A lung infection state early warning system based on medical big data

Abstract

This invention relates to the field of medical data processing technology, specifically to a lung infection status early warning system based on medical big data. The system captures the flow velocity response before and after expiratory pressure regulation, and uses regularized inversion to deconstruct the time-domain signal, which is disturbed by respiratory rate, into a standardized sequence of measured lung unit volumes. Subsequently, it constructs a theoretical distribution benchmark of purely elastic recoil using measured high-pressure data. By constructing a transmission cost matrix reflecting the shear characteristics of non-Newtonian fluids, it uses an optimal transmission algorithm to calculate the optimal transmission strategy that reshapes the theoretical distribution into the measured distribution. Finally, it analyzes the topological entropy and transmission energy consumption of this transmission strategy to extract obstruction features and generate an early warning. This invention utilizes a rheological model to specifically quantify the viscosity and focal distribution of fluid within the airway, effectively differentiating between physical pulmonary edema and biological airway obstruction at an early stage.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, and specifically to a lung infection status early warning system based on medical big data. Background Technology

[0002] During intensive monitoring of patients with pulmonary infections, mechanically ventilated patients face a high risk of ventilator-associated pneumonia (VAP) caused by airway secretion retention. The pathological nature of VAP is the formation of biofilms and focal obstruction of the bronchioles by high-viscosity purulent secretions. The rheological properties of this biofluid (high shear resistance) are significantly different from those of ordinary physical pulmonary edema fluid (low viscosity, diffuse distribution).

[0003] Current technologies can collect patients' physiological data during monitoring using chest imaging or conventional respiratory mechanics testing. Pressure-volume curves can be constructed to characterize the ventilation status of the patient's lungs, reflecting macroscopic morphological changes or overall compliance of lung tissue. However, this method cannot analyze the dynamic characteristics of airflow in complex airway networks at the microscopic level, making it difficult to distinguish between physical lung collapse and biological airway obstruction in the early stages of infection. Consequently, it cannot provide effective early warning results for clinical reference. Summary of the Invention

[0004] To address the technical problem that existing early warning systems cannot effectively distinguish between physical collapse and biological airway obstruction at the microscopic level in the early stages of infection, the present invention aims to provide a lung infection status early warning system based on medical big data. The specific technical solution adopted is as follows: This invention proposes a lung infection status early warning system based on medical big data, the system comprising: The actual expiratory flow rate data acquisition module is used to acquire the high-pressure expiratory flow rate sequence before the occurrence of the expiratory pressure regulation event and the low-pressure expiratory flow rate sequence after the occurrence. The measured data decomposition module is used to construct a time constant sequence characterizing the emptying state of different lung units. Based on the theoretical emptying flow rate of the time constant sequence, the high-pressure expiratory flow rate sequence and the low-pressure expiratory flow rate sequence are regularly solved to obtain the measured high-pressure gas volume sequence and the measured low-pressure gas volume sequence of each lung unit. The theory-measured data coupling module obtains the theoretical gas distribution probability of each lung unit based on the total gas volume change between the measured high-pressure gas volume sequence and the measured low-pressure gas volume sequence; it obtains the measured gas distribution probability of each lung unit based on the data proportion of each element in the measured low-pressure gas volume sequence; it obtains the transmission cost based on the time constant difference between different lung units and constructs a transmission cost matrix; and it jointly solves the transmission cost matrix, theoretical gas distribution probability, and measured gas distribution probability to obtain the optimal transmission strategy matrix. The pulmonary obstruction feature recognition module is used to statistically analyze the data distribution of elements in the optimal transmission strategy matrix to obtain pulmonary fluid obstruction features. The real-time early warning module is used to generate early warning signals based on the changes in total gas volume and the characteristics of pulmonary fluid obstruction.

[0005] Furthermore, the method for constructing the time constant sequence includes: The preset time interval is mapped to the logarithmic domain to obtain the mapped vertical axis range. A preset number of points are evenly selected on the vertical axis range, and the selected points are restored to the time domain to obtain the time constant sequence.

[0006] Furthermore, the method for obtaining the theoretical gas distribution probability includes: The first total gas capacity of the measured high-pressure gas capacity sequence is obtained, and the second total gas capacity of the measured low-pressure gas capacity sequence is obtained. The ratio of the second total gas capacity to the first total gas capacity is taken as the rate of change of gas capacity; For each lung unit, the product of the corresponding measured high-pressure gas capacity and the rate of change of the gas capacity is taken as the theoretical low-pressure gas capacity; the proportion of the theoretical low-pressure gas capacity under the second total gas capacity is taken as the theoretical gas distribution probability.

[0007] Furthermore, the method for obtaining the transmission cost includes: Based on the exponential growth law, the time constants in the time constant sequence are assigned rheological hindrance weights from smallest to largest. Between any two lung units, the maximum rheological retardation weight between the two lung units is used as a reference weight. The reference weight is then corrected using a preset resistance adjustment coefficient to obtain the rheological penalty coefficient between the two lung units. The rheological penalty coefficient is then weighted by the time constant difference between the two lung units to obtain the transmission cost.

