Artificial Intelligence-Based Dynamic Prediction Method for Childhood Pneumonia

Abstract

This invention discloses an artificial intelligence-based method for dynamic prediction of the course of pneumonia in children, relating to the field of medical information technology. The method includes S1, determining a perturbation testing time window that meets safety conditions based on the child's current condition and the ventilator's baseline ventilation parameters. Within this time window, the ventilator performs invasive ventilation according to a preset perturbation ventilation control sequence while maintaining stable overall oxygenation and ventilation levels. Simultaneously, the airway pressure waveform, airflow waveform, and exhaled volume curve output by the ventilator are continuously acquired at high temporal resolution. This invention organically combines ventilator perturbation ventilation testing, small airway digital twin modeling, and disease scenario simulation based on damage potential thresholds. It constructs an individualized, irreversible small airway remodeling risk field in the early stages of invasive ventilation in children with severe adenovirus pneumonia, thereby achieving advanced prediction and stratified follow-up of post-infectious obliterative bronchiolitis.

Description

Technical Field

[0001] This invention relates to the field of medical information technology, specifically to a method for dynamic prediction of the course of childhood pneumonia based on artificial intelligence. Background Technology

[0002] Severe adenovirus pneumonia is one of the most common types of severe viral pneumonia in pediatric intensive care units, with some children requiring invasive mechanical ventilation during the acute phase. A small number of children with severe adenovirus pneumonia develop post-infectious obliterative bronchiolitis (POB) months to years after the initial clinical symptoms subside, characterized by persistent airflow limitation and irreversible small airway remodeling, severely impacting their long-term quality of life. Current follow-up methods rely heavily on imaging, pulmonary function re-examination, and empirical judgment for rough stratification, lacking an individualized disease progression prediction system based on multi-source data from the early stages of mechanical ventilation. This makes it impossible to precisely assess the risk of long-term POB in the early stages of severe illness.

[0003] When those skilled in the art conduct long-term disease management for children with severe adenovirus pneumonia who have already received invasive mechanical ventilation, it is difficult to determine whether the child's small airways are highly vulnerable based solely on laboratory indicators, imaging scores, and experience impressions at a single time point, and it is also impossible to identify the potential irreversible small airway remodeling trend that exists against the backdrop of apparent clinical recovery.

[0004] On the other hand, existing risk scoring models are mostly statistical fits to the static characteristics of past cases, failing to incorporate ventilator waveforms, high-frequency monitoring data, and the dynamic response of inflammatory burden over time into a unified framework. They generally lack the ability to simulate various future treatment pathways and disease progression scenarios based on individualized digital models of the small airways, making it impossible to estimate cumulative irreversible damage from a future scenario perspective. Consequently, it is difficult to provide operable risk fields and stratified follow-up protocols for post-infectious obliterative bronchiolitis in the early stages of severe illness. Summary of the Invention

[0005] The purpose of this invention is to provide an artificial intelligence-based method for dynamic prediction of the course of childhood pneumonia, in order to solve the problems mentioned in the background art.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a method for dynamic prediction of the course of childhood pneumonia based on artificial intelligence, comprising S1, determining a perturbation test time window that meets safety conditions based on the child's current condition and the basic ventilation parameters of the ventilator, within which the ventilator performs invasive ventilation according to a preset perturbation ventilation control sequence while maintaining stable overall oxygenation and ventilation levels, and simultaneously continuously acquiring the time series of airway pressure waveform, airflow waveform, exhaled volume curve, and inflammatory-related indicators in exhaled gas, as well as synchronized chest wall lung sound signals and vital signs data at high temporal resolution, forming a time-synchronized multimodal raw data sequence;

[0007] S2. The lungs of the child are abstracted into multiple small airway alveolar units with compliance parameters, airway closure threshold pressure parameters, reopening hysteresis pressure parameters, and susceptibility to inflammatory remodeling parameters. By establishing a mathematical model describing the connection relationship and conduction characteristics between each small airway unit, the time series reflecting the lung mechanical response in the multimodal raw data is used as the observation output. The parameters of each small airway unit are solved through numerical optimization iteration, so that the model output and the observation output reach the preset fitting accuracy within the perturbation test time window, thereby obtaining an individualized small airway digital twin model of the child.

[0008] S3. Based on the digital twin model of small airways, define the instantaneous damage power expression form of each small airway unit at a given time point. Integrate and superimpose the instantaneous damage power of local stress level, local inflammatory load and small airway inflammatory remodeling susceptibility parameters on the instantaneous damage power. Obtain the total lung cumulative damage potential by integrating and superimposing the instantaneous damage power at each time point within the preset disease course time interval. At the same time, determine the individualized damage threshold corresponding to irreversible small airway remodeling based on the individualized small airway parameter structure of the child and clinical prior knowledge.

[0009] S4. Combining the standardized treatment pathway for severe adenovirus pneumonia, construct a set of disease scenarios that include different anti-inflammatory treatment intensities over time, different mechanical ventilation parameter adjustment pathways, and different infection control effect pathways. Input the treatment intensity time function, ventilation stress time function, and inflammatory source intensity time function corresponding to each disease scenario into the small airway digital twin model, and calculate the total lung cumulative damage potential corresponding to each disease scenario within the preset follow-up time interval.

[0010] S5. Based on the relationship between the cumulative damage potential of the whole lung and the individualized damage threshold under each disease course scenario, estimate the probability classification of post-infectious bronchiolitis obliterans under the current treatment pathway. At the same time, analyze the changing trend of cumulative damage potential under other feasible treatment pathways, construct a risk field map representing the relationship between the treatment pathway and the risk of post-infectious bronchiolitis obliterans, map the risk field map to the follow-up frequency, imaging and functional assessment time points and intervention intensity suggestions corresponding to different risk levels, and output them to the clinical decision-making end.

[0011] Specifically, S1 is:

[0012] S1-1. Based on the child's current arterial oxygen saturation, arterial blood pressure, heart rate, and basic ventilation parameters of the ventilator, and combined with the preset ventilation safety rules, a test time window with a duration between the preset minimum and preset maximum values ​​is calculated, and the upper and lower limits of airway pressure, tidal volume, and respiratory rate within this time window are determined. The upper and lower limits define the amplitude range of the perturbation ventilation control sequence.

