Acute brain injury patient pulmonary infection prediction method and system based on machine learning

Abstract

The invention relates to a lung infection prediction method and system for an acute brain injury patient based on machine learning, and relates to the field of artificial intelligence auxiliary diagnosis, and the method comprises the steps: carrying out the correlation sorting of a monitoring index set from large to small, and obtaining a monitoring index sequence; 2, extracting the first k monitoring indexes as input variables to train an infection prediction model to obtain a first infection prediction model; when the first prediction accuracy rate of the first infection prediction model is smaller than or equal to a prediction accuracy rate threshold value, using k + 1 to update the k value, and returning to the step 2 to execute circulation; and when the first prediction accuracy is greater than a prediction accuracy threshold, setting the first pulmonary infection prediction model as a target pulmonary infection prediction model to execute an infection prediction task. According to the invention, the technical problems of insufficient prediction accuracy, reliability and efficiency caused by incapability of effectively integrating multiple influence factors and poor adaptability and practicability of a traditional lung infection prediction method for the patient with acute brain injury can be solved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence-assisted diagnosis, and in particular to a method and system for predicting lung infection in patients with acute brain injury based on machine learning. Background Art

[0002] Acute brain injury (ABI) encompasses a range of neurological conditions that can lead to acute functional deficits, including ischemic or hemorrhagic stroke, subarachnoid hemorrhage due to aneurysm, and traumatic brain injury, resulting in approximately 12 million deaths each year.

[0003] Due to the uncertainty of the long-term functional prognosis of ABI, patients often have a longer hospital stay, and most require endotracheal intubation or tracheotomy, as well as mechanical ventilation and control of intracranial pressure. In addition, ABI is associated with inflammation and autonomic nervous system-mediated immune system changes, which may increase susceptibility to infection during and after hospitalization. Therefore, ABI patients with hospital-acquired infections are prone to lung infections, including ventilator-associated pneumonia and hospital-acquired pneumonia. Studies have shown that hospital-acquired pneumonia can significantly prolong the patient's hospital stay and increase mortality and disability rates.

[0004] Traditional lung infection detection methods generally include CPIS score, A2DS2 score, AIS-APS score, etc. Since there are many possible factors leading to lung infection in ABI patients, the above methods are not adaptable and practical enough in the process of analyzing lung infection in ABI patients, and have certain limitations. Therefore, how to improve the accuracy of predicting lung infection in ABI patients to assist doctors in improving the controllability of lung infection in ABI patients has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] The present invention aims to solve the technical problem that traditional methods for predicting lung infection in patients with acute brain injury cannot effectively integrate multiple influencing factors, have poor adaptability and practicality, and lead to insufficient prediction accuracy, reliability and efficiency. A method and system for predicting lung infection in patients with acute brain injury based on machine learning is provided to solve the problem.

[0006] The technical solution of the present invention to solve the above technical problems is as follows:

[0007] In a first aspect, the present invention provides a method for predicting lung infection in patients with acute brain injury based on machine learning, comprising: step one: sorting a set of monitoring indicators of patients with acute brain injury by correlation from large to small based on lung infection data of patients with acute brain injury, and obtaining a sequence of monitoring indicators of patients with acute brain injury; step two: extracting the first k monitoring indicators of patients with acute brain injury from the sequence of monitoring indicators of patients with acute brain injury as input variables, and training a prediction model for lung infection in patients with acute brain injury in combination with the lung infection data of patients with acute brain injury, and obtaining a first lung infection prediction model, wherein the initial value of k is equal to 1, and k is an integer; step three: when the first prediction accuracy of the first lung infection prediction model is less than or equal to the prediction accuracy threshold, using k+1 to update the k value, and returning to step two to execute a loop; step four: when the first prediction accuracy of the first lung infection prediction model is greater than the prediction accuracy threshold, setting the first lung infection prediction model as the target lung infection prediction model and building it into the platform server to perform the lung infection prediction task for patients with acute brain injury.

[0008] In a second aspect, the present invention provides a lung infection prediction system for patients with acute brain injury based on machine learning, including: a monitoring indicator correlation analysis module, which is used to sort the correlation of a set of monitoring indicators of patients with acute brain injury from large to small based on the lung infection data of patients with acute brain injury, and obtain a monitoring indicator sequence of patients with acute brain injury; a lung infection prediction model training module, which is used to extract the first k items of the monitoring indicators of patients with acute brain injury in the monitoring indicator sequence of patients with acute brain injury as input variables, and train the lung infection prediction model of patients with acute brain injury in combination with the lung infection data of patients with acute brain injury to obtain a first lung infection prediction model, wherein the initial value of k is equal to 1, and k is an integer; an infection prediction model loop training module, which is used to update the k value using k+1 when the first prediction accuracy of the first lung infection prediction model is less than or equal to the prediction accuracy threshold, and return to step two to execute the loop; a lung infection prediction module, which is used to set the first lung infection prediction model as the target lung infection prediction model when the first prediction accuracy of the first lung infection prediction model is greater than the prediction accuracy threshold, and is built into the platform server to perform the lung infection prediction task for patients with acute brain injury.