[0008] Furthermore, the method for solving the optimal transmission strategy matrix includes: The theoretical gas distribution probability of each lung unit constitutes the theoretical probability distribution vector, and the measured gas distribution probability of each lung unit constitutes the measured probability distribution vector. The theoretical probability distribution vector, the measured probability distribution vector, and the transmission cost matrix are input into the Sinkhorn solver, which outputs the optimal transmission strategy matrix.

[0009] Furthermore, the pulmonary fluid obstruction characteristics include the airway viscous resistance energy dissipation index and the dispersion of obstruction distribution; By statistically analyzing the gas distribution probability and corresponding transmission cost in the optimal transmission strategy matrix, the energy consumption index of airway viscous resistance is obtained; based on the disorder of elements in the optimal transmission strategy matrix, the dispersion of blockage distribution is obtained.

[0010] Furthermore, the method for obtaining the airway viscous resistance energy consumption index includes: Using the gas distribution probability as a weight, the gas distribution probability and the corresponding transmission cost are weighted and summed to obtain the airway viscous resistance energy consumption index.

[0011] Furthermore, the discreteness of the blocking distribution is the Shannon entropy of the optimal transmission strategy matrix.

[0012] Furthermore, the generation of the early warning signal based on changes in total gas volume and characteristics of pulmonary fluid obstruction includes: The total gas volume change, airway viscous resistance energy consumption index, and blockage distribution dispersion are input into the three-level early warning discrimination unit, and an early warning signal is output. The three-level early warning discrimination unit includes: Level 1 early warning mechanism: If the change in total gas volume is less than a preset change threshold, a lung consolidation / collapse early warning signal is fed back; otherwise, the level 2 early warning mechanism is activated. Level 2 warning mechanism: If the airway viscous resistance energy consumption index is less than the preset viscous threshold, a pulmonary edema risk warning signal will be fed back; otherwise, the level 3 warning mechanism will be activated. Three-level early warning mechanism: If the dispersion of the obstruction distribution is greater than or equal to a preset dispersion threshold, a diffuse inflammation early warning signal is fed back; otherwise, an airway secretion obstruction early warning signal is fed back.

[0013] Furthermore, after obtaining the high-pressure expiratory flow rate sequence and the low-pressure expiratory flow rate sequence, the following steps are also included: The first duration of the complete respiratory cycle corresponding to the high-pressure expiratory flow rate sequence and the second duration of the complete respiratory cycle corresponding to the low-pressure expiratory flow rate sequence are calculated. The minimum value among the first duration, the second duration, and the preset maximum analysis duration is set as the effective analysis duration. Based on the effective analysis duration, the high-pressure expiratory flow rate sequence and the low-pressure expiratory flow rate sequence are truncated and sampled.

[0014] The present invention has the following beneficial effects: This invention utilizes a measured data decomposition module and regularization to deconstruct macroscopic expiratory flow rate sequences into microscopic lung unit gas volume sequences. This approach overcomes the limitations of traditional monitoring methods, which can only acquire overall lung compliance and cannot penetrate to the bronchiolar level. It can eliminate time-domain interference such as respiratory rate fluctuations and accurately reconstruct the heterogeneous functional distribution of lung tissue at the microscopic time constant domain level.

[0015] Further, through a theory-measured data coupling module, the theoretical gas distribution probability is derived using the volume distribution under high pressure as a benchmark, and compared with the measured low-pressure gas distribution probability. Combined with a transport cost matrix constructed based on the time constant difference, the optimal transport strategy matrix is ​​obtained. This process essentially simulates the redistribution path of intrapulmonary gas under pressure changes, and the element distribution in the optimal transport strategy matrix directly maps the dynamic consumption and topological characteristics of gas flow. Based on this, the lung obstruction feature identification module can extract obstruction features reflecting the fluid properties within the airway by statistically analyzing the element data distribution in the optimal transport strategy matrix. Finally, the real-time warning module combines the total gas volume change reflecting the macroscopic physical state with the fluid obstruction features reflecting microscopic rheological properties, thereby accurately distinguishing between simple lung volume loss (physical collapse) and abnormal airflow obstruction (biological obstruction). Attached Figure Description

[0016] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a block diagram of a lung infection status early warning system based on medical big data, provided as an embodiment of the present invention. Detailed Implementation

[0018] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a lung infection status early warning system based on medical big data proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0020] The real-time early warning system provided in this invention can be applied to the digital medical ecosystem of ICUs, particularly relying on existing hospital medical big data integration platforms, such as Clinical Data Centers (CDRs) or ventilator IoT clusters. The system proposed in this invention can serve as an intelligent analysis node within a medical big data platform. This system intervenes in the waveform data streams of multiple networked ventilators in real time through a standard data interface. When the ICU big data center aggregates and detects an expiratory pressure regulation event at any bed, it marks it as a valid sample and triggers the system to perform data analysis and decoupling. The measured data decomposition module and the theory-measured data coupling module of this system can serve as core nodes for cloud or edge computing services, concurrently processing large amounts of measured data. The output optimal transmission strategy matrix and its pulmonary fluid obstruction characteristics can be written back to the patient's health record in the medical big data center, and fused with multimodal data such as imaging and laboratory data. The early warning system proposed in this invention not only enables real-time early warning for patients but also utilizes the big data platform to update patient health records, thereby achieving dynamic monitoring and risk stratification of the quality of airway management in hospitals or medical institutions.

