into the same compensation status identifier. After merging, the compensation intensity value is taken as the larger of the two values, so that temporally adjacent similar compensation critical events are identified as single continuous compensation peaks rather than independent events.
[0066] A trend-weighted assessment is conducted based on the compensation status indicators to generate a cardiopulmonary function monitoring report. The trend-weighted assessment revolves around the historical sequence of the compensation status indicators, analyzing the evolution characteristics of each compensation status indicator category within the monitoring period from three dimensions: frequency of occurrence, changes in time intervals, and trends in compensation intensity. A continuously increasing frequency of compensation status indicators corresponding to the high-burden compensation threshold indicates an accelerated rate of compensation resource depletion; a continuously decreasing frequency of compensation status indicators corresponding to the advanced compensation threshold indicates a gradual decline in the ability of proactive compensation regulation; and the simultaneous occurrence of both trends reflects an accelerated evolution of cardiopulmonary function towards decompensation. The trend slopes of the three dimensions are weighted and combined to form a comprehensive trend score. The weights for the frequency of occurrence, time interval changes, and compensation intensity value trends are 0.4, 0.3, and 0.3, respectively. Positive values indicate an overall trend of deterioration, while negative values indicate an improvement. Patients with acute infections experience a sharp rise in the comprehensive trend score to a high positive value during the peak of the infection, while patients with chronic progression show a long-term, slow, positive increase. The trend slope patterns of these two types of patients are clearly distinguishable in the cardiopulmonary function monitoring report. The main content of the cardiopulmonary function monitoring report includes the comprehensive trend score, the frequency distribution of each compensation status indicator category, a time-series summary of the comprehensive cardiopulmonary monitoring index, and statistical distribution of cardiopulmonary abnormality patterns. When the comprehensive trend score exceeds a preset positive threshold, an additional functional deterioration trend is indicated; a sustained negative trend confirms an additional functional improvement trend. The change in the comprehensive trend score between cardiopulmonary function monitoring reports of different periods is included in the report summary.
[0067] To implement the above-described method embodiments, a multi-parameter comprehensive monitoring method for cardiopulmonary function is proposed to achieve the corresponding functional and technical effects. See also... Figure 4 , Figure 4 This paper illustrates a structural block diagram of a cardiopulmonary function multi-parameter integrated monitoring system 400 provided in an embodiment of this application, comprising: The signal acquisition module 401 is used to acquire the patient's physiological signal set, which includes heart rate signal, respiratory signal and blood oxygen signal, and to perform multi-parameter coupling analysis on the physiological signal set to generate a cardiopulmonary function reference baseline. Feature extraction module 402 is used to perform fluctuation localization on the cardiopulmonary function reference baseline to extract cardiopulmonary abnormal features, perform cross-signal cross-validation on the cardiopulmonary abnormal features to generate a coupled validation set, and perform frequency band parsing to extract the cardiopulmonary coupled frequency band according to the coupled validation set; The state partitioning module 403 is used to calibrate the cardiopulmonary state identifier according to the cardiopulmonary coupling frequency band, and perform pattern clustering partitioning on the physiological signal set based on the cardiopulmonary state identifier to generate a cardiopulmonary state partitioning map. Risk assessment module 404 is used to perform offset quantification to generate offset amplitude based on the cardiopulmonary status zoning map and the cardiopulmonary function reference baseline, perform self-recovery ability assessment on the offset amplitude to generate recovery ability coefficient, and identify continuous abnormal time points to generate an abnormal risk set according to the recovery ability coefficient; The report output module 405 is used to generate a comprehensive cardiopulmonary monitoring index by performing joint hierarchical matching with the abnormal risk set and the cardiopulmonary status identifier, perform pattern recognition analysis on the comprehensive cardiopulmonary monitoring index to generate a cardiopulmonary abnormality pattern, identify the compensation peak of the cardiopulmonary abnormality pattern to generate a compensation status identifier, and perform trend weighted evaluation based on the compensation status identifier to generate a cardiopulmonary function monitoring report.
[0068] The aforementioned cardiopulmonary function multi-parameter integrated monitoring system 400 can implement one of the cardiopulmonary function multi-parameter integrated monitoring methods described in the above-described method embodiments. The options in the above method embodiments are also applicable to this embodiment and will not be detailed here. The remaining content of this application embodiment can be referred to the content of the above method embodiments, and will not be repeated in this embodiment.
