Construction method and equipment of refractory mycoplasma pneumonia prediction model

Abstract

The invention belongs to the field of intelligent medical treatment, and particularly relates to a construction method and equipment of an intractable mycoplasma pneumonia prediction model. The method comprises the following steps: acquiring a data set of baseline time and a follow-up time label of a mycoplasma pneumoniae patient; inputting the image data set into a pulmonary blood vessel segmentation model to obtain a pulmonary blood vessel network, obtaining a total pulmonary blood vessel volume based on the pulmonary blood vessel network, and obtaining blood vessel volumes of different cross section areas based on the blood vessel cross section areas in the pulmonary blood vessel network; and inputting the ratio of the blood vessel volume of different cross section areas to the total pulmonary blood vessel volume into a machine learning model to obtain a prediction label, and iteratively optimizing the machine learning model based on the difference between the prediction label and the follow-up time label to obtain an intractable mycoplasma pneumonia prediction model. According to the method, the lung CT image is segmented to obtain the blood vessel network graph, the percentage of the volume of blood vessels with different thicknesses in the total lung blood vessel volume is obtained through quantification, and the refractory pneumonia prediction model is constructed based on an innovative feature extraction mode.

Description

Technical Field

[0001] The present invention relates to the field of intelligent medicine, and more specifically, to a method, device, medium and program product for constructing a prediction model for refractory Mycoplasma pneumonia. Background Art

[0002] Mycoplasma pneumoniae (MP) is the smallest pathogen between bacteria and viruses. It can survive independently in cell-free culture media, lacks a cell wall, and takes on various forms, such as rods, spheres, or filaments. It is Gram-negative and facultatively anaerobic. It is primarily transmitted via droplets and can occur year-round, but is more common in winter and spring. Mycoplasma pneumoniae pneumonia (MPP) is a common pediatric disease. In non-epidemic years, MPP accounts for 10% to 20% of community-acquired pneumonia in children, while in epidemic years, this proportion can increase to 30%, and regional epidemics can occur every 3 to 7 years [1]. In recent years, there have been many reports of refractory mycoplasma pneumonia (RMPP). Conventional macrolide antibiotics are ineffective in treating this disease and may even aggravate the condition. Multiple system involvement may occur, and the disease is protracted and difficult to cure. In addition, the disease is difficult to identify in its early stages and difficult to treat. If not treated in time, it may leave pulmonary sequelae such as bronchiectasis, atelectasis, and obstructive bronchitis. In severe cases, it may be complicated by pulmonary embolism or even death.

[0003] Conventional CT imaging for MPP is as follows: Early X-rays of M. pneumoniae pneumonia may reveal interstitial pneumonia, increased and blurred lung markings, and lattice-like shadows. As the disease progresses, lung damage primarily occurs in the lower lung near the hilum, with extensive pulmonary infiltrates and consolidation, and a small number of children present with pleural effusions. Lesions can involve either unilateral or bilateral lung tissue. Studies have shown that lesions are more common in the lower lung fields than in the upper lung, and in the right lung than in the left. Chest radiographs of M. pneumoniae pneumonia may not show specific findings. If a child has severe symptoms but mild physical signs, the chest radiograph findings should be considered for M. pneumoniae infection. High-resolution CT can reveal interstitial lesions, air trapping, and enlarged hilar lymph nodes, and the likelihood of detecting a small pleural effusion increases. However, many patients are still reluctant to use CT to diagnose pneumonia directly, and this requires a gradual shift in perspective. RMPP presents with more severe pulmonary imaging findings than MPP. Chest X-rays may demonstrate large focal, segmental pneumonia, pulmonary consolidation with atelectasis, and varying degrees of pleural effusion (mostly moderate to large pleural effusions). Chest CT findings are more diverse, showing interstitial infiltrates (ground-glass opacities, reticular and irregular strip-like patterns), often involving the lung parenchyma, large patches of increased density (covering >2 / 3 of the lung lobes), accompanied by air bronchograms, and moderate to large pleural effusions. Severe cases may present with atelectasis and pulmonary embolism. Therefore, if children with MPP present with severe pulmonary imaging findings, the possibility of progression to refractory Mycoplasma pneumoniae pneumonia should be considered, and steroids should be administered as appropriate to prevent further deterioration, which may lead to atelectasis and bronchiectasis. However, these CT examinations are difficult to quantify, requiring highly experienced clinicians. Summary of the Invention

[0004] In view of the above problems, the present invention provides a method for constructing a prediction model for refractory Mycoplasma pneumoniae pneumonia, which extracts quantitative features based on CT images for the prediction of RMPP.

[0005] This application (first aspect) discloses a method for constructing a prediction model for refractory Mycoplasma pneumonia, comprising:

[0006] Obtaining a baseline dataset and follow-up time labels of patients with Mycoplasma pneumoniae, wherein the dataset includes an imaging dataset, and the labels include developing refractory Mycoplasma pneumoniae pneumonia and not developing refractory Mycoplasma pneumoniae pneumonia;

[0007] Inputting the image dataset into a pulmonary vascular segmentation model to obtain a pulmonary vascular network, obtaining a total pulmonary vascular volume based on the pulmonary vascular network, obtaining vascular volumes of different cross-sectional areas based on the cross-sectional areas of the vessels in the pulmonary vascular network, and obtaining a percentage of intravascular blood volume of the vessels of different cross-sectional areas to the total pulmonary blood volume based on a ratio of the vascular volumes of the different cross-sectional areas to the total pulmonary vascular volume;

[0008] The percentage of the intravascular blood volume of the different cross-sectional areas to the total lung blood volume is input into the machine learning model to obtain a prediction label. The machine learning model is iteratively optimized based on the difference between the prediction label and the follow-up time label to obtain a prediction model for refractory Mycoplasma pneumonia.