[0021] The following description, in conjunction with the accompanying drawings, details a specific scheme for a lung infection status early warning system based on medical big data provided by the present invention.

[0022] Please see Figure 1 The diagram illustrates a block diagram of a lung infection status early warning system based on medical big data according to an embodiment of the present invention. The system includes: an actual expiratory flow rate data acquisition module 101, a measured data decomposition module 102, a theoretical-measured data coupling module 103, a lung obstruction feature recognition module 104, and a real-time early warning module 105.

[0023] A step change in expiratory pressure is equivalent to applying a physical stimulus to the lungs: normal or simply collapsed alveoli will exhibit linear recoil conforming to the law of elasticity (volume change synchronized with pressure); however, in airways obstructed by viscous secretions, due to the shear-thinning non-Newtonian properties of the fluid, the emptying flow rate will exhibit significant hysteresis and nonlinear resistance. Therefore, the actual expiratory flow rate data acquisition module 101 acquires the high-pressure expiratory flow rate sequence before the expiratory pressure regulation event occurs and the low-pressure expiratory flow rate sequence after it occurs. The expiratory data from both states are used as the basis for analysis to effectively monitor the patient.

[0024] In one specific implementation of this invention, when the monitoring system detects a stepwise decrease in positive end-expiratory pressure (PEEP) and the pressure change exceeds a preset pressure change threshold, it is considered an expiratory pressure regulation event. The system can automatically lock one complete respiratory cycle before the event and one complete respiratory cycle after the event to extract expiratory flow rate data. A stepwise decrease in PEEP greater than 5 units is considered a occurrence, and the pressure change threshold is set to 2 units.

[0025] Preferably, in this embodiment of the invention, because the respiratory cycle duration of mechanically ventilated patients fluctuates irregularly before and after expiratory pressure regulation operations due to the influence of sedation depth, spontaneous breathing drive, or ventilator trigger sensitivity, the respiratory cycle duration before and after the expiratory pressure regulation event may differ. In this case, directly incorporating the high-pressure expiratory flow rate sequence and the low-pressure expiratory flow rate sequence into subsequent module calculations may lead to temporal inconsistencies, thereby affecting the calculation process. Therefore, to ensure the comparability of subsequent analyses, after obtaining the high-pressure expiratory flow rate sequence and the low-pressure expiratory flow rate sequence, the following steps are also included: The first duration of the complete respiratory cycle corresponding to the high-pressure expiratory flow rate sequence and the second duration of the complete respiratory cycle corresponding to the low-pressure expiratory flow rate sequence are calculated. The minimum value among the first duration, the second duration, and the preset maximum analysis duration is set as the effective analysis duration. Based on the effective analysis duration, the high-pressure and low-pressure expiratory flow rate sequences are truncated and sampled. In this embodiment, the maximum analysis duration is set to 1 second to exclude interference from invalid flow at the end of expiration. The truncating method involves extracting segments of the effective analysis duration starting from the initial moment and resampling using linear interpolation to obtain the preprocessed high-pressure and low-pressure expiratory flow rate sequences. The two sequences have the same length, and each element represents the expiratory flow rate at a given moment.

[0026] Since the macroscopic expiratory flow rate waveform is physically a superposition of the emptying processes of multiple lung units with different emptying states, such as severe obstruction and normal states, and because the emptying state can be represented by the time constant of the emptying process (i.e., the time constant is physically represented as the product of resistance and compliance, a well-known concept in respiratory physiology and mechanical ventilation), the measured data decomposition module 102 should first define a time constant sequence to characterize the emptying states of different lung units in order to decompose the expiratory flow rate data acquired by the actual expiratory flow rate data acquisition module 101 and determine the gas emptying situation of the lung units in each emptying state of the target patient. The time constant sequence stores the time constants corresponding to lung units in different emptying states, thus allowing the theoretical emptying flow rate generated by each lung unit at each specific moment to be obtained.

[0027] It should be noted that the method for obtaining the theoretical emptying velocity of a lung unit is well known in the field. Specifically, it can be obtained based on a first-order linear atrioventricular model. The theoretical emptying velocity of the lung unit in the m-th state at time k is... Can be defined as: ;in Let e ​​be the time constant of the lung unit in the m-th state, and let e be the natural constant. This represents the specific physical time corresponding to the k-th moment during the expiratory flow rate sequence acquisition process.

[0028] Ultimately, the theoretical expiratory flow rate for each lung unit at each time step can be obtained, which can be integrated into an N×M unit expiratory response matrix. Here, N represents the length of the expiratory flow rate sequence acquired by the actual expiratory flow rate data acquisition module 101, i.e., the number of time steps; and M is the length of the time constant sequence, i.e., the number of lung unit types. The elements in the unit expiratory response matrix... This represents the theoretical emptying velocity of the lung unit in the m-th state at the k-th time.

