Sensor-based pulmonary ventilation blood flow distribution real-time monitoring method and evaluation system

Abstract

The invention relates to the technical field of respiratory function monitoring, and discloses a pulmonary ventilation blood flow distribution real-time monitoring method and evaluation system based on a sensor. The method comprises the following steps: acquiring multi-modal data such as a blood oxygen saturation waveform, an airflow rate waveform and a thoracic pressure waveform of a lung area; performing time synchronization processing on the data to generate a time alignment sequence; dividing the data into an inspiration phase, an expiration phase and a transition phase according to a preset rule; extracting time-frequency domain features from each phase segment, and constructing a multi-dimensional respiration feature matrix; identifying an abnormal breathing mode and marking an abnormal section through breathing phase switching frequency analysis; decomposing the normal segment feature matrix by using a lung pathological feature library to generate a pathological feature vector; a pathological feature vector is optimized through a ventilation and blood flow coupling model; and carrying out abnormal region positioning based on the lung pathology propagation network, and generating a lung partition function evaluation result. The method realizes non-invasive and real-time monitoring, and is suitable for scenes such as clinical monitoring and rehabilitation evaluation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of respiratory function monitoring, in particular to a sensor-based real-time monitoring method and evaluation system for lung ventilation and blood flow distribution. BACKGROUND

[0002] Lung ventilation and blood flow distribution monitoring plays an important role in respiratory function evaluation and disease diagnosis. The efficiency of pulmonary gas exchange directly depends on the matching degree of alveolar ventilation and blood perfusion, and the imbalance of ventilation and blood flow ratio is often an important feature of early respiratory diseases. Traditional monitoring methods rely on devices such as spirometers and blood gas analyzers, which have obvious shortcomings in time resolution, continuous monitoring capability, and spatial distribution information acquisition.

[0003] Although the commonly used blood oxygen saturation monitoring in clinical practice can reflect the oxygenation state, it cannot directly provide the distribution difference between ventilation and blood flow. Imaging techniques such as CT and MRI can provide structural and functional information, but the devices are large in size and high in cost, which are not suitable for continuous bedside monitoring. Nuclear medicine methods can evaluate local ventilation and blood flow ratio, but there are problems of radioactive exposure and low time resolution.

[0004] With the development of sensing technology, wearable and implantable sensors can acquire multi-modal physiological signals related to respiration, including blood oxygen saturation, airflow rate, and chest pressure. These signals contain rich respiratory dynamics information, which can be used to analyze the characteristic changes in different phases of the respiratory process. However, how to effectively extract features related to lung ventilation and blood flow distribution from multi-source heterogeneous signals and establish an accurate functional evaluation model remains a technical challenge to be solved.

[0005] Existing signal processing methods have limitations in respiratory phase division, feature extraction, and multi-modal information fusion, making it difficult to capture subtle functional changes during the respiratory process. In addition, the differences in respiratory patterns under different pathological conditions are complex, and single signal analysis can easily lead to misjudgment. Therefore, establishing a method for real-time, continuous, and non-invasive monitoring of lung ventilation and blood flow distribution has important value for early detection and precise intervention of respiratory diseases. SUMMARY

[0006] The purpose of the present application is to provide a sensor-based real-time monitoring method and evaluation system for lung ventilation and blood flow distribution to solve the problems raised in the background.

[0007] To achieve the above-mentioned purpose, the present application provides a sensor-based real-time monitoring method for lung ventilation and blood flow distribution, which comprises:

[0008] Collecting multi-modal sensor data of the lung region, the multi-modal sensor data including blood oxygen saturation waveform data, airflow rate waveform data, and chest pressure waveform data;

[0009] time-synchronize the multi-modal sensor data to generate time-aligned sensor data sequences;

[0010] divide the time-aligned sensor data sequences into inhalation phase data segments, exhalation phase data segments and transition phase data segments according to preset respiratory phase division rules;

[0011] extract time-frequency domain feature parameters from each respiratory phase data segment to construct multi-dimensional respiratory feature matrices;

[0012] identify abnormal respiratory patterns and mark abnormal respiratory phase data segments based on respiratory phase switching frequency analysis results;

[0013] perform feature decomposition on the multi-dimensional respiratory feature matrices in normal respiratory phase data segments to generate pathological feature vectors according to a lung pathological feature library;

[0014] perform parameter optimization on the pathological feature vectors based on a ventilation-blood flow coupling model to obtain optimized pathological feature vectors;

[0015] perform abnormal region positioning on the optimized pathological feature vectors according to a lung pathological propagation network to generate lung sub-region function evaluation results.

[0016] Preferably, the time-synchronization of the multi-modal sensor data comprises:

[0017] align time stamps of the blood oxygen saturation waveform data, the airflow rate waveform data and the thoracic pressure waveform data using a clock synchronization protocol;

[0018] perform sliding window filtering on the multi-modal sensor data after time stamp alignment to remove high-frequency noise interference;

[0019] correct time delay differences between the multi-modal sensor data based on a cross-correlation analysis method;

[0020] resample the corrected multi-modal sensor data at a uniform sampling rate to generate time-aligned sensor data sequences.

[0021] Preferably, the extraction of time-frequency domain feature parameters from each respiratory phase data segment comprises:

[0022] extract rising slope and fluctuation amplitude features from the blood oxygen saturation waveform data in the inhalation phase data segment;

[0023] extract falling slope and oscillation frequency features from the airflow rate waveform data in the exhalation phase data segment;

[0024] extract pressure gradient and fluctuation period features from the thoracic pressure waveform data in the transition phase data segment;

[0025] The extracted time-frequency domain feature parameters are arranged in the order of respiratory phases to construct a multi-dimensional respiratory feature matrix.