[0009] The beneficial effects of the present invention are: by sorting the acute brain injury patient monitoring indicator set from large to small according to the correlation based on the acute brain injury patient lung infection data, a sequence of acute brain injury patient monitoring indicators is obtained; then the first k acute brain injury patient monitoring indicators of the acute brain injury patient monitoring indicator sequence are extracted as input variables, and the acute brain injury patient lung infection prediction model is trained in combination with the acute brain injury patient lung infection data to obtain a first lung infection prediction model, wherein the initial value of k is equal to 1, and k is an integer; then when the first prediction accuracy of the first lung infection prediction model is less than or equal to the prediction accuracy threshold, the k value is updated using k+1, and the cyclic training of the lung infection prediction model is performed; until the first prediction accuracy of the first lung infection prediction model is greater than the prediction accuracy threshold, the first lung infection prediction model is set as the target lung infection prediction model; and the target lung infection prediction model is built into the platform server to perform the acute brain injury patient lung infection prediction task. In other words, by analyzing key influencing variables, optimizing the selection of monitoring indicators, and combining machine learning to build a predictive model, more accurate result predictions can be achieved based on the least influencing variables, reducing computing power requirements, and significantly improving the efficiency, accuracy, and reliability of predicting lung infections in patients with acute brain injury, thereby effectively assisting doctors in identifying infection risks early. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A schematic diagram of a process for predicting lung infection in patients with acute brain injury based on machine learning provided by the present invention;

[0011] Figure 2 A schematic diagram of a set of monitoring indicators for patients with acute brain injury in the method for predicting lung infection in patients with acute brain injury based on machine learning provided by the present invention;

[0012] Figure 3 , Figure 4 A schematic diagram of the training prediction performance and verification prediction performance of multiple algorithms in the method for predicting lung infection in patients with acute brain injury based on machine learning provided by the present invention;

[0013] Figure 5 A schematic diagram of lung infection prediction by a platform server in the method for predicting lung infection in patients with acute brain injury based on machine learning provided by the present invention;

[0014] Figure 6 A schematic diagram of the structure of the machine learning-based lung infection prediction system for patients with acute brain injury provided by the present invention.

[0015] In the accompanying drawings, the components represented by the reference numerals are described as follows:

[0016] Monitoring index correlation analysis module 01, lung infection prediction model training module 02, infection prediction model cycle training module 03, lung infection prediction module 04. DETAILED DESCRIPTION

[0017] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0018] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0019] In the description of the present invention, the term "for example" is used to mean "used as an example, illustration or explanation". Any embodiment described as "for example" in the present invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any technician in the field to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes will not be elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed in the present invention.

[0020] Embodiment 1, as Figure 1 As shown, the embodiment of the present invention provides a method for predicting lung infection in patients with acute brain injury based on machine learning, which specifically includes the following steps:

[0021] Step 1: Based on the lung infection data of patients with acute brain injury, the set of monitoring indicators of patients with acute brain injury is sorted from large to small in terms of relevance to obtain a sequence of monitoring indicators of patients with acute brain injury.

[0022] Further, step one of the present invention also includes:

[0023] Obtain a first acute brain injury patient monitoring indicator of the acute brain injury patient monitoring indicator set.

[0024] Specifically, first, a set of monitoring indicators for patients with acute brain injury is obtained, such as Figure 2 As shown, the set of monitoring indicators for patients with acute brain injury includes at least the top 20 important monitoring indicators (which can be obtained by lightweight gradient boosting machine algorithm analysis), namely, tracheotomy time, antibiotic use time (the time from the time when patients with acute brain injury (ABI) start using antibiotics), blood sugar, age, mechanical ventilation time (the time interval from the time when patients with acute brain injury (ABI) start receiving mechanical ventilation treatment until the time when they stop using the ventilator), CRP (a marker of inflammatory response in the body, a higher CRP level may indicate the presence of active infection), GCS score, body temperature, lymphocyte percentage (the percentage of lymphocytes in the blood to the total cells), lymphocyte count (the absolute number of lymphocytes in the blood), gastric tube, albumin, proton pump inhibitors (whether proton pump inhibitors are used), intraoperative hypothermia (the lowest temperature during the operation when the core body temperature is lower than 36 degrees Celsius), total protein, red blood cell count, mean hemoglobin concentration, hematocrit, operation time (operation duration) and mean red blood cell volume.

[0025] Next, any one of the acute brain injury patient monitoring indicators is randomly selected from the acute brain injury patient monitoring indicator set as the first acute brain injury patient monitoring indicator, such as tracheotomy time or antibiotic use time.

[0026] Taking the first acute brain injury patient monitoring index as the independent variable and the lung infection severity coefficient as the dependent variable, correlation analysis and sample sorting are performed on the lung infection data of the acute brain injury patients to obtain the characteristic value sequence of the first acute brain injury patient monitoring index and the calibrated value sequence of the lung infection severity coefficient.

[0027] Furthermore, the present invention further comprises the following steps:

[0028] Obtain the first sample of the lung infection data of the acute brain injury patient, wherein the first sample of the lung infection data of the acute brain injury patient includes a first sample of the acute brain injury patient monitoring indicator characteristic value set and a first sample of the lung infection severity coefficient calibration value, and the first sample of the lung infection severity coefficient calibration value is the doctor calibration data; obtain the second sample of the lung infection data of the acute brain injury patient, wherein the second sample of the lung infection data of the acute brain injury patient includes a second sample of the acute brain injury patient monitoring indicator characteristic value set and a second sample of the lung infection severity coefficient calibration value, and the second sample of the lung infection severity coefficient calibration value is the doctor calibration data; delete the first acute brain injury patient monitoring indicator from the first sample of the acute brain injury patient monitoring indicator characteristic value set to obtain a first set of acute brain injury patient monitoring indicator characteristic values ​​to be analyzed; delete the first acute brain injury patient monitoring indicator from the second sample of the acute brain injury patient monitoring indicator characteristic value set to obtain a second set of acute brain injury patient monitoring indicator characteristic values ​​to be analyzed; when the first set of acute brain injury patient monitoring indicator characteristic values ​​to be analyzed is equal to The high-dimensional distribution distance of the second set of acute brain injury patient monitoring indicator characteristic values ​​to be analyzed is less than or equal to the distribution distance threshold, and the first acute brain injury patient monitoring indicator characteristic value of the first sample acute brain injury patient monitoring indicator characteristic value set and the second sample acute brain injury patient monitoring indicator characteristic value set are added to the first acute brain injury patient monitoring indicator characteristic value sequence, and the first sample lung infection severity coefficient calibration value and the second sample lung infection severity coefficient calibration value are added to the lung infection severity coefficient calibration value sequence, wherein the high-dimensional distribution distance is the N-dimensional Euclidean distance of the normalized values ​​of the first set of acute brain injury patient monitoring indicator characteristic values ​​to be analyzed and the second set of acute brain injury patient monitoring indicator characteristic values ​​to be analyzed, and N is the number of compared acute brain injury patient monitoring indicator attributes; otherwise, the first sample acute brain injury patient lung infection data and / or the second sample acute brain injury patient lung infection data are updated for cyclic analysis; until the first acute brain injury patient monitoring indicator characteristic value sequence meets the preset data volume, the first acute brain injury patient monitoring indicator characteristic value sequence and the lung infection severity coefficient calibration value sequence are output.