[0029] After obtaining the theoretical expiratory flow velocity matrix, regularization solutions can be applied to the high-pressure expiratory flow velocity sequence and the low-pressure expiratory flow velocity sequence respectively, thereby obtaining the measured high-pressure gas volume sequence and the measured low-pressure gas volume sequence for each lung unit. The regularization solution is specifically implemented using a regularized non-negative least squares optimization model. Taking the high-pressure expiratory flow velocity sequence as an example, this model is expressed by the following formula: in The measured high-pressure gas capacity sequence is to be solved. The value is Tikhonov regularization coefficient, which is set to 0.05 in this embodiment of the invention to suppress high-frequency oscillations caused by measurement noise; Empty the response matrix of the unit; This is a high-pressure expiratory flow rate sequence; argmin represents the candidate variables in the optimization process; argmin is the optimization mathematical operator used to represent the value of the independent variable that minimizes the objective function.

[0030] In the above model, considering that the decomposition process of measured data is actually a highly ill-conditioned inverse problem, taking any expiratory flow rate sequence as an example, it represents a macroscopic expiratory flow rate signal, which is generated by the linear superposition of countless lung unit emptying processes. However, because the unit emptying response matrix is ​​an ideal matrix with dense values ​​and highly correlated column vectors obtained from the time constant, the unconstrained least squares solution is extremely sensitive to measurement noise. Tiny noises in the measured signal will be amplified, resulting in meaningless negative values ​​or violent oscillations in the solution results. Therefore, the model introduces... By using Tikhonov regularization, ill-conditioned problems are transformed into well-conditioned problems, significantly reducing the condition number of solutions and suppressing high-frequency oscillations caused by noise. The smoothness prior ensures that the solution conforms to the continuous physiological characteristics of the lung function distribution (avoiding discrete spikes); at the same time, the non-negativity constraint This forces the physical properties of gas volume to be satisfied (it cannot be negative). This, in turn, enables the optimization model to reliably and stably extract physiologically significant data from noisy high-pressure expiratory flow sequences. This transforms macroscopic measurements into a quantitative characterization of microscopic lung function structures. The solution represents the distribution vector of the time constant of the lung unit under high pressure, where the m-th element represents the time constant under high pressure. The gas capacity contained in the lung units can therefore be used as a measured high-pressure gas capacity sequence. Similarly, the measured low-pressure gas capacity sequence can be obtained. It should be noted that the specific principles and processes of regularization are well-known superposition techniques to those skilled in the art, and will not be elaborated upon here.

[0031] Preferably, in a specific implementation of this invention, considering that the time constant span between lung units should be relatively large, uniform point selection cannot be directly performed within a preset time interval. For example, there might be 100% state difference between 0.1s and 0.2s, while only 1% change exists between 9.9s and 10s. Uniform point selection would result in severely insufficient resolution in normal lung units, while oversampling of obstructive lung units would be meaningless. Therefore, a mapping selection should be performed on a function that conforms to normal physiological distribution. The specific method for constructing the time constant sequence includes: A preset time interval is mapped to the logarithmic domain to obtain the mapped vertical axis range. A preset number of points are uniformly selected within this range, and the selected points are then restored to the time domain to obtain the time constant sequence. In other words, this embodiment of the invention selects the logarithmic domain as an analysis domain that conforms to the distribution of normal physiological data, ensuring consistent relative resolution of lung units in different states. Whether it is the minimum or maximum time constant, the proportion of physical property changes represented by adjacent sampling points remains constant, conforming to the exponential decay characteristics of biological signals. Specifically, in this embodiment of the invention, the logarithmic function of the logarithmic domain is selected... The function, with a preset time interval of 0.1s to 10s, after mapping, has its y-axis range change to -1 to 1. The preset number of points is 50, meaning 50 points are evenly selected between -1 and 1. These 50 points are then restored to the range of 0.1s to 10s to obtain the final time constant sequence. The final time constant sequence is not uniformly distributed throughout the time interval, but rather denser at the beginning and sparser at the end. The points are denser near small values, providing high resolution and revealing subtle differences, while the points are sparser near large values, resulting in lower resolution and avoiding meaningless sampling of large datasets.

[0032] In order to distinguish between increased lung tissue stiffness (such as changes in elastic recoil force caused by pulmonary edema) and increased resistance (such as airflow restriction caused by sputum obstruction) at the microscopic level, this invention constructs a rheological energy analysis framework based on optimal transport theory. It then assumes the transport cost information generated during the lung unit state transition at the microscopic level. That is, by quantifying the cost between the measured and ideal data, it indirectly characterizes the deviation between the measured data and the ideal data. Finally, it uses the cost information characterizing the deviation to determine the patient's lung state.