[0026] Preferably, the identifying abnormal respiratory patterns and marking abnormal respiratory phase data segments based on the switching frequency analysis results of the respiratory phases comprises:

[0027] The number of phase switching in a continuous respiratory cycle is calculated to obtain a real-time respiratory phase switching frequency;

[0028] The real-time respiratory phase switching frequency is compared with a normal respiratory frequency threshold;

[0029] When the real-time respiratory phase switching frequency exceeds a frequency abnormality threshold, a dynamic time warping algorithm is started;

[0030] The morphological similarity index between adjacent respiratory phase data segments is calculated by the dynamic time warping algorithm;

[0031] According to the comparison result of the morphological similarity index and a similarity threshold, an abnormal respiratory phase data segment is marked.

[0032] Preferably, the feature decomposition of the multi-dimensional respiratory feature matrix in the normal respiratory phase data segment comprises:

[0033] A historical pathological feature template is retrieved from a lung pathological feature library;

[0034] A matching degree index of the current multi-dimensional respiratory feature matrix and the historical pathological feature template is calculated;

[0035] When the matching degree index is greater than a matching threshold, a narrowband decomposition algorithm is used to extract a main pathological feature component;

[0036] When the matching degree index is less than the matching threshold, a wideband decomposition algorithm is used to extract a harmonic feature component;

[0037] The extracted main pathological feature component and harmonic feature component are combined to generate a pathological feature vector.

[0038] Preferably, the parameter optimization of the pathological feature vector based on the ventilation-blood flow coupling model comprises:

[0039] An amplitude-frequency coupling relationship model of blood oxygen saturation and blood flow rate is established;

[0040] A phase coupling relationship model of airflow rate and thoracic pressure is established;

[0041] The blood oxygen parameters in the pathological feature vector are corrected according to the amplitude-frequency coupling relationship model;

[0042] The airflow parameters in the pathological feature vector are compensated according to the phase coupling relationship model.

[0043] The corrected blood oxygen parameters and the compensated airflow parameters are combined to generate an optimized pathological feature vector.

[0044] Preferably, the abnormal region positioning on the optimized pathological feature vector comprises:

[0045] A pre-constructed lung pathological propagation network atlas is loaded;

[0046] An abnormal feature parameter is extracted from the optimized pathological feature vector;

[0047] The abnormal feature parameter is mapped to a corresponding node of the lung pathological propagation network atlas;

[0048] The propagation source of the abnormal feature parameter is traced in reverse along a pathological propagation path;

[0049] According to the abnormal degree of the nodes on the propagation path, a lung sub-zone function evaluation result is determined.

[0050] Preferably, the determination of the lung sub-zone function evaluation result comprises:

[0051] The number of abnormal feature parameters of each lung sub-zone node is counted;

[0052] The weighted sum of the abnormal feature parameters of each sub-zone node is calculated;

[0053] According to the weighted sum result, a function level of each sub-zone is divided;

[0054] An evaluation report containing the sub-zone function level information is generated.

[0055] Preferably, the method further comprises:

[0056] The lung pathological feature library is dynamically updated in real time during the monitoring process;

[0057] The ventilation-blood flow coupling model is optimized according to newly collected sensor data;

[0058] Based on the latest pathological feature library and the optimized coupling model, the monitoring accuracy is iteratively improved.

[0059] Preferably, the present application further comprises a sensor-based lung ventilation-blood flow distribution real-time monitoring system, which comprises a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor, when executing the computer program, implements the steps of the above-mentioned sensor-based lung ventilation-blood flow distribution real-time monitoring method.

[0060] Compared with the prior art, the present application has the following advantages:

[0061] Through multi-modal signal acquisition and fusion analysis, the dynamic changes in the respiratory process can be comprehensively captured. By synchronously processing signals such as blood oxygen saturation, airflow rate, and chest pressure, the consistency of data from different sources in the time dimension is ensured, providing a reliable basis for subsequent analysis.

[0062] The method introduces a respiratory phase division strategy, which subdivides the respiratory process into inspiration phase, expiration phase, and transition phase, and can more accurately reflect the physiological characteristics of different respiratory stages. The time-frequency domain feature parameters extracted in each phase form a multi-dimensional respiratory feature matrix, which can comprehensively characterize the respiratory kinetics.

[0063] Through respiratory phase switching frequency analysis, abnormal respiratory patterns can be identified and the corresponding data segments can be marked, which helps to find potential problems such as respiratory rhythm disorders. Combined with the lung pathological feature library, the abstract signal features can be converted into feature vectors with pathological significance.

[0064] The introduction of the ventilation-blood flow coupling model enables the optimization of pathological feature vectors in the physiological mechanism framework, improving the accuracy of feature expression. The application of lung pathological propagation network realizes the mapping from feature vector to abnormal region positioning, which can provide more intuitive functional evaluation results for clinical practice.

[0065] The entire method flow realizes a complete link from signal acquisition to functional evaluation, which can be continuously operated in a bedside environment and is suitable for various scenarios such as intensive care and rehabilitation monitoring. Its non-invasive nature improves patient comfort and compliance, avoiding the limitations of traditional examination methods.

