Non-invasive feature extraction method for alveolar proteinosis

Abstract

The invention discloses a non-invasive feature extraction method for alveolar proteinosis, which screens out key features strongly related to alveolar proteinosis from multi-dimensional clinical data through a minimum absolute contraction and selection operator regression method, avoids misdiagnosis or missed diagnosis caused by feature redundancy or irrelevant feature interference in a traditional method, and improves the accuracy of feature extraction. The method only depends on clinical data, does not need invasive examination, reduces patient pain and medical risks, automatically compresses coefficients of irrelevant features to zero through regularization constraint by means of the minimum absolute contraction and selection operator regression method, achieves feature selection, avoids overfitting, and screened key features have clear clinical significance. And doctors can understand diagnosis basis conveniently.

Description

Technical Field

[0001] The present invention relates to the technical field of extracting features of pulmonary alveolar proteinosis, and particularly to a non-invasive method for extracting features of pulmonary alveolar proteinosis. Background Art

[0002] Pulmonary alveolar proteinosis (PAP) is a rare lung disease characterized by the accumulation of a large amount of phospholipid-rich protein substances in the alveoli, leading to gas exchange disorders and dyspnea. Due to the diverse and non-specific clinical manifestations of PAP, its diagnosis often relies on invasive examination methods such as bronchoalveolar lavage (BAL) and lung biopsy. These methods not only cause pain to patients but also increase the risk of complications. Therefore, developing a non-invasive feature extraction method for assisting in the early diagnosis and disease monitoring of PAP has important clinical significance. Summary of the Invention

[0003] In view of this, the present invention proposes a non-invasive feature extraction method for pulmonary alveolar proteinosis, which can avoid invasive examinations and screen out key features strongly related to pulmonary alveolar proteinosis.

[0004] The technical solution of the present invention is implemented as follows:

[0005] A non-invasive feature extraction method for pulmonary alveolar proteinosis specifically includes:

[0006] Data collection step: Collect the clinical data of patients with interstitial lung diseases, where the clinical data includes age, gender, at least one tumor marker, at least one blood gas analysis index, and at least one imaging feature;

[0007] Data preprocessing step: Preprocess the collected clinical data, including missing value processing and data standardization processing;

[0008] Feature selection step: Use the least absolute shrinkage and selection operator regression method to screen out the key features related to pulmonary alveolar proteinosis from the preprocessed clinical data.

[0009] As a further optional solution of the non-invasive feature extraction method for pulmonary alveolar proteinosis, the preprocessing of the collected clinical data, including missing value processing and data standardization processing, specifically includes:

[0010] Missing value processing step: Process the missing values in the collected clinical data, and use a new interpolation algorithm for estimation. The new interpolation algorithm performs weighted averaging based on the changing trend of adjacent data points and the correlation between features;

[0011] Steps for data standardization: Standardize the clinical data after missing value processing, and use a new standardization formula to convert the data to a unified scale. The new standardization formula is as follows:

[0012]

[0013] In the formula, Z i represents the standardized eigenvalue, X i represents the original eigenvalue, μ i and σ i represent the mean and standard deviation of this feature respectively. α is a regulation coefficient, which is set according to the data distribution characteristics. IV i represents the information value of this feature, which is used to measure the prediction ability of the feature for the risk of pulmonary alveolar proteinosis.

[0014] As a further optional solution of the non-invasive feature extraction method for pulmonary alveolar proteinosis, the least absolute shrinkage and selection operator regression method is used to screen out the key features related to pulmonary alveolar proteinosis from the preprocessed clinical data, specifically including:

[0015] Construct a LASSO regression model;

[0016] Determine the regularization parameter of the LASSO regression model;

[0017] According to the regularization parameter, the LASSO regression model automatically performs sparsification processing on the preprocessed clinical data and screens out the key features related to pulmonary alveolar proteinosis.

[0018] As a further optional solution of the non-invasive feature extraction method for pulmonary alveolar proteinosis, the determination of the regularization parameter of the LASSO regression model specifically includes:

[0019] Divide the preprocessed clinical data into a training set and a validation set;

[0020] Based on the validation set, calculate the regularization parameter of the LASSO regression model. Among them, the specific calculation formula is:

[0021]

[0022] In the formula, the λ is the regularization parameter;

[0023] The CV MSE represents the mean square error obtained through cross-validation under a given λ value;

[0024] The MSE baseline represents the mean square error obtained by using the baseline model on the validation set, which is used to normalize CV MSE ;

[0025] The N n represents the number of non-zero coefficients in the LASSO regression model at a given λ value;

[0026] The N f represents the total number of features in the dataset;

[0027] β and γ are adjustment coefficients used to balance model error and sparsity, and are adjusted according to the data characteristics.