[0013] S1-2. Based on the preset waveform spectrum characteristics and amplitude range, generate a pressure control sequence or flow control sequence for each respiratory cycle within the test time window, so that the actual airway pressure waveform of each respiratory cycle has a slight disturbance on the basis of the baseline ventilation waveform, while ensuring that the child's average tidal volume and average oxygenation level remain within the target range set by the doctor throughout the entire test time window.

[0014] S1-3. Within the test time window, at a sampling frequency higher than the routine monitoring frequency, the airway pressure signal change curve over time, the airflow signal change curve over time, the exhaled volume time curve corresponding to each respiratory cycle, the concentration of inflammation-related indicators in exhaled gas change curve over time, the original acoustic signal of chest wall lung sounds, and the arterial blood oxygen saturation and hemodynamic indicators synchronized with the respiratory cycle are collected synchronously to generate a multimodal data sequence with timestamps for subsequent model inversion.

[0015] S1-4. Through the data interface between the system and the hospital's electronic medical record and laboratory information system, automatically extract the child's medication records, administration time, dosage information, and concurrent laboratory test results, including but not limited to inflammatory indicators such as white blood cell count, C-reactive protein, and cytokine levels, and pathogen detection results such as adenovirus load and viral nucleic acid quantification. Align the above data with the multimodal time series generated in S1-3 according to the timestamp.

[0016] According to the above technical solution, S2 specifically refers to:

[0017] S2-1, Abstractly divide the child's lungs into... Small airway alveolar units, for each small airway unit Initialize compliance parameters separately Airway closure threshold pressure parameter Reopening hysteresis pressure parameters and susceptibility parameters for inflammatory remodeling This forms a parameter vector containing all small airway unit parameters. ;

[0018] S2-2. Based on small airway alveolar units and the network connections between units, a mathematical model is established to describe the dynamic relationship between airway pressure input and lung volume output, yielding the result under a given parameter vector. and input airway pressure-time function Under these conditions, the alveolar unit of the small airway is obtained by solving the small airway network mechanical model over time. alveolar volume The model predicts the total volume of the entire lung as the sum of the volumes of each unit, i.e. The predicted gas flow rate is then obtained from the derivative of the total volume with respect to time. To obtain the given parameter vector Model output time function under input airway pressure time function conditions The curves of total lung compliance and airway resistance over time were further calculated by using the algebraic relationships between them.

[0019] S2-3. Obtain the whole lung airflow time function based on the expiratory volume-time curve obtained in S1. Integrating this signal over time yields the whole lung volume time function. Define the observation output time function as The selected time interval corresponding to the test time window is denoted as . The optimal parameter vector for each child is obtained by solving the problem. Iterative adjustment through numerical optimization algorithms Until the error The integral over the time interval is no greater than the preset convergence threshold, thus completing the individualized parameter inversion of the small airway digital twin model.

[0020] According to the above technical solution, S3 specifically refers to:

[0021] S3-1, For each small airway unit In time Define instantaneous damage power increment ,in Represents small airway unit In time The corresponding local stress level is calculated using the time series of airway pressure, airflow and exhaled volume collected in S1-3 as input, and the small airway digital twin model established and individualized inverted in S2. Small airway unit In time The corresponding local inflammatory load was obtained by inputting the time series of inflammatory marker concentrations in exhaled gas collected in S1-3 and the synchronous blood inflammatory markers into the small airway digital twin model, and then distributing it according to the local airflow distribution through the ventilation-perfusion mapping relationship within the model. To map local stress, local inflammatory load, and susceptibility parameters into functions of instantaneous injury power density;

[0022] S3-2, within the preset disease course time interval Within this process, the instantaneous damage power increments of all small airway units are integrated over time and summed unit by unit to obtain the total lung damage potential. ;

[0023] S3-3, Based on the individualized optimal parameter vector of the child The distribution of small and medium airway compliance, airway closure threshold pressure, reopening hysteresis pressure, and susceptibility to inflammatory remodeling parameters, combined with previous clinical experience, were used to determine the individualized damage threshold for the initiation of irreversible small airway remodeling. It is used to determine whether the cumulative damage potential of the whole lung exceeds the critical level of irreversible small airway remodeling.

[0024] According to the above technical solution, S4 specifically refers to:

[0025] S4-1. Construct a set of disease progression scenarios covering different treatment pathways. For each disease stage scenario Define the function of anti-inflammatory treatment intensity over time. Mechanical ventilation stress as a function of time and the function of inflammatory source intensity over time Together they describe the time intervals of the disease course. The external forces acting on the small airway network in this scenario, among which Medication records and inflammatory markers collected from S1-4 are input into the pharmacokinetic model, and the response curve of drug concentration over time is obtained by solving the model. The high-temporal-resolution ventilation waveforms of airway pressure, airflow, and expiratory volume collected by S1-3 are used as input. The mean mechanical stress of alveoli and small airways at each time point is calculated by the small airway digital twin model established and individually inverted by S2. The time series of concentrations of inflammation-related indicators in exhaled air collected by S1-3, the results of etiological detection, and the blood inflammation indicators were input into the small airway digital twin model. The local inflammatory load of the lungs was calculated by the ventilation and perfusion mapping relationship within the model, and combined with the infection control parameters set by the disease course scenario, including the onset time of antiviral treatment and the pathogen clearance rate.

[0026] S4-2, In each disease course scenario In the middle, , , Inputting a digital twin model of the small airways yields results over the disease course time interval. Local stress corresponding to each small airway unit With local inflammatory load Substituting the instantaneous injury power expression and the total lung cumulative injury potential energy expression in S3, the disease progression scenario is calculated. Total lung injury potential ;

[0027] S4-3, In the context of disease progression Upper definition of the probability measure of occurrence of disease course scenarios For each disease stage scenario Determine whether it exceeds the individualized damage threshold The probability of post-infectious bronchiolitis obliterans can be calculated using the following relationship. ,in In the context of disease progression above A probability measure defined for the independent variable.