[0066] Through multi-modal fusion and intelligent analysis, the method can reveal local lung function changes and help early detection of functional abnormalities, providing a reference for the development of personalized treatment plans. At the same time, the method has good scalability and can be combined with other physiological signal monitoring techniques to further enrich the evaluation dimensions. BRIEF DESCRIPTION OF DRAWINGS

[0067] Figure 1 The working principle diagram of the sensor-based lung ventilation-blood flow distribution real-time monitoring method described in the present application;

[0068] Figure 2 The flowchart for time-frequency domain feature parameter extraction;

[0069] Figure 3 The flowchart for multi-dimensional respiratory feature matrix feature decomposition. DETAILED DESCRIPTION

[0070] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0071] Please refer to Figure 1 The present application provides a sensor-based real-time monitoring method for pulmonary ventilation blood flow distribution, which comprises: collecting blood oxygen saturation waveform data, airflow rate waveform data and thoracic pressure waveform data of the lung region by a multi-modal sensor array deployed on the chest of a patient, and transmitting these data to a processing unit in real time at a high sampling rate. During the collection process, the sensor array covers the key areas of the lungs to ensure the comprehensiveness of the data. The multi-modal sensor data is time-synchronized, the time stamps of each waveform data are aligned using a clock synchronization protocol, high-frequency noise is removed through sliding window filtering, time delay differences are corrected based on cross-correlation analysis, and finally resampled at a unified sampling rate to generate a time-aligned sensor data sequence. According to a preset respiratory phase division rule, the time-aligned sequence is divided into inspiratory phase data segment, expiratory phase data segment and transition phase data segment; the division rule is based on the change trend of airflow rate and thoracic pressure, for example, the inspiratory phase is defined as the period when the airflow rate rises and the thoracic pressure increases negatively. Then, time-frequency domain feature parameters are extracted from each respiratory phase data segment, including the rising slope and fluctuation amplitude of the inspiratory phase blood oxygen saturation waveform, the falling slope and oscillation frequency of the expiratory phase airflow rate waveform, and the pressure gradient and fluctuation period of the transition phase thoracic pressure waveform, and these parameters are arranged in the order of respiratory phase to construct a multi-dimensional respiratory feature matrix. On this basis, abnormal respiratory patterns are identified based on the analysis results of respiratory phase switching frequency, the number of phase switching in consecutive respiratory cycles is calculated to obtain the real-time frequency, which is compared with the normal threshold value, if it is over the limit, the dynamic time warping algorithm is started to calculate the shape similarity, so as to mark the abnormal respiratory phase data segment.

[0072] For normal respiratory phase data segments, according to the lung pathological feature library, the feature is decomposed, the historical pathological feature template is called to calculate the matching degree, the narrowband or broadband decomposition algorithm is used to extract the main pathological feature component and the harmonic feature component, and the pathological feature vector is generated. Subsequently, based on the ventilation-blood flow coupling model, the pathological feature vector is optimized, the amplitude-frequency coupling relationship model of blood oxygen saturation and blood flow rate and the phase coupling relationship model of airflow rate and thoracic pressure are established, and the blood oxygen parameters and airflow parameters are corrected and compensated, and the optimized pathological feature vector is generated. Finally, according to the lung pathological propagation network, the abnormal area positioning is performed on the optimized vector, the pre-constructed network atlas is loaded, the abnormal feature parameters are extracted and mapped to the nodes, the source is traced back along the propagation path, the lung subregion function evaluation result is determined according to the node abnormality degree, including the number of abnormal parameters in each subregion, the weighted sum is calculated, the function level is divided, and the evaluation report is generated.

[0073] Embodiment 1: see Figure 2 The time synchronization processing starts from the time base alignment of the multi-channel data stream collected from different physical sensors. The blood oxygen saturation waveform data is derived from the pulse blood oxygen sensor, the airflow rate waveform data is derived from the respiratory flow rate sensor, and the thoracic pressure waveform data is obtained by the implantable or body surface pressure sensor. Since each sensor has an independent sampling clock and signal transmission path, there is subtle to millisecond level time jitter and fixed delay between the original data. A hardware trigger-based clock synchronization protocol is used as a primary alignment method. In the system initialization stage, a global timestamp synchronization signal is sent to all sensors, so that the time reference system of each data stream is unified to the system clock of the main processor. This protocol can use PTP precise time protocol variants or custom hardware synchronization pulse mechanism. The multi-modal data after time stamp preliminary alignment needs to be filtered by a sliding window to suppress high-frequency noise. The window width is dynamically configured according to the signal characteristics. For example, a narrower Hanning window is used for smoothing the blood oxygen saturation signal with rich frequency components, and a wider rectangular window is used for the relatively slow changing thoracic pressure signal. The filtering algorithm effectively attenuates sensor thermal noise and electromyographic interference while preserving the main trend of the signal.

[0074] After the initial filtering is completed, the time delay fine correction stage is entered, which calculates the relative delay amount between different signal channels based on the cross-correlation analysis method; Specifically, taking the airflow rate signal as the timing reference, the cross-correlation functions of the airflow rate signal and the blood oxygen saturation signal and the thoracic pressure signal are calculated respectively, and the inherent transmission delay between the signals is determined by finding the time shift corresponding to the peak value of the cross-correlation function. For the detected time delay difference, a polynomial interpolation algorithm is used to correct the signal channel that is ahead of or lagging behind, such as fractional delay filtering or spline interpolation reconstruction for the delayed channel, so that the data of each channel is strictly synchronized at the sample level. The corrected multi-modal data is finally converted into a timing sequence with a unified sampling rate by the resampling module, and the resampling process uses an anti-aliasing filter for preprocessing before performing linear interpolation or sine interpolation, ensuring that the output sequence is uniformly distributed in the time domain and retains the frequency domain characteristics of the original signal; The time-aligned sensor data sequence generated will be stored in the memory in the form of a ring buffer, providing high consistency input for subsequent respiratory phase division.