[0029] Specifically, first, obtain lung infection data of patients with acute brain injury, such as extracting relevant data of patients within a historical time period (such as within the past year) from the hospital's electronic health record system or clinical database. These data include basic information of patients (age, gender, length of hospitalization, type of disease, etc.), clinical monitoring indicators (tracheotomy time, mechanical ventilation time, blood sugar, etc.), treatment process (such as duration of antibiotic use, use of sedatives, type of surgery, whether tracheotomy was performed, etc.) and severity of lung infection (doctors determine the severity of pneumonia, such as mild, moderate or severe, based on lung images, clinical manifestations and laboratory test results).

[0030] Next, any one infection data is randomly selected from the lung infection data of the acute brain injury patient and set as the first sample lung infection data of the acute brain injury patient, wherein the first sample lung infection data of the acute brain injury patient includes a first sample acute brain injury patient monitoring indicator characteristic value set and a first sample lung infection severity coefficient calibration value, the first sample acute brain injury patient monitoring indicator characteristic value set includes various physiological, laboratory and clinical data of the acute brain injury patient (such as tracheotomy time, antibiotic use time, blood sugar, CRP, etc.); the first sample lung infection severity coefficient calibration value is the doctor calibration data, that is, the lung infection severity score calibrated by the clinician based on the patient's clinical manifestations, examination results and imaging manifestations, wherein the larger the severity coefficient calibration value, the more severe the lung infection, such as the first sample lung infection severity coefficient calibration value is 30, which represents moderate pneumonia. Then, any one infection data other than the first sample lung infection data of acute brain injury patients is randomly selected from the lung infection data of acute brain injury patients and set as the second sample lung infection data of acute brain injury patients, wherein the second sample lung infection data of acute brain injury patients includes a second sample acute brain injury patient monitoring indicator characteristic value set and a second sample lung infection severity coefficient calibration value, and the second sample lung infection severity coefficient calibration value is the doctor calibration data.

[0031] Further eliminate the first acute brain injury patient monitoring indicator (such as tracheotomy time or antibiotic use time) in the first sample acute brain injury patient monitoring indicator characteristic value set to obtain a first acute brain injury patient monitoring indicator characteristic value set to be analyzed; on the other hand, eliminate the first acute brain injury patient monitoring indicator in the second sample acute brain injury patient monitoring indicator characteristic value set to obtain a second acute brain injury patient monitoring indicator characteristic value set to be analyzed.

[0032] Next, the high-dimensional distribution distance between the first set of characteristic values ​​of monitoring indicators of patients with acute brain injury to be analyzed and the second set of characteristic values ​​of monitoring indicators of patients with acute brain injury to be analyzed is calculated, wherein the high-dimensional distribution distance is the N-dimensional Euclidean distance of the normalized values ​​of the first set of characteristic values ​​of monitoring indicators of patients with acute brain injury to be analyzed and the second set of characteristic values ​​of monitoring indicators of patients with acute brain injury to be analyzed, and N is the number of attributes of the compared monitoring indicators of patients with acute brain injury, that is, the number of features (number of dimensions) of the monitoring indicators of patients with acute brain injury to be compared; before calculating the Euclidean distance, the characteristic values ​​of the monitoring indicators are first normalized (such as standardization or Min-Max normalization) to ensure that characteristic values ​​of different dimensions have equal weights when calculating the distance.

[0033] Then set the distribution distance threshold, which can be set according to the actual scenario and is used to judge the similarity between the two data sets. If the high-dimensional distribution distance of the two data sets is less than or equal to this threshold, it is considered that their monitoring indicator characteristic value distributions are similar, and it is considered that the patients they represent have similar clinical characteristics and can be merged; on the contrary, if the high-dimensional distribution distance of the two data sets is greater than this threshold, they cannot be merged. Then, according to the distribution distance threshold, the high-dimensional distribution distance of the first set of monitoring indicator characteristic values ​​of patients with acute brain injury to be analyzed and the second set of monitoring indicator characteristic values ​​of patients with acute brain injury to be analyzed is judged. When the high-dimensional distribution distance is less than or equal to the distribution distance threshold, the first set of monitoring indicator characteristic values ​​of patients with acute brain injury in the first sample and the first set of monitoring indicator characteristic values ​​of patients with acute brain injury in the second sample are added to the first set of monitoring indicator characteristic values ​​of patients with acute brain injury; at the same time, the calibration value of the severity coefficient of lung infection of the first sample and the calibration value of the severity coefficient of lung infection of the second sample are added to the calibration value sequence of severity coefficient of lung infection.

[0034] If the high-dimensional distribution distance is greater than the distribution distance threshold, the first sample acute brain injury patient lung infection data and / or the second sample acute brain injury patient lung infection data are updated, that is, any set of sample data (including two sample data) is reselected from the acute brain injury patient lung infection data for cyclic analysis, and the sample data groups analyzed and compared each time are not exactly the same. Until the first acute brain injury patient monitoring indicator characteristic value sequence meets the preset data volume (which can be set according to demand, such as 500 data), the first acute brain injury patient monitoring indicator characteristic value sequence and the corresponding lung infection severity coefficient calibration value sequence are output.