[0028] As a further optional solution of the non-invasive feature extraction method for pulmonary alveolar proteinosis, the tumor markers include at least one of cancer antigen 153, carcinoembryonic antigen, and neuron-specific enolase, the blood gas analysis indexes include at least one of partial pressure of oxygen in blood, partial pressure of carbon dioxide in blood, and oxygen saturation, and the imaging features include at least one of ground-glass opacity, linear shadow, and lymph node enlargement.

[0029] A non-invasive feature extraction system for pulmonary alveolar proteinosis, comprising:

[0030] A data collection module for collecting clinical data of patients with interstitial lung diseases, where the clinical data includes age, gender, at least one tumor marker, at least one blood gas analysis index, and at least one imaging feature;

[0031] A data preprocessing module for preprocessing the collected clinical data, including missing value processing and data standardization processing;

[0032] A feature selection module for screening out key features related to pulmonary alveolar proteinosis from the preprocessed clinical data by using the least absolute shrinkage and selection operator regression method.

[0033] As a further optional solution of the non-invasive feature extraction system for pulmonary alveolar proteinosis, the system further includes a data verification unit for verifying the integrity and accuracy of the collected clinical data.

[0034] As a further optional solution of the non-invasive feature extraction system for pulmonary alveolar proteinosis, the data preprocessing module further includes an outlier detection and processing unit for identifying and processing outliers in the data.

[0035] A computing device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, the steps of any one of the above non-invasive feature extraction methods for pulmonary alveolar proteinosis are implemented.

[0036] A computer-readable storage medium stores a computer program thereon, and when the computer program is executed by a processor, the steps of any one of the above-described non-invasive feature extraction methods for pulmonary alveolar proteinosis are implemented.

[0037] The beneficial effects of the present invention are as follows: By using the least absolute shrinkage and selection operator regression method, key features strongly related to pulmonary alveolar proteinosis are screened out from multi-dimensional clinical data, avoiding misdiagnosis or missed diagnosis caused by feature redundancy or interference of irrelevant features in traditional methods. Depending only on clinical data without invasive examinations, it reduces the pain of patients and medical risks. The least absolute shrinkage and selection operator regression method automatically compresses the coefficients of irrelevant features to zero through regularization constraints to achieve feature selection and avoid overfitting. The selected key features have clear clinical significance, facilitating doctors to understand the diagnostic basis. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or in the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following-described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0039] Figure 1 It is a schematic flowchart of a non-invasive feature extraction method for pulmonary alveolar proteinosis of the present invention;

[0040] Figure 2 It is a schematic diagram of the composition of a non-invasive feature extraction system for pulmonary alveolar proteinosis of the present invention;

[0041] Figure 3 It is a schematic diagram of the composition of a computing device of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] The following will clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only some of the embodiments of the present invention, rather than all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0043] Refer to Figures 1 to 3 , a non-invasive feature extraction method for pulmonary alveolar proteinosis, specifically including:

[0044] Data collection step: Collect clinical data of patients with interstitial lung diseases, where the clinical data includes age, gender, at least one tumor marker, at least one blood gas analysis index, and at least one imaging feature;

[0045] Data preprocessing step: Preprocess the collected clinical data, including missing value handling and data standardization.

[0046] Feature selection step: Use the least absolute shrinkage and selection operator regression method to screen out the key features related to pulmonary alveolar proteinosis from the preprocessed clinical data.

[0047] In this embodiment, through the least absolute shrinkage and selection operator regression method, key features strongly related to pulmonary alveolar proteinosis are screened out from multi-dimensional clinical data, avoiding misdiagnosis or missed diagnosis caused by feature redundancy or interference of irrelevant features in traditional methods. Relying only on clinical data without invasive examinations reduces patient pain and medical risks. The least absolute shrinkage and selection operator regression method automatically compresses the coefficients of irrelevant features to zero through regularization constraints to achieve feature selection, avoiding overfitting. The selected key features have clear clinical significance, facilitating doctors to understand the diagnostic basis.

[0048] Preferably, the preprocessing of the collected clinical data, including missing value handling and data standardization, specifically includes:

[0049] Missing value handling step: Process the missing values in the collected clinical data, and use a new interpolation algorithm for estimation. The new interpolation algorithm performs weighted averaging based on the change trend of adjacent data points and the correlation between features.

