A data fusion modeling system for airway epithelial barrier status assessment

Abstract

This invention discloses a data fusion modeling system for assessing the state of the airway epithelial barrier, comprising the following steps: collecting time-series raw data in the physiological microenvironment of an airway epithelial cell culture model, and generating a synchronous multimodal barrier time-series dataset after processing; extracting and calculating multi-scale biophysical features characterizing the structure and function of the epithelial barrier from the synchronous multimodal barrier time-series dataset; constructing a multi-criteria weighted fusion model to generate a barrier health index; comparing the generated barrier health index with a preset barrier health index state threshold, and outputting an assessment report through a preset logical rule base. This invention has the following advantages and effects: through multimodal data fusion and multi-criteria weighted modeling, this invention achieves a comprehensive and quantitative assessment of the structure and function of the airway epithelial barrier, significantly improving the accuracy of barrier health status monitoring and early warning capabilities.

Description

Technical Field

[0001] This invention relates to the field of biomedical engineering technology, and in particular to a data fusion modeling system for assessing the state of the airway epithelial barrier. Background Technology

[0002] The airway epithelial barrier is a critical defense structure of the respiratory tract, and its functional integrity is essential to prevent pathogens, allergens, and harmful substances from invading deeper tissues. Impairment of barrier function is closely related to the occurrence and development of various respiratory diseases, such as asthma, chronic obstructive pulmonary disease, and pulmonary infections. Therefore, accurate assessment of the airway epithelial barrier status is of great significance in disease mechanism research, drug toxicity testing, and clinical diagnosis. Currently, the main methods for assessing barrier status include transmembrane resistance measurement, fluorescently labeled permeability assays, and cell morphology observation. However, the most critical problem with these methods in practice is that existing assessment techniques often rely on single or limited types of biological indicators, failing to comprehensively and quantitatively reflect the multi-scale structural and functional status of the epithelial barrier. For example, transmembrane resistance only provides electrical impedance information but cannot capture the dynamic morphological changes in intercellular connections; fluorescently labeled permeability assays can assess barrier permeability but are easily affected by label characteristics and difficult to quantify transport dynamics; cell morphology images, while visually displaying structural features, lack a direct correlation with functional parameters. This single-faceted or simply combined assessment approach leads to biased results and insufficient sensitivity, making it difficult to identify subtle damage or overall functional degradation of barriers in their early stages, thus limiting its application in real-time monitoring and early warning. Therefore, there is an urgent need for a new method that can integrate multimodal data and generate comprehensive quantitative indicators to address the core problems of incomplete and inaccurate assessments in existing technologies. Summary of the Invention

[0003] The purpose of this invention is to provide a data fusion modeling system for assessing the state of the airway epithelial barrier, in order to solve the problems mentioned in the background art.

[0004] The above-mentioned technical objective of the present invention is achieved through the following technical solution:

[0005] A data fusion modeling system for assessing the state of the airway epithelial barrier includes:

[0006] Data acquisition and synchronous preprocessing module: Collects time-series raw data in the physiological microenvironment of the airway epithelial cell culture model. The time-series raw data includes synchronously acquired transmembrane resistance time-series data, fluorescent marker permeability time-series data, and cell morphology time-series image data. The time-series raw data is subjected to quality inspection and outlier removal, and after being aligned with the time reference, a synchronous multimodal barrier time-series dataset is generated.

[0007] Multi-scale biophysical feature extraction module: Extracts and calculates multi-scale biophysical features that characterize the structure and function of the epithelial barrier from the synchronous multimodal barrier time-series dataset; the multi-scale biophysical features include: an impedance spectrum feature set extracted from transmembrane resistance time-series data, a permeability dynamics parameter set extracted from fluorescent marker permeability time-series data, and a cell morphology measurement set extracted from cell morphology time-series image data;

[0008] Multi-criteria weighted fusion model module: Constructs a multi-criteria weighted fusion model, taking the impedance spectrum feature set, permeability kinetic parameter set and cell morphology measurement set as inputs to generate a comprehensive and quantitative barrier health index; the barrier health index is a normalized scalar value used to intuitively reflect the overall functional status of the epithelial barrier;

[0009] Status assessment and report generation module: The generated barrier health index is compared with the preset barrier health index status threshold, and the specific trend discrimination features selected in the impedance spectrum feature set, permeability dynamic parameter set and cell morphology measurement set are combined with the preset logic rule base to output an assessment report.

[0010] By adopting the above technical solution, and simultaneously acquiring transmembrane resistance time-series data, fluorescent marker permeability time-series data, and cell morphology time-series image data, and performing rigorous quality checks and outlier removal, the reliability and temporal consistency of the multimodal data source are ensured. Impedance spectral feature sets, permeability dynamics parameter sets, and cell morphology measurement sets are extracted from these data, enabling a comprehensive characterization of the epithelial barrier state across multiple scales, including electrical properties, permeability dynamics, and structural morphology. A multi-criteria weighted fusion model is constructed to generate a normalized barrier health index, providing an intuitive and quantitative comprehensive indicator to reflect the overall functional state of the barrier. Finally, by comparing the index with a preset threshold and combining it with trend discrimination features to output an evaluation report, accurate classification of barrier status and early risk warning are achieved, effectively overcoming the problems of assessment bias and insufficient sensitivity caused by existing technologies relying on a single indicator.

[0011] Further settings include the following steps:

[0012] Model Validation and Dynamic Adjustment Module: During a specific validation cycle, two operations are performed in parallel: First, the current barrier health index is acquired and calculated using the data acquisition and synchronous preprocessing module, the multi-scale biophysical feature extraction module, and the multi-criteria weighted fusion model module, and used as the model calculation value; second, a baseline barrier integrity index at the same time point is acquired through offline biological sampling and analysis methods; the model calculation value is compared with the baseline barrier integrity index to calculate the model prediction deviation; and based on the model prediction deviation, the multi-criteria weighted fusion model is dynamically adjusted.

[0013] By adopting the above technical solution, model calculations and offline biological sampling are performed in parallel during a specific validation cycle to obtain model calculation values ​​and baseline barrier integrity indicators, and the model prediction bias is calculated, thus realizing real-time validation of the multi-criteria weighted fusion model. The model parameters are dynamically adjusted based on the prediction bias to ensure the accuracy and adaptability of the barrier health index. This feedback mechanism can continuously optimize model performance, improve its robustness and reliability in long-term monitoring, and thus enhance the practical value and stability of the entire assessment method in practical applications.

[0014] A further setting involves, within the data acquisition and synchronous preprocessing module, the steps for acquiring time-series raw data in the physiological microenvironment of the airway epithelial cell culture model include:

[0015] The system acquires transmembrane resistance time-series data, which includes complex impedance information obtained by multi-frequency AC excitation; it acquires fluorescent label permeability time-series data, which is obtained by monitoring the cumulative fluorescence intensity curve of the basal side fluorescence intensity over time after the introduction of the fluorescent label at a specific time point; and it acquires cell morphology time-series image data, which includes high-resolution cell images under bright field and / or specific fluorescent labels obtained by timed imaging.