[0028] According to the above technical solution, S5 specifically refers to:

[0029] S5-1, Based on the set of disease progression scenarios All disease course scenarios corresponding and The relationship between them maps the disease course scenario parameter space to a space with treatment path parameters as independent variables. and A multidimensional risk distribution map with ratios as the dependent variable forms a risk field representing the risk level of post-infectious obliterative bronchiolitis corresponding to different anti-inflammatory intensity pathways and ventilation strategy pathways.

[0030] S5-2. Based on the current treatment path of the child, the position of the disease course scenario in the risk field and the probability of post-infectious bronchiolitis obliterans. Children are categorized into pre-defined risk level ranges. For treatment pathways corresponding to higher risk levels, the treatment is tailored to each pathway within the risk field and the individualized injury threshold. The distance between them is used to screen alternative treatment pathways that have low risk and are feasible in clinical practice;

[0031] S5-3. Map the risk stratification results and alternative treatment pathway screening results to specific follow-up management strategies. This includes setting follow-up intervals, imaging re-examination time points, and pulmonary function assessment time points for different risk levels, determining whether to increase the frequency of remote monitoring and outpatient re-examinations, and linking the follow-up management strategies to the corresponding risk scenarios. Figure 1 And output it to the clinical decision-making end.

[0032] An AI-based dynamic prediction system for the course of childhood pneumonia includes:

[0033] The multimodal data acquisition module is used to apply a preset perturbation ventilation waveform to the lungs of the child within the safety boundary of invasive mechanical ventilation and simultaneously acquire high temporal resolution respiratory mechanics waveforms, exhaled gas indices, lung sound signals and vital signs data.

[0034] The small airway digital twin construction module is used to construct an individualized small airway unit network model for the child based on the multimodal data and obtain the damage potential index characterizing the irreversible small airway remodeling trend and the corresponding individualized damage threshold.

[0035] The disease course scenario simulation module is used to generate multiple disease course scenarios on the small airway digital twin model and calculate the cumulative damage potential energy under each scenario. Based on the relationship between the cumulative damage potential energy and the individualized damage threshold, it outputs the risk field of post-infectious obliterative bronchiolitis and the matching follow-up management strategy.

[0036] According to the above technical solution, the multimodal data acquisition module includes:

[0037] The perturbation test window determination unit is used to determine a perturbation test time window that meets preset safety conditions, as well as the corresponding upper and lower limits of ventilation pressure, tidal volume, and respiratory rate, based on the child's current vital signs and ventilation settings, provided that the child's condition is relatively stable and the ventilation parameters have been set by the clinician.

[0038] The perturbation waveform generation unit is used to generate a perturbation ventilation control sequence with preset spectrum characteristics within the safety boundary range and superimpose it onto the basic ventilation curve of the ventilator within the perturbation test time window to ensure that the overall gas exchange target remains stable while stimulating the small mechanical response of alveoli and small airways.

[0039] The high-frequency multimodal acquisition unit is used to continuously acquire airway pressure waveforms, airflow waveforms, exhaled volume time curves, time series of inflammation-related indicators in exhaled gas, chest wall lung sound signals, and hemodynamic indicators related to breathing at a sampling frequency higher than the conventional monitoring frequency within the perturbation test time window, and generate time-synchronized raw data sequences.

[0040] The small airway digital twin construction module includes:

[0041] Small airway unit network construction unit is used to abstractly divide the lungs of the child into multiple small airway alveolar units and assign compliance parameters, airway closure threshold pressure parameters, reopening hysteresis pressure parameters, and inflammatory remodeling susceptibility parameters to each small airway unit, and construct a mathematical model structure representing the connection relationship of the small airway network.

[0042] The parameter inversion unit is used to take the respiratory mechanical response in high-frequency multimodal data as the observation output. Based on the pre-established small airway network model structure, the parameters of each small airway unit are solved by numerical optimization algorithm so that the difference between the model output and the observation output within the perturbation test time window meets the preset convergence condition, thereby obtaining the individualized small airway network parameter set of the child.

[0043] The damage potential energy calculation unit is used to define the instantaneous damage power of each small airway unit based on the individualized small airway network parameters and to perform cumulative calculation over the entire disease course time interval to form the whole lung damage potential energy. At the same time, it determines the corresponding irreversible small airway remodeling damage threshold according to the individualized parameter set.

[0044] The disease progression scenario simulation module includes:

[0045] The disease progression scenario generation unit is used to generate a set of disease progression scenarios covering different treatment pathways and disease progression patterns based on the treatment guidelines for severe adenovirus pneumonia and the combination of anti-inflammatory strategies, ventilation strategies and infection control strategies that can be implemented by this medical institution.

[0046] The scenario simulation unit is used to input the corresponding anti-inflammatory intensity over time function, mechanical ventilation parameter over time function, and infection load over time function into the individualized small airway digital twin model under the disease course scenario set, and calculate the cumulative whole lung injury potential energy under each disease course scenario within a given follow-up time interval.

[0047] The follow-up strategy output unit is used to estimate the probability of post-infectious obliterative bronchiolitis based on the relationship between the cumulative whole lung injury potential and the individualized injury threshold under each disease stage scenario, and to construct a risk distribution map that changes with the treatment strategy. The risk distribution results are then mapped to different intensities of follow-up frequency, imaging re-examination time points, and functional assessment plans, and output to clinical users.

[0048] Compared with existing technologies, the beneficial effects achieved by this invention are as follows: Without adding additional invasive examinations, this invention utilizes a ventilator to apply short-term micro-amplitude ventilation disturbances within a preset safety boundary to obtain high temporal resolution lung mechanics and multimodal monitoring responses. By inverting and constructing child-specific small airway unit network parameters, a digital twin model capable of characterizing small airway vulnerability is formed. Then, disease scenarios composed of different anti-inflammatory strategies, ventilation strategies, and infection control strategies are superimposed on this model. The relationship between the cumulative irreversible small airway remodeling damage potential energy and the individual damage threshold under each scenario is calculated. The probability of post-infection obliterative bronchiolitis and the corresponding risk field are output, and a stratified follow-up and re-examination plan is automatically generated accordingly.