[0069] The purpose of the above embodiments is to reproduce and derive the technical solution of the present invention by way of example, and to fully describe the technical solution, purpose and effect of the present invention. The purpose is to enable the public to have a more thorough and comprehensive understanding of the disclosure of the present invention, and not to limit the scope of protection of the present invention.
Claims
1. A method for comprehensive monitoring of heart-lung function parameters, characterized in that, include: Collect the patient's physiological signal set, which includes heart rate signal, respiratory signal and blood oxygen signal, and perform multi-parameter coupling analysis on the physiological signal set to generate a cardiopulmonary function reference baseline; The process involves: extracting abnormal cardiopulmonary features by performing fluctuation localization on the cardiopulmonary function reference baseline; performing cross-signal cross-validation on the abnormal cardiopulmonary features to generate a coupled validation set; decomposing the abnormal cardiopulmonary features according to signal type to generate single-parameter abnormal sequences; performing phase lag detection on the single-parameter abnormal sequences to generate a lag time series distribution; performing coupling strength quantization on the lag time series distribution to generate a coupling strength distribution; filtering phase lag co-occurrence anomalies according to the coupling strength distribution to generate a coupled validation set; and performing frequency band analysis on the coupled validation set to extract the cardiopulmonary coupled frequency band. Cardiopulmonary state identifiers are calibrated according to the cardiopulmonary coupling frequency bands, and a cardiopulmonary state partition map is generated by performing pattern clustering partitioning on the physiological signal set based on the cardiopulmonary state identifiers. Based on the cardiopulmonary status zoning map and the cardiopulmonary function reference baseline, offset quantification is performed to generate offset amplitude, self-recovery capability assessment is performed on the offset amplitude to generate recovery capability coefficient, and continuous abnormal time points are identified and an abnormal risk set is generated according to the recovery capability coefficient. The abnormal risk set and the cardiopulmonary status identifier are used to perform joint hierarchical matching to generate a comprehensive cardiopulmonary monitoring index. Pattern recognition analysis is performed on the comprehensive cardiopulmonary monitoring index to generate cardiopulmonary abnormality patterns. Compensation peaks are identified in the cardiopulmonary abnormality patterns to generate compensation status identifiers. Trend weighted evaluation is performed based on the compensation status identifiers to generate a cardiopulmonary function monitoring report.
2. The method of claim 1, wherein, The step of generating a cardiopulmonary state partition map by performing pattern clustering partitioning on the physiological signal set based on the cardiopulmonary state identifier includes: The cardiopulmonary status identifiers are processed to perform state sequence merging to generate a state category set; A mapping sequence is generated by performing a time-series mapping between the physiological signal set and the state category set; Within the mapping sequence, identify densely transitioned state segments and implement priority boundary marking to generate a set of state boundary points; A cardiopulmonary state partition map is constructed using the set of state boundary points.
3. The method of claim 1, wherein, The step of performing an autonomous recovery capability assessment on the offset amplitude to generate a recovery capability coefficient includes: The offset amplitude is time-tracked to generate an offset time-series curve; Identify the recovery start point of the offset timing curve and generate a recovery start marker; The abnormal recovery rate is calculated based on the recovery start marker to generate a recovery rate value; The recovery capability coefficient is generated by comprehensively quantifying the recovery rate value and the offset amplitude.
4. The method according to claim 1, characterized in that, The step of generating an abnormal risk set by identifying persistent abnormal time points according to the recovery capability coefficient includes: The low-value intervals of the recovery coefficient are located to generate a set of low-recovery intervals; A pseudo-recovery marker set is generated based on the fluctuation characteristics of the low recovery interval set, which are identified by a brief rebound followed by a decline. The duration distribution of the pseudo-recovery marker set was statistically analyzed. The duration distribution and the pseudo-recovery marker set are integrated to generate an anomaly risk set.