[0009] Furthermore, the different cross-sectional areas include one or more of the following: 0mm 2 -5mm 2 , 5mm 2 ~10mm 2 , greater than 10mm 2 ;

[0010] Optionally, the different cross-sectional areas include: 1.25mm 2 -5mm 2 , 5mm 2 ~10mm 2 ;

[0011] Furthermore, the method for obtaining the pulmonary vascular segmentation model is:

[0012] Acquire a training set of lung images, wherein the image set includes annotations of lung lobe boundaries;

[0013] Iteratively training the UV-Net model based on the lung image set and lung lobe boundary annotation to obtain a trained pulmonary vascular segmentation model;

[0014] Furthermore, the machine learning model includes any one or more of the following: XGBoost, SVM, LightGBM, logistic regression, K-nearest neighbor algorithm, random forest, multi-layer perceptron;

[0015] Optionally, the machine learning model is XGBoost;

[0016] Optionally, the hyperparameters of the machine learning model are optimized by k-fold cross validation to obtain optimal hyperparameters.

[0017] Optionally, the data set further includes a clinical feature set, and the percentage of intravascular blood volume of different cross-sectional areas to total pulmonary blood volume and the clinical feature set are input into a machine learning model to obtain a prediction label, and the machine learning model is iteratively optimized based on the difference between the prediction label and the follow-up time label to obtain a prediction model for refractory Mycoplasma pneumonia; the clinical features include one or more of the following features: duration of azithromycin treatment before admission, body temperature, presence of severe pneumonia, CRP level, BV10%, albumin, γ-glutamyl transferase, platelet distribution width, serum prealbumin, heart rate, creatine kinase, platelet count, hemoglobin, BV5%, BV5-10%, adenosine deaminase, monocyte count, whether glucocorticoids were used before admission, and mixed infection;

[0018] Optionally, performing a SHAP analysis on the refractory Mycoplasma pneumonia prediction model to obtain the most influential K features, and reconstructing the model based on the most influential K features to obtain an optimized refractory Mycoplasma pneumonia prediction model, where K is a natural number greater than 1;

[0019] Optionally, K is 5, and the most influential K features are: duration of azithromycin treatment before admission, body temperature, presence of severe pneumonia, CRP level, and BV10%.

[0020] Furthermore, a baseline time dataset and follow-up time labels of Mycoplasma pneumoniae patients of different age groups were obtained, and after replacing the baseline time dataset of Mycoplasma pneumoniae patients of different age groups with the baseline time dataset of Mycoplasma pneumoniae patients, a prediction model for refractory Mycoplasma pneumoniae pneumonia of different age groups was trained.

[0021] The second aspect of the present application discloses a method for predicting a refractory Mycoplasma pneumonia prediction model, comprising:

[0022] Obtain images of patients with Mycoplasma pneumoniae pneumonia,

[0023] Based on the images, the percentage of intravascular blood volume in different cross-sectional areas to total lung blood volume is obtained;

[0024] The percentage of the intravascular blood volume of the different cross-sectional areas to the total lung blood volume is input into the refractory Mycoplasma pneumonia prediction model constructed according to any of the above-mentioned methods for constructing a refractory Mycoplasma pneumonia prediction model to obtain a prediction result.

[0025] Optionally, the age of the patient with Mycoplasma pneumonia is obtained at the same time, and the data is input into the refractory Mycoplasma pneumonia prediction model for the corresponding age group constructed according to the above method to obtain a prediction result.

[0026] A third aspect of the present application provides a method for predicting a model for refractory Mycoplasma pneumonia, the method comprising:

[0027] Obtain images of patients with Mycoplasma pneumoniae pneumonia,

[0028] Based on the image, the percentage of blood volume in blood vessels with a cross-sectional area less than 5 mm2 in the total lung blood volume;

[0029] If the percentage of intravascular blood volume with a cross-sectional area less than 5 mm2 to the total pulmonary blood volume is lower than the BV5 threshold, a result is obtained that the patient with Mycoplasma pneumonia has a high risk of developing refractory Mycoplasma pneumonia; otherwise, a result is obtained that the patient has a low risk of developing refractory Mycoplasma pneumonia.

[0030] Furthermore, the percentage of intravascular blood volume with a cross-sectional area less than 5 mm2 to the total lung blood volume was input into the classifier to obtain the prediction result.

[0031] A fourth aspect of the present application discloses a system for constructing a prediction model for refractory Mycoplasma pneumonia, comprising:

[0032] The first acquisition module 201 is used to obtain a baseline time dataset and a follow-up time label of a Mycoplasma pneumoniae patient, wherein the dataset includes an image dataset, and the labels include developing refractory Mycoplasma pneumoniae pneumonia and not developing refractory Mycoplasma pneumoniae pneumonia;

[0033] Feature extraction module 202: configured to input the image dataset into a pulmonary vascular segmentation model to obtain a pulmonary vascular network, obtain a total pulmonary vascular volume based on the pulmonary vascular network, obtain vascular volumes of different cross-sectional areas based on the cross-sectional areas of the vessels in the pulmonary vascular network, and obtain a percentage of blood volume in vessels of different cross-sectional areas to the total pulmonary blood volume based on a ratio of the volume of the vessels of different cross-sectional areas to the total pulmonary vascular volume;

[0034] Iterative training module 203: used to input the percentage of intravascular blood volume of the different cross-sectional areas to the total lung blood volume into the machine learning model to obtain a prediction label, and iteratively optimize the machine learning model based on the difference between the prediction label and the follow-up time label to obtain a refractory Mycoplasma pneumonia prediction model.

[0035] In a fifth aspect, the present application discloses a system for constructing a prediction model for refractory Mycoplasma pneumonia, comprising:

[0036] The second acquisition module is used to obtain images of patients with Mycoplasma pneumonia.

[0037] A second feature extraction module is used to obtain the percentage of intravascular blood volume of different cross-sectional areas to total lung blood volume based on the image;

[0038] The second prediction module: inputs the percentage of the intravascular blood volume of different cross-sectional areas to the total lung blood volume into the refractory Mycoplasma pneumonia prediction model constructed according to the construction method of the refractory Mycoplasma pneumonia prediction model to obtain a prediction result.

[0039] A sixth aspect of the present application discloses a system for constructing a prediction model for refractory Mycoplasma pneumonia, comprising:

[0040] The third acquisition module is used to obtain images of patients with Mycoplasma pneumonia.

