Model training method, respiratory tract infection pathogen identification method and identification system

Abstract

The invention relates to a model training method and a respiratory tract infection pathogen identification method and system, and the method comprises the steps: obtaining the clinical data of a respiratory tract infection pathogen; preprocessing the clinical data to obtain a training set; training a respiratory tract infection pathogen recognition model by using the training set to obtain the trained respiratory tract infection pathogen recognition model; wherein the trained respiratory tract infection pathogen identification model is used for identifying the respiratory tract infection pathogen based on the index detection data of the respiratory tract infection pathogen. Compared with the prior art, the method has the advantages of good recognition effect, low sampling requirement, high treatment efficiency and the like.

Description

Technical Field

[0001] The present invention relates to the field of medical technologies, and particularly to a model training method, a method and a system for identifying respiratory infection pathogens. Background Art

[0002] Currently, there is a new and unknown epidemic pattern and status of respiratory infectious diseases, and different respiratory infection pathogens circulate in turn, which is a severe challenge for clinical practice. To cope with this situation, more, faster and more accurate diagnostic methods for pathogens are particularly important.

[0003] Clinically, the diagnosis of common respiratory infections mainly relies on virus nucleic acid detection of respiratory throat swabs and antigen-antibody detection of serology. They all have their own advantages and disadvantages. Serological detection takes a short time and is convenient to detect, but has low sensitivity; RT-PCR has high sensitivity, but takes a slightly longer time, the experimental process is complex, requires professional instruments and infrastructure, and is relatively expensive. Moreover, even in symptomatic infected patients, the detection rate of nucleic acid detection is not 100%. Some reports have found that due to sampling reasons or manual operation factors, the sensitivity of RT-PCR in clinical applications is only 30%-60%. Therefore, it is very necessary to find more detection methods for respiratory infection pathogens.

[0004] Blood routine and C-reactive protein detection are the preferred examination indicators in the fever clinic. They are simple to operate and inexpensive. However, relying solely on these indicators is not enough for doctors to distinguish the pathogens of respiratory infection patients. In recent years, many studies have also found that the ratio of lymphocytes to monocytes (LMR), the ratio of neutrophils to lymphocytes (NLR), the mean platelet volume / platelet ratio (MPV / PLT), and the lymphocyte multiplied by platelet value (LYM*PLT) can also be used as a new inflammatory marker to predict the prognosis of infectious diseases. However, these data only utilize a small part of the information in the detection results, and the detection effect of respiratory infection pathogens is still poor. Summary of the Invention

[0005] The purpose of the present invention is to overcome the above-mentioned defects existing in the prior art and provide a model training method, a method and a system for identifying respiratory infection pathogens with good identification effect, low sampling requirements and high disposal efficiency.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] According to the first aspect of the present invention, a model training method is provided, and the method includes:

[0008] Obtain clinical data of respiratory infection pathogens;

[0009] Preprocess the clinical data to obtain a training set;

[0010] Train a respiratory tract infection pathogen recognition model using the training set to obtain the trained respiratory tract infection pathogen recognition model; wherein, the trained respiratory tract infection pathogen recognition model is used to recognize the respiratory tract infection pathogen based on the index detection data of the respiratory tract infection pathogen.

[0011] As a preferred technical solution, the respiratory tract infection pathogens include at least one of influenza A virus, influenza B virus, respiratory syncytial virus, adenovirus, human rhinovirus, Mycoplasma pneumoniae, novel coronavirus, fungi, and bacteria.

[0012] As a preferred technical solution, the obtaining of the clinical data of the respiratory tract infection pathogen includes:

[0013] Based on the clinical cohort of the respiratory tract infection pathogen, obtain the complete blood count (CBC) index data and C-reactive protein (CRP) index data of the enrolled population.

[0014] As a preferred technical solution, the preprocessing of the clinical data to obtain a training set includes:

[0015] Deduplicate the clinical data;

[0016] Encode the categorical features in the deduplicated clinical data and separate the categorical features from the labels.

[0017] As a preferred technical solution, the training of the respiratory tract infection pathogen recognition model using the training set to obtain the trained respiratory tract infection pathogen recognition model includes:

[0018] Build a model based on XGBClassifier, monitor the loss of multi-class classification using the multi-class classification negative log-likelihood function mlogloss, and use the fit method to train on the training set to obtain the trained respiratory tract infection pathogen recognition model.