5. The method according to claim 1, characterized in that, The step of generating a comprehensive cardiopulmonary monitoring index by jointly classifying and matching the abnormal risk set with the cardiopulmonary status identifier includes: Cross-state correlation analysis is performed on the abnormal risk set to generate a correlation strength distribution; The correlation strength distribution and the cardiopulmonary status identifier are combined with steady-state weight enhancement to generate a hierarchical weight set; The abnormal risk set is weighted and integrated according to the hierarchical weight set to generate a comprehensive abnormality quantity; The comprehensive abnormality was normalized to generate a comprehensive cardiopulmonary monitoring index.
6. The method according to claim 1, characterized in that, The step of identifying compensatory peaks and generating compensatory status identifiers for the cardiopulmonary abnormality pattern includes: The compensation intensity sequence is generated by tracking the compensation intensity of the aforementioned cardiopulmonary abnormality pattern. For the compensation intensity sequence, identify extreme inflection points and generate a peak candidate set; Based on the peak candidate set, peak stability verification is performed to generate valid peak points; The compensation status identifier is generated by calibrating the critical state of compensation using the effective peak points.
7. The method according to claim 2, characterized in that, The step of identifying densely populated state transition sections within the mapping sequence and generating a set of state boundary points by prioritizing boundary marking includes: Extract state mutation locations within the mapping sequence to generate a mutation point set; The mutation point set is subjected to inverse mutation magnitude weight mapping to generate a weak mutation weight set; Within the weak mutation weight set, high-weight segments are identified, and boundary expansion is performed to generate an expanded boundary set. The state boundary point set is determined based on the extended boundary set and the mutation point set.
8. The method according to claim 5, characterized in that, The step of performing cross-state association analysis on the abnormal risk set to generate an association strength distribution includes: The abnormal risk set is labeled with its state attribution to generate a state attribution set; For the state attribution set, extract the causal directions of heart rate abnormalities and respiratory abnormalities to generate an asymmetric causal matrix; Based on the aforementioned asymmetric causal matrix, directional weight allocation is performed to generate a weight allocation table; The weight allocation table is used to quantify the degree of cross-state association and generate an association strength distribution.
9. A comprehensive monitoring system for multiple parameters of cardiopulmonary function, characterized in that, include: The signal acquisition module is used to acquire the patient's physiological signal set, which includes heart rate signal, respiratory signal and blood oxygen signal, and to perform multi-parameter coupling analysis on the physiological signal set to generate a cardiopulmonary function reference baseline; The feature extraction module is used to extract cardiopulmonary abnormal features by performing fluctuation localization on the cardiopulmonary function reference baseline, and to perform cross-signal cross-validation on the cardiopulmonary abnormal features to generate a coupling validation set. This includes: decomposing the cardiopulmonary abnormal features according to signal type to generate a single-parameter abnormal sequence; performing phase lag detection on the single-parameter abnormal sequence to generate a lag time series distribution; performing coupling strength quantization on the lag time series distribution to generate a coupling strength distribution; filtering phase lag co-occurrence anomalies according to the coupling strength distribution to generate a coupling validation set; and performing frequency band analysis on the coupling validation set to extract the cardiopulmonary coupling frequency band. The state partitioning module is used to calibrate the cardiopulmonary state identifier according to the cardiopulmonary coupling frequency band, and perform pattern clustering partitioning on the physiological signal set based on the cardiopulmonary state identifier to generate a cardiopulmonary state partitioning map. The risk assessment module is used to generate an offset amplitude by performing offset quantification based on the cardiopulmonary status zoning map and the cardiopulmonary function reference baseline, perform self-recovery capability assessment on the offset amplitude to generate a recovery capability coefficient, and identify continuous abnormal time points and generate an abnormal risk set according to the recovery capability coefficient. The report output module is used to generate a comprehensive cardiopulmonary monitoring index by performing joint hierarchical matching with the abnormal risk set and the cardiopulmonary status identifier, perform pattern recognition analysis on the comprehensive cardiopulmonary monitoring index to generate cardiopulmonary abnormality patterns, identify the compensation peak of the cardiopulmonary abnormality patterns to generate compensation status identifiers, and perform trend weighted evaluation based on the compensation status identifiers to generate a cardiopulmonary function monitoring report.
Citation Information
Patent Citations
Cardiopulmonary sound preprocessing system based on dynamic filtering
CN121561263A
Personal authentication device and personal authentication program
JP2015042293A