[0041] A third feature extraction module is configured to obtain, based on the image, a percentage of the blood volume in blood vessels with a cross-sectional area less than 5 mm2 to the total lung blood volume;

[0042] The third prediction module is configured to obtain a prediction result based on the percentage of the intravascular blood volume with a cross-sectional area less than 5 mm2 to the total pulmonary blood volume. If the percentage of the intravascular blood volume with a cross-sectional area less than 5 mm2 to the total pulmonary blood volume is lower than the BV5 threshold, a result is obtained indicating that the patient with Mycoplasma pneumonia has a high risk of developing refractory Mycoplasma pneumonia; otherwise, a result is obtained indicating that the patient has a low risk of developing refractory Mycoplasma pneumonia.

[0043] In a seventh aspect, the present application discloses a computer device comprising: a memory and a processor; the memory is used to store program instructions; the processor is used to call the program instructions, and when the program instructions are executed, is used to execute the steps of the above method.

[0044] In an eighth aspect, the present application discloses a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above-mentioned method when the computer program is executed by a processor.

[0045] In a fifth aspect, the present application discloses a computer program product, comprising a computer program, which implements the steps of the above method when executed by a processor.

[0046] This application has the following beneficial effects:

[0047] (1) This application uses a pulmonary vascular segmentation model to segment lung CT images to obtain a vascular network map, and quantifies the percentage of blood vessel volumes of different thicknesses in the total pulmonary vascular volume based on the vascular network map. Based on an innovative feature extraction method, a prediction model for refractory pneumonia is constructed;

[0048] (2) Based on the feature extraction method of this application, important quantitative feature parameters were screened and determined in the process of constructing a prediction model for refractory pneumonia;

[0049] (3) The performance of the prediction model was improved by combining the quantitative feature parameters extracted in this application with traditional clinical features during model construction, and the model was further optimized through SHAP analysis of the constructed model. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0051] Figure 1 This is a schematic diagram of the method flow provided by the first aspect of the embodiment of the present invention;

[0052] Figure 2 is a schematic diagram of a program product provided by the fourth aspect of an embodiment of the present invention;

[0053] Figure 3 is a schematic diagram of a computer device provided by an embodiment of the present invention;

[0054] Figure 4 is a schematic diagram of the architecture of an exemplary computing device provided by an embodiment of the present invention;

[0055] Figure 5 is a schematic diagram of a storage medium provided by an embodiment of the present invention;

[0056] Figure 6 This is a distribution map of blood vessels of different cross-sectional areas obtained based on a pulmonary vascular segmentation model provided by an embodiment of the present invention;

[0057] Figure 7 Schematic diagram of the predicted AUC on a cross-validation set and an external test set of a clinical-BV fusion model and a clinical feature-only model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0058] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0059] In some of the processes described in the specification and claims of the present invention and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or may be executed in parallel. The serial numbers of the operations, such as S101, S102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., and do not represent the order of precedence, nor do they limit "first" and "second" to be different types.

[0060] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making any creative efforts shall fall within the scope of protection of the present invention.

[0061] Figure 1FIG. 4 is a flow chart of a method for predicting RMPP based on evaluating pulmonary microvascular changes, provided by an embodiment of the present invention. Specifically, the method comprises the following steps:

[0062] S101: Acquire a baseline dataset and follow-up time labels of a Mycoplasma pneumoniae patient, wherein the dataset includes an imaging dataset, and the labels include developing refractory Mycoplasma pneumoniae pneumonia and not developing refractory Mycoplasma pneumoniae pneumonia;

[0063] S102: inputting the image dataset into a pulmonary vascular segmentation model to obtain a pulmonary vascular network, obtaining a total pulmonary vascular volume based on the pulmonary vascular network, obtaining vascular volumes of different cross-sectional areas based on the cross-sectional areas of the vessels in the pulmonary vascular network, and obtaining a percentage of blood volume in the vessels of different cross-sectional areas to the total pulmonary blood volume based on a ratio of the volume of the vessels of different cross-sectional areas to the total pulmonary vascular volume;

[0064] S103: Inputting the percentage of the intravascular blood volume of the different cross-sectional areas to the total lung blood volume into the machine learning model to obtain a prediction label, and iteratively optimizing the machine learning model based on the difference between the prediction label and the follow-up time label to obtain a prediction model for refractory Mycoplasma pneumoniae pneumonia.

[0065] The solution of this application is based on the following research.

[0066] 1. Methods and Materials

[0067] The study was approved by the institutional ethics committees of both hospitals, and individual consent was not required for this retrospective analysis.

[0068] Study design and patient selection

[0069] This retrospective study included consecutive pediatric patients diagnosed with MPP from two medical institutions to form a cross-validation cohort and an external validation cohort.

[0070] The cross-validation cohort included patients admitted to the general inpatient department of a tertiary pediatric specialty hospital from July 2019 to December 2023. The external validation cohort included patients admitted to the pediatric department of a tertiary general hospital from January 2023 to April 2024.

[0071] Inclusion criteria were: (1) pediatric patients aged >28 days to ≤18 years; (2) clinical diagnosis of MPP; (3) baseline chest CT scan during hospitalization for all cohorts, with only the cross-validation cohort requiring a follow-up CT scan within 2 months of discharge. Exclusion criteria were: (1) baseline or follow-up CT image quality was insufficient for quantitative analysis; (2) no laboratory test data at all; (3) immunodeficiency or documented prior use of immunosuppressive therapy; and (4) requiring mechanical ventilation support during hospitalization. Finally, the cross-validation and external validation cohorts included 512 and 124 patients, respectively. The diagnostic criteria for MPP can be found in Supplementary Material 1.1.

[0072] 1.2. Data Collection and Chest CT Acquisition

[0073] All patients included in this study had a confirmed diagnosis of MPP. Clinical, etiological, and laboratory data were retrospectively collected from the digital hospital information system. Clinical variables included demographic characteristics, preexisting comorbidities, duration of fever and cough before admission, preadmission treatment, initial vital signs, and oxygen support requirements on admission. Disease severity was classified according to the 2023 Chinese Pediatric MPP Guidelines. Pathogen detection results (including viral, bacterial, and atypical pathogens) were recorded. Laboratory variables obtained on admission included complete blood count, liver and kidney function tests, inflammatory markers, coagulation tests, and electrolytes.

[0074] Clinical outcomes, complications, and hospitalizations (e.g., corticosteroids, intravenous immunoglobulin, bronchoscopy) were extracted from hospital discharge records. For outcomes, RMPP was defined as persistent fever after ≥7 days of macrolide treatment, clinical worsening, pulmonary radiographic progression, or extrapulmonary complications, and patients were divided into RMPP and non-RMPP groups.