[0033] Based on the above ideas, the system proposed in this embodiment of the invention uses the theory-measured data coupling module 103 to simulate and quantify the above assumptions. Since the measured high-pressure gas volume sequence represents the volume data of lung units in different states before the expiratory pressure regulation event, and the measured low-pressure gas volume sequence represents the volume data of lung units in different states after the expiratory pressure regulation event, the total gas volume change of both characterizes the degree of measured change caused by this expiratory pressure regulation event. By reconstructing the measured high-pressure gas volume sequence based on this degree of change, the theoretical gas distribution probability of the lung unit can be obtained, representing the distribution of gas volume in different lung units directly based on the overall change in an ideal scenario. The measured low-pressure gas volume sequence directly characterizes the actual gas volume distribution result of the lung unit after the expiratory pressure regulation event. Therefore, the measured gas distribution probability of each lung unit can be obtained directly based on the data proportion of each element in the measured low-pressure gas volume sequence in the overall sequence. That is, for any element, the element value is used as the numerator, and the cumulative value of the elements in the measured low-pressure gas volume sequence is used as the denominator to obtain the measured gas distribution probability corresponding to that element.

[0034] Preferably, in this embodiment of the invention, the method for obtaining the theoretical gas distribution probability includes: The first total gas capacity of the measured high-pressure gas capacity sequence is obtained, and the second total gas capacity of the measured low-pressure gas capacity sequence is obtained. The ratio of the second total gas capacity to the first total gas capacity is taken as the rate of change of gas capacity. That is, the rate of change of gas capacity before and after the adjustment event is obtained by taking the second total gas capacity as the numerator and the first total gas capacity as the denominator. This rate of change of gas capacity is an overall rate of change.

[0035] For each lung unit, the product of the corresponding measured high-pressure gas capacity and the rate of change of gas capacity is taken as the theoretical low-pressure gas capacity, which represents the actual gas capacity that the lung unit should have after this adjustment event in the absence of sputum; the proportion of the theoretical low-pressure gas capacity under the second total gas capacity is taken as the theoretical gas distribution probability.

[0036] Considering that purulent secretions from the lungs are non-Newtonian fluids exhibiting shear-thinning properties, the flow resistance they generate in the bronchopulmonary units with larger time constants is significantly greater than the resistance generated in the main airways with smaller time constants. Therefore, to quantify the resistance during the transmission process between lung units, the transmission cost between different lung units can be obtained based on the difference in time constants, thus constructing a transmission cost matrix. That is, the transmission cost matrix is ​​an M×M square matrix, where the elements... The transport cost represents the resistance cost incurred when gas is transferred from the i-th lung unit to the j-th lung unit. Therefore, the transport cost matrix, theoretical gas distribution probability, and measured gas distribution probability are jointly solved. The ideal gas distribution probability characterizes which lung units the gas should be distributed in if there is no sputum in the lungs; the measured gas distribution probability characterizes the actual locations where the observed gas has moved. By combining the transport cost matrix, a virtual transfer process can be constructed. Due to pathological changes, the gas that should have been in the i-th lung unit is actually in the j-th lung unit, and this pathological change is the "transport" analyzed in this process. Furthermore, by jointly solving the matrix, a "gas transfer scheme" that best reflects the actual situation can be obtained, i.e., the optimal transport strategy matrix.

[0037] Preferably, in this embodiment of the invention, considering the characteristics of non-Newtonian fluids, viscous sputum in the lungs tends to be more viscous when stationary or at low speeds, and becomes thinner at high speeds. Therefore, in lung units with a larger time constant, the gas flow rate is slower, resulting in a lower shear rate and thus higher viscous sputum viscosity; while in lung units with a smaller time constant, the gas flow rate is faster, resulting in a higher shear rate and correspondingly lower viscous sputum viscosity. In two lung units with significantly different time constants, the characteristics exhibited by the viscous sputum should show significant differences. To amplify these differences, this embodiment of the invention first assigns rheological hindrance weights to the time constants in the time constant sequence from smallest to largest based on an exponential growth law; that is, a larger time constant corresponds to a larger rheological hindrance weight, and the overall rheological hindrance weight exhibits an exponential growth, characterizing the degree of viscous hindrance of the viscous sputum in the lung unit corresponding to the time constant.

[0038] For any two lung units, the maximum rheological resistance weight between them can be used as a reference weight. Based on this reference weight, the weight data can be converted into a rheological penalty during the state transition process using a preset resistance adjustment coefficient, thus obtaining the rheological penalty coefficient between the two lung units. This rheological penalty coefficient is then weighted by the time constant difference between the two lung units to obtain the transport cost. In other words, the transport cost characterizes the physical energy barrier that must be overcome when a unit mass of gas is redistributed between different lung units.

[0039] As a specific example, in one implementation of this invention, for any time constant τ, the corresponding rheological hindrance weight... Represented as: ;in This is the minimum value among all time constants, used to further reflect the relative magnitude of τ, and then through a preset shear thinning coefficient. To achieve exponential growth The range is set between 1.2 and 1.8, and in this embodiment of the invention, it is set to 1.5.

[0040] As a specific example, in one implementation of this invention, for any two lung units, the rheological penalty coefficient... Represented as: ;in The preset resistance adjustment coefficient is set to 1 in this embodiment of the invention. The rheological hindrance weight for the i-th time constant. Let be the rheological hindrance weight for the j-th time constant, and max be the maximum value function. In this expression, the rheological hindrance weight is weighted using a drag adjustment coefficient, and then added to a positive integer 1 to obtain a rheological penalty coefficient greater than 1 to amplify the time constant difference. This rheological penalty coefficient represents that if the starting or ending point of the transmission path involves a large obstructive lung unit, the path cost of the entire process must be penalized, and the final cost should be amplified. That is, this rheological penalty coefficient can be regarded as the "road condition coefficient" between the transmission of two lung units.