[0035] Perform a Pearson correlation analysis on the first acute brain injury patient monitoring indicator characteristic value sequence and the lung infection severity coefficient calibration value sequence to obtain a first Pearson correlation coefficient, and add it into a Pearson correlation coefficient set, wherein the Pearson correlation coefficient set corresponds one-to-one to the acute brain injury patient monitoring indicator set; construct a Pearson correlation coefficient heat map based on the Pearson correlation coefficient set; sort the acute brain injury patient monitoring indicator set from large to small according to the Pearson correlation coefficient heat map to obtain the acute brain injury patient monitoring indicator sequence.

[0036] Specifically, a Pearson correlation analysis is performed on the characteristic value sequence of the first acute brain injury patient monitoring indicator and the pulmonary infection severity coefficient calibration value sequence. The Pearson correlation coefficient is a statistic that measures the strength and direction of the linear relationship between two variables, and its value range is -1 to +1, where +1 indicates a complete positive correlation, that is, when one variable increases, the other variable also increases proportionally; 0 indicates no linear relationship; -1 indicates a complete negative correlation, that is, when one variable increases, the other variable decreases proportionally. Among them, the characteristic value sequence of the first acute brain injury patient monitoring indicator is an independent variable, which represents the various clinical monitoring indicator data of the patient; the pulmonary infection severity coefficient calibration value sequence is a dependent variable, which represents the severity of the pulmonary infection calibrated by the doctor. By calculating the Pearson correlation coefficient between the two, the correlation between each monitoring indicator and the pulmonary infection severity coefficient can be evaluated, and the variable with a strong correlation with the severity of the pulmonary infection can be found; the first Pearson correlation coefficient is obtained, and the first Pearson correlation coefficient corresponds to the first acute brain injury patient monitoring indicator.

[0037] Then, using the same analysis method as that used to calculate the first Pearson correlation coefficient, correlation analysis is performed on other acute brain injury patient monitoring indicators in the acute brain injury patient monitoring indicator set to obtain a set of Pearson correlation coefficients for the acute brain injury patient monitoring indicator set, wherein the Pearson correlation coefficients and the acute brain injury patient monitoring indicators correspond one to one.

[0038] Then, a Pearson correlation coefficient heat map is constructed based on the Pearson correlation coefficient set, wherein a heat map is a visualization tool for displaying the relationship between data, which can help identify the correlation between different variables; based on the previously calculated Pearson correlation coefficient set, the Pearson correlation coefficient between each monitoring indicator and the severity coefficient of lung infection will be used as the value of the heat map, and the rows and columns of the heat map represent the monitoring indicators of patients with acute brain injury, respectively. The depth of the color indicates the strength of the correlation. Generally, the darker the color, the stronger the correlation, and the lighter the color, the weaker the correlation. In the Pearson correlation coefficient heat map, the horizontal and vertical axes are different monitoring indicators of patients with acute brain injury, and the color depth (heat) indicates the numerical value of the Pearson correlation coefficient. For example, if the Pearson correlation coefficient between a monitoring indicator (such as tracheotomy time) and the severity coefficient of lung infection is 0.8, then in the heat map, the correlation position of this item will be displayed as a darker color, indicating that it has a strong positive correlation with the severity of lung infection.

[0039] Then, according to the Pearson correlation coefficient heat map, the set of monitoring indicators for patients with acute brain injury is sorted from large to small in terms of correlation. From the heat map, it can be intuitively seen which monitoring indicators have a strong correlation with the severity coefficient of lung infection, and the monitoring indicators are sorted according to the size of the Pearson correlation coefficient, with the strongest correlated indicators in front and the weakest correlated indicators in the back, to obtain the monitoring indicator sequence for patients with acute brain injury. According to the analysis results, tracheotomy time, antibiotic use time, blood sugar level, mechanical ventilation time, and CRP are the top five monitoring indicators with the largest correlation.

[0040] By constructing a Pearson correlation coefficient heat map, we can clearly see the correlation between different monitoring indicators and the severity of lung infection, and quickly identify key influencing factors; by sorting and screening out monitoring indicators with strong correlation and removing monitoring indicators with weak correlation, we can improve the prediction accuracy of the machine learning model, while reducing unnecessary data redundancy and optimizing computing resources and model performance.

[0041] Step 2: Extract the first k acute brain injury patient monitoring indicators of the acute brain injury patient monitoring indicator sequence as input variables, and train the acute brain injury patient lung infection prediction model in combination with the acute brain injury patient lung infection data to obtain the first lung infection prediction model, wherein the initial value of k is equal to 1 and k is an integer.

[0042] Further, step 2 of the present invention also includes:

[0043] Based on the first k acute brain injury patient monitoring indicators of the acute brain injury patient monitoring indicator sequence, a monitoring indicator record value data set is extracted from the lung infection data of the acute brain injury patient; the monitoring indicator record value data set is traversed to perform lung infection probability identification to obtain a lung infection probability identification data set; the lung infection probability identification data set is used as supervision and the monitoring indicator record value data set is used as input data to set the data set for constructing a lung infection prediction model for acute brain injury patients.

[0044] Specifically, first, the first k acute brain injury patient monitoring indicators are extracted from the acute brain injury patient monitoring indicator sequence, where the initial value of k is equal to 1 and k is an integer, that is, the first most relevant indicator (such as tracheotomy time) is selected from the acute brain injury patient monitoring indicator sequence as the input feature of the model; then, the monitoring indicator record value data set of the first k (initial value is 1) acute brain injury patient monitoring indicators is extracted from the acute brain injury patient lung infection data. Then, the multiple monitoring indicator record value data in the monitoring indicator record value data set (such as different tracheotomy times) are respectively marked with the probability of lung infection. The probability of lung infection can be based on historical data or expert calibration, indicating the possibility of a patient having a lung infection, and a lung infection probability mark data set is obtained.

[0045] Then, the monitoring index record value data set is set as input data for model training, and the lung infection probability identification data set is set as supervision data for model training, so as to obtain a data set for constructing a lung infection prediction model for patients with acute brain injury.