[0075] Thin-slice (≤1.5 mm), noncontrast chest CT images were obtained from the Picture Archiving and Communication System for analysis. The earliest chest CT scan obtained during hospitalization was selected as the baseline CT. For the cross-validation cohort, baseline and follow-up chest CT scans were retrospectively collected within 2 months of discharge. For the external validation cohort, only the baseline chest CT scan was obtained.

[0076] Quantitative pulmonary vascular analysis

[0077] The pulmonary vessels were automatically segmented using a previously validated UV-Net-based segmentation model. The detailed method steps include:

[0078] This study used the deep learning framework UV-Net for 3D pulmonary vascular segmentation.

[0079] The model incorporates a U-shaped network architecture, consisting of a 2D encoder module and a 3D decoder module, enabling effective adaptation to data characteristics. To expand the receptive field and incorporate global context, the network incorporates an Atrous Spatial Pyramid Pooling (ASPP) module, which fuses multi-scale feature maps to preserve spatial and semantic details. For upsampling, the model uses PixelShuffle technology to restore spatial resolution while maintaining fine anatomical structure, ensuring continuity and topological consistency in the segmentation of pulmonary vessels and airways.

[0080] The segmentation process first involves the preliminary delineation of lung lobes with clear boundaries. Based on the delineated lung lobe boundaries, boundary representation (B-rep) data is constructed. The B-rep data structure contains multiple topological entities—faces, edges, half-edges, and vertices—as well as the connections between them, including parametric surfaces (including vascular surfaces in our study).

[0081] Topological UV-Net uses a face adjacency graph derived from B-rep to model topology, where vertices V represent faces in B-rep and edges E encode the connectivity between faces.

[0082] Airway segmentation was then performed to preserve the tree-like connectivity of the bronchial architecture.

[0083] Furthermore, the vessel segmentation module generates a complete and connected vascular network, enabling accurate quantification of vessel volume based on cross-sectional area.

[0084] During internal validation, the algorithm demonstrated high segmentation accuracy, with a Dice coefficient of 92.68% for arteries and 89.96% for veins. These results highlight the robustness and clinical applicability of UV-Net for pulmonary vascular analysis.

[0085] The segmented pulmonary vessels were then divided into three categories based on their cross-sectional area: cross-sectional area <5 mm2 (BV5), 5–10 mm2 (BV5–10), and >10 mm2 (BV10) ( Figure 6 The blood volume in each category was calculated and expressed as a percentage of the total lung blood volume, denoted as BV5%, BV5–10%, and BV10%, respectively.

[0086] Development and Validation of the RMPP Prediction Model

[0087] A RMPP prediction model was developed using clinical variables and PBV parameters.

[0088] First, candidate predictors collected at admission (including demographics, clinical symptoms, previous treatments, vital signs, laboratory tests, and BV parameters) were screened, and predictors with >40% missing data in the cross-validation cohort or external validation cohort were excluded.

[0089] To develop a predictive model for refractory mycoplasma pneumonia (RMPP), we evaluated five machine learning algorithms: extreme gradient boosting (XGBoost), support vector machine (SVM), lightweight gradient boosting machine (LightGBM), logistic regression (LR), and k-nearest neighbor (KNN). XGBoost, officially launched in 2016, is a state-of-the-art ensemble learning algorithm that builds multiple classification and regression trees within an optimized distributed gradient boosting framework. Compared to traditional machine learning algorithms, it offers higher complexity and superior performance. LightGBM is a high-performance distributed gradient boosting framework based on decision trees, widely used for ranking, classification, and other machine learning tasks. LR is a widely used linear classification model used to assess the relationship between a dependent variable and multiple independent variables, particularly well-suited for binary classification problems. SVM is a supervised learning algorithm that aims to construct an optimal hyperplane to separate positive and negative samples, and excels in binary classification and high-dimensional datasets. KNN is an instance-based, nonparametric learning algorithm that classifies samples based on their proximity to previously labeled instances. It is simple yet effective for multi-classification problems and those involving nonlinear decision boundaries.

[0090] For these models, we included 19 variables selected using least absolute shrinkage and selection operator regression, and optimized hyperparameters using 5-fold cross-validation based on average performance metrics. The final models were then retrained on the entire cross-validation cohort using the optimal hyperparameters identified. Receiver operating characteristic (ROC) curve analysis was used to comprehensively evaluate model performance, with key metrics including area under the ROC curve (AUC), accuracy, precision, sensitivity, specificity, and F1 score, performed on both the cross-validation and external validation cohorts. As shown in Table S1, XGBoost performed best in both datasets and was therefore selected as the final analysis algorithm.

[0091] All models were developed in R software (version 4.3.1) using the packages “caret,” “glmnet,” “knn,” “e1071,” “lightgbm,” and “xgboost.”

[0092] Based on the selected predictors, two models were constructed: (1) a clinical model containing 16 selected clinical variables and (2) a clinical BV model combining all 19 variables. For model development, model hyperparameters were optimized by 5-fold cross-validation, and the determined optimal hyperparameters were retrained on the full cross-validation cohort.

[0093] Among them, the 16 selected clinical variables were: duration of azithromycin treatment before admission, body temperature, presence of severe pneumonia, CRP level, BV10%, albumin, γ-glutamyltransferase, platelet distribution width, serum prealbumin, heart rate, creatine kinase, platelet count, hemoglobin, BV5%, BV5-10%, adenosine deaminase, monocyte count, use of glucocorticoids before admission, and mixed infection;

[0094] The 19-variable clinical-BV model included 16 selected clinical variables and 3 PBV parameters.

[0095] Model performance was evaluated internally and externally using receiver operating characteristic (ROC) curve analysis. Area under the ROC curve (AUC), accuracy, precision, sensitivity, specificity, and F1 score were calculated for both cohorts. Differences in AUC values ​​between the clinical and clinical BV models were compared using the DeLong test. SHAP (Shapley Additive Explanation) values ​​were calculated to quantify feature importance in the clinical BV model.