[0016] By adopting the above technical solutions, a rich and multidimensional source of original information is provided by acquiring transmembrane resistance time-series data containing complex impedance information, fluorescent marker permeability time-series data based on fluorescence intensity accumulation curves, and high-resolution cell morphology time-series image data. Multi-frequency AC excitation measurements can capture more detailed impedance spectrum characteristics, fluorescence intensity accumulation curves reflect permeability dynamics, and high-resolution images reveal subtle changes in cell morphology, thus providing a comprehensive and high-quality data foundation for subsequent feature extraction and improving the depth and breadth of the overall evaluation.

[0017] A further setting is that, in the data acquisition and synchronization preprocessing module, the steps for quality inspection and outlier removal of the raw time-series data include:

[0018] For the transmembrane resistance time-series data, its signal-to-noise ratio and phase angle are calculated, and data segments with a signal-to-noise ratio lower than a first preset threshold or a phase angle exceeding a reasonable physiological range are discarded; for the fluorescent label permeability time-series data, its baseline stability is checked and signal saturation is determined, and data segments with baseline drift exceeding a second preset threshold or signal saturation are discarded; for the cell morphology time-series image data, its focus sharpness is evaluated by calculating the image gradient, and its image quality is evaluated by calculating the local contrast, and image frames with sharpness lower than a third preset threshold or contrast lower than a fourth preset threshold are marked and discarded.

[0019] By adopting the above technical solutions, a strict data quality control mechanism is implemented. This mechanism calculates the signal-to-noise ratio and phase angle of transmembrane resistance time-series data and removes data segments with low signal-to-noise ratio or exceeding the physiological range. It also checks the baseline stability and signal saturation of fluorescent marker permeability time-series data and removes drifting or saturated data segments. Furthermore, it evaluates the focus sharpness and local contrast of cell morphology time-series image data and removes low-quality image frames. This operation effectively reduces the interference of noise and outliers, ensuring the accuracy and reliability of the input data for subsequent feature extraction and model construction, thereby improving the consistency and credibility of the entire evaluation method.

[0020] A further setting is that, in the multi-scale biophysical feature extraction module, the steps of extracting and calculating multi-scale biophysical features that can characterize the structure and function of the epithelial barrier from the synchronous multimodal barrier time-series dataset include:

[0021] From the transmembrane resistance time-series data, the sliding window average value of the steady-state transmembrane resistance, the linear regression slope changing with time, and the ratio of impedance amplitude under high-frequency and low-frequency excitation are extracted to form an impedance spectrum feature set.

[0022] Nonlinear fitting was performed on the fluorescence intensity accumulation curve in the permeability time series data of the fluorescent label to extract the time required for the fluorescence intensity accumulation curve to reach half of the plateau period, the maximum first derivative value of the curve in the initial stage, and the facilitated diffusion and active transport rate constants obtained by fitting with a two-compartment model, forming a permeability kinetic parameter set.

[0023] The cell morphology time-series image data is segmented and feature analyzed to quantify and calculate the cell boundary length per unit area, the continuity index of fluorescence signals of tight junction proteins between cells, and the aspect ratio distribution variance of the cell nucleus, thus forming a cell morphology metric set.

[0024] The characteristic parameters in the resulting impedance spectrum feature set, permeability kinetic parameter set, and cell morphology measurement set are standardized to eliminate dimensional differences and together constitute the multi-scale biophysical features.

[0025] By employing the above technical solutions, the sliding window average, linear regression slope, and impedance amplitude ratio of steady-state values ​​were extracted from transmembrane resistance time-series data, forming an impedance spectrum feature set reflecting changes in barrier electrical behavior. From fluorescent marker permeability time-series data, the half-time of the plateau phase, the maximum first derivative value, and the diffusion rate constant were extracted through nonlinear fitting, forming a permeability dynamics parameter set describing the dynamic process of barrier permeability. From cell morphology time-series image data, cell boundary length, tight junction protein continuity index, and the variance of the cell nucleus aspect ratio distribution were quantified, forming a set of cell morphological metrics characterizing barrier structural integrity. Standardizing these feature parameters constitutes multi-scale biophysical features, enabling comprehensive capture of key information about barrier status from different dimensions and providing rich and consistent input for the fusion model.

[0026] Further configuration involves a multi-criteria weighted fusion model module, which specifically includes the following steps:

[0027] Perform an initial weight allocation based on the analytic hierarchy process (AHP) to assign an initial weight coefficient to each feature parameter in the impedance spectrum feature set, the permeability kinetic parameter set, and the cell morphology metric set.

[0028] A principal component analysis model was constructed to perform dimensionality reduction on the standardized multi-scale biophysical features, and to extract the unprocessed features. One principal component; among which... The value is determined by the cumulative variance contribution rate exceeding a preset percentage threshold; the variance contribution rate of each principal component is allocated back to the original feature parameters according to the feature parameter loading ratio to obtain the contribution weight of each feature parameter based on data variability;

[0029] The initial weight coefficient of each feature parameter is weighted and combined with the contribution weight to calculate the final fusion weight of each feature parameter;

[0030] The value of each feature parameter is normalized by the min-max normalization method. Based on the normalized values ​​of all feature parameters and their corresponding final fusion weights, a weighted fusion calculation is performed to generate a barrier health index ranging from 0 to 1.

[0031] By adopting the above technical solution, an initial weight is assigned to each feature parameter using the analytic hierarchy process (AHP), and the final fusion weight is calculated by combining the data variability contribution weight extracted by principal component analysis for dimensionality reduction. This ensures that the weight allocation incorporates both expert judgment and data-driven principles. After standardizing the feature parameters, a weighted fusion is performed to generate a barrier health index ranging from 0 to 1, providing a standardized and comparable comprehensive indicator. This method makes the fusion model more scientific and objective, improves the accuracy and interpretability of the barrier health index, and effectively integrates the advantages of multi-source information.

[0032] A further setting involves, within the multi-criteria weighted fusion model module, performing an initial weight allocation based on the analytic hierarchy process (AHP). This step, assigning an initial weight coefficient to each feature parameter in the impedance spectrum feature set, the permeability kinetic parameter set, and the cell morphology metric set, includes:

[0033] A hierarchical model for assessing barrier health status is constructed. The hierarchical model uses the barrier health index as the target layer, the impedance spectrum feature set, the permeability dynamic parameter set and the cell morphology measurement set as the criterion layer, and the feature parameters in each feature set as the scheme layer.

[0034] Based on a preset judgment scale, a judgment matrix is ​​constructed between the criterion layer and the target layer, and between the scheme layer and the criterion layer. The judgment matrix is ​​subjected to a consistency check, and its feature vector is calculated after passing the check. After normalizing the feature vector, an initial weighting coefficient is assigned to each feature parameter in the impedance spectrum feature set, the permeability dynamic parameter set, and the cell morphology metric set.