[0019] As a preferred technical solution, the method further includes:

[0020] Evaluate the trained respiratory tract infection pathogen recognition model using a test set.

[0021] As a preferred technical solution, the evaluation parameters of the model evaluation include at least one of ROC curve, specificity, and sensitivity.

[0022] According to the second aspect of the present invention, a method for recognizing respiratory tract infection pathogens is provided, the method including:

[0023] Obtain the index detection data of the respiratory tract infection pathogen of the user;

[0024] Input the index detection data into a respiratory tract infection pathogen recognition model to obtain the respiratory tract infection pathogen recognition result output by the respiratory tract infection pathogen recognition model; wherein, the respiratory tract infection pathogen recognition model is a trained model obtained by using the first aspect or any one of the possible implementation manners of the first aspect.

[0025] As a preferred technical solution, the obtaining of the index detection data of the respiratory tract infection pathogen of the user includes:

[0026] Obtain the complete blood count (CBC) index data and C-reactive protein (CRP) index data of the respiratory tract infection pathogen of the user.

[0027] According to the third aspect of the present invention, there is provided a respiratory tract infection pathogen recognition system, which includes: a host computer deployed with a trained respiratory tract infection pathogen recognition model and an index detection device communicating with the host computer, wherein:

[0028] The index detection device is used to obtain the index detection data of the respiratory tract infection pathogen of the user and send the index detection data to the host computer;

[0029] The host computer is used to input the index detection data into the respiratory tract infection pathogen recognition model to obtain the respiratory tract infection pathogen recognition result output by the respiratory tract infection pathogen recognition model; wherein, the respiratory tract infection pathogen recognition model is a trained model obtained by using the first aspect or any one of the possible implementation manners of the first aspect.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] First, good recognition effect: In the embodiment of the present application, a respiratory tract infection pathogen recognition model is established, and the pathogens of respiratory tract infection patients are differentially diagnosed through conventional indexes such as blood cell count and CRP, which can effectively distinguish one of the nine pathogens from other different pathogens, and it is a rapid and simple method for early diagnosis of pathogens;

[0032] Second, reduce sampling requirements: The respiratory tract infection pathogen recognition model established in the embodiment of the present application relies on the parameters of blood specimens. Compared with throat swab and sputum specimens, blood specimens are not affected by sampling factors and can avoid false negatives caused by unqualified sampling;

[0033] III. Improving the disposal efficiency: The respiratory tract infection pathogen recognition model established in the embodiments of the present application differentiates and diagnoses the pathogens of patients with respiratory tract infections through routine indicators such as blood cell count and CRP. Compared with nucleic acid testing and bacterial culture, blood routine and CRP testing are simple, rapid, and low-cost, greatly improving the disposal efficiency of the fever clinic, enabling patients to receive accurate treatment more promptly, and reducing the queuing and waiting time for medical treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a schematic flowchart of the model training method provided by the embodiments of the present application;

[0035] Figure 2 It is the ROC curve of nine pathogen subgroups in a certain application scenario provided by the embodiments of the present application;

[0036] Figure 3 It is a two-dimensional clustering diagram in a certain application scenario provided by the embodiments of the present application;

[0037] Figure 4 It is a schematic structural diagram of the respiratory tract infection pathogen recognition system provided by the embodiments of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0038] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention. As used herein, "one embodiment" or "embodiment" refers to a specific feature, structure, or characteristic that may be included in at least one implementation manner of the present application.

[0039] Figure 1 It is a flowchart of the model training method provided in the embodiments of the present application. The present application provides the method operation steps as described in the embodiments or the flowchart, but based on routine or non-creative labor, more or fewer operation steps may be included. The step order listed in the embodiments is only one of the execution orders of numerous steps and does not represent the only execution order. This method should be implemented in software and / or hardware. Please refer to Figure 1 , the above model training method may include:

[0040] Step S110: Obtain the clinical data of respiratory tract infection pathogens;

[0041] Step S120: Preprocess the clinical data to obtain a training set;

[0042] Step S130: Train the respiratory tract infection pathogen recognition model using the training set to obtain a trained respiratory tract infection pathogen recognition model; wherein, the trained respiratory tract infection pathogen recognition model is used to recognize respiratory tract infection pathogens based on the index detection data of respiratory tract infection pathogens.

[0043] Optionally, the respiratory tract infection pathogens include at least one of influenza A virus, influenza B virus, respiratory syncytial virus, adenovirus, human rhinovirus, Mycoplasma pneumoniae, novel coronavirus, fungi, and bacteria.