[0041] As a specific example, in one implementation of this invention, the time constant is a mapping over the logarithmic field, specifically chosen as... The function, therefore, for any two lung units, has a transmission cost. The expression is: ;in This represents the resistance cost incurred when the lung unit in state i is transferred to the lung unit in state j. Let be the rheological penalty coefficient during the process of transferring lung units from the i-th state to the j-th state. The result of the logarithmic function of the time constant mapping value corresponding to the lung unit in the i-th state is given by the following: This is the result of the logarithmic function of the time constant mapping value corresponding to the lung unit in the j-th state. That is, in the above transmission cost expression, the rheological penalty coefficient is multiplied by the time constant difference. By weighting and considering the time constant difference as a uniform span distance reflected when the time constant is converted to the logarithmic domain, the overall transmission cost can be regarded as a "kilometers × road condition coefficient" that can vividly simulate the path cost under different lung unit states.

[0042] Preferably, in this embodiment of the invention, the method for solving the optimal transmission strategy matrix includes: The theoretical gas distribution probability of each lung unit constitutes a theoretical probability distribution vector, and the measured gas distribution probability of each lung unit constitutes a measured probability distribution vector. The theoretical probability distribution vector, the measured probability distribution vector, and the transmission cost matrix are input into the Sinkhorn solver, which outputs the optimal transmission strategy matrix. It should be noted that the Sinkhorn solver is a well-known technique in the art. This solver is used to efficiently solve the entropy-regularized optimal transmission problem, i.e., the goal is to find the minimum-cost transmission scheme between the theoretical probability distribution vector and the measured probability distribution vector. In analyzing the transmission process, a transmission cost matrix needs to be introduced to find the optimal transmission strategy matrix that minimizes the total cost. The specific solution algorithm is a well-known technique in the art and will not be elaborated here.

[0043] After obtaining the optimal transport strategy matrix, it represents the optimal gas mass ratio distribution resulting from the transfer of the ideal gas distribution probability to the measured gas distribution probability. The optimal transport strategy matrix can be viewed as a "distribution map." If the elements within it are scattered, it indicates that the gas is moving randomly during transport; a uniformly distributed matrix better reflects diffuse lesions. If the elements are clustered and have large values, it indicates that the gas is forcibly squeezed into certain lung units by a specific mechanism, reflecting severe local obstruction. Therefore, the lung obstruction feature identification module 104 can obtain the patient's lung fluid obstruction characteristics by statistically analyzing the element data distribution in the optimal transport strategy matrix. Furthermore, the real-time warning module 105 can generate a real-time warning signal by combining the changes in total gas volume before and after the expiratory pressure regulation event with the lung fluid obstruction characteristics. That is, changes in total gas volume can macroscopically represent abnormal states, while lung fluid obstruction characteristics can microscopically represent abnormal lung states; the combination of the two enables accurate and effective warning signal feedback.

[0044] Preferably, in this embodiment of the invention, the pulmonary fluid obstruction characteristics include the airway viscous resistance energy consumption index and the obstruction distribution dispersion.

[0045] The airway viscous resistance energy consumption index mainly reflects the energy required for gas transmission under the current optimal transmission strategy matrix. It can characterize the resistance driven by changes in gas distribution. Therefore, it needs to be obtained in conjunction with the transmission cost, that is, by statistically analyzing the gas distribution probability and the corresponding transmission cost in the optimal transmission strategy matrix to obtain the airway viscous resistance energy consumption index.

[0046] The obstruction distribution dispersion reflects the dispersion of resistance distribution caused by gas changes under the current optimal transport strategy matrix. The dispersion characterizes the current patient's lesion category. That is, the obstruction distribution dispersion is obtained based on the disorder of elements in the optimal transport strategy matrix.

[0047] Furthermore, methods for obtaining the airway viscous resistance energy consumption index include: Using the gas distribution probability as a weight, the gas distribution probability and the corresponding transmission cost are weighted and summed to obtain the airway viscous resistance energy consumption index. It should be noted that the optimal transmission strategy matrix and the transmission cost matrix have the same size, but the optimal transmission strategy matrix only has element values ​​at the elements corresponding to the optimal transmission strategy; the gas distribution probability of elements at other positions is 0. Therefore, the product between elements at the same position in the two matrices can be calculated, and then the products are summed to obtain the weighted sum.

[0048] Furthermore, the dispersion of the obstruction distribution is represented by the Shannon entropy of the optimal transport strategy matrix. A higher Shannon entropy indicates a more scattered distribution of elements in the matrix, signifying a global, diffuse drift in gas distribution, corresponding to physical pulmonary edema characteristics. Conversely, a lower Shannon entropy indicates a highly sparse distribution of elements, with gas distribution changes concentrated on a few specific pathways, implying focal airway obstruction, corresponding to infectious mucus plugging characteristics. The specific methods for obtaining Shannon entropy are well-known to those skilled in the art and will not be elaborated upon here.