[0035] By adopting the above technical solution, a hierarchical model is constructed with the barrier health index as the target layer, three feature sets as the criterion layer, and each feature parameter as the scheme layer. Based on a preset judgment scale, a judgment matrix is ​​constructed to perform consistency checks and weight calculations, and a scientific and reasonable initial weight is assigned to each feature parameter. This structured method ensures the systematicness and logic of the weight allocation, avoids subjective arbitrariness, and enables the initial weight to accurately reflect the relative importance of different features to the barrier health status, laying a solid foundation for the subsequent fusion model.

[0036] Further settings include a status assessment and report generation module, which specifically includes the following steps:

[0037] A preset barrier health index state threshold is invoked, which includes a healthy state threshold, a slightly damaged state threshold, and a moderately damaged state threshold. The barrier health index is compared with the healthy state threshold, the slightly damaged state threshold, and the moderately damaged state threshold to determine a baseline state level. The baseline state level includes a normal stable state, a slightly unstable state, a moderately unstable state, a severely unstable state, and a barrier collapse state.

[0038] Select specific trend discrimination features and mark abnormal feature states;

[0039] The baseline state level is combined with the characteristic abnormal state and input into the logical rule base for matching query. Conflicts are resolved according to preset priority rules, and a unique and definite final state classification result is output.

[0040] Based on the final state classification results, corresponding graded early warning signals are generated and an evaluation report consisting of the final state classification results and the graded early warning signals is output.

[0041] By adopting the above technical solution, the baseline status level is determined by comparing the barrier health index with a preset health status threshold, and abnormal status is marked by combining selected trend discrimination features. The system then uses a logical rule base for matching and conflict resolution to output a unique final status classification and graded early warning signal. This multi-dimensional assessment mechanism not only considers static index values ​​but also incorporates dynamic trend features, enabling it to more sensitively identify early changes and potential risks in the barrier, thereby providing a more comprehensive and accurate status assessment and timely early warning.

[0042] Further settings are available in the status assessment and report generation module:

[0043] The specific trend discrimination features include: extracting the linear regression slope of the steady-state value of transmembrane resistance over time from the impedance spectrum feature set, defined as the transmembrane resistance decrease slope; extracting the maximum first derivative value of the fluorescence intensity accumulation curve in the initial stage from the permeability kinetic parameter set, defined as the maximum permeability slope; and extracting the continuity index of the fluorescence signal of intercellular tight junction proteins from the cell morphology measurement set, defined as the tight junction continuity index.

[0044] The marking of characteristic anomalous states includes:

[0045] For the slope of the transmembrane resistance decrease, if its moving average value within multiple consecutive preset time windows is lower than a preset negative sensitivity threshold, it is marked as a characteristic abnormal state with a continuous negative development trend; for the maximum permeability slope, if its moving average value within multiple consecutive preset time windows is higher than a preset positive sensitivity threshold, it is marked as a characteristic abnormal state with a continuous positive growth trend; for the tight connection continuity index, if its single change in two adjacent monitoring periods exceeds a preset jump sensitivity threshold, it is marked as a characteristic abnormal state with a significant negative jump trend.

[0046] By employing the above technical solutions, key dynamic changes in barrier function can be captured by extracting the slope of transmembrane resistance decrease from the impedance spectrum feature set, the maximum slope of permeability from the permeability kinetic parameter set, and the tight junction continuity index from the cell morphology measurement set as trend discrimination features. These features are then labeled with continuous negative development trends, continuous positive growth trends, and significant negative jump trends. This trend analysis enhances the sensitivity of the assessment, enabling early identification of subtle damage or deterioration trends in the barrier, thereby achieving more proactive monitoring and early warning.

[0047] Further setup involves the model validation and dynamic adjustment module, which includes the following steps:

[0048] During a specific verification cycle, the first verification operation and the second verification operation are executed in parallel. The first verification operation involves acquiring and calculating the current barrier health index using the data acquisition and synchronous preprocessing module, the multi-scale biophysical feature extraction module, and the multi-criteria weighted fusion model module, and using it as the model calculation value. The second verification operation involves acquiring the baseline barrier integrity index at the same time point through offline biological sampling and analysis methods.

[0049] The model prediction deviation is calculated by comparing the calculated value with the benchmark barrier integrity index; wherein the model prediction deviation is the absolute or relative difference between the calculated value and the benchmark barrier integrity index.

[0050] Based on the model prediction bias, a Bayesian update algorithm is used to dynamically adjust the final fusion weights in the multi-criteria weighted fusion model. The Bayesian update algorithm uses the final fusion weights before adjustment as the prior distribution and the benchmark barrier integrity index as the observed value to infer the updated final fusion weights.

[0051] By adopting the above technical solution, model calculation and offline biological sampling are performed in parallel during the validation cycle to obtain the calculated model value and the baseline barrier integrity index. The model prediction bias is calculated, and the final fusion weights are dynamically adjusted using the Bayesian update algorithm, thereby realizing the model's self-optimization and continuous improvement. The Bayesian update combines the prior weights with the observed values ​​to infer the updated weights, enabling the model to adapt to data changes and improve prediction accuracy, thus ensuring the reliability and accuracy of the barrier health index in long-term use.

[0052] In summary, the present invention has the following beneficial effects: by using multimodal data fusion and multi-criteria weighted modeling, the present invention achieves a comprehensive and quantitative assessment of the structure and functional status of the airway epithelial barrier, which significantly improves the accuracy of barrier health status monitoring and early warning capabilities. Attached Figure Description

[0053] Figure 1 This is the main flowchart of an embodiment;

[0054] Figure 2 This is a flowchart illustrating the data acquisition and synchronization preprocessing module in the embodiment.

[0055] Figure 3 This is a flowchart illustrating the multi-scale biophysical feature extraction module in the embodiment.

[0056] Figure 4This is a flowchart illustrating the multi-criteria weighted fusion model module in the embodiment.

[0057] Figure 5 This is a flowchart illustrating the status assessment and report generation module in the embodiment.

[0058] Figure 6 This is a flowchart illustrating the model verification and dynamic adjustment module in the embodiment. Detailed Implementation

[0059] The present invention will be further described in detail below with reference to the accompanying drawings.

[0060] As attached Figures 1-6 As shown;

[0061] This embodiment discloses a data fusion modeling system for assessing the state of the airway epithelial barrier, including the following steps:

[0062] Data acquisition and synchronous preprocessing module: Collects time-series raw data in the physiological microenvironment of the airway epithelial cell culture model. The time-series raw data includes synchronously acquired transmembrane resistance time-series data, fluorescent marker permeability time-series data, and cell morphology time-series image data. The time-series raw data is subjected to quality inspection and outlier removal, and after being aligned with the time reference, a synchronous multimodal barrier time-series dataset is generated.