[0044] Optionally, the above step S110 may include: obtaining the complete blood count (CBC) index data and C-reactive protein (CRP) index data of the enrolled population based on the clinical cohort of respiratory tract infection pathogens. This implementation method is as follows:

[0045] Establish nine clinical cohorts according to the nucleic acid detection of influenza A virus, influenza B virus, respiratory syncytial virus, adenovirus, human rhinovirus, Mycoplasma pneumoniae, novel coronavirus, and fungi, as well as the results of sputum bacterial culture;

[0046] Record the complete blood count (CBC) and C-reactive protein (CRP) index data of the enrolled population in the clinical cohort. All detection indexes are baseline results before receiving any form of treatment.

[0047] It can be understood that the above-mentioned enrolled population can exclude the following criteria: no clear pathogen; mixed infection; receiving drug or infusion treatment before detection; for the same patient with multiple detection results, the first detection result is included.

[0048] Optionally, the above step S120 may include: removing duplicates from the clinical data; encoding the categorical features in the de-duplicated clinical data and separating the categorical features from the labels (categorical columns).

[0049] Optionally, the above step S130 may include: building a model based on XGBClassifier, monitoring the loss of multi-class classification using the multi-class classification negative log-likelihood function mlogloss, and training on the training set using the fit method to obtain a trained respiratory tract infection pathogen recognition model. This implementation method is as follows:

[0050] Using the Pycharm 2024.2 software, programming through Python (3.12), the XGBoost classification model (XGBClassifier) was used for modeling, and the data was trained and predicted through this algorithm. When modeling, the outdated use_label_encoder was removed, and eval_metric was set to mlogloss to monitor the loss of multi-classification. The model was trained on the training set using the fit method to learn the relationship between features and labels.

[0051] It can be understood that the above XG-Boost (Extreme gradient boosting) is a type of algorithm called Ensemble Learning, which integrates the prediction results of multiple classifiers as the final prediction result. XG-Boost is one of the most popular machine learning tools currently, and it has unique advantages, such as fast running speed, parallelization. Most importantly, its operation principle can avoid the influence of missing data, which is difficult for many other algorithms to achieve, and it has been widely used clinically.

[0052] Optionally, the above model training method may further include: using the test set to evaluate the trained respiratory tract infection pathogen recognition model.

[0053] Optionally, the evaluation parameters for the above model evaluation include at least one of the ROC curve, specificity, and sensitivity.

[0054] It can be understood that the diagnosis of pathogens in clinically diagnosed respiratory tract infection patients mainly relies on bacterial culture and virus nucleic acid detection. Bacterial culture takes a long time. Although RT-PCR is effective, with high sensitivity and specificity, it has a high inspection cost, takes a long time, and has high hardware requirements. The embodiment of the present application established a respiratory tract infection pathogen recognition model CBCC, which can effectively distinguish one of the nine pathogens from other different pathogens through routine indicators such as blood cell count and CRP for the pathogens of respiratory tract infection patients. It is a rapid and simple method for early diagnosis of pathogens, which can enable patients to receive timely and accurate treatment earlier and has high economic benefits.

[0055] The following provides a specific application scenario to introduce the effectiveness of the above respiratory tract infection pathogen recognition model:

[0056] A total of 21,429 patients with respiratory tract infections who visited a certain hospital between January 2023 and March 2024 were selected. The diagnosis of pathogens was based on six respiratory pathogen nucleic acid tests (influenza A virus, influenza B virus, respiratory syncytial virus, adenovirus, human rhinovirus, and Mycoplasma pneumoniae), nucleic acid test for novel coronavirus, three fungal nucleic acid tests (Aspergillus spp., Cryptococcus neoformans, and Pneumocystis jirovecii), and sputum bacterial culture results. All patients underwent blood routine and C-reactive protein (CRP) tests, and all test indicators were baseline results before receiving any form of treatment.

[0057] Exclusion criteria were: (1) no clear pathogen, (2) mixed infection, (2) having received drug or infusion treatment before testing, (4) for the same patient with multiple test results, the first test result was included. According to the above criteria, a total of 5,611 patients with respiratory tract infections were finally enrolled and divided into 9 subgroups according to pathogen test results: (1) 810 cases of influenza A virus, (2) 1,432 cases of influenza B virus, (3) 209 cases of respiratory syncytial virus, (4) 77 cases of adenovirus, (5) 353 cases of human rhinovirus, (6) 95 cases of Mycoplasma pneumoniae, (7) 1,155 cases of novel coronavirus, (8) 129 cases of fungi, and (9) 1,351 cases of bacteria.

