Intelligent respiratory disease detection system based on lung function measurement indexes

Abstract

The invention discloses an intelligent respiratory disease detection system based on lung function measurement indexes, and relates to the technical field of intelligent medical detection. The problems that diagnosis standards are not uniform and depend on experience, primary doctors are prone to missed diagnosis or misdiagnosis due to insufficient experience, and the efficiency of the detection process is low are solved. According to the method, a machine learning model is constructed based on lung function indexes, parameters are optimized through five-fold cross validation, automatic classification of asthma, COPD and ILD / DPLD is achieved, the overall diagnosis accuracy rate reaches 81.3%, the limitation that traditional diagnosis depends on doctor experience is broken through, and the basic medical diagnosis capacity is especially improved. Healthy people are rapidly eliminated through the FVC / FEV1 ratio, invalid calculation is reduced, meanwhile, multi-model weighted fusion is adopted for complex cases, diagnosis robustness is improved, it is ensured that diagnosis results conform to medical logic, the recognition degree of doctors is enhanced, a data basis is provided for clinical research and curative effect tracking, and respiratory disease diagnosis is promoted to develop towards the direction of automation and precision.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical testing technology, and in particular to an intelligent respiratory disease detection system based on lung function measurement indicators. Background Technology

[0002] Chronic respiratory diseases are mainly divided into obstructive and restrictive respiratory diseases. Among them, the diagnosis of chronic obstructive pulmonary disease (COPD) mainly relies on pulmonary function tests, but current technologies have the following limitations:

[0003] 1. Only chronic obstructive pulmonary disease has a clear gold standard for pulmonary function diagnosis. Other diseases lack quantitative diagnostic indicators and rely on medical imaging or blood gas analysis, which carries the risk of missed diagnosis. Furthermore, pulmonary function indicators need to be judged in combination with clinical experience, and primary healthcare personnel are prone to misdiagnosis due to insufficient experience.

[0004] 2. The time-series characteristics of lung function indicators and the correlation between multiple indicators have not been effectively explored, making it difficult to improve diagnostic accuracy through data-driven approaches. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent respiratory disease detection system based on lung function measurement indicators, to construct a mapping relationship between lung function indicators and disease categories, to achieve automated and accurate diagnosis of respiratory diseases, to achieve automated classification and diagnosis of obstructive and restrictive respiratory diseases, and to significantly improve the respiratory disease diagnosis capabilities of primary healthcare institutions, thereby solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] Intelligent respiratory disease detection systems based on lung function measurement indicators include:

[0008] The data preprocessing module is configured to integrate a pulmonary function testing instrument to acquire patients' pulmonary function test data. At the same time, it collects patients' basic information, calculates the ratio of pulmonary function indicators before and after medication to the predicted values, and performs preprocessing to construct a feature set that meets the model input requirements.

[0009] The model validation and optimization module is configured to train each machine learning model based on the preprocessed feature set, validate the diagnostic results of each machine learning model, identify the accuracy and feasibility of each machine learning model based on the validation results, and determine the optimal classification decision model.

[0010] The intelligent classification decision module is configured to use a defined classification decision model to predict the classification of respiratory diseases and diagnose the patient's respiratory disease category based on the prediction results.

[0011] Furthermore, the data preprocessing module includes:

[0012] The data acquisition unit is configured to establish a data interface with the pulmonary function testing instrument to acquire the patient's pulmonary function test data in real time, including pulmonary function index values ​​before and after medication, and to collect the patient's basic information simultaneously.

[0013] The ratio calculation unit is configured to extract corresponding physiological parameters based on the patient's basic information, determine the patient's expected values, calculate the ratio of each lung function index value before and after medication to the patient's expected values, and construct a one-dimensional feature vector of the lung function index.

[0014] The feature set generation unit is configured to standardize the one-dimensional feature vector and format the processed feature set into a two-dimensional matrix according to the model input requirements. Each row corresponds to a patient sample, each column corresponds to a lung function index, and label vectors corresponding to respiratory disease classifications are generated simultaneously.