[0049] Preferably, in this embodiment of the invention, generating a warning signal based on changes in total gas volume and characteristics of pulmonary fluid obstruction includes: The total gas volume change, airway viscous resistance energy consumption index, and blockage distribution dispersion are input into the three-level early warning discrimination unit, and an early warning signal is output. The three-level early warning discrimination unit includes: Level 1 warning mechanism: If the change in total gas volume is less than the preset change threshold, it indicates that the lung tissue compliance is extremely poor, and obvious structural abnormalities can be reflected macroscopically. Therefore, it is necessary to provide a warning signal for lung consolidation / collapse. Conversely, it indicates that the macroscopic lung structure is relatively stable, and a more accurate judgment needs to be made using a microscopic discrimination mechanism. In this case, it is necessary to enter the Level 2 warning mechanism.

[0050] It should be noted that, in this embodiment of the invention, the method for quantifying the change in total gas volume during the discrimination process is as follows: the first total gas volume under high pressure is subtracted from the second total gas volume, and the difference is used as the numerator; the pressure change amplitude generated by the expiratory pressure regulation event is used as the denominator, thus obtaining the change in total gas volume. It should also be noted that the pressure change amplitude of the expiratory pressure regulation event can be directly obtained through detection, and details will not be elaborated further.

[0051] Level 2 warning mechanism: If the airway viscosity resistance energy consumption index is less than the preset viscosity threshold, it indicates that the fluid resistance in the airway is low and the change in gas distribution is mainly driven by elastic factors, which is consistent with the characteristics of physical edema. Then, a warning signal of pulmonary edema risk is fed back. Conversely, it indicates that the gas has overcome significant additional shear resistance during the redistribution process, and there may be high viscosity abnormal secretions in the airway, thus entering the level 3 warning mechanism. The three-level early warning mechanism states that if the dispersion of the obstruction distribution is greater than or equal to the preset dispersion threshold, it indicates that the optimal transmission strategy matrix exhibits a high dispersion state, the resistance distribution is relatively uniform, and the lesion is diffusely distributed, which may be a pathological feature of diffuse airway inflammation or early ARDS. Therefore, a diffuse inflammation early warning signal is fed back. Conversely, it indicates that the abnormal resistance is concentrated in a few specific paths, which is a focal distribution, consistent with the typical mucus plugging characteristics of ventilator-associated pneumonia. An airway secretion obstruction early warning signal is fed back.

[0052] It should be noted that the change threshold, viscosity threshold, and dispersion threshold can all be obtained through retrospective analysis based on historical data. For the change threshold, a group of patients with severe lung collapse / consolidation and a group of patients with non-severe lung collapse / consolidation can be divided in a large database. The change threshold can be obtained by analyzing the distribution boundary between the two groups. In one specific implementation of this invention, it can be set to 20 units. For the viscosity threshold, a group of patients with simple physical pulmonary edema and a group of patients diagnosed with sputum plugs can be divided in a large database. The viscosity threshold is obtained by statistically analyzing the intersection point of the viscous resistance energy consumption index between the two groups. In one specific implementation of this invention, it can be set to 0.4. For the dispersion threshold, after normalizing the dispersion of the obstruction distribution, it can be directly set to a dispersion threshold of 0.6. That is, a normalized obstruction distribution dispersion greater than or equal to 0.6 is considered a high value, and vice versa, achieving the purpose of threshold screening.

[0053] In summary, the system of this invention captures the flow response before and after expiratory pressure regulation, and uses regularized inversion to deconstruct the time-domain signal disturbed by respiratory rate into a standardized measured lung unit volume sequence. Subsequently, a theoretical distribution benchmark of purely elastic recoil is constructed using measured high-pressure data. By constructing a transport cost matrix reflecting the shear characteristics of non-Newtonian fluids, an optimal transport strategy is calculated using an optimal transport algorithm to reshape the theoretical distribution into the measured distribution. Finally, the topological entropy and transport energy consumption of this transport strategy are analyzed to extract obstruction features and generate early warnings. This invention utilizes a rheological model to specifically quantify the viscosity and focal distribution of fluid within the airway, effectively differentiating between physical pulmonary edema and biological airway obstruction at an early stage.

[0054] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0055] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A lung infection status early warning system based on medical big data, characterized in that, The system includes: The actual expiratory flow rate data acquisition module is used to acquire the high-pressure expiratory flow rate sequence before the occurrence of the expiratory pressure regulation event and the low-pressure expiratory flow rate sequence after the occurrence. The measured data decomposition module is used to construct a time constant sequence characterizing the emptying state of different lung units. Based on the theoretical emptying flow rate of the time constant sequence, the high-pressure expiratory flow rate sequence and the low-pressure expiratory flow rate sequence are regularly solved to obtain the measured high-pressure gas volume sequence and the measured low-pressure gas volume sequence of each lung unit. The theory-measured data coupling module obtains the theoretical gas distribution probability of each lung unit based on the total gas volume change between the measured high-pressure gas volume sequence and the measured low-pressure gas volume sequence; it obtains the measured gas distribution probability of each lung unit based on the data proportion of each element in the measured low-pressure gas volume sequence; it obtains the transmission cost based on the time constant difference between different lung units and constructs a transmission cost matrix; and it jointly solves the transmission cost matrix, theoretical gas distribution probability, and measured gas distribution probability to obtain the optimal transmission strategy matrix. The pulmonary obstruction feature recognition module is used to statistically analyze the data distribution of elements in the optimal transmission strategy matrix to obtain pulmonary fluid obstruction features. The real-time early warning module is used to generate early warning signals based on the changes in total gas volume and the characteristics of pulmonary fluid obstruction.

