Lung health monitoring and pneumoconiosis risk assessment method

Abstract

The invention discloses a lung health monitoring and pneumoconiosis risk assessment method. The method comprises the following steps: establishing a quantitative mapping model of biomarker concentration and pneumoconiosis development stage; constructing an exhalant identification sensor array, and establishing a conduction model of response signals and a pneumoconiosis development stage; collecting sample data; preprocessing the data; constructing a dual-path feature enhancement strategy to perform feature extraction, and fusing time domain features and frequency domain features based on feature dimensions determined by a conduction model to construct a more comprehensive high-dimensional feature space; constructing a multi-modal diagnosis model by taking a long-short term memory network or a Transform network as a core structure, and training the multi-modal diagnosis model by utilizing the training set and the test set to obtain a lung health assessment model; and performing online monitoring and outputting a pneumoconiosis risk assessment result. According to the method, the lung health condition of the person to be monitored can be efficiently and accurately evaluated, early warning can be carried out according to the actual risk condition, and reliable technical support can be provided for early prevention and diagnosis of pneumoconiosis and daily health management.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent monitoring, and particularly relates to a lung health monitoring and pneumoconiosis risk assessment method. BACKGROUND

[0002] Pneumoconiosis is the most serious occupational lung disease in the world, which is a systemic disease mainly caused by diffuse fibrosis of lung tissue due to long-term inhalation of productive dust. In many industries such as mining, construction, mechanical processing and metal smelting, a large number of people are exposed to high-concentration dust environment for a long time, resulting in a high incidence of pneumoconiosis.

[0003] Pneumoconiosis is characterized by incurability, latency and long disease cycle, and can only be inhibited as soon as possible by early detection and early treatment. At present, the diagnosis of pneumoconiosis mainly relies on X-ray chest radiography combined with expert reading. However, medical imaging is difficult to make an accurate diagnosis of early pneumoconiosis, and the heterogeneity of lung deposits formed by contact with different dusts also increases the difficulty of imaging diagnosis. Therefore, it is urgent to develop an economic, reliable and non-invasive lung health monitoring and pneumoconiosis risk assessment to realize early assessment and risk warning of the lung health of miners. SUMMARY

[0004] In view of the problems in the prior art, the application provides a lung health monitoring and pneumoconiosis risk assessment method, which has high intelligent degree. The method can reveal the dynamic evolution law of multi-omics exhaled gas biomarkers in the development process of pneumoconiosis by introducing high-frequency quartz crystal microbalance array sensing technology combined with multi-element information fusion and artificial intelligence recognition model, can realize efficient and accurate assessment of the lung health status of the monitored person, and can give early warning according to the actual risk situation, and can provide reliable technical support for early prevention, diagnosis and daily health management of pneumoconiosis.