[0015] Furthermore, the model validation optimization module includes:

[0016] The model training unit is configured to input the preprocessed feature set into each machine learning model, optimize the model parameters through cross-validation mechanism, and establish a mapping relationship between lung function indicators and disease categories.

[0017] The model evaluation unit is configured to calculate the average accuracy and standard deviation of each machine learning model based on the results of 5 cross-validations, and select machine learning models with good stability that have an accuracy higher than 75% and a standard deviation lower than 5%.

[0018] The model determination unit is configured to select the machine learning model with the highest single-class accuracy based on the screening results as the base model. The base models selected in each round are weighted according to their classification accuracy, and the final classification decision model is generated through weighted fusion to obtain the prediction results and verify whether the model performance has improved.

[0019] Furthermore, the model training unit optimizes the model parameters through a cross-validation mechanism, specifically including:

[0020] Five-fold cross-validation is used to divide the feature set into five equal subsets. In each round, four subsets are selected to form the training set, and the remaining one subset is used as the test set. Five rounds of cross-validation are performed.

[0021] The training set is input into multiple machine learning models in sequence, and the training process is started simultaneously. Each machine learning model is trained iteratively 5 times, and the parameter combination and test set performance index are recorded for each training.

[0022] Furthermore, the model training unit establishes a mapping relationship between lung function indicators and disease categories, expressed by the following formula:

[0023] X = (x1, x2, ..., x N)

[0024] X→y

[0025] In the formula, y represents the label vector corresponding to the respiratory disease classification; X represents the one-dimensional feature vector of the lung function index; x i (i = 1, 2, ..., N) represents the specific lung function indicators.

[0026] Furthermore, the model evaluation unit specifically includes:

[0027] Based on test set performance metrics, the performance of each machine learning model is evaluated using the test set. The accuracy, precision, recall, and F1 score of each machine learning model are calculated, and the confusion matrix of each machine learning model is generated to evaluate the classification performance of each machine learning model for three types of respiratory diseases.

[0028] Furthermore, the model determination unit also includes: acquiring a small amount of newly added lung function data and verifying the accuracy of the classification decision model based on the confusion matrix of the classification decision model.

[0029] Furthermore, the intelligent classification decision-making module includes:

[0030] The model classification unit is configured to perform inference operations on the newly acquired feature set based on the trained classification decision model, and output the disease category prediction result.

[0031] The results output unit is configured to generate a structured diagnostic report based on the prediction results and integrate preliminary screening rules to achieve rapid determination of health status.

[0032] Furthermore, the model classification unit also includes:

[0033] For each lung function test, if FVC / FEV1 > 70%, the patient is directly diagnosed as healthy.

[0034] If FVC / FEV1 < 70%, the model inference process is initiated, and the ratio of each lung function index value before and after medication to the patient's predicted value is obtained. This ratio is then input into the classification decision model to predict the respiratory disease classification and obtain the diagnosis result of the patient's respiratory disease category.

[0035] Furthermore, in the structured diagnostic report, if COPD is predicted, the clinical gold standard of "FEV1 / FVC < 70%" is automatically cited; if asthma is predicted, "bronchodilator test recommended to verify reversible airflow limitation" is suggested.

[0036] Compared with the prior art, the beneficial effects of the present invention are:

[0037] A machine learning model based on lung function indicators was constructed, and parameters were optimized using five-fold cross-validation to achieve automated classification of asthma, COPD, and ILD / DPLD, with an overall diagnostic accuracy of 81.3%. This overcomes the limitations of traditional reliance on physician experience and particularly enhances the diagnostic capabilities of primary healthcare institutions. The FVC / FEV1 ratio is used to quickly exclude healthy individuals, reducing unnecessary calculations. Furthermore, multi-model weighted fusion is employed for complex cases to improve diagnostic robustness, ensure that diagnostic results conform to medical logic, enhance physician acceptance, provide a data foundation for clinical research and efficacy tracking, and promote the automation and precision of respiratory disease diagnosis. Attached Figure Description