[0063] The specific implementation process is as follows: Time-series raw data were collected in the physiological microenvironment of the airway epithelial cell culture model, including simultaneous acquisition of transmembrane resistance time-series data, fluorescent label permeability time-series data, and cell morphology time-series image data. This data collection is used to subsequently generate a synchronous multimodal barrier time-series dataset, serving as the basis for assessing the epithelial barrier status. Transmembrane resistance time-series data acquisition was achieved through multi-frequency AC excitation measurement. The specific process is as follows: An impedance analyzer was used to apply an AC signal with a frequency range of 1Hz to 100kHz to measure the complex impedance of the airway epithelial cell layer, recording the data sequence of its amplitude and phase angle changes over time, thereby obtaining dynamic information reflecting the barrier resistance characteristics. Fluorescent label permeability time-series data acquisition was completed using a fluorescence spectroscopy monitoring system. The specific process is as follows: A fluorescent label with a known molecular weight (such as FITC-glucan) was introduced into the top chamber of the airway epithelial cell model. A microplate reader was used to periodically detect the fluorescence intensity in the basal chamber, generating a cumulative fluorescence intensity curve. This curve is used to quantify the permeation rate and accumulation amount of the label across the barrier. Cell morphology time-series image data were acquired using an inverted microscope combined with a high-resolution CCD camera. The specific process is as follows: bright field images and / or fluorescence images under specific fluorescent labels (such as ZO-1 antibody-labeled tight junction proteins) were captured at preset time points (e.g., every 30 minutes) to obtain high-resolution sequence images of cell monolayer morphology, cell boundary structure and protein distribution.

[0064] The acquired time-series raw data undergoes quality inspection and outlier removal, specifically as follows: For transmembrane resistance time-series data, the signal-to-noise ratio (SNR) and phase angle are calculated. If the SNR is lower than a first preset threshold (preferably, the first preset threshold is 20 dB) or the phase angle exceeds a reasonable physiological range (e.g., -90° to 90°), the data segment is determined to have noise interference or measurement error and is removed. For fluorescent label permeability time-series data, the baseline stability is checked and signal saturation is determined. If the baseline drift exceeds a second preset threshold (preferably, the first preset threshold is 10% of the initial fluorescence intensity value) or signal saturation occurs (e.g., the fluorescence intensity value is close to the detector's maximum range), the data segment is determined to be unreliable and is removed. For cell morphology time-series image data, the image focus sharpness is evaluated by calculating the image gradient, and the image quality is evaluated by calculating the local contrast. If the sharpness is lower than a third preset threshold (preferably, the third preset threshold is 0.1) or the contrast is lower than a fourth preset threshold (preferably, the fourth preset threshold is 50), the image frame is determined to be blurry or of poor quality and is marked and removed. After quality inspection and outlier removal, a precise time synchronization protocol (such as the PTP protocol) is used to unify the timestamps of transmembrane resistance time-series data, fluorescent marker permeability time-series data, and cell morphology time-series image data to the same macroscopic time base. For data streams with inconsistent sampling frequencies (e.g., transmembrane resistance data sampling rate of 1Hz, fluorescence data of 0.1Hz, and image data of 0.033Hz), a resampling algorithm based on cubic spline interpolation is used to interpolate all data onto a unified high-frequency time axis (e.g., 1kHz) to achieve microsecond-level time synchronization. Finally, the transmembrane resistance time-series data, fluorescent marker permeability time-series data, and cell morphology time-series image data, after time base alignment, are associated and packaged according to timestamps to generate a structured synchronized multimodal barrier time-series dataset. This dataset is stored in time series form, with each time point containing the corresponding impedance value, fluorescence intensity value, and image index, ensuring strict alignment of multimodal data in the time dimension and providing a consistent and reliable data foundation for subsequent feature extraction and fusion modeling.

[0065] Multi-scale biophysical feature extraction module: Extracts and calculates multi-scale biophysical features that characterize the structure and function of the epithelial barrier from the synchronous multimodal barrier time-series dataset; the multi-scale biophysical features include: an impedance spectrum feature set extracted from transmembrane resistance time-series data, a permeability kinetic parameter set extracted from fluorescent marker permeability time-series data, and a cell morphology metric set extracted from cell morphology time-series image data.

[0066] The specific implementation process is as follows: The process of extracting the impedance spectrum feature set from the transmembrane resistance time series data is as follows: The transmembrane resistance time series data is a complex impedance sequence obtained through multi-frequency AC excitation measurement, containing impedance amplitude and phase angle information. First, the steady-state value of the transmembrane resistance is averaged using a sliding window. A sliding window of size N (e.g., N=10 time points) is used to calculate the arithmetic mean of the steady-state value of the transmembrane resistance within each window, in order to smooth instantaneous fluctuations and capture long-term trends. The formula is expressed as: ;in, Indicates a point in time The average value of the calculated steady-state transmembrane resistance over a sliding window; Indicates the size of the sliding window; Indicates a specific point in time. The steady-state value of the transmembrane resistance was measured. Next, the slope of the linear regression of the transmembrane resistance over time was calculated. By performing a least-squares linear fit on the time-series data of the transmembrane resistance, the fitting equation was obtained. ;in, Indicates a point in time The steady-state value of the transmembrane resistance; Represents the intercept constant; This represents the slope of the linear regression; a positive value indicates increased barrier resistance, while a negative value indicates decreased barrier resistance. Finally, the ratio of impedance amplitude under high-frequency and low-frequency excitation is calculated, and a high-frequency excitation point (e.g., frequency) is selected. ) and low-frequency excitation points (e.g., frequency) impedance amplitude and Calculate the ratio ;in, The impedance ratio represents the ratio of high-frequency to low-frequency impedance amplitudes. This value reflects the capacitive properties of the barrier and is related to the tightness of intercellular connections. The three characteristic parameters mentioned above—the sliding window mean, the linear regression slope, and the impedance amplitude ratio—together constitute the impedance spectrum feature set, used to characterize the electrophysiological properties of the barrier.

[0067] The process of extracting the permeability kinetic parameter set from fluorescent label permeability time-series data is as follows: Fluorescent label permeability time-series data is obtained by monitoring the permeation process of fluorescent labels (such as FITC-glucan) in the barrier through cumulative fluorescence intensity curves. First, the cumulative fluorescence intensity curves are nonlinearly fitted using a sigmoid function (such as the Logistic function) to extract the time parameter required for the cumulative fluorescence intensity curve to reach half of the plateau period. This parameter represents the time it takes for the marker to penetrate to half of its maximum accumulation, reflecting the permeability rate. Secondly, the maximum first derivative value of the curve in the initial stage is calculated by numerically differentiating the fluorescence intensity accumulation curve over the initial time interval (e.g., the first 30 minutes) to obtain the maximum value of the first derivative. , indicating fluorescence intensity Regarding time The first derivative of the value represents the peak permeation rate and is used to assess the initial permeability of the barrier. Finally, a two-compartment model is used to fit the cumulative fluorescence intensity curve. The two-compartment model incorporates facilitated diffusion and active transport mechanisms, and its differential equation is: ;in, Indicates the concentration of the marker in the basal chamber Over time The rate of change; and The labeled concentrations in the top chamber and the base chamber are represented, respectively. The facilitated diffusion rate constant is estimated by fitting using the nonlinear least squares method. and active transport rate constant The time parameter required for the cumulative fluorescence intensity curve to reach half of its plateau phase was extracted. The maximum first derivative value of the curve in the initial stage Facilitated diffusion rate constant and active transport rate constant Together they constitute a set of permeability dynamic parameters, used to quantify the molecular permeability dynamics of the barrier.