[0075] After the time synchronization data sequence enters the respiratory phase division module, the feature extraction process is started; This process designs differentiated feature extraction strategies for the inspiratory phase, expiratory phase and transition phase data segments obtained by division. The inspiratory phase data segment mainly focuses on the dynamic response characteristics of the blood oxygen saturation waveform, and the feature extraction algorithm first identifies the rising edge interval of the waveform, and fits the rising slope parameter by the least squares method, which reflects the rate of oxygen exchange in the alveoli; At the same time, the fluctuation amplitude of the waveform in this interval is calculated, that is, the average difference between adjacent peak and valley values, which is used to represent the stability of the blood oxygen concentration. The analysis of the expiratory phase data segment focuses on the airflow rate waveform, and when extracting its features, the data interval from the expiratory starting point to the end point is located first, and the linear regression is used to calculate the flow rate decline slope to evaluate the degree of airway patency; Then the expiratory waveform is subjected to short-time Fourier transform, and the dominant oscillation frequency component is extracted, which is closely related to the bronchial dynamics state. The transition phase data segment focuses on the morphological features of the thoracic pressure waveform, and the gradient value of the pressure change curve is calculated to reflect the sharpness of the pleural cavity pressure conversion; At the same time, the periodicity characteristics of the fluctuation are analyzed by using the autocorrelation function, and whether there is an abnormal respiratory rhythm is identified.

[0076] The extracted time-frequency domain feature parameters are organized in a structured feature matrix in the order of respiratory phases. The row dimension of the matrix corresponds to the serially collected respiratory cycle numbers, and the column dimension sequentially arranges the feature values of the inspiratory phase slope, the inspiratory phase amplitude, the expiratory phase slope, the expiratory phase frequency, the transition phase gradient, and the transition phase cycle, etc. An abnormal value detection mechanism is introduced in the matrix construction process to mark and interpolate replace the feature values that obviously exceed the physiological range. The finally generated multi-dimensional respiratory feature matrix will serve as the input data source for subsequent abnormal pattern recognition and pathological feature decomposition. The dimension of the matrix can be dynamically expanded with the monitoring time length, while the time sequence correlation of the original waveform data is preserved. The entire feature extraction process is implemented in a pipeline architecture, allowing real-time feature calculation while continuous data collection is in progress, effectively supporting the low-latency monitoring requirements of the system.

[0077] In the cooperative work of time sequence synchronization and feature extraction, the system maintains a configurable parameter library for storing key threshold values of the algorithm, such as the sliding window length, the cross-correlation search range, the feature effective value interval, etc. These parameters can be dynamically adjusted through a configuration interface according to different application scenarios (such as intensive care, exercise pulmonary function test). All signal processing algorithms use fixed-point operation optimization to ensure real-time execution efficiency on embedded platforms. The data stream carries a quality identification bit when passing between different processing modules, facilitating subsequent modules to perform fault-tolerant processing on low-quality data segments. The technical path described in this embodiment constructs a high-quality data foundation for ventilation and blood flow distribution analysis through fine time sequence alignment and multi-dimensional feature extraction. Its modular design allows customization of feature extraction strategies for specific pathological states.

[0078] Embodiment 2: refer to Figure 3 The abnormal respiratory pattern recognition process starts with the statistical calculation of the number of phase switching times within a continuous respiratory cycle. A respiratory cycle is defined as the complete time interval from the start of one inspiration phase to the start of the next inspiration phase. The detection of phase switching events is achieved by analyzing the cross characteristics of the airflow rate waveform and the chest pressure waveform. The system slides a detection window on the time-aligned sensor data sequence. The window width can be configured to include several respiratory cycles. Within each window, the waveform zero-crossing points, peak points, or threshold crossing points are identified as the marker points of phase switching. The real-time respiratory phase switching frequency is obtained by counting the number of effective switching times per unit time. This frequency value is filtered by moving average to eliminate the interference caused by accidental fluctuations. The calculated real-time frequency is immediately compared with the preset normal respiratory frequency threshold. The normal threshold range is set based on the physiological parameter database of the target population, such as the typical frequency range of adult resting state. The system allows clinical doctors to adjust the upper and lower limits of the threshold according to the individual differences of patients.

[0079] When the real-time respiratory phase switching frequency exceeds the frequency abnormality threshold, the system automatically activates a dynamic time warping algorithm for in-depth analysis; this algorithm processes the shape comparison between adjacent respiratory phase data segments, first normalizing the amplitude of the data segments to eliminate the influence of individual differences, and then finding the optimal bending path between the two sequences through dynamic programming, which minimizes the cumulative distance metric. The shape similarity index is obtained by calculating the normalized distance under the warping path, and the lower the index value, the more similar the shapes of the two respiratory phases; the system compares the similarity index with an adaptive similarity threshold, which is established by analyzing the shape variation characteristics of historical normal respiratory data segments. The comparison result is directly used to mark abnormal respiratory phase data segments, and the marked abnormal segments are temporarily excluded from the subsequent pathological feature analysis process, while the system records the timestamp and feature parameters of the abnormality for generating a monitoring report. The entire abnormality recognition process uses a multi-level verification mechanism to avoid false positives caused by motion artifacts or transient interference, ensuring the clinical reliability of the marked results.