[0038] Figure 1 This is a logic diagram of the detection algorithm of the present invention;

[0039] Figure 2 This is a flowchart of the model training process based on five-fold cross-validation of the present invention;

[0040] Figure 3 This is a distribution diagram on a three-dimensional axis formed by the three lung function indicators of this invention;

[0041] Figure 4 This is a confusion matrix diagram of the detection results of the LR, KNN, BNB, GNB, Random Forest, and CNN models of this invention;

[0042] Figure 5 This is a confusion matrix diagram of the detection results of the SVM, KNN, decision tree, XGBoost, LightGBM, and MLP models of the present invention;

[0043] Figure 6 This is a confusion matrix diagram of the detection results of the optimal classification decision model of the present invention;

[0044] Figure 7 This is a block diagram of the intelligent respiratory disease detection system based on lung function measurement indicators of the present invention. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] To address the technical challenges of having a definitive gold standard for pulmonary function diagnosis only for chronic obstructive pulmonary disease (COPD), while other diseases lack quantifiable diagnostic indicators and rely on medical imaging or blood gas analysis (leading to potential missed diagnoses), and the need for comprehensive clinical experience in interpreting pulmonary function indicators, which can result in misdiagnosis by primary care physicians due to insufficient experience, please refer to [link to relevant documentation]. Figure 1-7 This embodiment provides the following technical solution:

[0047] Intelligent respiratory disease detection systems based on lung function measurement indicators include:

[0048] The data preprocessing module is configured to integrate a pulmonary function testing instrument to acquire patients' pulmonary function test data, including indicators such as FEV1, FVC, and PEF. At the same time, it collects patients' basic information, calculates the ratio of pulmonary function indicators before and after medication to the predicted values, and performs preprocessing to construct a feature set that meets the model input requirements.

[0049] The model validation and optimization module is configured to train each machine learning model based on the preprocessed feature set, validate the diagnostic results of each machine learning model, identify the accuracy and feasibility of each machine learning model based on the validation results, and determine the optimal classification decision model for respiratory disease classification and detection.

[0050] The intelligent classification decision module is configured to use a defined classification decision model to predict the classification of respiratory diseases and diagnose the patient's respiratory disease category based on the prediction results, including one of asthma, COPD, or interstitial lung disease (ILD / DPLD).

[0051] In this embodiment, the data comes from publicly available research, including pulmonary function test data collected by the research team from patients with a certain type of respiratory disease in hospitals. A total of 1325 patient data points are included, and each data point contains the information shown in the table below:

[0052] Table 1 explains the meaning of each item in each data entry from the public dataset.

[0053]

[0054]

[0055] In this embodiment, the lung function indicators that can be used as features in the above dataset mainly include the actual values ​​of PEF, FVC, and FEV1 before and after using bronchodilators (a commonly used drug for COPD patients to dilate the airway and relieve dyspnea), and the ratio of actual to expected values. These indicators can effectively reflect the test subject's ventilation level and ventilation status, as well as the magnitude of their ventilation capacity relative to the expected values.

[0056] The label, or disease / symptom tag classification, includes the disease or symptom diagnosed by the hospital and afflicted by the patient. Labels categorized as obstructive or restrictive diseases, or other diseases or pathological symptoms, include Asthma, COPD, ILD, DPLD, Sarcodiosis, OSA, and Chest Pain. Asthma and COPD are obstructive respiratory diseases, while ILD and DPLD are restrictive respiratory diseases, exhibiting certain characteristics in pulmonary function tests. Sarcodiosis is an inflammatory disease, chest pain is a pathological symptom, and OSA is sleep apnea syndrome; a single pulmonary function test is usually insufficient to reflect their characteristics. Therefore, this experiment only diagnosed and classified Asthma, COPD, and ILD / DPLD (ILD / DPLD were grouped into one category). The distribution of all samples from the data source for the three respiratory diseases on the three-dimensional axis composed of the three pulmonary function indicators (PEF, FEV1, FVC) is shown below. Figure 3 As shown:

[0057] (a) Values ​​of three pulmonary function indicators before medication (bronchodilator);

[0058] (b) Values ​​of three pulmonary function indicators after medication (bronchodilator);

[0059] (c) The ratio of the three pulmonary function indicators before medication (bronchodilator) to the predicted values;

[0060] (d) The ratio of the values ​​of the three pulmonary function indicators after medication (bronchodilator) to the predicted values.