[0049] Compared with experience-based stratification schemes that rely solely on static indicators, this invention can identify high-risk children earlier and more precisely, focusing limited follow-up resources on individuals who truly face long-term irreversible remodeling risks. This helps optimize the overall management pathway for severe adenovirus pneumonia, reduces the missed detection rate of post-infectious bronchiolitis obliterans, and addresses insufficient follow-up, thereby effectively solving the technical problem in existing technologies that cannot utilize multi-source data from early mechanical ventilation for long-term disease prediction. Attached Figure Description

[0050] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0051] Figure 1 This is a schematic diagram of the overall modular structure of the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Please see Figure 1 This invention provides a technical solution: a method for dynamic prediction of the course of childhood pneumonia based on artificial intelligence, comprising:

[0054] S1. Based on the child's current condition and the basic ventilation parameters of the ventilator, a perturbation test time window that meets safety conditions is determined. Within this time window, the ventilator performs invasive ventilation according to the preset perturbation ventilation control sequence and maintains the overall oxygenation and ventilation levels stable. At the same time, the airway pressure waveform, airflow waveform, exhaled volume curve, time series of inflammation-related indicators in exhaled gas, and synchronous chest wall lung sound signals and vital signs data are continuously acquired with high temporal resolution to form a time-synchronized multimodal raw data sequence.

[0055] S2. The lungs of the child are abstracted into multiple small airway alveolar units with compliance parameters, airway closure threshold pressure parameters, reopening hysteresis pressure parameters, and susceptibility to inflammatory remodeling parameters. By establishing a mathematical model describing the connection relationship and conduction characteristics between each small airway unit, the time series reflecting the lung mechanical response in the multimodal raw data is used as the observation output. The parameters of each small airway unit are solved through numerical optimization iteration, so that the model output and the observation output reach the preset fitting accuracy within the perturbation test time window, thereby obtaining an individualized small airway digital twin model of the child.

[0056] S3. Based on the digital twin model of small airways, define the instantaneous damage power expression form of each small airway unit at a given time point. Integrate and superimpose the instantaneous damage power of local stress level, local inflammatory load and small airway inflammatory remodeling susceptibility parameters on the instantaneous damage power. Obtain the total lung cumulative damage potential by integrating and superimposing the instantaneous damage power at each time point within the preset disease course time interval. At the same time, determine the individualized damage threshold corresponding to irreversible small airway remodeling based on the individualized small airway parameter structure of the child and clinical prior knowledge.

[0057] S4. Combining the standardized treatment pathway for severe adenovirus pneumonia, a set of disease scenarios is constructed, including different anti-inflammatory treatment intensities over time, different mechanical ventilation parameter adjustment pathways, and different infection control effect pathways. The treatment intensity time function, ventilation stress time function, and inflammatory source intensity time function corresponding to each disease scenario are input into the small airway digital twin model, and the total lung cumulative damage potential corresponding to each disease scenario is calculated within the preset follow-up time interval.

[0058] S5. Based on the relationship between the cumulative damage potential of the whole lung and the individualized damage threshold under each disease course scenario, estimate the probability classification of post-infectious bronchiolitis obliterans under the current treatment pathway, and analyze the changing trend of cumulative damage potential under other feasible treatment pathways. Construct a risk field map representing the relationship between the treatment pathway and the risk of post-infectious bronchiolitis obliterans. Map the risk field map to the follow-up frequency, imaging and functional assessment time points and intervention intensity suggestions corresponding to different risk levels and output them to the clinical decision-making end.

[0059] S1 specifically refers to:

[0060] S1-1. Based on the child's current arterial oxygen saturation, arterial blood pressure, heart rate, and basic ventilation parameters of the ventilator, and combined with the preset ventilation safety rules, a test time window with a duration between the preset minimum and preset maximum values ​​is calculated, and the upper and lower limits of airway pressure, tidal volume, and respiratory rate within this time window are determined. The upper and lower limits limit the amplitude range of the perturbation ventilation control sequence.

[0061] S1-2. Based on the preset waveform spectrum characteristics and amplitude range, generate a pressure control sequence or flow control sequence for each respiratory cycle within the test time window, so that the actual airway pressure waveform of each respiratory cycle has a slight disturbance on the basis of the baseline ventilation waveform, while ensuring that the child's average tidal volume and average oxygenation level remain within the target range set by the doctor throughout the entire test time window.

[0062] S1-3. Within the test time window, at a sampling frequency higher than the routine monitoring frequency, the airway pressure signal change curve over time, the airflow signal change curve over time, the exhaled volume time curve corresponding to each respiratory cycle, the concentration of inflammation-related indicators in exhaled gas change curve over time, the original acoustic signal of chest wall lung sounds, and the arterial blood oxygen saturation and hemodynamic indicators synchronized with the respiratory cycle are collected synchronously to generate a multimodal data sequence with timestamps for subsequent model inversion.

[0063] S1-4. Through the data interface between the system and the hospital's electronic medical record and laboratory information system, automatically extract the child's medication records, administration time, dosage information, and concurrent laboratory test results, including but not limited to inflammatory indicators, such as white blood cell count, C-reactive protein, and cytokine levels, and etiological test results, such as adenovirus load and viral nucleic acid quantification, and align the above data with the multimodal time series generated in S1-3 according to the timestamp.