[0058] For respiratory pathogen nucleic acid detection, six respiratory pathogen qualitative detection reagents, nucleic acid qualitative detection reagent for novel coronavirus (2019-nCoV RNA), nucleic acid detection reagents for Aspergillus spp., Cryptococcus neoformans, and Pneumocystis jirovecii were used, and all adopted real-time fluorescence quantitative PCR detection technology. For sputum culture, blood agar plates, chocolate plates, and MacConkey plates were used for culture and identification.

[0059] The results of model evaluation for the trained respiratory tract infection pathogen recognition model were as follows:

[0060] The ROC curve, specificity, and sensitivity of the validation set were plotted to demonstrate the classification ability and performance of the model. All statistics were programmed using Python (3.12). The normality test was performed using scipy.stats.normaltest by calculating the skewness and kurtosis of the sample data and comparing them with the expected values of the normal distribution. To determine whether the numerical data within each category conforms to the normal distribution, normal distribution data is represented by the mean ± standard deviation (SD). Non-normal distribution data is represented by the median and quartiles (25th and 75th percentiles). The non-parametric test for multiple group comparisons (Kruskal-Wallis test) was performed using Python scipy.stats.kruskal to compare the median differences between different categories. A P < 0.05 was considered statistically significant.

[0061] The efficacy analysis of the above-mentioned respiratory tract infection pathogen recognition model is as follows:

[0062] As Figure 2 shown in the comparison of the ROC curves of the nine pathogen subgroups, Figure 2 in the figure, diagrams 1-9 correspond to influenza A virus, influenza B virus, respiratory syncytial virus, adenovirus, human rhinovirus, Mycoplasma pneumoniae, novel coronavirus, fungi, and bacteria, respectively. The specific diagnostic efficacy parameters are shown in Table 1.

[0063] The area under the curve (AUC) of influenza A virus was 0.71, the specificity was 74%, and the sensitivity was 62%; the AUC of influenza B virus was 0.85, the specificity was 70%, and the sensitivity was 85%; the AUC of respiratory syncytial virus was 0.83, the specificity was 63%, and the sensitivity was 80%; the AUC of adenovirus was 0.72, the specificity was 73%, and the sensitivity was 62%; the AUC of human rhinovirus was 0.78, the specificity was 62%, and the sensitivity was 80%; the AUC of Mycoplasma pneumoniae was 0.83, the specificity was 79%, and the sensitivity was 86%; the AUC of the novel coronavirus was 0.67, the specificity was 68%, and the sensitivity was 66%; the AUC of fungi was 0.85, the specificity was 65%, and the sensitivity was 83%; the AUC of bacteria was 0.94, the specificity was 83%, and the sensitivity was 95%. It can be seen that the XG-BOOST model established based on blood routine and CRP parameters can effectively distinguish nine different pathogens in patients with respiratory tract infections. As Figure 3As shown, in order to more intuitively and realistically reflect the classification accuracy, a two-dimensional clustering map was made using T-SNE. The points of different colors in the figure represent different types of pathogens. When high-dimensional data is reduced to low-dimensional data, there will be a certain amount of information loss. However, this is already the best visualization method today. It can be seen from the figure that different regions are clearly distinguishable. The above Figure 3 The diagrams 1-9 in it respectively correspond to influenza A virus, influenza B virus, respiratory syncytial virus, adenovirus, human rhinovirus, Mycoplasma pneumoniae, novel coronavirus, fungi, and bacteria.

[0064] Table 1 Diagnostic efficacy parameters of the respiratory tract infection pathogen recognition model

[0065] Subgroup Specificity Sensitivity AUC Area 1 0.74 0.62 0.71 2 0.70 0.85 0.85 3 0.63 0.80 0.83 4 0.73 0.62 0.72 5 0.62 0.80 0.78 6 0.79 0.86 0.83 7 0.68 0.66 0.67 8 0.65 0.83 0.85 9 0.83 0.95 0.94

[0066] Based on the same inventive concept, an embodiment of the present application further provides a method for recognizing respiratory tract infection pathogens, which includes:

[0067] Obtain the index detection data of the respiratory tract infection pathogen of the user;

[0068] Input the index detection data into the respiratory tract infection pathogen recognition model to obtain the respiratory tract infection pathogen recognition result output by the respiratory tract infection pathogen recognition model; wherein, the respiratory tract infection pathogen recognition model is a trained model obtained by using the above model training method.