[0050] Data standardization step: Standardize the clinical data after missing value handling, and use a new standardization formula to convert the data to a unified scale. The new standardization formula is:

[0051]

[0052] In the formula, Z i represents the standardized feature value, X i represents the original feature value, μ i and σ i respectively represent the mean and standard deviation of this feature, α is a regulation coefficient, which is set according to the data distribution characteristics, and IV i represents the information value of this feature, which is used to measure the prediction ability of the feature for the risk of pulmonary alveolar proteinosis.

[0053] In this embodiment, a new interpolation algorithm based on the change trend of adjacent data points and the correlation weighting average between features is adopted, which can estimate missing values more accurately. Compared with traditional methods (such as mean filling, median filling, etc.), this algorithm takes into account the time series characteristics of the data and the correlation between variables, thereby improving the accuracy of the estimated values. By effectively filling the missing values, the integrity of the data is enhanced, the analysis bias caused by data missing is reduced, and the subsequent data analysis and modeling are made more reliable; the correlation between features is used for weighted average to ensure the consistency of the filled data with the original data in statistical characteristics and avoid the additional noise introduced by improper filling methods. The data is transformed to a unified scale through a new normalization formula, eliminating the influence of different feature dimensions on the analysis results. For example, indicators with different dimensions such as age, blood pressure, and blood sugar are transformed into dimensionless standard values, facilitating subsequent comprehensive analysis and comparison. The information value IV is introduced into the formula. i As a regularization coefficient, it can measure the predictive ability of features for the target variable (such as the risk of pulmonary alveolar proteinosis). The regularization coefficient is set according to the data distribution characteristics, making the normalization process more flexible and adaptive. Through the combination of the regularization coefficient and the data distribution characteristics, the normalization method has stronger adaptability to different data sets and improves the robustness of data processing.

[0054] Preferably, the least absolute shrinkage and selection operator regression method is adopted to screen out the key features related to pulmonary alveolar proteinosis from the preprocessed clinical data, specifically including:

[0055] Construct a LASSO regression model;

[0056] Determine the regularization parameter of the LASSO regression model;

[0057] According to the regularization parameter, the LASSO regression model automatically performs sparsification processing on the preprocessed clinical data and screens out the key features related to pulmonary alveolar proteinosis.

[0058] In this embodiment, LASSO compresses the regression coefficients of non-critical features to zero through a regularization parameter, and only retains the features strongly correlated with PAP. For example, if the clinical data contains 50 variables (such as blood indicators, imaging parameters, gene expression, etc.), LASSO may only retain 5-10 core features, significantly reducing the data dimension, reducing redundant information, avoiding overfitting, and improving the generalization ability of the model. Compared with traditional methods (such as stepwise regression), LASSO does not require manual intervention and can quickly complete feature screening. For example, in a clinical cohort containing 1000 samples, LASSO can complete feature selection within seconds, while stepwise regression may take several hours; the regularization mechanism of LASSO reduces the sensitivity of the model to noisy data. For example, when there are 10% outliers in the data, the prediction fluctuation range of the LASSO model may only be 1 / 3 of that of the full-feature model. The sparsified feature matrix significantly reduces storage and computational requirements. For example, after the original 50-dimensional data is reduced to 5 dimensions by LASSO, the subsequent model training time may be shortened by 80%. LASSO alleviates the small-sample problem through regularization. For example, in a cohort of only 50 PAP patients, LASSO can still effectively screen features, while traditional methods may fail due to overfitting.

[0059] Preferably, determining the regularization parameter of the LASSO regression model specifically includes:

[0060] Dividing the preprocessed clinical data into a training set and a validation set;

[0061] Based on the validation set, calculate the regularization parameter of the LASSO regression model, where the specific calculation formula is:

[0062]

[0063] In the formula, λ is the regularization parameter;

[0064] CV MSE represents the mean squared error obtained through cross-validation at a given λ value;

[0065] MSE baseline represents the mean squared error obtained using the baseline model on the validation set, and is used to normalize CV MSE ;

[0066] N n represents the number of non-zero coefficients in the LASSO regression model at a given λ value;

[0067] N f represents the total number of features in the dataset;

[0068] β and γ are adjustment coefficients used to balance model error and sparsity, and are adjusted according to data characteristics.