[0061] In this embodiment, the data preprocessing module includes:

[0062] The data acquisition unit is configured to establish a data interface with the pulmonary function testing instrument to acquire the patient's pulmonary function test data in real time, including the values ​​of various pulmonary function indicators such as FEV1, FVC, and PEF before and after medication (such as FEV1PreVal and FVCPostVal), and simultaneously collect basic information such as patient ID and diagnosis results (Label).

[0063] The ratio calculation unit is configured to extract corresponding physiological parameters such as gender, age, height, and weight based on the patient's basic information, determine the patient's predicted value, calculate the ratio of each lung function index value before and after medication to the patient's predicted value (e.g., FEV1PrePercent = FEV1PreVal / predicted value × 100%), and construct a one-dimensional feature vector of lung function index.

[0064] The feature set generation unit is configured to standardize the one-dimensional feature vector and format the processed feature set into a two-dimensional matrix according to the model input requirements. Each row corresponds to a patient sample, each column corresponds to a lung function index, and label vectors corresponding to respiratory disease classifications are generated simultaneously.

[0065] In this embodiment, based on a large amount of pulmonary function test data from patients, the model can be continuously optimized as the data accumulates, improving the accuracy and reliability of diagnosis. The pulmonary function test data is processed and analyzed to provide diagnostic results, thereby significantly shortening diagnosis time and improving diagnostic efficiency. Based on a machine learning model, the characteristics of patients with respiratory diseases can be effectively identified, and a comprehensive judgment can be made in conjunction with the patient's pulmonary function test data, thereby improving diagnostic accuracy and reducing the possibility of misdiagnosis and missed diagnosis. It can serve as an auxiliary tool for doctors' diagnosis, providing them with more comprehensive diagnostic information, helping them make more accurate diagnoses, reducing reliance on doctors' experience, lowering diagnostic costs, and increasing the accessibility of diagnosis.

[0066] In this embodiment, the model validation optimization module includes:

[0067] The model training unit is configured to input the preprocessed feature set into each machine learning model, optimize the model parameters through cross-validation, and establish a mapping relationship between lung function indicators and disease categories, as expressed in the following formula:

[0068] X = (x1, x2, ..., x N )

[0069] X→y

[0070] In the formula, y represents the label vector corresponding to the respiratory disease classification; X represents the one-dimensional feature vector of the lung function index; x i (i = 1, 2, ..., N) represents the specific lung function indicators, including FEV1, FVC, PEF, etc. The specific indicators selected will be determined based on the maximum number of indicators available in the collected dataset.

[0071] In this embodiment, the model training unit optimizes the model parameters through a cross-validation mechanism, specifically including:

[0072] Five-fold cross-validation is used to divide the feature set into five equal subsets. In each round, four subsets are selected to form the training set, and the remaining one subset is used as the test set. Five rounds of cross-validation are performed.

[0073] The training set is sequentially input into various machine learning models such as logistic regression (LR), k-nearest neighbors (KNN), random forest (RF), convolutional neural network (CNN), and multilayer perceptron (MLP), and the training process is started simultaneously. Each machine learning model is trained iteratively 5 times, and the parameter combination and test set performance index are recorded for each training.

[0074] The model evaluation unit is configured to calculate the average accuracy and standard deviation of each machine learning model based on the results of 5 cross-validations, and select machine learning models with good stability, such as logistic regression, MLP, and SVM, with an accuracy higher than 75% and a standard deviation lower than 5%.