[0069] Optionally, the above-mentioned obtaining the index detection data of the respiratory tract infection pathogen of the user includes: obtaining the complete blood count CBC index data and C-reactive protein CRP index data of the respiratory tract infection pathogen of the user.

[0070] Please refer to Figure 4 , based on the same inventive concept, an embodiment of the present application further provides a respiratory tract infection pathogen recognition system 200, which includes: a host computer 210 deployed with a trained respiratory tract infection pathogen recognition model and an index detection device 220 communicating with the host computer 210, wherein:

[0071] The index detection device 220 is used to obtain the index detection data of the respiratory tract infection pathogen of the user and send the index detection data to the host computer 210;

[0072] The host computer 210 is used to input the index detection data into the respiratory tract infection pathogen recognition model to obtain the respiratory tract infection pathogen recognition result output by the respiratory tract infection pathogen recognition model; wherein, the respiratory tract infection pathogen recognition model is a trained model obtained by using the above model training method.

[0073] As described above, it is only the specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. A model training method, characterized in that: The method comprises: Obtain clinical data on pathogens of respiratory infections; Preprocessing the clinical data to obtain a training set; The training set is used to train a respiratory infection pathogen recognition model to obtain the trained respiratory infection pathogen recognition model; wherein the trained respiratory infection pathogen recognition model is used to identify the respiratory infection pathogen based on the indicator detection data of the respiratory infection pathogen.

2. The model training method according to claim 1, characterized in that: The respiratory tract infection pathogens include: at least one of influenza A virus, influenza B virus, respiratory syncytial virus, adenovirus, human rhinovirus, Mycoplasma pneumoniae, new coronavirus, fungi and bacteria.

3. The model training method according to claim 1, characterized in that: The clinical data of respiratory tract infection pathogens are obtained, including: Based on the clinical cohort of the respiratory tract infection pathogen, the complete blood cell count (CBC) indicator data and C-reactive protein (CRP) indicator data of the enrolled population were obtained.

4. The model training method according to claim 1, characterized in that: Preprocess the clinical data to obtain a training set, including: Deduplication of the clinical data; The classification features in the clinical data after deduplication are encoded, and the classification features are separated from the labels.

5. The model training method according to claim 1, characterized in that: The method of training the respiratory tract infection pathogen recognition model using the training set to obtain the trained respiratory tract infection pathogen recognition model includes: Modeling was performed based on XGBClassifier, and the negative log-likelihood function mlogloss of multi-class classification was used to monitor the loss of multi-class classification. The fit method was used to perform training on the training set to obtain the trained respiratory infection pathogen recognition model.

6. The model training method according to any one of claims 1 to 5, characterized in that: The method further comprises: The test set is used to evaluate the trained respiratory infection pathogen recognition model.

7. The model training method according to claim 6, characterized in that: The evaluation parameters of the model evaluation include: at least one of ROC curve, specificity and sensitivity.

8. A method for identifying pathogens of respiratory tract infection, characterized in that: The method comprises: Obtain the indicator detection data of the user's respiratory infection pathogens; The indicator detection data is input into a respiratory infection pathogen identification model to obtain a respiratory infection pathogen identification result output by the respiratory infection pathogen identification model; wherein the respiratory infection pathogen identification model is a trained model obtained by the method described in any one of claims 1 to 7.

9. The method for identifying respiratory tract infection pathogens according to claim 8, characterized in that: The step of obtaining the index detection data of the user's respiratory tract infection pathogens includes: Obtain the user's complete blood cell count (CBC) indicator data and C-reactive protein (CRP) indicator data of respiratory tract infection pathogens.

10. A respiratory tract infection pathogen identification system, characterized in that: The system includes: a host computer equipped with a trained respiratory infection pathogen recognition model and an indicator detection device communicating with the host computer, wherein: The index detection device is used to obtain the index detection data of the user's respiratory tract infection pathogens and send the index detection data to the host computer; The host computer is used to input the indicator detection data into the respiratory infection pathogen identification model to obtain the respiratory infection pathogen identification result output by the respiratory infection pathogen identification model; wherein the respiratory infection pathogen identification model is a trained model obtained by the method described in any one of claims 1 to 7.

Images