[0064] S2 specifically refers to:

[0065] S2-1, Abstractly divide the child's lungs into... Small airway alveolar units, for each small airway unit Initialize compliance parameters separately Airway closure threshold pressure parameter Reopening hysteresis pressure parameters and susceptibility parameters for inflammatory remodeling This forms a parameter vector containing all small airway unit parameters. ;

[0066] This approach utilizes the perturbation-response concept from cybernetics and materials mechanics. Within the baseline ventilation parameters and safety boundaries set by the physician, it applies limited, controllable-spectrum perturbations to each respiratory cycle. This allows each lung to undergo a dynamic mechanical examination over those few minutes, revealing the true compliance, closure threshold, and hysteresis characteristics of the small airway network from subtle differences in their responses. This step in this protocol upgrades the ventilator from a treatment device to an active detector, serving as the original driving force for all subsequent individualized modeling. Compared to conventional read-only monitoring data methods, the originality of this step lies in transplanting the loading and unloading perturbation concept from materials / structural testing to the invasive ventilation scenario of critically ill children. Furthermore, it ensures clinical usability through safety boundary constraints. Instead of relying on adding more indicators, it uses a small, almost non-invasive perturbation to subtly transform static monitoring into an active exploration test of the small airways.

[0067] S2-2. Based on small airway alveolar units and the network connections between units, a mathematical model is established to describe the dynamic relationship between airway pressure input and lung volume output, yielding the result under a given parameter vector. and input airway pressure-time function Under these conditions, the alveolar unit of the small airway is obtained by solving the small airway network mechanical model over time. alveolar volume The model predicts the total volume of the entire lung as the sum of the volumes of each unit, i.e. The predicted gas flow rate is then obtained from the derivative of the total volume with respect to time. To obtain the given parameter vector Model output time function under input airway pressure time function conditions The curves of total lung compliance and airway resistance over time were further calculated by using the algebraic relationships between them.

[0068] S2-3. Obtain the whole lung airflow time function based on the expiratory volume-time curve obtained in S1. Integrating this signal over time yields the whole lung volume time function. Define the observation output time function as The selected time interval corresponding to the test time window is denoted as . The optimal parameter vector for each child is obtained by solving the problem. Iterative adjustment through numerical optimization algorithms Until the error The integral over the time interval is not greater than the preset convergence threshold, thus completing the individualized parameter inversion of the small airway digital twin model;

[0069] Most practitioners of this art focus on simple mechanical parameters at the whole-lung level, such as estimating one or two overall compliance or resistance indices using plateau pressure, pressure, and volume curves. At most, they set a few summary parameters in the model. Few have attempted to explicitly construct a small airway unit network and then perform unit-by-unit parameter inversion in a pediatric pneumonia + invasive ventilation scenario. The principle of S2-3 is to treat the high temporal resolution multimodal mechanical response elicited by S1-2 / S1-3 as the observed output of an inverse problem, constructing an optimization problem. Through numerical iteration, the compliance, closure threshold, reopening hysteresis, and inflammatory susceptibility of each small airway unit are derived, resulting in a child-specific small airway digital twin fingerprint. In this approach, this step plays a central role in translating coarse whole-lung monitoring information into refined small airway network parameters, forming the basis for all subsequent damage potential calculations and disease progression simulations. Compared to the conventional approach of directly using a few lung indicators as features, the originality of this branch lies in the fact that it uses intensive care data as an observation source for solving the inverse problem of small airway networks. It truly approximates the invisible small airway structure into a computable, individualized network model. Its remarkable achievement is that it uses readily available data from the ICU to build a bridge: from macroscopic respiratory waveforms to the digital twin of microscopic small airway mechanics.

[0070] S3 specifically refers to:

[0071] S3-1, For each small airway unit In time Define instantaneous damage power increment ,in Represents small airway unit In time The corresponding local stress level is calculated using the time series of airway pressure, airflow and exhaled volume collected in S1-3 as input, and the small airway digital twin model established and individualized inverted in S2. Small airway unit In time The corresponding local inflammatory load was obtained by inputting the time series of inflammatory marker concentrations in exhaled gas collected in S1-3 and the synchronous blood inflammatory markers into the small airway digital twin model, and then distributing it according to the local airflow distribution through the ventilation-perfusion mapping relationship within the model. To map local stress, local inflammatory load, and susceptibility parameters into functions of instantaneous injury power density;

[0072] The principle of S3-1 is to define the instantaneous damage power increment for each small airway unit, and then obtain the total lung damage potential energy through time integration and unit summation. A damage threshold dependent on individual parameter vectors is also defined, concretizing the criterion of whether irreversible small airway remodeling is progressing as whether the potential energy exceeds the individual threshold. In this scheme, this step plays the role of unifying various chaotic physiological quantities onto a physical quantity highly correlated with PIBO, providing a clear objective function for subsequent scenario simulation and risk assessment. Compared with conventional weighted scoring methods, the originality of this step lies in its physical and biological implications: it respects the interaction between mechanics and inflammation while explicitly injecting individual susceptibility, viewing long-term lung injury as an energy process accumulating over time. It no longer simply states that a few indicators are dangerously high, but rather explains why, in this child, this combination of stress and inflammation just happened to cross the irreversible threshold over such a long period.

[0073] S3-2, within the preset disease course time interval Within this process, the instantaneous damage power increments of all small airway units are integrated over time and summed unit by unit to obtain the total lung damage potential. ;

[0074] S3-3, Based on the individualized optimal parameter vector of the child The distribution of small and medium airway compliance, airway closure threshold pressure, reopening hysteresis pressure, and susceptibility to inflammatory remodeling parameters, combined with previous clinical experience, were used to determine the individualized damage threshold for the initiation of irreversible small airway remodeling. It is used to determine whether the cumulative damage potential of the whole lung exceeds the critical level of irreversible small airway remodeling.

[0075] S4 specifically refers to:

[0076] S4-1. Construct a set of disease progression scenarios covering different treatment pathways. For each disease stage scenario Define the function of anti-inflammatory treatment intensity over time. Mechanical ventilation stress as a function of time and the function of inflammatory source intensity over time Together they describe the time intervals of the disease course. The external forces acting on the small airway network in this scenario, among which Medication records and inflammatory markers collected from S1-4 are input into the pharmacokinetic model, and the response curve of drug concentration over time is obtained by solving the model. The high-temporal-resolution ventilation waveforms of airway pressure, airflow, and expiratory volume collected by S1-3 are used as input. The mean mechanical stress of alveoli and small airways at each time point is calculated by the small airway digital twin model established and individually inverted by S2. The time series of concentrations of inflammation-related indicators in exhaled air collected by S1-3, the results of etiological detection, and the blood inflammation indicators were input into the small airway digital twin model. The local inflammatory load of the lungs was calculated by the ventilation and perfusion mapping relationship within the model, and combined with the infection control parameters set by the disease course scenario, including the onset time of antiviral treatment and the pathogen clearance rate.