[0075] The model determination unit is configured to select the machine learning model with the highest single-class accuracy as the base model based on the screening results. For example, in the first round of Testi, MLP has the highest accuracy against Asthma, and in the second round of Testi, CNN has the highest accuracy against COPD. The base models selected in each round are weighted according to their classification accuracy, and the final classification decision model is generated through weighted fusion. The prediction results are obtained to verify whether the model performance has improved. A small amount of newly added lung function data is obtained, and the accuracy of the classification decision model is verified based on the confusion matrix of the classification decision model.

[0076] In this embodiment, model parameters were optimized through cross-validation, various machine learning models were evaluated, and the best model was selected based on their average accuracy and standard deviation. Through multiple cross-validations, a stable model with high accuracy and low standard deviation was selected to ensure the reliability of the diagnostic results. For the base models selected in different rounds, weights were assigned according to their classification accuracy, and the final classification decision model was generated through weighted fusion, which further improved the overall performance and increased the speed and accuracy of diagnosis.

[0077] In this embodiment, the model evaluation unit specifically includes:

[0078] Based on the test set performance metrics, the performance of each machine learning model was evaluated using the test set. Accuracy, Precision, Recall, and F1 score were calculated for each model, and a confusion matrix was generated. The classification performance of each machine learning model for three respiratory diseases—Asthma, COPD, and ILD / DPLD—was evaluated. The performance evaluation results are shown in the table below.

[0079] Table 2. Accuracy statistics of the optimal group for each model under the five-fold cross-validation test.

[0080]

[0081]

[0082] Table 3. Precision statistics of the best-performing model group under the five-fold cross-validation test.

[0083]

[0084]

[0085] Table 4. Recall statistics for the best-performing model group under the five-fold cross-validation test.

[0086]

[0087]

[0088] Table 5. Statistics of F1 scores for the model with the best accuracy under the five-fold cross-validation test.

[0089]

[0090]

[0091] In this embodiment, the test results of each model include the accuracy of various machine learning models, as shown in Tables 2 to 5, combined with... Figure 4 Figure 5 As shown in the results above, the LogisticRegression model achieved the highest overall recognition accuracy of 81.3%. Furthermore, it achieved the best results on both Asthma and ILD / DPLD F1-Scores. Considering Recall and Precision, the LogisticRegression model was deemed the optimal choice for comprehensive evaluation and was therefore used in the detection algorithm. Using a small amount of collected pulmonary function data—3 groups of healthy individuals and 3 groups of COPD patients, all collected by a hospital pulmonary function instrument—the best test set data from the cross-validation was combined for testing. The results are as follows. Figure 6 As shown.

[0092] The performance evaluation of each machine learning model includes:

[0093] For the logistic regression model, the adaptive step-size gradient descent method is used to update the weight parameters, and the optimal regularization coefficient is determined by minimizing the validation set loss function.

[0094] For the KNN model, a dynamic search algorithm is used to traverse the number of neighbors (k = 3 to k = 10) and select the k value that has the highest accuracy on the validation set.

[0095] For the random forest model, a hierarchical parameter tuning strategy is adopted, first optimizing the depth of the decision tree, then optimizing the number of trees, and evaluating the generalization ability through the out-of-bag error (OOBError).

[0096] In this embodiment, the intelligent classification decision module includes:

[0097] The model classification unit, configured to perform inference operations on the newly acquired feature set based on the trained classification decision model, outputs disease category prediction results, and also includes:

[0098] For each lung function test, if FVC / FEV1 > 70%, the patient is directly diagnosed as healthy.

[0099] If FVC / FEV1 < 70%, the model inference process is initiated, and the ratio of each lung function index value before and after medication to the patient's predicted value is obtained. This ratio is then input into the classification decision model to predict the respiratory disease classification and obtain the diagnosis result of the patient's respiratory disease category.

[0100] The results output unit is configured to generate a structured diagnostic report based on the prediction results and integrate preliminary screening rules to achieve rapid determination of health status. In the structured diagnostic report, if COPD is predicted, the clinical gold standard of "FEV1 / FVC<70%" is automatically referenced; if asthma is predicted, "bronchodilator test is recommended to verify reversible airflow limitation".