[0005] In order to achieve the above purpose, the application provides a lung health monitoring and pneumoconiosis risk assessment method, which comprises the following steps: Step 1: Establishing a quantitative mapping model of biomarker concentration and pneumoconiosis development stage; S11: Cross-sectional sample collection operations are carried out at different places, and exhaled gas samples of patients with different mines, different types of work and different disease stages are collected; S12: Analyze the metabolome characteristics of volatile organic compounds in the exhaled air sample by gas chromatography-mass spectrometry, detect the proteome expression profile in the exhaled air sample by liquid chromatography-mass spectrometry, and analyze the characteristic changes of free DNA and miRNA in the exhaled air sample by metagenomic sequencing; from the three omics levels of metabolomics, proteomics and genomics, a variety of biomarkers associated with the progression of pneumoconiosis are screened out, and a quantitative mapping model of the concentration of a variety of biomarkers and different stages of pneumoconiosis development is established; Step two: Construct an exhaled material recognition sensor array and establish a conduction model of response signal and pneumoconiosis development stage; S21: Establish a high-frequency QCM performance prediction model by combining the influence of electrode thickness, electrode material and electrode structure on the working frequency and quality factor of the quartz crystal microbalance chip, use the high-frequency QCM performance prediction model to predict the preparation parameters that affect the sensitivity and stability of the chip, and prepare the quartz crystal microbalance sensor based on the preparation parameters, and then construct an exhaled material recognition sensor array based on the quartz crystal microbalance sensor; S22: Place the exhaled material recognition sensor array in an environment with a variety of biomarker concentrations, obtain a variety of response signals, and use the standard curve method to establish a conversion model of response signal and biomarker concentration, and then based on the quantitative mapping model and the conversion model, a conduction model of a variety of response signals and different stages of pneumoconiosis development is established; Step three: Sample data collection; Use the exhaled material recognition sensor array to collect a large amount of exhaled air samples from confirmed pneumoconiosis patients, dust-exposed populations and healthy controls, and obtain a large amount of response signals; based on the large amount of response signals, a sample data set is constructed, and the sample data set is divided into a training set and a test set according to a set proportion; Step four: Data preprocessing; Based on the centering processing and baseline dynamic calibration technology, the sample data is denoised to remove environmental background interference and non-specific signals; Step five: Construct a double-path feature enhancement strategy for feature extraction; S51: Extract the signal peak value, valley value, response curve area, signal slope, response time and recovery time characteristics in the sample data by the sliding window method, and use the Savitzky-Golay smoothing filter to remove random noise to obtain time domain characteristics; S52: Use fast Fourier transform to obtain the frequency spectrum energy distribution of the sample data, and further use continuous wavelet transform to analyze the time-frequency characteristics of the drift signal to extract frequency domain characteristics in different frequency ranges; S53: Fuse the time domain characteristics and frequency domain characteristics based on the feature dimension determined by the conduction model to construct a more comprehensive high-dimensional feature space; Step six: constructing a lung health assessment model based on a machine learning method; S61: constructing a multi-modal diagnostic model with a long short-term memory network or a Transformer network as the core structure, training the multi-modal diagnostic model using the training set and the test set, and integrating the quantitative relationship established by the conduction model into the training process of the multi-modal diagnostic model; in the training process, a self-adaptive gradient optimization algorithm is used to dynamically adjust the model parameters in real time, and a periodic self-supervised learning mechanism is used to continuously correct the model output, so that the model parameters are adaptively optimized and updated in the prediction process, realizing accurate tracking and real-time compensation of the drift trend; at the same time, combined with the transfer learning technology, the model parameters are pre-trained through historical data collected under different environmental conditions, so as to quickly adapt to and handle the differences in signal drift under different environments, significantly enhancing the generalization ability and robustness of the drift compensation model; at the same time, the test set is used for verification, and the mean square error, root mean square error, accuracy and stability evaluation indexes are used to comprehensively and quantitatively evaluate the real-time performance, accuracy and robustness of the model, ensuring that the drift compensation mechanism works cooperatively with the exhaled substance recognition sensor array; at the same time, the reinforcement learning technology is introduced to construct an intelligent decision-making module of Deep Q-Learning or Actor-Critic algorithm, which monitors the output deviation, drift trend and compensation effect of the exhaled substance recognition sensor array in real time, and defines them as reward feedback signals, and automatically selects the best model parameters and adjustment direction between different states through intelligent decision-making strategies to maximize the accuracy and stability of drift compensation; finally, the lung health assessment model is obtained; Step seven: online monitoring; S71: making the detection personnel exhale gas to the exhaled substance recognition sensor array, and the exhaled substance recognition sensor array outputs real-time response signals after contacting the exhaled gas; S72: sequentially performing denoising and feature extraction on the real-time response signals to obtain a high-dimensional feature space; S73: inputting the high-dimensional feature space as input data into the lung health assessment model for prediction, and finally outputting a dust lung risk assessment result, and generating risk warning information based on the risk situation of the assessment result.

[0006] As a preferred, in S11 of step one, the collection process of the exhaled gas sample is as follows: S11-1: sample grouping; recruit the required number of people in coal mining enterprises and occupational disease prevention and treatment hospitals, and divide the detection objects into a dust lung group, a healthy control group and a professional exposure control group, wherein the dust lung group is a dust-exposed worker diagnosed with different stages of dust lung, the healthy control group is a healthy person without dust exposure history, and the professional exposure control group is a worker exposed to dust but not diagnosed with dust lung; S12-2: Sample collection; first, standardize the behavior of the detection object, then use a polytetrafluoroethylene gas bag to collect the exhaled breath of the object, collect multiple breaths, collect 3L of exhaled breath; use a constant temperature condensation device to collect 10 mL of exhaled breath condensate from each subject.