[0077] S4-2, In each disease course scenario In the middle, , , Inputting a digital twin model of the small airways yields results over the disease course time interval. Local stress corresponding to each small airway unit With local inflammatory burden Substituting the instantaneous injury power expression and the total lung cumulative injury potential energy expression in S3, the disease progression scenario is calculated. Total lung injury potential ;

[0078] S4-3, In the context of disease progression Upper definition of the probability measure of occurrence of disease course scenarios For each disease stage scenario Determine whether it exceeds the individualized damage threshold The probability of post-infectious bronchiolitis obliterans can be calculated using the following relationship. ,in In the context of disease progression above A probability measure defined for the independent variable;

[0079] First, a set of disease progression scenarios is defined, with each scenario representing a different treatment path. Three time functions are defined for each scenario. Then, the S3 damage potential framework is run under each scenario to calculate the cumulative damage potential of the entire lung. Based on an empirical weight distribution, the system statistically analyzes whether an individualized damage threshold is exceeded and calculates the probability of post-infectious obliterative bronchiolitis. In this approach, this set of steps transforms a single prediction into a dynamic risk field covering multiple treatment paths, enabling physicians not only to assess the risk under the current path but also to compare whether adjusting the treatment plan reduces the risk. The originality of this step compared to conventional static scoring methods lies in its formalization of the uncertainty of future disease progression into multiple dynamic scenario inputs, quantifying the evolution of damage energy under each path through a digital twin model, and then using empirical weights to obtain the overall risk. Using a pediatric small airway digital twin, the complex clinical treatment decision problem is transformed into a computable energy and probability problem, allowing physicians to see the consequences of different treatment choices over the next few months in the risk field early in the ICU.

[0080] S5 specifically refers to:

[0081] S5-1, Based on the set of disease progression scenarios All disease course scenarios corresponding and The relationship between them maps the disease course scenario parameter space to a space with treatment path parameters as independent variables. and A multidimensional risk distribution map with ratios as the dependent variable forms a risk field representing the risk level of post-infectious obliterative bronchiolitis corresponding to different anti-inflammatory intensity pathways and ventilation strategy pathways.

[0082] S5-2. Based on the current treatment path of the child, the position of the disease course scenario in the risk field and the probability of post-infectious bronchiolitis obliterans. Children are categorized into pre-defined risk level ranges. For treatment pathways corresponding to higher risk levels, the treatment is tailored to each pathway within the risk field and the individualized injury threshold. The distance between them is used to screen alternative treatment pathways that have low risk and are feasible in clinical practice;

[0083] S5-3. Map the risk stratification results and alternative treatment pathway screening results to specific follow-up management strategies. This includes setting follow-up intervals, imaging re-examination time points, and pulmonary function assessment time points for different risk levels, determining whether to increase the frequency of remote monitoring and outpatient re-examinations, and linking the follow-up management strategies to the corresponding risk scenarios. Figure 1 And output it to the clinical decision-making end;

[0084] An AI-based dynamic prediction system for the course of childhood pneumonia includes:

[0085] The multimodal data acquisition module is used to apply a preset perturbation ventilation waveform to the lungs of the child within the safety boundary of invasive mechanical ventilation and simultaneously acquire high temporal resolution respiratory mechanics waveforms, exhaled gas indices, lung sound signals and vital signs data.

[0086] The small airway digital twin construction module is used to construct an individualized small airway unit network model for children based on multimodal data and obtain the damage potential index characterizing the irreversible small airway remodeling trend and the corresponding individualized damage threshold.

[0087] The disease course scenario simulation module is used to generate various disease course scenarios on the small airway digital twin model and calculate the cumulative damage potential under each scenario. Based on the relationship between the cumulative damage potential and the individualized damage threshold, it outputs the risk field of post-infectious obliterative bronchiolitis and the matching follow-up management strategy.

[0088] The multimodal data acquisition module includes:

[0089] The perturbation test window determination unit is used to determine a perturbation test time window that meets preset safety conditions, as well as the corresponding upper and lower limits of ventilation pressure, tidal volume, and respiratory rate, based on the child's current vital signs and ventilation settings, provided that the child's condition is relatively stable and the ventilation parameters have been set by the clinician.

[0090] The perturbation waveform generation unit is used to generate a perturbation ventilation control sequence with preset spectrum characteristics within the safety boundary range and superimpose it onto the basic ventilation curve of the ventilator within the perturbation test time window to ensure that the overall gas exchange target remains stable while stimulating the small mechanical response of alveoli and small airways.

[0091] The high-frequency multimodal acquisition unit is used to continuously acquire airway pressure waveforms, airflow waveforms, exhaled volume time curves, time series of inflammation-related indicators in exhaled gas, chest wall lung sound signals, and hemodynamic indicators related to breathing at a sampling frequency higher than the conventional monitoring frequency within the perturbation test time window, and generate time-synchronized raw data sequences.

[0092] The small airway digital twin building blocks include:

[0093] Small airway unit network construction unit is used to abstractly divide the lungs of the child into multiple small airway alveolar units and assign compliance parameters, airway closure threshold pressure parameters, reopening hysteresis pressure parameters, and inflammatory remodeling susceptibility parameters to each small airway unit, and construct a mathematical model structure representing the connection relationship of the small airway network.

[0094] The parameter inversion unit is used to take the respiratory mechanical response in high-frequency multimodal data as the observation output. Based on the pre-established small airway network model structure, the parameters of each small airway unit are solved by numerical optimization algorithm so that the difference between the model output and the observation output within the perturbation test time window meets the preset convergence condition, thereby obtaining the individualized small airway network parameter set of the child.