2. The lung infection status early warning system based on medical big data according to claim 1, characterized in that, The method for constructing the time constant sequence includes: The preset time interval is mapped to the logarithmic domain to obtain the mapped vertical axis range. A preset number of points are evenly selected on the vertical axis range, and the selected points are restored to the time domain to obtain the time constant sequence.

3. The lung infection status early warning system based on medical big data according to claim 1, characterized in that, The method for obtaining the theoretical gas distribution probability includes: The first total gas capacity of the measured high-pressure gas capacity sequence is obtained, and the second total gas capacity of the measured low-pressure gas capacity sequence is obtained. The ratio of the second total gas capacity to the first total gas capacity is taken as the rate of change of gas capacity; For each lung unit, the product of the corresponding measured high-pressure gas capacity and the rate of change of the gas capacity is taken as the theoretical low-pressure gas capacity; the proportion of the theoretical low-pressure gas capacity under the second total gas capacity is taken as the theoretical gas distribution probability.

4. The lung infection status early warning system based on medical big data according to claim 1, characterized in that, The method for obtaining the transmission cost includes: Based on the exponential growth law, the time constants in the time constant sequence are assigned rheological hindrance weights from smallest to largest. Between any two lung units, the maximum rheological retardation weight between the two lung units is used as a reference weight. The reference weight is then corrected using a preset resistance adjustment coefficient to obtain the rheological penalty coefficient between the two lung units. The rheological penalty coefficient is then weighted by the time constant difference between the two lung units to obtain the transmission cost.

5. A lung infection status early warning system based on medical big data according to claim 1, characterized in that, The method for solving the optimal transmission strategy matrix includes: The theoretical gas distribution probability of each lung unit constitutes the theoretical probability distribution vector, and the measured gas distribution probability of each lung unit constitutes the measured probability distribution vector. The theoretical probability distribution vector, the measured probability distribution vector, and the transmission cost matrix are input into the Sinkhorn solver, which outputs the optimal transmission strategy matrix.

6. A lung infection status early warning system based on medical big data according to claim 1, characterized in that, The pulmonary fluid obstruction characteristics include the airway viscous resistance energy dissipation index and the dispersion of obstruction distribution; By statistically analyzing the gas distribution probability and corresponding transmission cost in the optimal transmission strategy matrix, the energy consumption index of airway viscous resistance is obtained; based on the disorder of elements in the optimal transmission strategy matrix, the dispersion of blockage distribution is obtained.

7. A lung infection status early warning system based on medical big data according to claim 6, characterized in that, The method for obtaining the airway viscous resistance energy consumption index includes: Using the gas distribution probability as a weight, the gas distribution probability and the corresponding transmission cost are weighted and summed to obtain the airway viscous resistance energy consumption index.

8. A lung infection status early warning system based on medical big data according to claim 6, characterized in that, The discreteness of the blocking distribution is the Shannon entropy of the optimal transmission strategy matrix.

9. A lung infection status early warning system based on medical big data according to claim 6, characterized in that, Early warning signals are generated based on changes in total gas volume and characteristics of pulmonary fluid obstruction, including: The total gas volume change, airway viscous resistance energy consumption index, and blockage distribution dispersion are input into the three-level early warning discrimination unit, and an early warning signal is output. The three-level early warning discrimination unit includes: Level 1 early warning mechanism: If the change in total gas volume is less than a preset change threshold, a lung consolidation / collapse early warning signal is fed back; otherwise, the level 2 early warning mechanism is activated. Level 2 warning mechanism: If the airway viscous resistance energy consumption index is less than the preset viscous threshold, a pulmonary edema risk warning signal will be fed back; otherwise, the level 3 warning mechanism will be activated. Three-level early warning mechanism: If the dispersion of the obstruction distribution is greater than or equal to a preset dispersion threshold, a diffuse inflammation early warning signal is fed back; otherwise, an airway secretion obstruction early warning signal is fed back.

10. A lung infection status early warning system based on medical big data according to claim 1, characterized in that, After obtaining the high-pressure expiratory flow rate sequence and the low-pressure expiratory flow rate sequence, the following steps are also included: The first duration of the complete respiratory cycle corresponding to the high-pressure expiratory flow rate sequence and the second duration of the complete respiratory cycle corresponding to the low-pressure expiratory flow rate sequence are calculated. The minimum value among the first duration, the second duration, and the preset maximum analysis duration is set as the effective analysis duration. Based on the effective analysis duration, the high-pressure expiratory flow rate sequence and the low-pressure expiratory flow rate sequence are truncated and sampled.

Images