[0007] As a preferred, in step one S12, the process of establishing a quantitative mapping model of biomarker concentration and pneumoconiosis development stage is as follows: S12-1: Sample detection and analysis; gas chromatography-mass spectrometry and liquid chromatography-mass spectrometry are used to analyze inorganic gas molecules and volatile organic compounds in exhaled breath; unlabeled quantitative proteomics technology is used, combined with nanoliter liquid chromatography-tandem mass spectrometry to analyze protein / polypeptide expression profiles in exhaled breath condensate; exosome miRNA sequencing and DNA methylation chip analysis are used to analyze nucleic acid components in exhaled breath; biomarkers associated with pneumoconiosis are identified from three omics levels of metabolomics, proteomics, and genomics, and differential molecules significantly related to pneumoconiosis progression are screened; S12-2: Data integration and bioinformatics analysis; construct a pneumoconiosis multi-omics correlation matrix, use weighted gene co-expression network analysis to identify co-variable modules, screen core marker groups with causal correlation based on causal reasoning, and establish a quantitative mapping model of biomarker concentration and pneumoconiosis development stage.

[0008] As a preferred, in step one S12-2, object behavior standardization includes fasting, mouthwash, sitting rest for 10 minutes, and excluding a series of physiological and pathological factors interference.

[0009] Further, in order to eliminate errors and improve the consistency of measurement results, after denoising processing, data standardization methods based on minimum-maximum standardization and Z-score standardization are used to process the data.

[0010] The application provides a dust lung exhaled gas biomarker sensing technology based on a high-frequency quartz crystal microbalance array and a multi-element information fusion dust lung risk early warning method. On this basis, a conversion model of the response signal and the biomarker concentration is established by using a standard curve method, and a conduction model of multiple response signals and different dust lung development stages is established based on the quantitative mapping model and the conversion model, so that the feature dimension determined based on the conduction model can provide a basis for feature extraction and fusion of the double-path feature enhancement strategy, ensure that the feature extraction process is highly relevant to the dust lung pathological mechanism, and greatly improve the prediction accuracy of the subsequent model. After data preprocessing, environmental background interference and non-specific signals can be effectively removed, and the purity of the obtained signals can be effectively ensured. In addition, in the model training process, an adaptive gradient optimization algorithm is used to dynamically adjust the model parameters in real time, which can realize accurate tracking and real-time compensation of the drift trend. At the same time, combined with the transfer learning technology, the model parameters are pre-trained by using historical data collected under different environmental conditions, so as to quickly adapt to and process the differences in signal drift under different environments, significantly enhance the generalization ability and robustness of the drift compensation model; at the same time, the reinforcement learning technology is introduced, which can maximize the accuracy and stability of drift compensation, and effectively ensure the monitoring accuracy. Finally, the lung health evaluation model is used for real-time monitoring of the exhaled gas of the detected person, which can efficiently and accurately evaluate the lung health status.

[0011] The method has high intelligence and low implementation cost, and can reveal the dynamic evolution law of multiple omics exhaled gas biomarkers in the development process of dust lung, can efficiently and accurately evaluate the lung health status of the monitored person, and can give early warning according to the actual risk situation, and can provide reliable technical support for early prevention, diagnosis and daily health management of dust lung disease. BRIEF DESCRIPTION OF DRAWINGS

[0012] Figure 1 is the overall flowchart of the evaluation method in the application; Figure 2 This is a flowchart of the screening process for biomarkers in exhaled breath according to the present invention. Detailed Implementation

[0013] The invention will now be further described with reference to the accompanying drawings.