[0095] The damage potential energy calculation unit is used to define the instantaneous damage power of each small airway unit based on individualized small airway network parameters and to perform cumulative calculation over the entire disease course time interval to form the whole lung damage potential energy. At the same time, it determines the corresponding irreversible small airway remodeling damage threshold based on the individualized parameter set.

[0096] The disease progression scenario simulation module includes:

[0097] The disease progression scenario generation unit is used to generate a set of disease progression scenarios covering different treatment pathways and disease progression patterns based on the treatment guidelines for severe adenovirus pneumonia and the combination of anti-inflammatory strategies, ventilation strategies and infection control strategies that can be implemented by this medical institution.

[0098] The scenario simulation unit is used to input the corresponding anti-inflammatory intensity over time, mechanical ventilation parameters over time, and infection load over time into the individualized small airway digital twin model under a set of disease scenarios, and calculate the cumulative whole lung injury potential energy under each disease scenario within a given follow-up time interval.

[0099] The follow-up strategy output unit is used to estimate the probability of post-infectious obliterative bronchiolitis based on the relationship between the cumulative whole lung injury potential and the individualized injury threshold under each disease stage scenario, and to construct a risk distribution map that changes with the treatment strategy. The risk distribution results are then mapped to different intensities of follow-up frequency, imaging re-examination time points, and functional assessment plans, and output to clinical users.

[0100] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "include," "contain," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0101] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for dynamic prediction of the course of childhood pneumonia based on artificial intelligence, characterized by: include: S1. Based on the child's current condition and the basic ventilation parameters of the ventilator, a perturbation test time window that meets safety conditions is determined. Within this time window, the ventilator performs invasive ventilation according to the preset perturbation ventilation control sequence and maintains the overall oxygenation and ventilation levels stable. At the same time, the airway pressure waveform, airflow waveform, exhaled volume curve, time series of inflammation-related indicators in exhaled gas, and synchronous chest wall lung sound signals and vital signs data are continuously acquired with high temporal resolution to form a time-synchronized multimodal raw data sequence. S2. The lungs of the child are abstracted into multiple small airway alveolar units with compliance parameters, airway closure threshold pressure parameters, reopening hysteresis pressure parameters, and susceptibility to inflammatory remodeling parameters. By establishing a mathematical model describing the connection relationship and conduction characteristics between each small airway unit, the time series reflecting the lung mechanical response in the multimodal raw data is used as the observation output. The parameters of each small airway unit are solved through numerical optimization iteration, so that the model output and the observation output reach the preset fitting accuracy within the perturbation test time window, thereby obtaining an individualized small airway digital twin model of the child. S3. Based on the digital twin model of small airways, define the instantaneous damage power expression form of each small airway unit at a given time point. Integrate and superimpose the instantaneous damage power of local stress level, local inflammatory load and small airway inflammatory remodeling susceptibility parameters on the instantaneous damage power. Obtain the total lung cumulative damage potential by integrating and superimposing the instantaneous damage power at each time point within the preset disease course time interval. At the same time, determine the individualized damage threshold corresponding to irreversible small airway remodeling based on the individualized small airway parameter structure of the child and clinical prior knowledge. S4. Combining the standardized treatment pathway for severe adenovirus pneumonia, construct a set of disease scenarios that include different anti-inflammatory treatment intensities over time, different mechanical ventilation parameter adjustment pathways, and different infection control effect pathways. Input the treatment intensity time function, ventilation stress time function, and inflammatory source intensity time function corresponding to each disease scenario into the small airway digital twin model, and calculate the total lung cumulative damage potential corresponding to each disease scenario within the preset follow-up time interval. S5. Based on the relationship between the cumulative damage potential of the whole lung and the individualized damage threshold under each disease course scenario, estimate the probability classification of post-infectious bronchiolitis obliterans under the current treatment pathway. At the same time, analyze the changing trend of cumulative damage potential under other feasible treatment pathways, construct a risk field map representing the relationship between the treatment pathway and the risk of post-infectious bronchiolitis obliterans, map the risk field map to the follow-up frequency, imaging and functional assessment time points and intervention intensity suggestions corresponding to different risk levels, and output them to the clinical decision-making end.

2. The method for dynamic prediction of the course of childhood pneumonia based on artificial intelligence according to claim 1, characterized in that: Specifically, S1 is: S1-1. Based on the child's current arterial oxygen saturation, arterial blood pressure, heart rate, and basic ventilation parameters of the ventilator, and combined with the preset ventilation safety rules, a test time window with a duration between the preset minimum and preset maximum values ​​is calculated, and the upper and lower limits of airway pressure, tidal volume, and respiratory rate within this time window are determined. The upper and lower limits define the amplitude range of the perturbation ventilation control sequence. S1-2. Based on the preset waveform spectrum characteristics and amplitude range, generate a pressure control sequence or flow control sequence for each respiratory cycle within the test time window, so that the actual airway pressure waveform of each respiratory cycle has a slight disturbance on the basis of the baseline ventilation waveform, while ensuring that the child's average tidal volume and average oxygenation level remain within the target range set by the doctor throughout the entire test time window. S1-3. Within the test time window, at a sampling frequency higher than the conventional monitoring frequency, the airway pressure signal change curve over time, the airflow signal change curve over time, the exhaled volume time curve corresponding to each respiratory cycle, the concentration of inflammation-related indicators in exhaled gas change curve over time, the original acoustic signal of chest wall lung sounds, and the arterial blood oxygen saturation and hemodynamic indicators synchronized with the respiratory cycle are synchronously collected to generate a multimodal data sequence with timestamps. S1-4. Through the data interface between the system and the hospital's electronic medical record and laboratory information system, automatically extract the child's medication records, administration time, dosage information, and concurrent laboratory test results, including but not limited to inflammatory indicators such as white blood cell count, C-reactive protein, and cytokine levels, and pathogen detection results such as adenovirus load and viral nucleic acid quantification. Align the above data with the multimodal data sequence generated in S1-3 according to the timestamp.