[0096] Statistical analysis

[0097] Statistical analyses were performed using R software (version 4.3.1). Categorical variables were compared with the chi-square test, and continuous variables were compared with the Student's t test (for normally distributed data) or the Wilcoxon rank-sum test (for nonnormally distributed data). Multiple comparisons were corrected with the Holm method. Missing data were imputed using the multiple imputation method in the "mice" package in R.

[0098] The association between PBV parameters and RMPP was investigated in the cross-validation cohort. Numerical variables were standardized using z-score standardization, and univariate logistic regression was performed to initially examine associations. Clinically relevant variables with <40% missing data were selected by LASSO regression (“glmnet” package) and included in the multivariate model to independently evaluate BV parameters. In addition, subgroup analysis was performed in the 5-10 year age group based on the age distribution in the cross-validation cohort (Figure S1), age-related immune differences, and the higher prevalence of MPP in children older than 5 years. Pearson correlations further evaluated the associations between significant BV parameters and key laboratory indices (D-dimer, CRP, LDH) and fever duration.

[0099] XGBoost models were developed using the "caret" and "xgboost" packages. SHAP analysis was performed using the "SHAPforxgboost" package. ROC analysis was performed using the "pROC" package. P values ​​less than 0.05 were considered statistically significant.

[0100] 2. Results

[0101] Comparison of clinical and laboratory features between RMPP and non-RMPP

[0102] As shown in Table 1, the cross-validation cohort included 512 patients (256 men and 256 women; median age, 6.42 years [IQR, 4.33–8.25]), of whom 265 were classified as non-RMPP and 247 as RMPP. Patients in the RMPP group had significantly longer durations of fever and cough before admission and a higher incidence of severe / critical pneumonia. In addition, these patients were more likely to require oxygen support and have elevated temperature, pulse, and respiratory rate (all p ≤ 0.01). With regard to clinical outcomes (Table S2), the RMPP group had a significantly longer length of hospital stay, a prolonged total duration of fever, and a higher incidence of pulmonary complications compared with the non-RMPP group (all p ≤ 0.03).

[0103] There were no statistically significant differences between the groups with regard to etiology (Table S3) (all p > 0.05). Laboratory test results (Table S4) showed that patients with RMPP had significantly higher CRP levels, erythrocyte sedimentation rate (ESR), neutrophil count, LDH levels, and D-dimer levels (all p ≤ 0.03).

[0104] Analysis of PBV parameters at baseline and follow-up

[0105] At baseline, the BV5% of the RMPP group was significantly lower than that of the non-RMPP group (58.50% vs. 60.63%, p = 0.007), while the BV5-10% and BV10% of the RMPP group were significantly higher than those of the non-RMPP group (median BV5-10%: 15.74% vs. 15.13%, p = 0.03; median BV10%: 20.90% vs. 19.39%, p = 0.004).

[0106] The median interval between follow-up CT scans was 16.00 days [13.00, 23.00]. Follow-up CT scans showed a significant increase in BV5% relative to baseline (p < 0.001). In contrast, both BV5-10% and BV10% values ​​on follow-up CT were significantly lower than those at baseline (all p < 0.01), suggesting that pulmonary blood distribution may recover over time. Notably, during the follow-up period, the BV5% in the RMPP group remained lower than that in the non-RMPP group (median: 62.42% vs. 63.55%, p = 0.03), and the BV10% was higher than that in the non-RMPP group (Table 1).

[0107] Table 1. Comparison of quantitative lung blood volume parameters during follow-up in patients with refractory and non-refractory mycoplasma pneumonia in the cross-validation cohort

[0108]

[0109] Table Notes: Unless otherwise noted, data are presented as the number of patients, with percentages in parentheses. BV5%, BV5-10%, and BV10% represent the percentage of pulmonary blood volume contained within vessels with a cross-sectional area of ​​less than 5 square millimeters, 5-10 square millimeters, and greater than 10 square millimeters, respectively. * indicates p < 0.05.

[0110] Correlation Analysis between RMPP and Quantitative Characteristics of Pulmonary Blood Volume and Its Association with RMPP-Related Risk Factors

[0111] As shown in Table 2, in the cross-validation cohort, univariate analysis showed that each PBV parameter was significantly associated with the development of RMPP (all p ≤ 0.03). In subsequent multivariate logistic regression analysis, after adjusting for clinical covariates, BV 5% remained an independent protective factor for RMPP (odds ratio [OR] = 0.70, 95% confidence interval [CI]: 0.54–0.89, p = 0.005), while BV 10% became an independent risk factor (OR = 1.49, 95% CI: 1.17–1.92, p = 0.002). Similarly, in patients aged 5-10 years, after adjusting for relevant covariates, BV5% continued to show a protective association with RMPP (OR = 0.68; 95% CI: 0.49-0.83; p = 0.018), while BV10% remained a significant risk factor (OR = 1.58; 95% CI: 1.15-2.19; p = 0.005).

[0112] Table 2. Univariate and multivariate logistic regression analysis of quantitative pulmonary blood volume parameters associated with refractory Mycoplasma pneumonia in the cross-validation cohort.

[0113]

[0114]

[0115] Table Notes: The adjusted model for the entire cohort included the following covariates: duration of azithromycin treatment before admission, severe pneumonia, temperature, platelet distribution width, albumin, serum prealbumin, prothrombin time, thrombin time, serum phosphorus, sodium, uric acid, and cystatin C. For patients aged >5 years and <10 years, the adjusted model included the following covariates: duration of azithromycin treatment before admission, duration of fever before admission, severe pneumonia, temperature, platelet distribution width, albumin, adenosine deaminase, prothrombin time, serum phosphorus, uric acid, and cystatin C. Abbreviation: CI, confidence interval. BV5%, BV5–10%, and BV10% indicate the cross-sectional area between 1.25 and 5 mm, respectively. 2 , 5–10 mm 2 and greater than 10mm 2 The percentage of intravascular pulmonary blood volume to total pulmonary blood volume. * indicates p < 0.05.

[0116] In addition, if Figure 4 As shown, correlation analysis showed that BV5% was significantly negatively correlated with key inflammatory and coagulation biomarkers (including NR, CRP, and D-dimer) (r = -0.12, -0.13, and -0.19, respectively; all p < 0.01). In contrast, BV10% was significantly positively correlated with these biomarkers (r = 0.14, 0.14, and 0.21, respectively; all p < 0.01).