[0014] like Figure 1 As shown, the present invention provides a method for lung health monitoring and pneumoconiosis risk assessment, comprising the following steps: Step 1: Establish a quantitative mapping model between biomarker concentrations and the developmental stages of pneumoconiosis; S11: Conduct cross-sectional sample collection operations at different locations to collect exhaled air samples from patients in different mines, different occupations, and different stages of the disease; this will ensure that the exhaled air samples can cover a wider range of people, which is conducive to the subsequent construction of a more comprehensive and accurate biomarker spectrum of pneumoconiosis exhaled air. S12: The metabolomic characteristics of volatile organic compounds (VOCs) in exhaled breath samples were analyzed using gas chromatography-mass spectrometry (GC-MS), the protein / peptide expression profiles in exhaled breath samples were detected using liquid chromatography-mass spectrometry (LC-MS), and the characteristic changes of free DNA and miRNA in exhaled breath samples were analyzed using metagenomic sequencing. Multiple biomarkers associated with the progression of pneumoconiosis were screened from the three omics levels of metabolomics, proteomics, and genomics. At the same time, quantitative mapping models of the concentration of multiple biomarkers and different stages of pneumoconiosis development were established. Step 2: Construct an array of exhaled matter recognition sensors and establish a transmission model of response signals and the development stages of pneumoconiosis; S21: Combining the influence of electrode thickness, electrode material, and electrode structure on the working frequency and quality factor of quartz crystal microbalance chip, a high-frequency QCM performance prediction model is established. The high-frequency QCM performance prediction model is used to predict the preparation parameters that affect the chip sensitivity and stability. Based on the preparation parameters, the quartz crystal microbalance sensor is prepared. Then, an exhaled matter identification sensor array based on the quartz crystal microbalance sensor is constructed to achieve high specificity and high sensitivity identification of pneumoconiosis markers. S22: Place the exhaled matter identification sensor array in an environment with multiple biomarker concentrations to obtain the corresponding multiple response signals, and use the standard curve method to establish a conversion model between the response signals and biomarker concentrations. Then, based on the quantitative mapping model and the conversion model, establish a transmission model between multiple response signals and different stages of pneumoconiosis development. Step 3: Sample data collection; A massive amount of exhaled air samples were collected from confirmed pneumoconiosis patients, people with occupational dust exposure, and healthy controls using an exhaled air recognition sensor array to obtain a massive amount of response signals; a sample dataset was constructed based on the massive amount of response signals, and the sample dataset was divided into training set and test set according to a set ratio; Step four: data preprocessing; Based on the centralized processing and baseline dynamic calibration technology, the sample data is denoised to remove environmental background interference and non-specific signals; As a preferred, the data after denoising is further processed based on the data standardization method of minimum-maximum standardization and Z-score standardization, to eliminate the system error between different devices and sensors, and improve the consistency of measurement results.

[0015] Step five: construct a double-path feature enhancement strategy for feature extraction; S51: Extract the signal peak value, valley value, response curve area, signal slope, response time and recovery time characteristics in the sample data by sliding window method, and use Savitzky-Golay smoothing filter to remove random noise, to improve the accuracy and stability of time domain feature extraction, and obtain time domain features; S52: Use fast Fourier transform (FFT) to obtain the frequency spectrum energy distribution of the sample data, and further use continuous wavelet transform (CWT) to analyze the time-frequency characteristics of the drift signal, and extract the frequency domain features in different frequency ranges; S53: Based on the feature dimension determined by the conduction model, the time domain features and frequency domain features are fused to construct a more comprehensive high-dimensional feature space; Step six: construct a lung health assessment model based on machine learning method; Long Short-Term Memory Networks or Transformers A multimodal diagnostic model is constructed with a network as its core structure to fully reveal the association mechanism between candidate biomarkers and pulmonary fibrosis. The model is trained using training and testing sets, and the quantitative relationships established by the transmission model are integrated into the training process. During training, an adaptive gradient optimization algorithm is used to dynamically adjust model parameters in real time, and a periodic self-supervised learning mechanism is used to continuously correct the model output. This allows the model parameters to adaptively optimize and update during prediction, achieving accurate tracking and real-time compensation of drift trends. Simultaneously, transfer learning technology is combined to pre-train model parameters using historical data collected under different environmental conditions, enabling rapid adaptation and handling of signal drift differences in various environments, significantly enhancing the generalization ability and robustness of the drift compensation model. The model is validated using a test set, employing mean squared error (MSE), root mean square error (RMSE), accuracy, and stability evaluation metrics to comprehensively and quantitatively assess the model's real-time performance, accuracy, and robustness. This ensures that the drift compensation mechanism and the exhaled matter recognition sensor array work together to maintain long-term stability and high-precision operation in complex real-world environments. Furthermore, reinforcement learning technology is introduced to construct a Deep Learning module. The intelligent decision-making module of the Q-Learning or Actor-Critic algorithm monitors the output deviation, drift trend and compensation effect of the exhaled matter recognition sensor array in real time, and defines it as a reward feedback signal. Through intelligent decision-making strategy, it automatically selects the best model parameters and adjusts the direction between different states to maximize the accuracy and stability of drift compensation; finally, a lung health assessment model is obtained. Step 7: Online monitoring; S71: The operator exhales gas into the exhaled gas identification sensor array, and the exhaled gas identification sensor array outputs a real-time response signal after contacting the exhaled gas; S72: Denoising and feature extraction are performed sequentially on the real-time response signal to obtain a high-dimensional feature space; S73: The high-dimensional feature space is input into the lung health assessment model for prediction. Finally, the pneumoconiosis risk assessment result is output, and risk warning and reminder information is generated based on the risk status of the assessment result.