[0101] In this embodiment, by setting a rule that directly diagnoses a patient as healthy when FVC / FEV1 > 70%, the system can quickly identify obviously healthy individuals, avoiding these patients from entering the subsequent complex model inference process, thus improving the efficiency of the overall diagnostic process and automating patient triage. For patients with FVC / FEV1 < 70%, the system only initiates model inference and uses a preprocessed feature set to ensure the quality of the data input to the model, allowing it to focus more on identifying potential respiratory diseases and improving the accuracy of subsequent diagnoses. This not only speeds up the identification of healthy individuals but also ensures the accuracy of suspected case diagnoses, thereby improving the overall efficiency and reliability of respiratory disease diagnosis.

[0102] In this embodiment, based on the above analysis, we employ the machine learning models listed above and conduct experiments sequentially. Five-fold cross-validation is used on all samples in the dataset to verify which model achieves better classification performance in the current scenario. Then, for the three models with the best performance in each test, their classification accuracy is used as weights, and a multi-expert evaluation model (MOE) and a hybrid multi-expert model (MMOE) are added to verify whether they can improve the model's performance.

[0103] The training and evaluation use five-fold cross-validation, which randomly divides the dataset into five equal parts. Each time, one part is used as the sample set for validation and testing, i.e., the test set Testi for the i-th cross-validation. The other four parts are used for training, i.e., the training set Trainini. After training, the test set is used for validation to evaluate the model's accuracy, precision, recall, and F1 score.

[0104] After each i-th cross-validation training, the machine learning model (such as MLP, CNN, etc.) that performs best in classification accuracy for each class (Asthma, COPD, ILD / DPLD) in the test set Testi of the i-th cross-validation is selected as the base model of the final detection algorithm. It is then tested in combination with a small amount of collected lung function data.

[0105] For each lung function test indicator, when FVC / FEV1 > 70%, the patient is directly diagnosed as healthy; if FVC / FEV1 < 70%, the FVC, FEV1, and PEF indicators of the lung function test after medication need to be input, combined with the FVC, FEV1, and PEF indicators of the lung function test before medication, to calculate the ratio with the predicted value, and then send it to a module using a certain machine learning model to obtain a diagnosis of a certain disease.

[0106] The test results show that the final recognition algorithm based on LogisticRegression has a recognition accuracy of 80.6%. It has a high accuracy in classifying COPD and asthma, and can detect the normal class with almost 100% accuracy. However, its recognition effect on interstitial lung disease is average.

[0107] Employing machine learning-based algorithms, it can intelligently detect asthma, COPD, and interstitial lung disease (ILD / DPLD). When tested on publicly available datasets, the detection accuracy rate reaches 81.3%.

[0108] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An intelligent respiratory disease detection system based on lung function measurement indicators, characterized in that, include: The data preprocessing module is configured to integrate a pulmonary function testing instrument to acquire patients' pulmonary function test data. At the same time, it collects patients' basic information, calculates the ratio of pulmonary function indicators before and after medication to the predicted values, and performs preprocessing to construct a feature set that meets the model input requirements. The model validation and optimization module is configured to train each machine learning model based on the preprocessed feature set, validate the diagnostic results of each machine learning model, identify the accuracy and feasibility of each machine learning model based on the validation results, and determine the optimal classification decision model. The intelligent classification decision module is configured to use a defined classification decision model to predict the classification of respiratory diseases and diagnose the patient's respiratory disease category based on the prediction results.

2. The intelligent respiratory disease detection system based on lung function measurement indicators as described in claim 1, characterized in that, The data preprocessing module includes: The data acquisition unit is configured to establish a data interface with the pulmonary function testing instrument to acquire the patient's pulmonary function test data in real time, including pulmonary function index values ​​before and after medication, and to collect the patient's basic information simultaneously. The ratio calculation unit is configured to extract corresponding physiological parameters based on the patient's basic information, determine the patient's expected values, calculate the ratio of each lung function index value before and after medication to the patient's expected values, and construct a one-dimensional feature vector of the lung function index. The feature set generation unit is configured to standardize the one-dimensional feature vector and format the processed feature set into a two-dimensional matrix according to the model input requirements. Each row corresponds to a patient sample, each column corresponds to a lung function index, and label vectors corresponding to respiratory disease classifications are generated simultaneously.