[0069] In this embodiment, LASSO regression can automatically perform feature selection through L1 regularization, compressing the coefficients of unimportant features to zero, which helps reduce the complexity of the model and improve the interpretability of the model. By selecting the optimal λ value through cross-validation, it can ensure that the most important features for prediction are retained during the feature selection process while avoiding overfitting. Using cross-validation (CV) to evaluate the model performance (mean squared error MSE) at different λ values, the λ value that makes the model perform best on the validation set can be selected, which helps improve the generalization ability of the model. By comparing the MSE of the LASSO model with the baseline model (usually the full-feature model), the improvement in the performance of the LASSO model can be quantified. LASSO regression has high computational efficiency when dealing with high-dimensional data and can quickly screen out important features. Although cross-validation may require multiple trainings and validations of the model, the sparsity property of LASSO makes the computational cost of each training relatively small. The introduction of the regularization parameter λ effectively controls the complexity of the model and prevents the occurrence of overfitting. Especially in the case of a small sample size or a large number of features, LASSO regression can significantly improve the stability of the model. β and γ, as adjustment coefficients, can be adjusted according to the specific data characteristics, enabling the LASSO model to perform well on different data sets.

[0070] Preferably, the tumor markers include at least one of cancer antigen 153, carcinoembryonic antigen, and neuron-specific enolase, the blood gas analysis indicators include at least one of partial pressure of oxygen in blood, partial pressure of carbon dioxide in blood, and oxygen saturation, and the imaging features include at least one of ground-glass opacity, linear shadow, and lymph node enlargement.

[0071] In this embodiment, by combining multiple tumor markers (such as cancer antigen 153, carcinoembryonic antigen, and neuron-specific enolase), different types of tumors can be detected more comprehensively because different markers may be expressed differently in different types of cancers. Blood gas analysis indicators (such as partial pressure of oxygen in blood, partial pressure of carbon dioxide in blood, and oxygen saturation) can reflect the respiratory function and acid-base balance status of the patient, helping to evaluate the overall health status of the patient, especially the common respiratory system complications in cancer patients. Imaging features (such as ground-glass opacity, linear shadow, and lymph node enlargement) directly reflect the morphological characteristics of the lesions and are of great significance for the localization and qualitative diagnosis of tumors.

[0072] A non-invasive feature extraction system for pulmonary alveolar proteinosis, comprising:

[0073] A data collection module for collecting clinical data of patients with interstitial lung disease, where the clinical data includes age, gender, at least one tumor marker, at least one blood gas analysis indicator, and at least one imaging feature;

[0074] A data preprocessing module for preprocessing the collected clinical data, including missing value processing and data standardization processing;

[0075] A feature selection module for screening out key features related to pulmonary alveolar proteinosis from the preprocessed clinical data by using the least absolute shrinkage and selection operator regression method.

[0076] Preferably, the system further includes a data verification unit for verifying the integrity and accuracy of the collected clinical data.

[0077] In this embodiment, it is checked whether the clinical data contains all required fields (such as age, gender, tumor markers, blood gas analysis indicators, imaging features), missing values or abnormal formats are identified. If a certain patient record lacks the value of "blood oxygen saturation", the system will mark it as incomplete data; data anomalies are detected through a rule engine (such as age exceeding a reasonable range, abnormal high / low concentration of tumor markers). If the value of "carcinoembryonic antigen (CEA)" of a certain patient exceeds 1000 ng / mL (the normal range is usually <5 ng / mL), the system will trigger an anomaly alarm; ensure the consistency of information such as timestamps and patient IDs from different data sources (such as imaging reports and laboratory results). If the CT examination date is 3 months earlier than the blood test date, the system may prompt a conflict in data timeliness; by removing or correcting abnormal data (such as extreme values, incorrect entries), it is avoided that the model is overfitted or underfitted due to noisy data. High-quality data ensures that the features selected by LASSO regression truly reflect the PAP pathological mechanism rather than data noise. Instead of manually checking item by item, the verification unit can complete the integrity check of a dataset of thousands within seconds.

[0078] Preferably, the data preprocessing module further includes an outlier detection and processing unit for identifying and processing outliers in the data.

[0079] In this embodiment, machine learning models (such as isolation forest, One-Class SVM) are used to detect outliers in the data. The value of "partial pressure of oxygen in blood" of a certain patient is 900 mmHg (normal range: 80 - 100 mmHg) and is marked as an outlier; processing strategies are selected according to the type of anomaly, including correction: replacing with median / mean (such as correcting the "age" entered as -5 to a reasonable value), deletion: removing extreme outliers (such as "tumor marker concentration" exceeding 10 times the upper detection limit), and retention and marking: retaining and annotating outliers that may have clinical significance (such as rare diseases complicated with PAP).

[0080] A computing device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of any one of the above-mentioned non-invasive feature extraction methods for pulmonary alveolar proteinosis are implemented.