[0016] As a preferred option, such as Figure 2 As shown, in step S11 of step one, the process of collecting exhaled breath samples is as follows: S11-1: Sample grouping; The required number of people were recruited from coal mining enterprises and occupational disease prevention and treatment hospitals. The subjects were divided into a pneumoconiosis group, a healthy control group, and an occupational exposure control group. The pneumoconiosis group consisted of workers diagnosed with dust exposure at different stages of pneumoconiosis development. The healthy control group consisted of healthy people with no history of dust exposure. The occupational exposure control group consisted of workers exposed to dust but not diagnosed with pneumoconiosis. S12-2: Sample collection; first standardize the detection object behavior, then use a polytetrafluoroethylene gas bag to collect the exhaled breath of the object, collect multiple breaths, and collect 3L of exhaled breath; use a constant temperature condensation device (-20°C, flow rate 15 L / min) to collect 10 mL of exhaled breath condensate from each subject.

[0017] As a preferred, in step one S12, the process of establishing a quantitative mapping model of biomarker concentration and pneumoconiosis development stage is as follows: S12-1: Sample detection and analysis; use gas chromatography-mass spectrometry (GC-MS) and liquid chromatography-mass spectrometry (LC-MS) to analyze inorganic gas molecules and volatile organic compounds in exhaled breath; use label-free quantitative proteomics technology combined with nanoscale liquid chromatography-tandem mass spectrometry (NanoLC-MS / MS) to analyze protein / polypeptide group expression profiles in exhaled breath condensate; use exosome miRNA sequencing (small RNA-Seq) and DNA methylation chip analysis to analyze nucleic acid components in exhaled breath; identify biomarkers associated with pneumoconiosis from three omics levels of metabolomics, proteomics, and genomics, and screen for differentially expressed molecules significantly associated with pneumoconiosis progression; Analyze the metabolomics characteristics of volatile organic compounds (VOCs) in exhaled breath using gas chromatography-mass spectrometry, detect protein / polypeptide group expression profiles in exhaled breath condensate using liquid chromatography-mass spectrometry, and use metagenomic sequencing to analyze characteristic changes in free DNA and miRNA in exhaled breath condensate; S12-2: Data integration and bioinformatics analysis; construct a pneumoconiosis multi-omics correlation matrix, use weighted gene co-expression network analysis (WGCNA) to identify co-variance modules, screen for core marker groups with causal associations based on causal reasoning (Bayesian network, Mendelian randomization), and establish a quantitative mapping model of biomarker concentration and pneumoconiosis development stage.

[0018] As a preferred, in step one S12-2, object behavior standardization includes fasting, mouthwash, sitting quietly for 10 minutes, and excluding interference from a series of physiological and pathological factors.

[0019] As a preferred, in step two, the process of constructing an exhaled breath recognition sensor array based on a quartz crystal microbalance sensor is as follows: Based on the equivalent mass sensitivity model, use multi-physics simulation software to perform pre-simulation of electrode design and optimization to predict the impact of electrode thickness and material selection on QCM chip performance, and comprehensively analyze the specific impact of electrode thickness and material selection on the working frequency and Q value of the QCM chip. This study focuses on high-frequency QCM chips fabricated under precisely controlled laboratory conditions. By accurately controlling electroplating parameters, the thickness of the electrode plating layer is precisely controlled. Micro-nano fabrication techniques are employed to precisely fabricate complex electrode layouts. The regulatory mechanisms of fabrication parameters, such as electrode plating thickness and electrode layout, on chip performance are analyzed. Advanced physical characterization techniques, including scanning electron microscopy (SEM) and X-ray diffraction (XRD), are used to analyze the microstructure and electrode surface characteristics of the QCM chip in detail. Simultaneously, network analyzers and impedance analyzers are used for electrical testing to measure the chip's frequency response and Q-value. Statistical and machine learning techniques, such as random forests or support vector machines (SVM), are used to analyze the experimental data. Based on the analysis results, a high-frequency QCM performance prediction model is established. This model will be used to predict the impact of different fabrication parameters on the chip's sensitivity and stability, thereby guiding the optimization of electrode materials and design.