[0068] The process of extracting cell morphology metrics from time-series cell morphology image data is as follows: Time-series cell morphology image data includes high-resolution sequence images under bright field and specific fluorescent labels (such as ZO-1 antibody labeling). First, the images are preprocessed and segmented. A U-Net-based deep learning model is used to segment cell boundaries and nuclei, and the length of the cell boundary per unit area is calculated. The length of the cell boundary per unit area is obtained by counting the total number of segmented cell boundary pixels and dividing by the image area. ;in, Indicates the length of the cell boundary; This represents the total number of pixels representing cell boundaries in the image. The total pixel area of ​​the image is represented; cell boundary length reflects the complexity of cell shape and the degree of intercellular contact. Secondly, the continuity index of the fluorescence signal of tight junction proteins is quantified. Tight junction signals are extracted from ZO-1 fluorescence images, and the continuity index is defined by calculating the local variance and global continuity of the signal intensity. ;in, Indicates a continuity index; Represents the local signal variance. The value represents the square of the global signal mean, and a higher continuity exponent indicates a more continuous connection. Finally, the variance of the aspect ratio distribution of cell nuclei is analyzed, and the aspect ratio is calculated for each cell nucleus. ;in, This indicates the aspect ratio of the cell nucleus; This represents the length of the major axis after fitting the ellipse to the cell nucleus; This represents the length of the minor axis after fitting the ellipse to the cell nucleus. Then, the variance of the aspect ratio for all cell nuclei is calculated. High variance indicates strong heterogeneity in cell nuclear morphology, which may indicate cell stress or differentiation status. The variance of cell boundary length per unit area, tight junction continuity index, and cell nuclear aspect ratio distribution together constitute a set of cell morphological measures used to assess the structural integrity of the barrier.

[0069] Finally, all feature parameters in the resulting impedance spectrum feature set, permeability kinetic parameter set, and cell morphology metric set were standardized to eliminate dimensional differences. A min-max normalization method was used to map each feature parameter value to the [0,1] interval. The standardized feature parameters collectively constitute a multi-scale biophysical feature vector, providing consistent and comparable input for the subsequent multi-criteria weighted fusion model.

[0070] Multi-criteria weighted fusion model module: Construct a multi-criteria weighted fusion model, taking the impedance spectrum feature set, permeability dynamic parameter set and cell morphology measurement set as inputs to generate a comprehensive and quantitative barrier health index; the barrier health index is a normalized scalar value used to intuitively reflect the overall functional status of the epithelial barrier.

[0071] The specific implementation process is as follows: First, an initial weight allocation based on the analytic hierarchy process (AHP) is performed, assigning an initial weight coefficient to each feature parameter in the impedance spectrum feature set, permeability kinetic parameter set, and cell morphology measurement set. The AHP, by constructing a hierarchical structure model and judgment matrix, can systematically integrate expert knowledge, ensuring the rationality and consistency of the weight allocation. In the specific implementation, a hierarchical structure model for assessing the health status of the barrier is constructed. This model uses the barrier health index as the target layer (highest layer), the impedance spectrum feature set, permeability kinetic parameter set, and cell morphology measurement set as the criterion layers (middle layers), and the specific feature parameters in each feature set as the scheme layer (bottom layer). For example, the impedance spectrum feature set includes the sliding window average of the steady-state transmembrane resistance, the slope of the linear regression of transmembrane resistance over time, and the ratio of impedance amplitude under high-frequency and low-frequency excitation; the permeability kinetic parameter set includes the time required for the cumulative fluorescence intensity curve to reach half of the plateau phase, the maximum first derivative of the curve in the initial stage, the facilitated diffusion rate constant, and the active transport rate constant; the cell morphology metric set includes the cell boundary length per unit area, the continuity index of fluorescence signals of tight junction proteins between cells, and the variance of the aspect ratio distribution of the cell nucleus. Based on a preset judgment scale (such as the 1-9 scale), a judgment matrix is ​​constructed between the criterion layer and the target layer, as well as a judgment matrix between the scheme layer and the criterion layer. For example, the criterion layer judgment matrix is ​​used to compare the relative importance of the impedance spectrum feature set, the permeability kinetic parameter set, and the cell morphology metric set to the barrier health index; the scheme layer judgment matrix is ​​used to compare the relative importance of each feature parameter under the same criterion layer. For each judgment matrix, a consistency check is performed, and the consistency ratio is calculated. If the consistency ratio is less than 0.1, the consistency check is passed, indicating that the judgment matrix is ​​reasonable. Then, the eigenvectors of the judgment matrices are calculated and normalized to obtain the initial weight coefficients for each characteristic parameter. For example, through calculation, an initial weight of 0.15 might be assigned to the sliding window average value of the transmembrane resistance steady-state value, an initial weight of 0.10 to the linear regression slope of the transmembrane resistance over time, an initial weight of 0.05 to the ratio of impedance amplitude under high-frequency and low-frequency excitation, and so on, ensuring that the sum of the initial weight coefficients for all characteristic parameters is 1.

[0072] Secondly, a principal component analysis model is constructed to reduce the dimensionality of the standardized multi-scale biophysical features. Specifically, the standardized feature parameters are arranged into a feature matrix, its covariance matrix is ​​calculated, and the eigenvalues ​​and eigenvectors are solved. The eigenvalues ​​are then sorted from largest to smallest, and the top eigenvalues ​​are selected. One principal component, of which The value is determined by the cumulative variance contribution rate exceeding a preset percentage threshold (e.g., 85%). The variance contribution rate of each principal component represents its ability to explain the variability of the original data. Subsequently, the variance contribution rate of each principal component is allocated back to the original feature parameters according to the proportion of the feature parameter's loading on that principal component, resulting in the contribution weight of each feature parameter based on data variability. For example, if the variance contribution rate of the first principal component is 50%, and the sliding window average value of the transmembrane resistance steady-state value has a loading of 0.3 on that principal component, then the contribution weight obtained by this feature parameter from the first principal component is 50% × 0.3 = 0.15; this process is repeated for all principal components and feature parameters, and normalized to obtain the final contribution weight of each feature parameter. Next, the initial weight coefficient of each feature parameter is weighted and combined with the contribution weight to calculate the final fusion weight of each feature parameter. Specifically, a linear weighting method is used, and a combination coefficient (e.g., 0.5 to balance subjective and objective weights) is set, expressed by the formula: ;in, Indicates the first The final fusion weights of the feature parameters; Represents the combination coefficients; Representing characteristic parameters The initial weighting coefficients; Representing characteristic parameters The contribution weights are determined by this step, which incorporates expert experience and takes into account the variability of the data itself, making the weight allocation more scientific and adaptable.