[0080] After completing the abnormal respiratory pattern recognition, the system performs feature decomposition on the normal respiratory phase data segments that are not marked; this process starts with retrieving historical pathological feature templates from the lung pathological feature library, which stores respiratory feature patterns under various typical lung pathological states in the form of a matrix, with each pathological state corresponding to a set of feature vector templates. The system calculates the matching degree index of the current multi-dimensional respiratory feature matrix with each historical pathological feature template, using a comprehensive evaluation method that combines weighted Euclidean distance and angle similarity, with different feature parameters being assigned different weights according to their clinical importance. The comparison result of the matching degree index with the preset matching threshold determines the specific algorithm path of feature decomposition, when the index is greater than the matching threshold, it indicates that the current respiratory feature is highly similar to a certain pathological state, at which point the narrowband decomposition algorithm is used to focus on analyzing the main pathological feature components. The narrowband decomposition algorithm first reduces the dimension of the feature matrix through principal component analysis, identifying the main components with the highest contribution rate, and then performs band-pass filtering on the main components to extract the feature frequency band related to the target pathology; the passband range of the filter is dynamically adjusted according to the spectral characteristics of the historical pathological feature templates, ensuring that the extracted main pathological feature components have the maximum discriminability. When the matching degree index is less than the matching threshold, it indicates that the current respiratory feature does not match any known pathological state, and the system switches to the wideband decomposition algorithm to explore potential new feature combinations; the wideband decomposition divides the feature space into multiple sub-bands through wavelet packet transformation of the feature matrix, and then calculates the energy entropy and mutual information index of each sub-band to extract harmonic feature components that may reflect early pathological changes.

[0081] The system keeps monitoring the data quality during the feature decomposition process, and initiates a reprocessing mechanism for data segments with too low signal-to-noise ratio or obvious artifacts. The main pathological feature components and harmonic feature components obtained by decomposition are sorted according to the contribution degree and combined to generate a pathological feature vector, and each element in the vector is accompanied by a reliability score index. The entire feature decomposition process adopts an incremental learning strategy, and the newly generated reliable decomposition results can be selectively updated to the lung pathological feature library after expert confirmation, so that the system has the ability to continuously optimize the recognition accuracy over time. After the pathological feature vector is generated, it enters the parameter optimization stage, providing refined input data for subsequent ventilation-blood flow coupling analysis. The abnormality recognition and feature decomposition module adopts a parallel pipeline architecture design, which can continuously process respiratory phase data segments without interrupting real-time data acquisition; the system allocates independent cache areas to each processing thread to avoid data competition and loss. All algorithm parameters are managed through configuration files, allowing quick adjustment of recognition sensitivity and decomposition granularity according to different monitoring scenarios (such as postoperative monitoring and chronic disease management). The module running state and intermediate results are recorded in real time through the system log, providing complete data traceability for clinical review and algorithm optimization. The technical solution described in this embodiment builds an intelligent analysis path that can distinguish between normal physiological variations and real pathological changes through multi-level abnormality detection and adaptive feature decomposition, providing key technical support for accurate assessment of lung ventilation-blood flow distribution abnormalities.

[0082] Example 3: The parameter optimization process starts from the construction stage of the ventilation-blood flow coupling model, which consists of two interrelated sub-models, namely the amplitude-frequency coupling model describing the relationship between blood oxygen saturation and blood flow rate, and the phase coupling model describing the relationship between airflow rate and thoracic pressure. The establishment of the amplitude-frequency coupling relationship model is based on the principle of blood oxygen dynamics. There is an amplitude modulation relationship between the low-frequency oscillation component in the blood oxygen saturation waveform and the periodic fluctuation of the blood flow rate. The model quantifies this correlation by establishing a transfer function of blood oxygen saturation change to microvascular blood perfusion state. The structure of this transfer function takes into account the nonlinear characteristics of the oxygenated hemoglobin dissociation curve and the compliance influence of the pulmonary capillary bed. The phase coupling relationship model is derived from respiratory mechanics, analyzing the dynamic changes of the phase difference between the airflow rate waveform and the thoracic pressure waveform. The model considers thoracic pressure as the main driving force for airflow, and airflow resistance and lung elastic characteristics jointly determine the degree of phase lag. Model parameters are initialized by analyzing the statistical characteristics of a large number of data samples in normal breathing state.

[0083] After the model is built, the system starts to optimize the pathological feature vector. First, the amplitude-frequency coupling model is used to process the blood oxygen parameters. The pathological feature vector contains parameters such as mean saturation, fluctuation amplitude, and rising slope extracted from the blood oxygen saturation waveform. These parameters may deviate due to sensor errors or physiological noise in pathological conditions. The correction process uses the amplitude-frequency coupling model to generate a theoretical estimation sequence of blood oxygen saturation, which is calculated based on the current blood flow rate parameter. Then, the actual blood oxygen parameters are compared with the theoretical estimation value, and the adaptive correction is performed for the parameters with significant differences. The correction algorithm uses an iterative approximation method to adjust the value of the blood oxygen parameter to minimize the residual error between the actual value and the theoretical value. The residual error calculation considers both amplitude difference and waveform shape similarity. The compensation of airflow parameters is based on the phase coupling relationship model. The peak flow rate and expiratory decline slope in the pathological feature vector are input into the phase model together with the chest pressure parameter to calculate the theoretical airflow parameter value. The difference between the actual measurement value and the theoretical value is attributed to changes in airway resistance or respiratory muscle function. The compensation algorithm adjusts the phase properties of the airflow parameter to reduce this difference.

[0084] A core relationship is involved in the parameter optimization process to quantify the coupling effect:

[0085]

[0086] Where: represents the coupling strength coefficient, which evaluates the interaction degree between blood oxygen parameters and blood flow parameters; represents the duration of a complete respiratory cycle; is a function of blood oxygen saturation over time; is a function of blood flow rate over time; is the inherent physiological delay time between the two types of signals; the integral operation covers the entire respiratory cycle, and the differential operation captures the instantaneous change rate of blood oxygen saturation. The calculation result of this relationship is used to guide the direction and amplitude of parameter correction. The comparison of the coupling strength coefficient with the preset threshold determines whether to start the intensive correction process.