[0117] 2.4. XGBoost-based RMPP prediction enhanced by quantitative PBV parameters

[0118] To improve the predictive performance of RMPP, we incorporated clinical variables and quantitative PBV features into the XGBoost-based model. Feature selection by LASSO regression identified 19 variables, including 16 clinical features and 3 BV-derived features.

[0119] Table 3. Performance indicators of XGBoost models in cross-validation cohort and external validation cohort

[0120]

[0121]

[0122] Table Notes: This table shows the performance metrics of the Extreme Gradient Boosting (XGBoost) model in the clinical-blood volume model and the clinical model. The clinical-blood volume model incorporated all 19 variables (16 clinical variables and 3 quantitative features derived from blood volume) selected by LASSO regression, while the clinical model used only 16 clinical variables. Metrics include area under the curve (AUC) and its 95% confidence interval (CI), accuracy, sensitivity, specificity, precision, and F1 score, which are presented in both the cross-validation cohort and the external validation cohort. The p-value was derived using the DeLong test, which was used to evaluate the difference in AUC between the clinical-blood volume model and the clinical model in the cross-validation cohort and the external validation cohort. *Indicates p < 0.05.

[0123] As shown in Table 3, Figure 7 As shown, in the cross-validation cohort, the clinical-BV model had a significantly higher AUC than the clinical model (AUC: 0.91 vs. 0.88, p < 0.001), and had higher accuracy, precision, sensitivity, specificity, and F1 score.

[0124] Table 4 Comparison of prediction AUC of different indicators

[0125] AUC Cutoff value Remark BV5% 0.57 0.60 BV10% 0.58 0.18 BV5-10% 0.56 0.17 BV5% and BV10% 0.78 Build a model using Xgboost BV5%, BV10%, BV5-10% 0.81 Build a model using Xgboost The five most influential characteristics combined 0.90 Build a model using Xgboost

[0126] SHAP analysis of the clinic-BV model showed that the five most influential features in the cross-validation cohort were duration of azithromycin treatment before admission, body temperature, presence of severe pneumonia, CRP level, and BV10%, highlighting the importance of inflammatory markers and BV features.

[0127] To assess the generalizability of the model, we evaluated its performance in an independent external validation cohort of 124 patients (median age: 7.00 years [IQR: 5.00–9.00], 52% male). In this cohort, the clinical-BV model achieved a greater area under the predictive analysis (AUC) than the clinical model, although the difference was not statistically significant (AUC: 0.84 vs. 0.81, p = 0.09). However, the model performed well in terms of accuracy, precision, specificity, and F1 score. As shown in Table 4, after SHAP analysis of the clinical-BV model, the AUC for the five most influential factors reached 0.90, demonstrating that the combination of the quantitative PBV parameters discovered in this application with clinical indicators can achieve better predictive results.

[0128] 3. Discussion

[0129] In this study, patients with RMPP demonstrated a consistent pattern of microvascular remodeling on CT scans, characterized by a decrease in BV5% (microvessels <5 mm²) and an increase in BV10% (large vessels >10 mm²) at baseline and follow-up. These changes suggest altered pulmonary vascular volume distribution, a finding also observed in patients with acute pulmonary microcirculatory disturbances during COVID-19. Furthermore, multivariate analysis further revealed that BV5%, which was negatively correlated with inflammatory and coagulation biomarkers, was an independent protective predictor of RMPP. Conversely, BV10% emerged as an independent risk factor and positively correlated with these same biomarkers. When integrated into an XGBoost-based model, these quantitative PBV parameters significantly improved RMPP prediction in the cross-validation cohort compared with a model using clinical variables alone. Although the difference in AUC in the external validation cohort was not statistically significant, the clinical-BV model still outperformed the clinical model in terms of accuracy, precision, specificity, and F1 score. These findings suggest that incorporating PBV parameters can enhance the predictive performance of RMPP and maintain clinical applicability in independent cohorts.

[0130] In patients with RMPP, we observed a significant decrease in BV5% and a concomitant increase in BV5–10% and BV10%. Consistent with these CT findings, histopathological analysis in a mouse model of recurrent MP infection revealed thickening of the pulmonary arteriolar walls and narrowing of the lumen, leading to elevated pulmonary artery pressures. Microscopically, pulmonary arterioles are defined as vessels with a diameter less than 500 μm, corresponding to the subset of vessels classified as BV5% on CT. The concordance between microscopic and imaging observations further suggests that patients with RMPP indeed exhibit a reduction in pulmonary microvascular volume.

[0131] Given that the pathogenesis of MPP involves direct toxicity, immune-mediated injury, and vascular inflammation / thrombosis, we hypothesized that the observed decrease in BV5% (reflecting reduced microvascular volume in RMPP patients) might be related to microvascular endothelial inflammation or thrombosis triggered by enhanced coagulation and inflammatory states. The observed increase in BV10% might, however, reflect compensatory dilation of larger vessels due to reduced microvascular volume. This hypothesis is supported by three lines of indirect evidence. First, RMPP patients in the cross-validation cohort had significantly elevated D-dimer, CRP, ESR, and neutrophil count. Second, correlation analysis further confirmed these findings, revealing an inverse correlation between BV5% and inflammatory or coagulation markers. Finally, all documented vascular thrombotic events occurred in the RMPP group. Furthermore, SHAP value analysis of the XGBoost model revealed that BV10% was more significant than BV5%. This observation suggests that large vessel dilation is an active compensatory response, rather than merely a passive consequence.

[0132] Compared with traditional pneumonia volumetric analysis (such as consolidation volume), quantitative pulmonary vascular analysis can reveal more subtle pathophysiological changes. Traditional volumetric analysis primarily reflects parenchymal tissue damage but provides limited insight into the underlying pathogenic mechanisms. In contrast, pulmonary vascular analysis, particularly by reducing BV by 5%, effectively captures the reduction in microvascular volume seen in patients with RMPP. This reduction highlights specific pathophysiological processes in RMPP, in which enhanced coagulation and inflammation may affect the pulmonary microvasculature, leading to reduced microvascular volume. Furthermore, incorporating BV parameters into clinical prediction models (including a clinical BV model (constructed from 16 clinical features + 3 BV parameters) and the most influential 5-feature model) improved early identification of RMPP. The clinical BV model demonstrated superior performance to the clinical model alone (AUC: 0.91 vs. 0.88), enabling timely administration of immunomodulatory agents or second-line antibiotics and potentially preventing complications such as necrotizing pneumonia. Furthermore, these quantitative vascular measurements can be obtained with routine CT scans, eliminating the need for additional testing and radiation exposure. Furthermore, RMPP patients continued to demonstrate a 5% relative decrease in BV during follow-up (median duration: 16 days) compared to non-RMPP patients, which is similar to the pattern of vascular recovery observed in post-COVID-19 studies, in which 87.4% of patients still had residual vascular abnormalities at 6 months. This further emphasizes the need for continued vascular monitoring after symptom resolution.