[0046] A data set is constructed based on the lung infection prediction model for patients with acute brain injury, and the first lung infection prediction model is trained.

[0047] Furthermore, the present invention further comprises the following steps:

[0048] The data set for constructing a prediction model for lung infection in patients with acute brain injury includes M groups of data for constructing a prediction model for lung infection in patients with acute brain injury, and a first training weight is configured for each group of data for constructing a prediction model for lung infection in patients with acute brain injury in the M groups of data for constructing a prediction model for lung infection in patients with acute brain injury, wherein the first training weight is equal to one-M; and a prediction loss function for lung infection in patients with acute brain injury is constructed: Among them, L(y i ,y i0 ) represents the training loss value of the i-th group, y i Represents the infection probability prediction value of the i-th group training, y i0 represents the infection probability supervision value of the i-th group training, δ represents the demarcation threshold, and w iCharacterize the weight of the i-th group of training data; according to the prediction loss function of lung infection in patients with acute brain injury, call the M groups of lung infection prediction model construction data of patients with acute brain injury and the first training weight to train the first regression decision tree to obtain the first lung infection prediction decision tree, wherein the first lung infection prediction decision tree has a first training group number whose loss value is greater than or equal to the loss threshold.

[0049] Specifically, first, the data set for constructing a prediction model for lung infection in patients with acute brain injury is equally divided into M groups of data for constructing a prediction model for lung infection in patients with acute brain injury, wherein the number of constructed data in each group is the same, M is an integer greater than 1, and the specific value of M can be set according to actual conditions, such as 10; then, a first training weight is configured for each group of data for constructing a prediction model for lung infection in patients with acute brain injury of the M groups of data for constructing a prediction model for lung infection in patients with acute brain injury, and the first training weight is equal to one-M, for example, assuming that M is 10, the first training weight is 1 / 10.

[0050] Next, a prediction loss function for lung infection in patients with acute brain injury is constructed. In the prediction loss function for lung infection in patients with acute brain injury, L(y i ,y i0 ) represents the training loss value of the i-th group, y i Represents the infection probability prediction value of the i-th group training, y i0 represents the infection probability supervision value of the i-th group training, δ represents the demarcation threshold, and w i Characterize the weight of the i-th group of training data. The loss function is the core of optimizing model parameters during training. It reflects the gap between the model prediction results and the actual supervision value. The goal of constructing the loss function is to minimize the prediction error and thus improve the accuracy of the model.

[0051] Among them, when performing algorithm simulation, 7 algorithms are selected to predict the model accuracy, such as Figure 3 , Figure 4As shown in the figure, after research and analysis, the AUC values ​​of the seven algorithms are as follows: XGBoost (AUC = 0.956), LightGBM model (AUC = 0.950), RandomForest (Random Forest) (AUC = 0.918), AdaptiveBoost model (AUC = 0.913), decision tree (AUC = 0.878), TabNet model (AUC = 0.841) and BP model (AUC = 0.815); the validation set found that the XGBoost model had the highest accuracy (0.96), while the random forest and adaptive Boost models had the highest AUC values ​​(AUC = 0.883). Therefore, it was finally decided to select XGBoost (gradient boosting tree) to construct the first lung infection prediction model, indicating that the model is superior to other models in the accuracy of predicting lung infection.

[0052] Next, a first regression decision tree is constructed based on XGBoost (gradient boosted tree). The XGBoost model is usually used for data mining. Compared with deep learning algorithms, it is not easy to overfit on limited data sets and has lower processing requirements, but it performs well under various variable conditions; compared with deep learning models, it does not need to extract larger data sets and is more practical; the first regression decision tree is used to predict the probability or severity of lung infection based on the monitoring indicators of patients with acute brain injury; then the M group of lung infection prediction model construction data for patients with acute brain injury and the first training weight (1 / M) are retrieved, and the first regression decision tree is trained according to the lung infection prediction loss function for patients with acute brain injury until the loss value of the loss function is less than or equal to the loss threshold (which can be set based on the model accuracy requirements), then the training is stopped to obtain the first lung infection prediction decision tree, wherein the first lung infection prediction decision tree has a first training group number whose loss value is greater than or equal to the loss threshold, that is, during the training process, the number of training groups whose loss value is greater than or equal to the set loss threshold will be recorded to represent the maximum number of training groups before stopping the training.

[0053] When the first number of training groups is greater than or equal to the convergence group number threshold, M training accuracy rates of the M groups of acute brain injury patients' lung infection prediction model construction data in the first lung infection prediction decision tree are obtained, and a second training weight is configured for the M groups of acute brain injury patients' lung infection prediction model construction data according to the M training accuracy rates.

[0054] Furthermore, the present invention further comprises the following steps:

[0055] Use 1 to subtract the M training accuracy rates to obtain M training error rates; add the M training error rates to obtain a training error rate sum; calculate the ratio of the M training error rates to the training error rate sum to obtain M training weights, and add them into the second training weight.

[0056] Specifically, a convergence group number threshold is configured, and the convergence group number threshold is a preset parameter used to determine whether the training process converges; when the first training group number is greater than or equal to the convergence group number threshold, M training accuracy rates of the M groups of acute brain injury patient lung infection prediction model construction data in the first lung infection prediction decision tree are obtained. Further, a second training weight is configured for the M groups of acute brain injury patient lung infection prediction model construction data according to the M training accuracy rates.

[0057] First, use 1 to subtract the M training accuracy rates, and set the difference between 1 and the training accuracy rate as the training error rate to obtain M training error rates. Then, add and sum the M training error rates to obtain the training error rate sum value; then, calculate the ratio of the M training error rates to the training error rate sum value, set the ratio as the training weight, and obtain M training weights. For example, assuming that the M training accuracy rates are 0.95, 0.93, and 0.92, respectively, the M training error rates are 0.05, 0.07, and 0.08, respectively, the training error rate sum value is 0.2, and the M training weights are 0.25, 0.35, and 0.4, respectively; and set the M training weights as the second training weight. By adjusting the weight of the training data according to the error rate of each training stage during the model training process, the training stage with a larger error can obtain a higher weight, which helps the model pay more attention to the part with a larger error, thereby improving the overall prediction ability of the model.