[0087] The optimized blood oxygen parameters and airflow parameters are merged after standardization, and the merging operation is not a simple parameter splicing but considers the physiological relevance between parameters; the merging algorithm assigns appropriate weight coefficients to different parameters, and the determination of the weight coefficients is based on the sensitivity index of the parameters to pathological state changes in historical data. The optimized pathological feature vector generated in the merging process not only contains the corrected parameter values, but also adds parameter reliability indicators and coupling coordination indicators, which provide more decision-making basis for subsequent abnormal area positioning. The entire parameter optimization process adopts a closed-loop control strategy, and the optimization results are fed back to the coupling model parameter update module, enabling the model to gradually adapt to individual physiological characteristic changes.

[0088] In the specific implementation of parameter optimization, the system independently performs optimization calculation for each respiratory cycle while maintaining a cross-cycle state variable library to save long-term trend information; the optimization algorithm is very sensitive to changes in signal quality, and when it detects a decrease in cycle data quality, it automatically reduces the optimization intensity of the cycle parameters to avoid amplifying noise into the feature vector. All threshold values and control parameters required in the optimization process are stored in a configurable parameter table, and clinical engineers can adjust the aggressiveness and conservatism of the optimization algorithm according to the actual situation of the patient. The parameter optimization technology described in this embodiment significantly improves the physiological consistency and clinical interpretability of the pathological feature vector by establishing a physiologically reasonable coupling model and performing targeted correction of feature parameters, providing purified input data for accurate assessment of lung sub-region functional abnormalities. The optimized feature vector more truly reflects the essential characteristics of ventilation and blood flow matching, enabling subsequent positioning analysis to be based on more reliable parameters for judgment.

[0089] Example 4: The abnormal area positioning process starts with the loading operation of the pre-constructed lung pathological propagation network graph, which represents the anatomical connections and pathological propagation paths between different regions of the lung in a directed graph structure; the graph nodes correspond to the anatomical sub-regions of the lung (such as the upper anterior segment and the lower dorsal segment), and the edges represent the direction in which pathological abnormalities may propagate (such as the propagation path from the upper respiratory tract to the lower respiratory tract). The graph data is derived from lung anatomy databases and clinical pathology research literature, containing physiological characteristic parameters and common pathological correlation information for each region. The system extracts abnormal feature parameters from the parameter-optimized pathological feature vector, including blood oxygen abnormality index, airflow restriction coefficient, pressure fluctuation factor, and other numerical indicators, each parameter having a reliability score passed from the optimization process.

[0090] The extracted abnormal feature parameters need to be mapped to the corresponding nodes of the lung pathological transmission network atlas. The mapping process is based on the physiological correlation rules between parameters and lung regions. For example, the blood oxygen abnormality index is mainly mapped to the lung region node with dense alveoli, while the airflow limitation coefficient is preferentially associated with the node region where the bronchus is located. The mapping algorithm calculates an abnormal integral value for each node, which integrates direct mapping parameters and indirect impact parameters. After mapping, the system starts the reverse tracing analysis along the pathological transmission path. The tracing algorithm starts from the node with abnormal integral value and searches for possible transmission sources along the reverse direction of the network edge. The search process considers the resistance coefficient and historical transmission probability on the transmission path, and assigns a credibility weight to each potential path.

[0091] Table 1: Lung partition abnormality evaluation results

[0092]

[0093] Referring to Table 1, the lung partition function evaluation results are determined according to the node abnormality degree on the transmission path, and the abnormality degree is evaluated by the abnormal integral value of the node and the correlation strength of the adjacent node. When counting the number of abnormal feature parameters of each node in the lung, the system distinguishes between primary parameters and secondary parameters. The primary parameters refer to indicators that directly reflect the function of the region, and the secondary parameters refer to indirect indicators affected by other regions. When calculating the weighted sum of abnormal feature parameters of each partition node, the weight coefficient is dynamically adjusted according to the correlation between parameter type and partition function. For example, for lung partitions important for blood oxygen exchange function, the weight of blood oxygen related parameters is correspondingly increased. The function level is divided based on the numerical range of the weighted abnormal integral, and the system divides the lung function status into four levels: normal, mild abnormality, moderate abnormality and severe abnormality. The level threshold is obtained by analyzing a large amount of clinical data. The generation of the evaluation report adopts a templating method, and the report content includes a partition function level summary table, an abnormal region distribution diagram and a trend analysis of key abnormal parameters. The report format supports clinical standard terminology, which is convenient for direct integration into the electronic medical record system.

[0094] In a specific application example, consider a monitoring case of suspected lung ventilation inequality, the system identifies from the optimized pathological feature vector that the airflow limitation coefficient is consistently high in multiple consecutive respiratory cycles, and the blood oxygen abnormality index is significantly elevated in the expiratory phase. These abnormal parameters are mapped to the corresponding nodes of the pathological propagation network graph, and the upper right lobe front node shows the highest abnormal integral value. Reverse tracing analysis finds that the abnormality may originate from the lower right lobe basal segment node and propagate upward along the bronchial tree to the right upper lobe area. The evaluation result table shows that both the lower right lobe basal segment and the right upper lobe front segment are moderate or above abnormality, which is consistent with the common ventilation inequality distribution pattern in clinical practice. The entire positioning and evaluation process uses an incremental update strategy, and when new respiratory cycle data arrives, the system only recalculates the changed area to avoid the computational overhead of full-scale repeated analysis. The evaluation result is accompanied by a confidence index, which is calculated based on parameter quality, data integrity, and model matching degree. The system provides a visualization interface for the evaluation results, displaying the lung sub-region functional status in the form of a heat map, with color depth corresponding to the abnormality degree level. The technical solution described in this embodiment realizes precise mapping from global features to local abnormalities by combining anatomical knowledge and dynamic propagation models, providing intuitive and operable lung function evaluation results for clinical practice.