[0133] 4. Conclusion

[0134] This study demonstrates that RMPP in children is associated with significant pulmonary microvascular alterations, characterized by decreased BV5% and increased BV10%. These quantitative PBV parameters can serve as independent predictors of RMPP, enhance the performance of XGBoost-based prediction models, and can be easily integrated into routine CT workflows. Assessing pulmonary vascular changes through routine CT imaging is a noninvasive approach that can help identify high-risk patients and guide early risk stratification and intervention.

[0135] In some embodiments, the BV5% and BV10% of the training set and the corresponding labels are obtained, and the first classifier is obtained by iterative training as a prediction model for refractory Mycoplasma pneumonia. When used, the BV5% and BV10% of the subject are obtained and input into the first classifier to obtain a predicted label.

[0136] In some embodiments, the training set of pre-hospital azithromycin treatment duration, body temperature, presence of severe pneumonia, CRP level, and BV10% is obtained. ， The second classifier is iteratively trained to obtain the corresponding labels as a prediction model for refractory Mycoplasma pneumonia. When used, the duration of azithromycin treatment, body temperature, presence of severe pneumonia, CRP level and BV10% of the subjects before admission are obtained and input into the second classifier to obtain the predicted label.

[0137] Figure 3 is a schematic diagram of a computer device provided by an embodiment of the present invention, such as Figure 3 As shown, the device 2000 may include: one or more processors 2010, and one or more memories 2020; wherein the memories store computer-readable codes, and when the computer-readable codes are run by the one or more processors, they may execute the method described above.

[0138] The processor in this embodiment can be an integrated circuit chip with signal processing capabilities. The above-mentioned processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, or a discrete hardware component. It can implement or execute the various methods, operations, and logic block diagrams disclosed in the embodiments of the present disclosure. The general-purpose processor can be a microprocessor or any conventional processor, etc., and can be an X86 architecture or an ARM architecture.

[0139] In general, various example embodiments of the present disclosure may be implemented in hardware or dedicated circuitry, software, firmware, logic, or any combination thereof. Certain aspects may be implemented in hardware, while other aspects may be implemented in firmware or software that may be executed by a controller, microprocessor, or other computing device. When various aspects of the disclosed embodiments are illustrated or described as block diagrams, flow charts, or using some other graphical representation, it will be understood that the blocks, devices, systems, techniques, or methods described herein may be implemented, as non-limiting examples, in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or a controller or other computing device, or some combination thereof.

[0140] For example, the method or apparatus according to the embodiment of the present disclosure may also be implemented by Figure 4 The architecture of the computing device 3000 shown in FIG. Figure 4 As shown, the computing device 3000 may include a bus 3010, one or more CPUs 3020, a read-only memory (ROM) 3030, a random access memory (RAM) 3040, a communication port 3050 connected to a network, an input / output component 3060, a hard disk 3070, etc. The storage device in the computing device 3000, such as the ROM 3030 or the hard disk 3070, may store various data or files used for processing and / or communication of the method provided in the present disclosure, as well as program instructions executed by the CPU. The computing device 3000 may also include a user interface 3080. Of course, Figure 4 The architecture shown is only exemplary and can be omitted according to actual needs when implementing different devices. Figure 4One or more components of a computing device are shown.

[0141] The embodiment of the present invention further provides a computer-readable storage medium, such as Figure 5 As shown, it is a schematic diagram of a storage medium 4000 provided in an embodiment of the present invention, and computer-readable instructions 4010 are stored on the computer storage medium 4020. When the computer-readable instructions 4010 are executed by the processor, the method according to the embodiment of the present disclosure described with reference to the above figures can be executed. The computer-readable storage medium in the embodiment of the present disclosure can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. The non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM) or a flash memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus random access memory (DR RAM). It should be noted that the memory of the methods described herein is intended to include, but is not limited to, these and any other suitable types of memory. It should be noted that the memory of the methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0142] The present disclosure also provides a computer program product or a computer program, which implements the steps of the above method when executed by a processor, such as Figure 2 As shown, the computer program product or computer program includes:

[0143] The first acquisition module 201 is used to obtain a baseline time dataset and a follow-up time label of a Mycoplasma pneumoniae patient, wherein the dataset includes an image dataset, and the labels include developing refractory Mycoplasma pneumoniae pneumonia and not developing refractory Mycoplasma pneumoniae pneumonia;

[0144] Feature extraction module 202: configured to input the image dataset into a pulmonary vascular segmentation model to obtain a pulmonary vascular network, obtain a total pulmonary vascular volume based on the pulmonary vascular network, obtain vascular volumes of different cross-sectional areas based on the cross-sectional areas of the vessels in the pulmonary vascular network, and obtain a percentage of blood volume in vessels of different cross-sectional areas to the total pulmonary blood volume based on a ratio of the volume of the vessels of different cross-sectional areas to the total pulmonary vascular volume;

[0145] Iterative training module 203: used to input the percentage of intravascular blood volume of the different cross-sectional areas to the total lung blood volume into the machine learning model to obtain a prediction label, and iteratively optimize the machine learning model based on the difference between the prediction label and the follow-up time label to obtain a refractory Mycoplasma pneumonia prediction model.

[0146] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architectures, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the module, program segment, or a part of code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified functions or operations, or can be implemented using a combination of dedicated hardware and computer instructions.