[0073] Finally, based on the standardized values ​​of all feature parameters and their corresponding final fusion weights, a weighted fusion calculation is performed to generate the barrier health index. First, the value of each feature parameter is mapped to the [0,1] interval using a min-max normalization method. Then, the barrier health index is calculated using a weighted summation formula: ;in, Indicates the barrier health index; Indicates the total number of characteristic parameters; Representing characteristic parameters The standardized value is the value of the barrier health index. Since the sum of all weights is 1 and the standardized value is in the range [0,1], the barrier health index will also fall between 0 and 1, with a higher value indicating a better barrier health status. For example, if all feature parameters are in an ideal state, the barrier health index is close to 1; if most feature parameters are abnormal, the index tends to be close to 0. As a comprehensive indicator, this index can effectively integrate multimodal data, intuitively reflect the overall functional status of the epithelial barrier, and provide a reliable basis for subsequent status assessment and early warning.

[0074] Status assessment and report generation module: The generated barrier health index is compared with the preset barrier health index status threshold, and the specific trend discrimination features selected in the impedance spectrum feature set, permeability dynamic parameter set and cell morphology measurement set are combined with the preset logic rule base to output an assessment report.

[0075] The specific implementation process is as follows: First, the system calls the preset barrier health index state thresholds, which are pre-set based on a large amount of historical experimental data, clinical validation, and expert consensus. The system compares the calculated barrier health index with the preset thresholds level by level to determine a baseline state level. The division of the baseline state level provides a preliminary quantitative assessment of the overall barrier function, laying the foundation for subsequent refined discrimination based on specific trend discrimination features. The baseline state levels include normal stable state, slightly unstable state, moderately unstable state, severely unstable state, and barrier collapse state. The correspondence between the barrier health index state thresholds and the baseline state levels is shown in Table 1:

[0076] Table 1. Correspondence between Barrier Health Index Thresholds and Baseline Status Levels

[0077]

[0078] Next, the system selects specific trend discrimination features and marks abnormal states. These specific trend discrimination features include: extracting the linear regression slope of the steady-state transmembrane resistance value over time from the impedance spectrum feature set, defined as the transmembrane resistance decrease slope; extracting the maximum first derivative of the fluorescence intensity accumulation curve in the initial stage from the permeability kinetic parameter set, defined as the maximum permeability slope; and extracting the continuity index of the fluorescence signal of tight junction proteins from the cell morphology measurement set, defined as the tight junction continuity index. For the transmembrane resistance decrease slope, the system calculates its moving average over multiple consecutive preset time windows (e.g., every 3 time points constitute one window, and the time window length can be dynamically adjusted according to the experimental cycle). If the moving average is consistently lower than a preset negative sensitivity threshold (e.g., -0.05 Ω / ...), the system will determine the trend. If the permeability slope is higher than a preset positive sensitivity threshold (e.g., 0.1 RFU / minute), it is marked as a characteristic anomaly with a continuous negative trend, indicating that the barrier resistance is continuously deteriorating. For the maximum permeability slope, the system calculates its moving average over multiple consecutive preset time windows. If the moving average is higher than a preset positive sensitivity threshold (e.g., 0.1 RFU / minute), it is marked as a characteristic anomaly with a continuous positive growth trend, reflecting an abnormal increase in barrier permeability. For the tight junction continuity index, the system compares the single change in two adjacent monitoring periods (e.g., every 6 hours). If the change exceeds a preset jump sensitivity threshold (e.g., a decrease of more than 0.15), it is marked as a characteristic anomaly with a significant negative jump trend, suggesting that the tight junction structure may be suffering acute damage.

[0079] The system then combines the baseline state level with the characteristic abnormal states and inputs them into the logical rule base for matching queries. The logical rule base is a predefined set of rules based on domain expert knowledge (such as pathophysiological mechanisms and preclinical model validation data) and statistical analysis of a large amount of historical experimental data. It employs "IF-THEN" production rules to map multi-source information (baseline state level and dynamic trend characteristics) to the final state classification. For example, for a barrier with a baseline state level of "slight instability," if its three trend discriminant features—"slope of decrease in transmembrane resistance," "maximum slope of permeability," and "tight connectivity continuity index"—are all marked as abnormal, this indicates that although the current overall index is acceptable, several key aspects of the barrier are continuously deteriorating. Based on this, the system will upgrade the final state classification to "moderate instability" to achieve early risk warning. When the input combination conditions simultaneously satisfy multiple rules, the system will activate a conflict resolution mechanism. The preset priority rules are: 1. Priority based on the number of feature anomalies: Under the same baseline state level, the more feature anomalies triggered, the higher the priority of the corresponding rule; 2. Priority based on state severity: If the triggered rules point to different final states, the rule pointing to the more severe state has higher priority. Through this mechanism, the system can ensure that for any input combination, it outputs a unique and definite final state classification result.

[0080] Finally, based on the final state classification results, the system maps and generates corresponding graded early warning signals and outputs a structured comprehensive assessment report. This report aims to provide users with a clear overview of the state and decision support. The graded early warning signals use an internationally recognized five-level color coding and text labeling system: green represents a normal stable state, blue represents a slightly unstable state requiring attention, yellow represents a moderately unstable state requiring an early warning, orange represents a severely unstable state triggering an alarm, and red represents a barrier collapse state requiring a severe alarm. The output assessment report is a comprehensive document, with core content including the final state classification conclusions, prominent early warning signal color labels, and a systematic integration of key data and in-depth analysis. The report clearly lists the current barrier health index values, the specific values ​​of each trend discrimination feature, and their anomaly markers, forming a summary of key parameters; it also provides time-series charts of transmembrane resistance, fluorescence permeability curves, and cell morphology images to visually present the data dynamics. In addition, the report will automatically generate a brief mechanism analysis based on the final state and specific abnormal characteristics, such as indicating that "the current state is driven by a continuous decrease in resistance and a significant increase in permeability, suggesting that both the barrier structure and function are damaged"; and will attach preliminary action recommendations according to the state level, from "green" state (no action required), to "yellow" state (increase monitoring frequency), and then to "red" state (immediate manual intervention required for confirmation).

[0081] Model Validation and Dynamic Adjustment Module: During a specific validation cycle, two operations are performed in parallel: First, the current barrier health index is acquired and calculated using the data acquisition and synchronous preprocessing module, the multi-scale biophysical feature extraction module, and the multi-criteria weighted fusion model module, and used as the model calculation value; second, a baseline barrier integrity index at the same time point is acquired through offline biological sampling and analysis methods; the model calculation value is compared with the baseline barrier integrity index to calculate the model prediction deviation; and based on the model prediction deviation, the multi-criteria weighted fusion model is dynamically adjusted.