[0095] Example 5: The dynamic update process starts with the quality assessment and screening of newly collected sensor data. The system performs real-time quality scoring on the continuous input multi-modal sensor data stream, with scoring criteria including signal signal-to-noise ratio, waveform integrity, and consistency with previous data. Data segments with quality scores above the threshold are marked as valid samples and enter the update pipeline, while poor quality data is temporarily cached for further verification or discarded. After the new data undergoes the feature extraction and optimization process consistent with the previous embodiment, a new set of candidate pathological feature vectors is generated. These candidate vectors are compared with existing feature templates in the lung pathological feature library for similarity. The update strategy for the lung pathological feature library uses a gradual incremental learning mechanism. When the similarity between a newly generated pathological feature vector and existing templates in the library is below the preset update threshold, the vector is considered as a potential new template candidate. The decision to add a new template to the library is based on its discriminability from existing templates and its stability in consecutive monitoring periods. The decision algorithm analyzes the frequency and consistency of the vector in different time windows. For a confirmed new template, the system assigns an initial weight and establishes an association with existing templates based on distance metrics in the feature space and clinical pathological classification information. After the template library structure changes, the system automatically recalculates the feature matching threshold and decomposition algorithm parameters to maintain the coordination of the entire feature processing pipeline.

[0096] The optimization process of the ventilation-flow coupling model is synchronized with the incoming of new data, and the model parameters are adjusted using the sliding window least square estimation algorithm, which recursively updates the model coefficients based on the monitoring data in the latest period. The optimization process focuses on the residual sequence between the model predicted values and the actual measured values, and the statistical characteristics of the residuals are used to evaluate the adaptability of the current state of the model; when the residual distribution deviates systematically or the variance increases significantly, the model structure adjustment process is triggered. The model structure optimization involves the reselection of the order of the coupling relationship and the addition or deletion of nonlinear terms, and the structure search is based on the comprehensive judgment of information criterion and cross-validation error.

[0097] In a specific application example, consider a scenario of long-term monitoring of a patient with chronic respiratory diseases, the system gradually accumulates individualized data of the patient during continuous weeks of operation. In the initial monitoring stage, the system mainly relies on the built-in general lung pathological feature library for analysis and judgment, as the patient's data continues to accumulate, the system begins to identify the patient's unique respiratory pattern characteristics; for example, the patient usually shows a slight airflow limitation pattern in the early morning period, and this pattern may be classified as an abnormal state in the general feature library. When the system detects that this pattern repeatedly appears at a high frequency and does not completely match the clinical symptoms, it will start the feature library update process, store this pattern as an individualized feature template in the feature library, and adjust the related judgment threshold. The optimization of the coupling model also reflects the individualized adaptation process, and the model will gradually adjust the phase relationship parameters between airflow rate and chest pressure for the patient's unique respiratory mechanics characteristics. For example, due to the chest shape or breathing habit, the pressure-flow relationship of some patients may have systematic differences with the standard model, and through continuous data learning, the model can gradually capture this individual characteristic, thereby providing more accurate parameter correction. The entire update and optimization process uses a two-way verification mechanism, and any modification to the feature library or model needs to be verified forward on new data and backward on historical data to ensure that the modification does not cause the degradation of system performance.

[0098] The system maintains an update log recording all modification operations on the feature library and the model, and the log content includes the modification time, the modification content, the modification basis data segment, and the performance evaluation index after the modification. The clinical user can view the historical trajectory of the self-optimization of the system through the interface, and has the final confirmation right to major modifications to prevent the system from producing false updates due to data quality problems or abnormal situations. The system also provides a manual intervention function, allowing clinical experts to directly import verified feature templates or model parameters to accelerate the optimization process of the system in specific application scenarios. Iterative improvement of monitoring accuracy is achieved through periodic performance evaluation, and the system regularly tests the performance indicators of the current configuration on the reserved validation data set, including the sensitivity and specificity of anomaly identification, the accuracy of partition positioning, etc. The performance evaluation results are compared with the historical best performance, and when performance degradation or stagnation is found, the reinforcement learning process is started to automatically adjust the hyperparameters in the update strategy. The entire dynamic updating system adopts a modular design, and the update strategy, optimization algorithm, and validation process can be replaced or upgraded according to actual application requirements, maintaining the openness and scalability of the technical system. The self-optimization mechanism described in this embodiment enables the monitoring system to continuously adapt to changes in the patient's physiological state and the evolving needs of the clinical environment, forming an intelligent monitoring solution with long-term value.