[0058] According to the prediction loss function of lung infection in patients with acute brain injury, the M groups of lung infection prediction model construction data for patients with acute brain injury and the second training weights are retrieved to train a second regression decision tree to obtain a second lung infection prediction decision tree, wherein the second lung infection prediction decision tree has a second number of training groups with a loss value greater than or equal to a loss threshold.

[0059] Specifically, a second regression decision tree is constructed based on XGBoost (gradient boosting tree), the M group of acute brain injury patients' lung infection prediction model construction data and the second training weights are retrieved, and the second regression decision tree is trained according to the acute brain injury patients' lung infection prediction loss function until the loss value is less than the loss threshold, then the training is stopped to obtain a trained second lung infection prediction decision tree, wherein the second lung infection prediction decision tree has a second number of training groups with a loss value greater than or equal to the loss threshold, that is, the maximum number of training groups before stopping the training.

[0060] Based on the second number of training groups, cyclic training is performed until the Qth number of training groups is less than the convergence group number threshold, and the first lung infection prediction decision tree, the second lung infection prediction decision tree, and the Qth lung infection prediction decision tree are merged to generate the first lung infection prediction model.

[0061] Furthermore, the present invention further comprises the following steps:

[0062] Obtain a first verification accuracy of the first lung infection prediction decision tree, a second verification accuracy of the second lung infection prediction decision tree, and so on, up to the Qth verification accuracy of the Qth lung infection prediction decision tree; according to the first verification accuracy, the second verification accuracy, and so on, up to the Qth verification accuracy, construct a weighted mean function of the outputs of the first lung infection prediction decision tree, the second lung infection prediction decision tree, and so on, up to the Qth lung infection prediction decision tree, construct a fully connected layer, merge the first lung infection prediction decision tree, the second lung infection prediction decision tree, and so on, to generate the first lung infection prediction model.

[0063] Specifically, the M group of acute brain injury patients' lung infection prediction model construction data and the dynamic training weights of the previous training stage are retrieved to continue the iterative training of the regression decision tree until the Qth training group number is less than the convergence group number threshold, then the training is stopped, and the first lung infection prediction decision tree, the second lung infection prediction decision tree, and so on are obtained.

[0064] Next, the first lung infection prediction decision tree, the second lung infection prediction decision tree, and the Qth lung infection prediction decision tree are respectively verified and trained through the verification data set to obtain the first verification accuracy of the first lung infection prediction decision tree, the second verification accuracy of the second lung infection prediction decision tree, and the Qth verification accuracy of the Qth lung infection prediction decision tree. Further, output credible weights are configured according to the first verification accuracy, the second verification accuracy, and the Qth verification accuracy, wherein the sum of the Q output credible weights is 1, and the weight value is positively correlated with the verification accuracy. The larger the verification accuracy, the larger the corresponding credible weight. The coefficient of variation method can be used to configure the credible weights. Then, a weighted mean function of the outputs of the first lung infection prediction decision tree, the second lung infection prediction decision tree, and the Qth lung infection prediction decision tree is constructed according to the Q credible weights, and a fully connected layer is constructed according to the weighted mean function. The first lung infection prediction decision tree, the second lung infection prediction decision tree, and the Qth lung infection prediction decision tree are merged to generate the first lung infection prediction model, wherein the output of the first lung infection prediction model is the weighted calculation result of the outputs of the first lung infection prediction decision tree, the second lung infection prediction decision tree, and the Qth lung infection prediction decision tree.

[0065] By weighted calculation results, a new prediction model is generated by fusion. This model combines the prediction results of multiple decision trees and assigns different weights according to their respective accuracies. A prediction model with strong robustness, high accuracy and quantifiable credibility can be constructed. This model can effectively predict the risk of lung infection in patients with acute brain injury and provide strong support for clinical decision-making.

[0066] Furthermore, the present invention further comprises the following steps:

[0067] Set the stop training L, where L represents the upper limit number of decision trees that can be merged; when the Yth training group number is greater than or equal to the convergence group number threshold, determine whether Y is greater than or equal to L; if Y is greater than or equal to L, generate the first lung infection prediction model, and configure the first prediction accuracy to be less than the prediction accuracy threshold; if Y is less than L, continue to execute the loop.

[0068] Specifically, first, set the stop training L, where L represents the upper limit of the number of decision trees that can be merged. The value of L is determined according to the specific task requirements, data characteristics, and computing resources. Setting L can prevent the model from being overly complicated and the training time from being too long. When the number of training groups Y is greater than or equal to the convergence group number threshold, determine whether Y is greater than or equal to L. If Y is greater than or equal to L, that is, a sufficient number of decision trees have been generated, then proceed to the next step to generate the first lung infection prediction model, and set the first prediction accuracy of this model to be lower than the set prediction accuracy threshold. If Y is less than L, continue the training process and add more decision trees.

[0069] By setting the stop training L, we can prevent overfitting in the training process and avoid insufficient generalization ability caused by excessive model complexity. At the same time, when the maximum number of mergeable decision trees is reached, we can avoid adding unnecessary computing resources, thereby saving training time and computing overhead.

[0070] Step three: When the first prediction accuracy of the first lung infection prediction model is less than or equal to the prediction accuracy threshold, use k+1 to update the k value and return to step two to execute the loop.