[0020] Targeting specific exhaled biomarkers for pneumoconiosis, this study employs a strategy combining multi-scale computation and experimental validation to precisely screen and analyze the response mechanisms of targeted sensor materials for pneumoconiosis-specific exhaled biomarkers. First, in the material screening stage, analysis of variance (ANOVA) was used to preliminarily evaluate the differences in response to the target analyte among different materials. Then, partial least squares discriminant analysis (PLS-DA) was used to clarify the classification contribution weight of each sensor material to the pneumoconiosis diagnostic model. Principal component analysis (PCA) and linear discriminant analysis (LDA) were used to cluster the cross-response data of the sensor array to quantify the discriminative power. Finally, SHAP (SHapley Additive Explanations) analysis was used to clarify the specific contribution and mechanism of action of a single sensor in the diagnostic model prediction, analyzing the contribution of each sensor to the classification model and eliminating those with a contribution < With 5% redundant units, a complete multi-level quantitative system for material evaluation is formed. Secondly, molecular docking simulation is used to predict the molecular-level interaction between candidate sensing materials and target volatile organic compounds and biomolecules, clarify the interaction sites and binding modes, and density functional theory (DFT) is used to analyze the binding energy, orbital electron distribution and charge transfer characteristics of the material and the target molecules in depth, so as to predict the specific recognition ability of the material and initially screen out sensing candidate materials with specific recognition performance. Based on the above prediction results, candidate materials are synthesized by solvent thermal method, chemical vapor deposition (CVD) and other preparation processes, and the microstructure and crystal characteristics of the materials are characterized by transmission electron microscopy (TEM) and X-ray diffraction (XRD). The elemental composition and chemical state of the material surface are analyzed by X-ray photoelectron spectroscopy (XPS), the topological structure and fine spatial distribution characteristics of active sites on the material surface are measured by atomic force microscopy (AFM), and the response sensitivity, selectivity and stability of the sensing material to the target analyte are determined by fluorescence response, redox response, electrical conductivity and oscillation response dynamics testing. The non-covalent interaction mechanisms such as hydrogen bonding, charge transfer and electrostatic interaction are determined by dynamic response data, and the response signal generation and conduction model is established. Based on the above quantitative contribution, intelligent algorithms such as genetic algorithm (GA) or particle swarm optimization (PSO) are further applied, with classification accuracy (Accuracy) and feature redundancy as the fitness function, to iteratively select the optimal subset and optimize the sensing array structure combination, in order to achieve high specificity and high sensitivity recognition of pneumoconiosis markers, and to construct an exhaled breath recognition sensor array based on a quartz crystal microbalance sensor.

[0021] The application provides a dust pneumo exhaled gas biomarker sensing technology based on a high-frequency quartz crystal microbalance array and a multi-element information fusion dust pneumo risk early warning method. On this basis, a conversion model of the response signal and the biomarker concentration is established by using a standard curve method, and a conduction model of multiple response signals and different dust pneumo development stages is established based on the quantitative mapping model and the conversion model, so that the feature dimension determined based on the conduction model can provide a basis for feature extraction and fusion of the double-path feature enhancement strategy, ensure that the feature extraction process is highly relevant to the dust pneumo pathological mechanism, and greatly improve the prediction accuracy of the subsequent model. After data preprocessing, environmental background interference and non-specific signals can be effectively removed, and the purity of the obtained signals can be effectively ensured. Then, the double-path feature enhancement strategy is used for feature extraction, which can process different dimensional feature information in parallel, significantly improving the feature extraction capability in complex scenes. At the same time, combined with the transfer learning technology, the model parameters are pre-trained by using historical data collected under different environmental conditions, so as to quickly adapt to and process the differences in signal drift under different environments, significantly enhance the generalization ability and robustness of the drift compensation model; at the same time, the reinforcement learning technology is introduced, which can maximize the accuracy and stability of drift compensation, effectively ensuring the monitoring accuracy. Finally, the lung health evaluation model is used for real-time monitoring of the exhaled gas of the detected person, which can efficiently and accurately evaluate the lung health status.