[0099] It should be noted that the relational terms herein such as first and second and the like are used solely to distinguish one entity or action from another, without necessarily requiring or implying any such actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0100] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A sensor-based method for real-time monitoring of pulmonary ventilation and blood flow distribution, characterized in that, include: Multimodal sensor data of the lung region is collected, including blood oxygen saturation waveform data, airflow rate waveform data, and intrathoracic pressure waveform data; The multimodal sensor data is time-synchronized to generate a time-aligned sensor data sequence. According to the preset respiratory phase division rules, the time-aligned sensor data sequence is divided into inspiratory phase data segment, expiratory phase data segment and transition phase data segment; Time-frequency domain feature parameters are extracted from each respiratory phase data segment to construct a multidimensional respiratory feature matrix; Based on the analysis results of respiratory phase switching frequency, abnormal respiratory patterns are identified and abnormal respiratory phase data segments are marked. Based on the lung pathology feature database, feature decomposition is performed on the multidimensional respiratory feature matrix in the normal respiratory phase data segment to generate pathological feature vectors. The pathological feature vector is optimized by performing parameter optimization based on the ventilation-perfusion coupling model to obtain the optimized pathological feature vector. Based on the lung pathology propagation network, abnormal regions are located in the optimized pathological feature vectors to generate lung regional functional assessment results.

2. The sensor-based real-time monitoring method for lung ventilation and blood flow distribution according to claim 1, characterized in that, The time synchronization processing of the multimodal sensor data includes: The timestamps of blood oxygen saturation waveform data, airflow rate waveform data, and intrathoracic pressure waveform data are aligned using a clock synchronization protocol. A sliding window filter is applied to the timestamp-aligned multimodal sensor data to remove high-frequency noise interference; Based on cross-correlation analysis, the time delay differences between multimodal sensor data are corrected; The calibrated multimodal sensor data is resampled at a uniform sampling rate to generate a time-aligned sensor data sequence.

3. The sensor-based real-time monitoring method for lung ventilation and blood flow distribution according to claim 1, characterized in that, The extraction of time-frequency domain feature parameters from each respiratory phase data segment includes: Extract the rise slope and fluctuation amplitude features from the blood oxygen saturation waveform data in the inspiratory phase data segment; Extract the descent slope and oscillation frequency features from the airflow rate waveform data in the expiratory phase data segment; Pressure gradient and fluctuation period characteristics are extracted from the intrathoracic pressure waveform data in the transition phase data segment; The extracted time-frequency domain feature parameters are arranged in order of respiratory phase to construct a multidimensional respiratory feature matrix.

4. The sensor-based real-time monitoring method for lung ventilation and blood flow distribution according to claim 1, characterized in that, The process of identifying abnormal breathing patterns and marking abnormal breathing phase data segments based on respiratory phase switching frequency analysis results includes: The number of phase transitions within a continuous respiratory cycle is calculated to obtain the real-time respiratory phase transition frequency; The real-time respiratory phase switching frequency is compared with the normal respiratory rate threshold. When the real-time respiratory phase switching frequency exceeds the frequency abnormality threshold, the dynamic time warping algorithm is activated. The morphological similarity index between adjacent respiratory phase data segments is calculated using a dynamic time warping algorithm. Based on the comparison results of morphological similarity index and similarity threshold, abnormal respiratory phase data segments are marked.

5. The sensor-based real-time monitoring method for lung ventilation and blood flow distribution according to claim 1, characterized in that, The feature decomposition of the multidimensional respiratory feature matrix in the normal respiratory phase data segment includes: Retrieve historical pathological feature templates from the lung pathology feature database; Calculate the matching degree index between the current multidimensional respiratory feature matrix and the historical pathological feature template; When the matching degree index is greater than the matching threshold, the narrowband decomposition algorithm is used to extract the main pathological feature components. When the matching degree index is less than the matching threshold, the broadband decomposition algorithm is used to extract the harmonic feature components. The extracted main pathological feature components and harmonic feature components are combined to generate a pathological feature vector.

6. The sensor-based real-time monitoring method for lung ventilation and blood flow distribution according to claim 1, characterized in that, The parameter optimization of the pathological feature vector based on the ventilation-perfusion coupling model includes: Establish an amplitude-frequency coupling model of blood oxygen saturation and blood flow velocity; Establish a phase coupling model between airflow rate and intrathoracic pressure; The blood oxygen parameters in the pathological feature vector are corrected based on the amplitude-frequency coupling relationship model; The airflow parameters in the pathological feature vector are compensated based on the phase coupling relationship model; The corrected blood oxygen parameters and the compensated airflow parameters are combined to generate an optimized pathological feature vector.

7. The sensor-based real-time monitoring method for lung ventilation and blood flow distribution according to claim 1, characterized in that, The abnormal region localization of the optimized pathological feature vector includes: Load a pre-constructed network map of lung pathology transmission; Extract abnormal feature parameters from the optimized pathological feature vector; Map abnormal feature parameters to corresponding nodes in the lung pathology transmission network map; Tracing the source of the abnormal characteristic parameters back along the path of pathological transmission; The results of functional assessment of different lung regions are determined based on the degree of abnormality at the nodes along the transmission pathway.

8. The sensor-based real-time monitoring method for lung ventilation and blood flow distribution according to claim 7, characterized in that, The determination of lung regional functional assessment results includes: Count the number of abnormal characteristic parameters in each region of the lung; Calculate the weighted sum of the abnormal feature parameters of each partition node; The functional levels of each partition are determined based on the weighted sum results; Generate an evaluation report that includes information on the functional levels of each partition.

9. The sensor-based real-time monitoring method for lung ventilation and blood flow distribution according to claim 1, characterized in that, Also includes: The lung pathology feature database is dynamically updated during real-time monitoring. The ventilation-perfusion coupling model was optimized based on newly acquired sensor data; Based on the latest pathological feature database and the optimized coupling model, the monitoring accuracy is iteratively improved.

10. A sensor-based real-time monitoring system for lung ventilation and blood flow distribution, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the sensor-based real-time monitoring method for lung ventilation and blood flow distribution as described in any one of claims 1 to 9.

Images