3. The method for dynamic prediction of the course of childhood pneumonia based on artificial intelligence according to claim 2, characterized in that: Specifically, S2 is: S2-1, Abstractly divide the child's lungs into... Small airway alveolar units, for each small airway unit Initialize compliance parameters separately Airway closure threshold pressure parameter Reopening hysteresis pressure parameters and susceptibility parameters for inflammatory remodeling This forms a parameter vector containing all small airway unit parameters. ; S2-2. Based on small airway alveolar units and the network connections between units, a mathematical model is established to describe the dynamic relationship between airway pressure input and lung volume output, yielding the result under a given parameter vector. and input airway pressure-time function Under these conditions, the alveolar unit of the small airway is obtained by solving the small airway network mechanical model over time. alveolar volume The model predicts the total volume of the entire lung as the sum of the volumes of each unit, i.e. The predicted gas flow rate is then obtained from the derivative of the total volume with respect to time. To obtain the given parameter vector Model output time function under input airway pressure time function conditions The curves of total lung compliance and airway resistance over time were further calculated by using the algebraic relationships between them. S2-3. Obtain the whole lung airflow time function based on the expiratory volume-time curve obtained in S1. Integrating this signal over time yields the whole lung volume time function. Define the observation output time function as The selected time interval corresponding to the test time window is denoted as . The optimal parameter vector for each child is obtained by solving the problem. Iterative adjustment through numerical optimization algorithms Until the error The integral over the time interval is no greater than the preset convergence threshold, thus completing the individualized parameter inversion of the small airway digital twin model.

4. The method for dynamic prediction of the course of childhood pneumonia based on artificial intelligence according to claim 3, characterized in that: Specifically, S3 is: S3-1, For each small airway unit In time Define instantaneous damage power increment ,in Represents small airway unit In time The corresponding local stress level is calculated using the time series of airway pressure, airflow and exhaled volume collected in S1-3 as input, and the small airway digital twin model established and individualized inverted in S2. Small airway unit In time The corresponding local inflammatory load was obtained by inputting the time series of inflammatory marker concentrations in exhaled gas collected in S1-3 and the synchronous blood inflammatory markers into the small airway digital twin model, and then distributing it according to the local airflow distribution through the ventilation-perfusion mapping relationship within the model. To map local stress, local inflammatory load, and susceptibility parameters into functions of instantaneous injury power density; S3-2, within the preset disease course time interval Within this process, the instantaneous damage power increments of all small airway units are integrated over time and summed unit by unit to obtain the total lung damage potential. ; S3-3, Based on the individualized optimal parameter vector of the child The distribution of small and medium airway compliance, airway closure threshold pressure, reopening hysteresis pressure, and susceptibility to inflammatory remodeling parameters, combined with previous clinical experience, were used to determine the individualized damage threshold for the initiation of irreversible small airway remodeling. It is used to determine whether the cumulative damage potential of the whole lung exceeds the critical level of irreversible small airway remodeling.

5. The method for dynamic prediction of the course of childhood pneumonia based on artificial intelligence according to claim 4, characterized in that: Specifically, S4 is: S4-1. Construct a set of disease progression scenarios covering different treatment pathways. For each disease stage scenario Define the function of anti-inflammatory treatment intensity over time. Mechanical ventilation stress as a function of time and the function of inflammatory source intensity over time Together they describe the time intervals of the disease course. The external forces acting on the small airway network in this scenario, among which Medication records and inflammatory markers collected from S1-4 are input into the pharmacokinetic model, and the response curve of drug concentration over time is obtained by solving the model. The high-temporal-resolution ventilation waveforms of airway pressure, airflow, and expiratory volume collected by S1-3 are used as input. The mean mechanical stress of alveoli and small airways at each time point is calculated by the small airway digital twin model established and individually inverted by S2. The time series of concentrations of inflammation-related indicators in exhaled air collected by S1-3, the results of etiological detection, and the blood inflammation indicators were input into the small airway digital twin model. The local inflammatory load of the lungs was calculated by the ventilation and perfusion mapping relationship within the model, and combined with the infection control parameters set by the disease course scenario, including the onset time of antiviral treatment and the pathogen clearance rate. S4-2, In each disease course scenario In the middle, , , Inputting a digital twin model of the small airways yields results over the disease course time interval. Local stress corresponding to each small airway unit With local inflammatory load Substituting the instantaneous injury power expression and the total lung cumulative injury potential energy expression in S3, the disease progression scenario is calculated. Total lung injury potential ; S4-3, In the context of disease progression Upper definition of the probability measure of occurrence of disease course scenarios For each disease stage scenario Determine whether it exceeds the individualized damage threshold The probability of post-infectious bronchiolitis obliterans can be calculated using the following relationship. ,in In the context of disease progression above A probability measure defined for the independent variable.

6. The method for dynamic prediction of the course of childhood pneumonia based on artificial intelligence according to claim 5, characterized in that: Specifically, S5 is: S5-1, Based on the set of disease progression scenarios All disease course scenarios corresponding and The relationship between them maps the disease course scenario parameter space to a space with treatment path parameters as independent variables. and A multidimensional risk distribution map with ratios as the dependent variable forms a risk field representing the risk level of post-infectious obliterative bronchiolitis corresponding to different anti-inflammatory intensity pathways and ventilation strategy pathways. S5-2. Based on the current treatment path of the child, the position of the disease course scenario in the risk field and the probability of post-infectious bronchiolitis obliterans. Children are categorized into pre-defined risk level ranges. For treatment pathways corresponding to higher risk levels, the treatment is tailored to each pathway within the risk field and the individualized injury threshold. The distance between them is used to screen alternative treatment pathways that have low risk and are feasible in clinical practice; S5-3. Map the risk stratification results and alternative treatment pathway screening results into specific follow-up management strategies, including setting follow-up intervals, imaging re-examination time points, and pulmonary function assessment time points for different risk levels, determining whether it is necessary to increase the frequency of remote monitoring and outpatient re-examination, and outputting the follow-up management strategies and corresponding risk field views to the clinical decision-making end.

Images