3. The intelligent respiratory disease detection system based on lung function measurement indicators as described in claim 1, characterized in that, The model validation and optimization module includes: The model training unit is configured to input the preprocessed feature set into each machine learning model, optimize the model parameters through cross-validation mechanism, and establish a mapping relationship between lung function indicators and disease categories. The model evaluation unit is configured to calculate the average accuracy and standard deviation of each machine learning model based on the results of 5 cross-validations, and select machine learning models with good stability that have an accuracy higher than 75% and a standard deviation lower than 5%. The model determination unit is configured to select the machine learning model with the highest single-class accuracy based on the screening results as the base model. The base models selected in each round are weighted according to their classification accuracy, and the final classification decision model is generated through weighted fusion to obtain the prediction results.

4. The intelligent respiratory disease detection system based on lung function measurement indicators as described in claim 3, characterized in that, The model training unit optimizes model parameters through cross-validation, specifically including: Five-fold cross-validation is used to divide the feature set into five equal subsets. In each round, four subsets are selected to form the training set, and the remaining one subset is used as the test set. Five rounds of cross-validation are performed. The training set is input into multiple machine learning models in sequence, and the training process is started simultaneously. Each machine learning model is trained iteratively 5 times, and the parameter combination and test set performance index are recorded for each training.

5. The intelligent respiratory disease detection system based on lung function measurement indicators as described in claim 4, characterized in that, The model training unit establishes a mapping relationship between lung function indicators and disease categories, expressed by the following formula: X=(x1,x2,...,x N ) X→y In the formula, y represents the label vector corresponding to the respiratory disease classification; X represents the one-dimensional feature vector of the lung function index; x i (i = 1, 2, ..., N) represents the specific lung function indicators.

6. The intelligent respiratory disease detection system based on lung function measurement indicators as described in claim 5, characterized in that, The model evaluation unit specifically includes: Based on test set performance metrics, the performance of each machine learning model is evaluated using the test set. The accuracy, precision, recall, and F1 score of each machine learning model are calculated, and the confusion matrix of each machine learning model is generated to evaluate the classification performance of each machine learning model for three types of respiratory diseases.

7. The intelligent respiratory disease detection system based on lung function measurement indicators as described in claim 6, characterized in that, The model determination unit also includes: acquiring a small amount of newly added lung function data and verifying the accuracy of the classification decision model based on the confusion matrix of the classification decision model.

8. The intelligent respiratory disease detection system based on lung function measurement indicators as described in claim 1, characterized in that, The intelligent classification decision-making module includes: The model classification unit is configured to perform inference operations on the newly acquired feature set based on the trained classification decision model, and output the disease category prediction result. The results output unit is configured to generate a structured diagnostic report based on the prediction results and integrate preliminary screening rules to achieve rapid determination of health status.

9. The intelligent respiratory disease detection system based on lung function measurement indicators as described in claim 8, characterized in that, The model classification unit also includes: For each lung function test, if FVC / FEV1 > 70%, the patient is directly diagnosed as healthy. If FVC / FEV1 < 70%, the model inference process is initiated, and the ratio of each lung function index value before and after medication to the patient's predicted value is obtained. This ratio is then input into the classification decision model to predict the respiratory disease classification and obtain the diagnosis result of the patient's respiratory disease category.

10. The intelligent respiratory disease detection system based on lung function measurement indicators as described in claim 9, characterized in that, In the structured diagnostic report, if COPD is predicted, the clinical gold standard of "FEV1 / FVC < 70%" is automatically cited; if asthma is predicted, a bronchodilator test is recommended to verify reversible airflow limitation.

Images