[0071] Specifically, when the first prediction accuracy of the first lung infection prediction model is less than or equal to the prediction accuracy threshold, the prediction accuracy threshold is used to determine whether the performance of the model is good enough, and can be set according to the prediction accuracy requirements. If the accuracy of the model is lower than or equal to the threshold, it means that the model has not yet reached the expected prediction ability, that is, it is impossible to accurately predict lung infection through one monitoring indicator alone. In this case, k+1 updates the k value, that is, increases the number of selected monitoring indicators, and enhances the prediction ability of the model by using more input features (such as more monitoring indicators). After updating the k value, return to step 2 (i.e., extract a new sequence of monitoring indicators for patients with acute brain injury and train a new prediction model), train the model again by adding input features (or adjusting training data), and continuously improve the accuracy of the prediction model by adding features, adjusting parameters or changing the training set until the prediction accuracy of the model reaches the set threshold.

[0072] Step 4: When the first prediction accuracy of the first lung infection prediction model is greater than the prediction accuracy threshold, the first lung infection prediction model is set as the target lung infection prediction model and built into the platform server to perform the lung infection prediction task for patients with acute brain injury.

[0073] Specifically, when the first prediction accuracy of the first lung infection prediction model is greater than the prediction accuracy threshold, the prediction accuracy of the characterization model meets the requirements, indicating that the model can better predict the risk of lung infection in patients with acute brain injury, and at the same time can achieve more accurate result prediction based on the least influencing variables, such as lung infection prediction based on tracheotomy time, antibiotic use time, blood sugar level, mechanical ventilation time and CRP. At this time, the first lung infection prediction model is set as the target lung infection prediction model. Then the target lung infection prediction model is built into the platform server, and the lung infection prediction task of patients with acute brain injury is performed through the target lung infection prediction model, such as Figure 5 As shown, for example, users only need to input these parameters through the platform interface to trigger the underlying algorithm interface and provide diagnosis results in a timely manner. At the same time, in order to make the diagnosis results more intuitive, a dynamic visualization display is designed. On the one hand, the pie chart displays the probability distribution of disease in real time, so that users can immediately understand the patient's condition; on the other hand, trend charts are provided for individual patients to dynamically track their disease progression. This intuitive visualization analysis not only improves the accuracy of diagnosis, but also significantly improves decision-making support for clinical treatment.

[0074] The method for predicting lung infection in patients with acute brain injury based on machine learning provided in the embodiment of the present invention has at least the following technical effects:

[0075] 1. By constructing a Pearson correlation coefficient heat map, we can clearly see the correlation between different monitoring indicators and the severity of lung infection, and quickly identify key influencing factors; by sorting and screening out monitoring indicators with strong correlation and removing monitoring indicators with weak correlation, we can improve the prediction accuracy of the machine learning model, while reducing unnecessary data redundancy and optimizing computing resources and model performance.

[0076] 2. During the model training process, by adjusting the weight of the training data according to the error rate of each training stage, the training stage with larger errors can be given a higher weight, which helps the model pay more attention to the part with larger errors, thereby improving the overall prediction ability of the model.

[0077] 3. By analyzing key influencing variables, optimizing the selection of monitoring indicators, and combining machine learning to build a prediction model, more accurate result predictions can be achieved based on the least influencing variables, reducing computing power requirements, and significantly improving the efficiency, accuracy, and reliability of lung infection prediction in patients with acute brain injury, thereby effectively assisting doctors in identifying infection risks early.

[0078] Embodiment 2, as Figure 6As shown, based on the same inventive concept as the method for predicting lung infection in patients with acute brain injury based on machine learning provided in Example 1, the embodiment of the present invention also provides a system for predicting lung infection in patients with acute brain injury based on machine learning, including:

[0079] The monitoring indicator correlation analysis module 01 is used to sort the acute brain injury patient monitoring indicator set from large to small according to the correlation based on the acute brain injury patient lung infection data, and obtain the acute brain injury patient monitoring indicator sequence; the lung infection prediction model training module 02 is used to extract the first k acute brain injury patient monitoring indicators of the acute brain injury patient monitoring indicator sequence as input variables, and train the acute brain injury patient lung infection prediction model in combination with the acute brain injury patient lung infection data to obtain the first lung infection prediction model, wherein the initial value of k is equal to 1, and k is an integer; the infection prediction model loop training module 03 is used to update the k value using k+1 when the first prediction accuracy of the first lung infection prediction model is less than or equal to the prediction accuracy threshold, and return to step 2 to execute the loop; the lung infection prediction module 04 is used to set the first lung infection prediction model as the target lung infection prediction model when the first prediction accuracy of the first lung infection prediction model is greater than the prediction accuracy threshold, and the built-in platform server performs the acute brain injury patient lung infection prediction task.

[0080] Furthermore, the machine learning-based lung infection prediction system for acute brain injury patients is also used to: obtain a first acute brain injury patient monitoring indicator of the acute brain injury patient monitoring indicator set; use the first acute brain injury patient monitoring indicator as an independent variable and the lung infection severity coefficient as a dependent variable to perform correlation analysis sample sorting on the lung infection data of the acute brain injury patient to obtain a first acute brain injury patient monitoring indicator characteristic value sequence and a lung infection severity coefficient calibration value sequence; perform Pearson correlation analysis on the first acute brain injury patient monitoring indicator characteristic value sequence and the lung infection severity coefficient calibration value sequence to obtain a first Pearson correlation coefficient, and add it to the Pearson correlation coefficient set, wherein the Pearson correlation coefficient set corresponds one-to-one to the acute brain injury patient monitoring indicator set; construct a Pearson correlation coefficient heat map based on the Pearson correlation coefficient set; and sort the acute brain injury patient monitoring indicator set by correlation from large to small based on the Pearson correlation coefficient heat map to obtain the acute brain injury patient monitoring indicator sequence.