[0022] The method has high intelligence and low implementation cost, and can reveal the dynamic evolution law of multi-omics exhaled gas biomarkers in the development process of dust pneumo, efficiently and accurately evaluate the lung health status of the monitored person, and give early warning according to the actual risk, which can provide reliable technical support for early prevention, diagnosis and daily health management of dust pneumo.

Claims

1. A method for monitoring lung health and assessing the risk of pneumoconiosis, characterized in that, Includes the following steps: Step 1: Establish a quantitative mapping model between biomarker concentrations and the developmental stages of pneumoconiosis; S11: Conduct cross-sectional sample collection operations at different locations to collect exhaled breath samples from patients in different mines, different occupations, and different stages of their illness; S12: The metabolomic characteristics of volatile organic compounds in exhaled breath samples were analyzed using gas chromatography-mass spectrometry (GC-MS), the protein / peptide expression profiles in exhaled breath samples were detected using liquid chromatography-mass spectrometry (LC-MS), and the characteristic changes of free DNA and miRNA in exhaled breath samples were analyzed using metagenomic sequencing. Multiple biomarkers associated with the progression of pneumoconiosis were screened from the three omics levels of metabolomics, proteomics, and genomics. At the same time, quantitative mapping models of the concentration of multiple biomarkers and different stages of pneumoconiosis development were established. Step 2: Construct an array of exhaled matter recognition sensors and establish a transmission model of response signals and the development stages of pneumoconiosis; S21: Combining the influence of electrode thickness, electrode material and electrode structure on the working frequency and quality factor of quartz crystal microbalance chip, a high-frequency QCM performance prediction model is established. The high-frequency QCM performance prediction model is used to predict the preparation parameters that affect the chip sensitivity and stability. Based on the preparation parameters, the quartz crystal microbalance sensor is prepared, and then an exhaled matter identification sensor array based on the quartz crystal microbalance sensor is constructed. S22: Place the exhaled matter identification sensor array in an environment with multiple biomarker concentrations to obtain the corresponding multiple response signals, and use the standard curve method to establish a conversion model between the response signals and biomarker concentrations. Then, based on the quantitative mapping model and the conversion model, establish a transmission model between multiple response signals and different stages of pneumoconiosis development. Step 3: Sample data collection; A massive amount of exhaled air samples were collected from confirmed pneumoconiosis patients, people with occupational dust exposure, and healthy controls using an array of exhaled air recognition sensors to obtain a large number of response signals. A sample dataset is constructed based on massive response signals, and the sample dataset is divided into a training set and a test set according to a set ratio. Step 4: Data preprocessing; The sample data is denoised using centralized processing and baseline dynamic calibration techniques to remove environmental background interference and non-specific signals. Step 5: Construct a dual-path feature enhancement strategy for feature extraction; S51: The peak value, valley value, area under the response curve, slope, response time and recovery time features of the sample data are extracted by the sliding window method, and the Savitzky-Golay smoothing filter is used to remove random noise to obtain the time domain features; S52: The spectral energy distribution of the sample data is obtained by using Fast Fourier Transform, and the time-frequency characteristics of the drift signal are further analyzed by continuous wavelet transform to extract frequency domain characteristics in different frequency ranges. S53: Based on the feature dimensions determined by the transmission model, time-domain features and frequency-domain features are fused to construct a more comprehensive high-dimensional feature space; Step Six: Construct a lung health assessment model based on machine learning methods; A multimodal diagnostic model is constructed using a long short-term memory network or a Transformer network as the core structure. The multimodal diagnostic model is trained using training and test sets, and the quantitative relationship established by the transmission model is integrated into the training process of the multimodal diagnostic model. During training, an adaptive gradient optimization algorithm is used to dynamically adjust model parameters in real time, and a periodic self-supervised learning mechanism is used to continuously correct the model output. This allows the model parameters to be adaptively optimized and updated during prediction, achieving accurate tracking and real-time compensation of drift trends. Simultaneously, transfer learning technology is used to pre-train model parameters using historical data collected under different environmental conditions. This allows for rapid adaptation to and handling of signal drift differences under varying environments, significantly enhancing the generalization ability and robustness of the drift compensation model. Furthermore, validation is performed using a test set, employing mean square error, root mean square error, accuracy, and stability evaluation metrics to comprehensively and quantitatively assess the model's real-time performance, accuracy, and robustness, ensuring the drift compensation mechanism works collaboratively with the exhaled matter recognition sensor array. Additionally, reinforcement learning technology is introduced to construct an intelligent decision-making module based on DeepQ-Learning or Actor-Critic algorithms. This module monitors the output deviation, drift trend, and compensation effect of the exhaled matter recognition sensor array in real time, defining these as reward feedback signals. Through intelligent decision-making strategies, the optimal model parameters and adjustment direction are automatically selected between different states to maximize the accuracy and stability of drift compensation. Ultimately, a lung health assessment model was obtained; Step 7: Online monitoring; S71: The operator exhales gas into the exhaled gas identification sensor array, and the exhaled gas identification sensor array outputs a real-time response signal after contacting the exhaled gas; S72: Denoising and feature extraction are performed sequentially on the real-time response signal to obtain a high-dimensional feature space; S73: The high-dimensional feature space is input into the lung health assessment model for prediction. Finally, the pneumoconiosis risk assessment result is output, and risk warning and reminder information is generated based on the risk status of the assessment result.