[0081] A computer-readable storage medium stores a computer program. When the computer program is executed by a processor, the steps of any one of the above-mentioned non-invasive feature extraction methods for pulmonary alveolar proteinosis are implemented.

[0082] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A non-invasive feature extraction method for pulmonary alveolar proteinosis, characterized in that, Specifically include: Data collection step: Collect the clinical data of patients with interstitial lung disease, where the clinical data includes age, gender, at least one tumor marker, at least one blood gas analysis index, and at least one imaging feature; Data preprocessing step: Preprocess the collected clinical data, including missing value processing and data standardization processing; Feature selection step: Use the least absolute shrinkage and selection operator regression method to screen out the key features related to pulmonary alveolar proteinosis from the preprocessed clinical data.

2. The non-invasive feature extraction method for pulmonary alveolar proteinosis according to claim 1, wherein The preprocessing of the collected clinical data, including missing value processing and data standardization processing, specifically includes: Missing value processing step: Process the missing values in the collected clinical data, and use a new interpolation algorithm for estimation. The new interpolation algorithm performs weighted averaging based on the change trend of adjacent data points and the correlation between features; Data standardization processing step: Standardize the clinical data after missing value processing, and use a new standardization formula to convert the data to a unified scale. The new standardization formula is: In the formula, Zi represents the standardized feature value, Xi represents the original feature value, μi and σi respectively represent the mean and standard deviation of this feature, α is an adjustment coefficient, which is set according to the data distribution characteristics, and IVi represents the information value of this feature, which is used to measure the prediction ability of the feature for the risk of pulmonary alveolar proteinosis.

3. The non-invasive feature extraction method for pulmonary alveolar proteinosis according to claim 2, characterized in that, The method of using the least absolute shrinkage and selection operator regression method to screen out the key features related to pulmonary alveolar proteinosis from the preprocessed clinical data specifically includes: Construct a LASSO regression model; Determine the regularization parameter of the LASSO regression model; According to the regularization parameter, the LASSO regression model automatically performs sparsification processing on the preprocessed clinical data, and screens out the key features related to pulmonary alveolar proteinosis.

4. The non-invasive feature extraction method for pulmonary alveolar proteinosis according to claim 3, characterized in that The determination of the regularization parameter of the LASSO regression model specifically includes: Divide the preprocessed clinical data into a training set and a validation set; Based on the validation set, calculate the regularization parameter of the LASSO regression model. The specific calculation formula is: In the formula, λ is the regularization parameter; CVMSE represents the mean square error obtained through cross-validation at a given λ value; MSEbaseline represents the mean square error obtained by using the baseline model on the validation set, which is used to normalize CVMSE; Nn represents the number of non-zero coefficients in the LASSO regression model at a given λ value; Nf represents the total number of features in the dataset; β and γ are adjustment coefficients, which are used to balance the model error and sparsity, and are adjusted according to the data characteristics.

5. The non-invasive feature extraction method for pulmonary alveolar proteinosis according to claim 4, wherein, The tumor markers include at least one of cancer antigen 153, carcinoembryonic antigen, and neuron-specific enolase. The blood gas analysis indicators include at least one of partial pressure of oxygen in blood, partial pressure of carbon dioxide in blood, and oxygen saturation. The imaging features include at least one of ground-glass opacity, linear shadow, and lymph node enlargement.

6. A non-invasive feature extraction system for pulmonary alveolar proteinosis, characterized in that, Include: A data collection module for collecting clinical data of patients with interstitial lung disease, where the clinical data includes age, gender, at least one tumor marker, at least one blood gas analysis index, and at least one imaging feature; A data preprocessing module for preprocessing the collected clinical data, including missing value processing and data standardization processing; A feature selection module for screening out key features related to pulmonary alveolar proteinosis from the preprocessed clinical data by using the least absolute shrinkage and selection operator regression method.

7. The non-invasive feature extraction system for pulmonary alveolar proteinosis according to claim 6, characterized in that, The system further includes a data verification unit for verifying the integrity and accuracy of the collected clinical data.

8. The non-invasive feature extraction system for pulmonary alveolar proteinosis according to claim 7, characterized in that, The data preprocessing module further includes an outlier detection and processing unit for identifying and processing outliers in the data.

9. A computing device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the non-invasive feature extraction method for pulmonary alveolar proteinosis according to any one of claims 1-5 are implemented.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the storage medium. When the computer program is executed by the processor, the steps of the non-invasive feature extraction method for pulmonary alveolar proteinosis according to any one of claims 1-5 are implemented.

Images