[0082] The specific implementation process is as follows: During a specific validation cycle, the system executes two validation operations in parallel to evaluate the accuracy and reliability of the multi-criteria weighted fusion model and makes dynamic adjustments accordingly. The first validation operation involves acquiring and calculating the current barrier health index in real time, using the data acquisition and synchronous preprocessing module, the multi-scale biophysical feature extraction module, and the multi-criteria weighted fusion model module as the model calculation value. Specifically, the system synchronously acquires transmembrane resistance time-series data, fluorescent marker permeability time-series data, and cell morphology time-series image data. After quality inspection and time alignment, a synchronous multimodal barrier time-series dataset is generated. Multi-scale biophysical features (including impedance spectrum feature set, permeability dynamic parameter set, and cell morphology measurement set) are then extracted, and the barrier health index is calculated through the multi-criteria weighted fusion model. This index is a normalized scalar value used to quantify the overall functional state of the epithelial barrier. The second validation step involves obtaining baseline barrier integrity indices at the same time point through offline biological sampling and analysis methods. Offline methods include using a transmembrane resistance meter to detect endpoint resistance values, calculating apparent permeability coefficients through fluorescent label permeation experiments, or using immunofluorescence staining combined with microscopic observation to assess the continuity of tight junction proteins. These offline indices serve as the gold standard to validate the accuracy of the model's calculated values.

[0083] The model prediction bias is calculated by comparing the calculated values ​​with the baseline barrier integrity index. The model prediction bias is defined as the absolute or relative difference between the calculated values ​​and the baseline barrier integrity index, and the specific formula is: Absolute Difference or relative difference ;in, This represents the calculated value from the model. This represents the baseline barrier integrity index. The system presets a deviation tolerance threshold (for example, the absolute difference threshold is set to 0.05, and the relative difference threshold is set to 10%). If the calculated model prediction deviation exceeds this threshold, the weight adjustment mechanism is triggered.

[0084] Based on model prediction bias, the system employs a Bayesian update algorithm to dynamically adjust the final fusion weights in the multi-criteria weighted fusion model. In this algorithm, the system treats the final fusion weight vector before adjustment as a prior distribution, which is set as a multivariate Gaussian distribution. Its mean vector is the currently used final fusion weight vector, and the covariance matrix is ​​initialized based on historical data or expert experience. Simultaneously, the system uses the baseline barrier integrity index obtained from offline measurements as the observed value and establishes an observation model: this observed value is considered a Gaussian distribution with the barrier health index calculated by the model as the mean and a preset observation noise variance as the variance. The magnitude of the observation noise variance can be determined based on repeatability data from multiple experiments. Using Bayes' theorem, the system combines the prior distribution with the observation likelihood function to derive the posterior distribution of the final fusion weights. The specific update process is as follows: First, the system obtains the values ​​of all standardized feature parameters at the current time, forming a feature vector. Next, the mean vector of the posterior distribution is calculated using the Bayesian update formula. This calculation integrates the prior weight vector, the prior covariance matrix, the current feature vector, the observation noise variance, and the difference between the model's calculated values ​​and the baseline observations. This calculation essentially corrects the prior weights, with the correction magnitude depending on prior uncertainty, observation noise, and the performance of the current feature parameters. The calculated posterior mean vector serves as the updated final fused weight vector. Subsequently, the system normalizes this vector to ensure that the sum of all weight components is 1, thus obtaining the adjusted final fused weights that can be directly used for the next calculation of the Barrier Health Index. This complete dynamic adjustment process allows the model to adaptively calibrate the weight allocation based on offline validation data, continuously improving the accuracy and robustness of the Barrier Health Index prediction, ultimately ensuring that the evaluation results are consistent with the actual biological state.

[0085] This specific embodiment is merely an explanation of the present invention and is not intended to limit the invention. After reading this specification, those skilled in the art can make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they are within the scope of the claims of the present invention.

Claims

1. A data fusion modeling system for airway epithelial barrier status assessment, characterized by, The method comprises the following steps: a data acquisition and synchronization preprocessing module: collecting time-series raw data in the physiological microenvironment of the airway epithelial cell culture model, the time-series raw data including synchronously collected transmembrane resistance time-series data, fluorescent marker permeability time-series data, and cell morphology time-series image data; performing quality inspection and outlier rejection on the time-series raw data, and generating a synchronized multi-modal barrier time-series data set after aligning the time reference; a multi-scale biophysical feature extraction module: extracting and calculating multi-scale biophysical features capable of representing the structure and function of the epithelial barrier from the synchronized multi-modal barrier time-series data set; the multi-scale biophysical features include impedance spectrum feature sets extracted from transmembrane resistance time-series data, permeability kinetic parameter sets extracted from fluorescent marker permeability time-series data, and cell morphology metric sets extracted from cell morphology time-series image data; a multi-criteria weighted fusion model module: constructing a multi-criteria weighted fusion model, taking the impedance spectrum feature sets, permeability kinetic parameter sets, and cell morphology metric sets as inputs to generate a comprehensive and quantitative barrier health index; the barrier health index is a normalized scalar value that intuitively reflects the overall functional state of the epithelial barrier; a state evaluation and report generation module: comparing the generated barrier health index with preset barrier health index state thresholds, and combining selected specific trend discrimination features in the impedance spectrum feature sets, permeability kinetic parameter sets, and cell morphology metric sets, outputting an evaluation report through a preset logic rule base; in the data acquisition and synchronization preprocessing module, the step of collecting time-series raw data in the physiological microenvironment of the airway epithelial cell culture model comprises: collecting transmembrane resistance time-series data containing complex impedance information measured by multi-frequency alternating current excitation; collecting fluorescent marker permeability time-series data by monitoring the fluorescence intensity accumulation curve of the fluorescence intensity on the basal side over time after introducing a fluorescent marker at a specific time point; collecting cell morphology time-series image data containing high-resolution cell images under bright field and / or specific fluorescent markers obtained by timed shooting; the state evaluation and report generation module specifically comprises the following steps: calling preset barrier health index state thresholds, the barrier health index state thresholds including health state thresholds, mild impairment state thresholds, and moderate impairment state thresholds; comparing the barrier health index with the health state thresholds, mild impairment state thresholds, and moderate impairment state thresholds to determine a reference state level; the reference state level includes normal stable state, slight instability state, moderate instability state, severe instability state, and barrier collapse state; selecting specific trend discrimination features and performing feature abnormal state labeling operations; combining the reference state level and the feature abnormal state to input into the logic rule base for matching query, and performing conflict resolution according to the preset priority rules to output a unique and determined final state classification result; According to the final state classification result, a corresponding hierarchical early warning signal is generated, and an evaluation report composed of the final state classification result and the hierarchical early warning signal is output.