2. The method for lung health monitoring and pneumoconiosis risk assessment according to claim 1, characterized in that, In step S11 of step one, the process of collecting exhaled breath samples is as follows: S11-1: Sample grouping; The required number of people were recruited from coal mining enterprises and occupational disease prevention and treatment hospitals. The subjects were divided into a pneumoconiosis group, a healthy control group, and an occupational exposure control group. The pneumoconiosis group consisted of workers diagnosed with dust exposure at different stages of pneumoconiosis development. The healthy control group consisted of healthy people with no history of dust exposure. The occupational exposure control group consisted of workers exposed to dust but not diagnosed with pneumoconiosis. S12-2: Sample collection; First, the behavior of the test subjects is standardized, and then the exhaled air of the subjects is collected using a polytetrafluoroethylene gas bag. Multiple breaths are collected continuously, and 3L of exhaled air is collected; 10 mL of exhaled air condensate is collected from each subject using a constant temperature condensation device.

3. The method for lung health monitoring and pneumoconiosis risk assessment according to claim 1, characterized in that, In step S12 of step one, the process of establishing a quantitative mapping model between biomarker concentration and the development stage of pneumoconiosis is as follows: S12-1: Sample detection and analysis; Gas chromatography-mass spectrometry (GC-MS) and liquid chromatography-mass spectrometry (LC-MS) were used to analyze inorganic gas molecules and volatile organic compounds in exhaled breath; Label-free quantitative proteomics technology, combined with nano-liquid chromatography-tandem mass spectrometry (LC-MS / MS), was used to analyze the protein / peptide expression profiles in exhaled breath condensate; Exosomal miRNA sequencing and DNA methylation microarrays were used to analyze the nucleic acid components in exhaled breath; Biomarkers associated with pneumoconiosis were identified from the three omics levels of metabolomics, proteomics, and genomics, and differentially expressed molecules significantly associated with the progression of pneumoconiosis were screened; S12-2: Data integration and bioinformatics analysis; constructing a multi-omics association matrix for pneumoconiosis, using weighted gene co-expression network analysis to identify covariate modules, screening core biomarker groups with causal associations based on causal reasoning, and establishing a quantitative mapping model between biomarker concentration and pneumoconiosis development stage.

4. The method for lung health monitoring and pneumoconiosis risk assessment according to claim 2, characterized in that, In step S12-2 of step one, the normalization of object behavior includes fasting, rinsing the mouth, and sitting quietly for 10 minutes to eliminate interference from a series of physiological and pathological factors.

5. The method for lung health monitoring and pneumoconiosis risk assessment according to claim 1, characterized in that, In step four, after denoising, the data is processed using data normalization methods based on min-max normalization and Z-score normalization.

Images