[0147] In general, various example embodiments of the present disclosure may be implemented in hardware or dedicated circuitry, software, firmware, logic, or any combination thereof. Certain aspects may be implemented in hardware, while other aspects may be implemented in firmware or software that may be executed by a controller, microprocessor, or other computing device. When various aspects of the disclosed embodiments are illustrated or described as block diagrams, flow charts, or using some other graphical representation, it will be understood that the blocks, devices, systems, techniques, or methods described herein may be implemented, as non-limiting examples, in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or a controller or other computing device, or some combination thereof.

[0148] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0149] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0150] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0151] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0152] The exemplary embodiments of the present disclosure described in detail above are merely illustrative and not restrictive. Those skilled in the art will appreciate that various modifications and combinations may be made to these embodiments or their features without departing from the principles and spirit of the present disclosure, and such modifications should fall within the scope of the present disclosure.

Claims

1. A method for constructing a prediction model for refractory Mycoplasma pneumonia, characterized in that: The method comprises: Obtaining a baseline dataset and follow-up time labels of patients with Mycoplasma pneumoniae, wherein the dataset includes an imaging dataset, and the labels include developing refractory Mycoplasma pneumoniae pneumonia and not developing refractory Mycoplasma pneumoniae pneumonia; Inputting the image dataset into a pulmonary vascular segmentation model to obtain a pulmonary vascular network, obtaining a total pulmonary vascular volume based on the pulmonary vascular network, obtaining vascular volumes of different cross-sectional areas based on the cross-sectional areas of the vessels in the pulmonary vascular network, and obtaining a percentage of intravascular blood volume of the vessels of different cross-sectional areas to the total pulmonary blood volume based on a ratio of the vascular volumes of the different cross-sectional areas to the total pulmonary vascular volume; The percentage of the intravascular blood volume of the different cross-sectional areas to the total lung blood volume is input into the machine learning model to obtain a prediction label. The machine learning model is iteratively optimized based on the difference between the prediction label and the follow-up time label to obtain a prediction model for refractory Mycoplasma pneumonia.

2. The method for constructing a prediction model for refractory Mycoplasma pneumonia according to claim 1, wherein: The different cross-sectional areas include one or more of the following: 0mm 2 -5mm 2 , 5mm 2 ~10mm 2 , greater than 10mm 2 ; Optionally, the different cross-sectional areas include: 1.25mm 2 -5mm 2 , 5mm 2 ~10mm 2 .

3. The method for constructing a prediction model for refractory Mycoplasma pneumonia according to claim 1, wherein: The method for obtaining the pulmonary vascular segmentation model is: Acquire a training set of lung images, wherein the image set includes annotations of lung lobe boundaries; The UV-Net model is iteratively trained based on the lung image set and lung lobe boundary annotation to obtain a trained pulmonary vascular segmentation model.

4. The method for constructing a prediction model for refractory Mycoplasma pneumonia according to claim 1, wherein: The machine learning model includes any one or more of the following: XGBoost, SVM, LightGBM, logistic regression, K-nearest neighbor algorithm, random forest, and multi-layer perceptron; Optionally, the machine learning model is XGBoost; Optionally, optimizing the hyperparameters of the machine learning model by k-fold cross validation to obtain optimal hyperparameters; Optionally, the data set further includes a clinical feature set, and the percentage of intravascular blood volume of different cross-sectional areas to total pulmonary blood volume and the clinical feature set are input into a machine learning model to obtain a prediction label, and the machine learning model is iteratively optimized based on the difference between the prediction label and the follow-up time label to obtain a prediction model for refractory Mycoplasma pneumonia; the clinical features include one or more of the following features: duration of azithromycin treatment before admission, body temperature, presence of severe pneumonia, CRP level, BV10%, albumin, γ-glutamyl transferase, platelet distribution width, serum prealbumin, heart rate, creatine kinase, platelet count, hemoglobin, BV5%, BV5-10%, adenosine deaminase, monocyte count, whether glucocorticoids were used before admission, and mixed infection; Optionally, performing a SHAP analysis on the refractory Mycoplasma pneumonia prediction model to obtain the most influential K features, and reconstructing the model based on the most influential K features to obtain an optimized refractory Mycoplasma pneumonia prediction model, where K is a natural number greater than 1; Optionally, K is 5, and the most influential K features are: duration of azithromycin treatment before admission, body temperature, presence of severe pneumonia, CRP level, and BV10%.

5. The method for constructing a prediction model for refractory Mycoplasma pneumonia according to claim 1, wherein: A baseline time dataset and follow-up time labels of Mycoplasma pneumoniae patients of different age groups are obtained, and after replacing the baseline time dataset of Mycoplasma pneumoniae patients of different age groups with the baseline time dataset of Mycoplasma pneumoniae patients of different age groups, a prediction model for refractory Mycoplasma pneumoniae pneumonia of different age groups is trained.

6. A method for predicting a refractory Mycoplasma pneumonia prediction model, characterized in that: The method comprises: Obtain images of patients with Mycoplasma pneumoniae pneumonia, Based on the images, the percentage of intravascular blood volume in different cross-sectional areas to total lung blood volume is obtained; The percentage of the intravascular blood volume of different cross-sectional areas to the total lung blood volume is input into the refractory Mycoplasma pneumonia prediction model constructed according to the method according to any one of claims 1 to 5 to obtain a prediction result. Optionally, the age of the patient with Mycoplasma pneumonia is obtained at the same time, and the data is input into the refractory Mycoplasma pneumonia prediction model for the corresponding age group constructed according to the method of claim 5 to obtain a prediction result.

7. A method for predicting a refractory Mycoplasma pneumonia prediction model, characterized in that: The method comprises: Obtain images of patients with Mycoplasma pneumoniae pneumonia, The cross-sectional area based on the image is less than 5 mm 2 The percentage of intravascular blood volume to total pulmonary blood volume is greater than 10 mm 2 The percentage of intravascular blood volume to total pulmonary blood volume; The image is obtained with a cross-sectional area of ​​less than 5 mm 2 The percentage of intravascular blood volume to total pulmonary blood volume is greater than 10 mm 2 The percentage of intravascular blood volume to total lung blood volume is input into the classifier to obtain the results of whether patients with Mycoplasma pneumonia have a high or low risk of developing refractory Mycoplasma pneumonia.

8. A computer device, characterized in that: The device comprises: a memory and a processor; the memory is used to store a computer program; and the processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that A computer program is stored thereon, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images