2. The data fusion modeling system for airway epithelial barrier status evaluation of claim 1, wherein, Further comprising steps: The model verification and dynamic adjustment module: in a specific verification period, two operations are performed in parallel: one is to obtain and calculate the current barrier health index according to the data acquisition and synchronous preprocessing module, the multiscale biophysical feature extraction module and the multi-criteria weighted fusion model module, as the model calculation value; the second is to obtain the reference barrier integrity index at the same time point through offline biological sampling and analysis method; the model calculation value is compared with the reference barrier integrity index, and the model prediction deviation is calculated; based on the model prediction deviation, the multi-criteria weighted fusion model is dynamically adjusted.

3. The data fusion modeling system for airway epithelial barrier status evaluation of claim 1, wherein, In the data acquisition and synchronous preprocessing module, the steps of quality inspection and outlier rejection of the time series raw data include: For the transmembrane resistance time series data, the signal-to-noise ratio and the phase angle are calculated, and the data segments with signal-to-noise ratio below the first preset threshold or phase angle beyond the reasonable physiological range are rejected; for the fluorescence marker permeability time series data, the baseline stability is checked and whether there is signal saturation is judged, and the data segments with baseline drift exceeding the second preset threshold or signal saturation are rejected; for the cell morphology time series image data, the focusing clarity is evaluated by calculating the image gradient, and the image quality is evaluated by calculating the local contrast, and the image frames with clarity below the third preset threshold or contrast below the fourth preset threshold are marked and rejected.

4. The data fusion modeling system for airway epithelial barrier status evaluation of claim 1, wherein, In the multiscale biophysical feature extraction module, the steps of extracting and calculating the multiscale biophysical features capable of representing the structure and function of the epithelial barrier from the synchronous multi-modal barrier time series data set include: From the transmembrane resistance time series data, the sliding window average of the transmembrane resistance steady-state value, the linear regression slope of the time variation, and the ratio of the impedance amplitude under high-frequency and low-frequency excitation are extracted to form an impedance spectrum feature set; Nonlinear fitting is performed on the fluorescence intensity accumulation curve in the fluorescence marker permeability time series data to extract the time required for the fluorescence intensity accumulation curve to reach half of the plateau, the maximum first derivative value of the curve in the initial stage, and the facilitated diffusion and active transport rate constants obtained by fitting with a two-compartment model to form a permeability kinetic parameter set; Image segmentation and feature analysis are performed on the cell morphology time series image data to quantitatively calculate the cell boundary length per unit area, the continuity index of the intercellular tight junction protein fluorescence signal, and the long-width ratio distribution variance of the cell nucleus to form a cell morphology measurement set; The feature parameters in the impedance spectrum feature set, the permeability kinetic parameter set and the cell morphology measurement set are standardized to eliminate dimensional differences, and together constitute the multiscale biophysical features.

5. The data fusion modeling system for airway epithelial barrier status evaluation of claim 4, wherein, The multi-criteria weighted fusion model module specifically includes steps: An initial weight distribution based on the analytic hierarchy process is performed to assign an initial weight coefficient to each feature parameter in the impedance spectrum feature set, the permeability kinetic parameter set and the cell morphology measurement set; The principal component analysis model is constructed, the multi-scale biophysical characteristics after normalization are dimensionally reduced, and the first principal component is extracted; wherein, the value is determined by the cumulative variance contribution rate exceeding the preset percentage threshold; the variance contribution rate of each principal component is proportionally distributed back to the original characteristic parameters according to the load of the characteristic parameters, and the contribution weight of each characteristic parameter based on data variability is obtained; The initial weight coefficient of each characteristic parameter is combined with the contribution weight, and the final fusion weight of each characteristic parameter is calculated; The value of each characteristic parameter is normalized by the minimum-maximum normalization method to obtain its standardized value; based on the standardized values of all characteristic parameters and the corresponding final fusion weight, weighted fusion calculation is performed to generate a barrier health index ranging from 0 to 1.

6. The data fusion modeling system for airway epithelial barrier status evaluation of claim 5, wherein, In the multi-criteria weighted fusion model module, the initial weight distribution based on the analytic hierarchy process is performed, and an initial weight coefficient is assigned to each characteristic parameter in the impedance spectrum feature set, the permeability dynamics parameter set and the cell morphology measurement set. A hierarchical model for evaluating the barrier health state is constructed, the hierarchical model takes the barrier health index as the target layer, takes the impedance spectrum feature set, the permeability dynamics parameter set and the cell morphology measurement set as the criterion layer, and takes the characteristic parameters in each feature set as the scheme layer; Based on the preset judgment scale, the judgment matrix of the criterion layer to the target layer and the scheme layer to the criterion layer is constructed; the judgment matrix is subjected to consistency check, the characteristic vector is calculated after passing, and the characteristic vector is normalized to assign an initial weight coefficient to each characteristic parameter in the impedance spectrum feature set, the permeability dynamics parameter set and the cell morphology measurement set.

7. The data fusion modeling system for airway epithelial barrier status evaluation of claim 1, wherein, In the state evaluation and report generation module: The specific trend discrimination features include: extracting the linear regression slope of the transmembrane resistance steady-state value changing with time from the impedance spectrum feature set, defined as the transmembrane resistance decline slope; extracting the maximum first derivative value of the fluorescence intensity accumulation curve in the initial stage from the permeability dynamics parameter set, defined as the permeability maximum slope; extracting the continuity index of the intercellular tight junction protein fluorescence signal from the cell morphology measurement set, defined as the tight junction continuity index; The feature abnormal state marking operation includes: For the transmembrane resistance decline slope, if the moving average value in the continuous multiple preset time windows is lower than the preset negative sensitive threshold, it is marked as a feature abnormal state with a sustained negative development trend; for the permeability maximum slope, if the moving average value in the continuous multiple preset time windows is higher than the preset positive sensitive threshold, it is marked as a feature abnormal state with a sustained positive growth trend; for the tight junction continuity index, if the single change amount in adjacent two monitoring periods exceeds the preset jump sensitive threshold, it is marked as a feature abnormal state with a significant negative jump trend.

8. The data fusion modeling system for airway epithelial barrier status evaluation of claim 5, wherein, The model verification and dynamic adjustment module specifically includes the steps of: In a specific verification period, the first verification operation and the second verification operation are performed in parallel; wherein the first verification operation is to obtain and calculate the current barrier health index as a model calculation value according to the data acquisition and synchronization preprocessing module, the multiscale biophysical feature extraction module and the multi-criteria weighted fusion model module; the second verification operation is to obtain the reference barrier integrity index at the same time point through an offline biological sampling and analysis method; The model calculation value is compared with the reference barrier integrity index to calculate a model prediction deviation; wherein the model prediction deviation is an absolute difference or a relative difference between the model calculation value and the reference barrier integrity index; Based on the model prediction deviation, a Bayesian updating algorithm is used to dynamically adjust the final fusion weight in the multi-criteria weighted fusion model; wherein the Bayesian updating algorithm takes the final fusion weight before adjustment as a prior distribution, takes the reference barrier integrity index as an observation value, and obtains the updated final fusion weight through inference.

Images