[0081] Further, the machine learning-based lung infection prediction system for patients with acute brain injury is also used to: obtain the first sample of lung infection data of the acute brain injury patient, wherein the first sample of lung infection data of the acute brain injury patient includes a first sample of acute brain injury patient monitoring indicator characteristic value set and a first sample of lung infection severity coefficient calibration value, and the first sample of lung infection severity coefficient calibration value is the doctor calibration data; obtain the second sample of lung infection data of the acute brain injury patient, wherein the second sample of lung infection data of the acute brain injury patient includes a second sample of acute brain injury patient monitoring indicator characteristic value set and a second sample of lung infection severity coefficient calibration value, and the second sample of lung infection severity coefficient calibration value is the doctor calibration data; delete the first acute brain injury patient monitoring indicator from the first sample of acute brain injury patient monitoring indicator characteristic value set to obtain a first set of acute brain injury patient monitoring indicator characteristic values ​​to be analyzed; delete the first acute brain injury patient monitoring indicator from the second sample of acute brain injury patient monitoring indicator characteristic value set to obtain a second set of acute brain injury patient monitoring indicator characteristic values ​​to be analyzed; when the first to be analyzed acute The high-dimensional distribution distance between the characteristic value set of monitoring indicators of patients with acute brain injury and the characteristic value set of monitoring indicators of the second patient with acute brain injury to be analyzed is less than or equal to the distribution distance threshold, the first characteristic value of the monitoring indicator of patients with acute brain injury of the first sample characteristic value set of monitoring indicators of patients with acute brain injury and the first characteristic value set of monitoring indicators of patients with acute brain injury of the second sample characteristic value set of monitoring indicators of patients with acute brain injury are added to the characteristic value sequence of the first patient with acute brain injury monitoring indicators, and the calibration value of the severity coefficient of lung infection of the first sample and the calibration value of the severity coefficient of lung infection of the second sample are added to the calibration value sequence of the severity coefficient of lung infection, wherein the high-dimensional distribution distance is the N-dimensional Euclidean distance of the normalized values ​​of the characteristic value set of the first patient with acute brain injury monitoring indicators to be analyzed and the characteristic value set of the second patient with acute brain injury monitoring indicators to be analyzed, and N is the number of compared attributes of the monitoring indicators of patients with acute brain injury; otherwise, the lung infection data of the first sample patient with acute brain injury and / or the lung infection data of the second sample patient with acute brain injury are updated for cyclic analysis; until the characteristic value sequence of the first patient with acute brain injury monitoring indicators meets the preset data amount, the characteristic value sequence of the first patient with acute brain injury monitoring indicators and the calibration value sequence of the severity coefficient of lung infection are output.

[0082] Furthermore, the machine learning-based lung infection prediction system for patients with acute brain injury is also used to: extract a monitoring indicator record value data set from the lung infection data of patients with acute brain injury based on the first k acute brain injury patient monitoring indicators in the acute brain injury patient monitoring indicator sequence; traverse the monitoring indicator record value data set to identify the probability of lung infection and obtain a lung infection probability identification data set; use the lung infection probability identification data set as supervision and the monitoring indicator record value data set as input data to set a data set for constructing a lung infection prediction model for patients with acute brain injury; and train the first lung infection prediction model based on the data set constructed by the lung infection prediction model for patients with acute brain injury.

[0083] Furthermore, the machine learning-based lung infection prediction system for patients with acute brain injury is also used for: the lung infection prediction model construction data set for patients with acute brain injury includes M groups of lung infection prediction model construction data for patients with acute brain injury, and a first training weight is configured for each group of lung infection prediction model construction data for patients with acute brain injury in the M groups of lung infection prediction model construction data for patients with acute brain injury, and the first training weight is equal to one-M; and a lung infection prediction loss function for patients with acute brain injury is constructed:

[0084] Among them, L(y i ,y i0 ) represents the training loss value of the i-th group, y i Represents the infection probability prediction value of the i-th group training, y i0 represents the infection probability supervision value of the i-th group training, δ represents the demarcation threshold, and w iA weight that represents the i-th group of training data; according to the prediction loss function of lung infection in patients with acute brain injury, the M groups of lung infection prediction model construction data for patients with acute brain injury and the first training weight are retrieved to train a first regression decision tree to obtain a first lung infection prediction decision tree, wherein the first lung infection prediction decision tree has a first training group number with a loss value greater than or equal to a loss threshold; when the first training group number is greater than or equal to a convergence group number threshold, M training accuracy rates of the lung infection prediction model construction data for the M groups of patients with acute brain injury in the first lung infection prediction decision tree are obtained, and the lung infection prediction model for the M groups of patients with acute brain injury is evaluated according to the M training accuracy rates. The prediction model construction data configures the second training weight; according to the acute brain injury patient lung infection prediction loss function, the M groups of acute brain injury patient lung infection prediction model construction data and the second training weight are called to train the second regression decision tree to obtain a second lung infection prediction decision tree, wherein the second lung infection prediction decision tree has a second training group number with a loss value greater than or equal to a loss threshold; based on the second training group number, the training is cyclically performed until the Qth training group number is less than the convergence group number threshold, and the first lung infection prediction decision tree, the second lung infection prediction decision tree until the Qth lung infection prediction decision tree are merged to generate the first lung infection prediction model.

[0085] Furthermore, the machine learning-based lung infection prediction system for patients with acute brain injury is also used to: subtract the M training accuracy rates from 1 to obtain M training error rates; add the M training error rates to obtain a training error rate sum; calculate the ratio of the M training error rates to the training error rate sum to obtain M training weights, and add them into the second training weight.

[0086] Furthermore, the machine learning-based lung infection prediction system for patients with acute brain injury is also used to: obtain a first verification accuracy of the first lung infection prediction decision tree, a second verification accuracy of the second lung infection prediction decision tree, and up to the Qth verification accuracy of the Qth lung infection prediction decision tree; based on the first verification accuracy, the second verification accuracy, and up to the Qth verification accuracy, construct a weighted mean function of the outputs of the first lung infection prediction decision tree, the second lung infection prediction decision tree, and up to the Qth lung infection prediction decision tree, construct a fully connected layer, merge the first lung infection prediction decision tree, the second lung infection prediction decision tree, and up to the Qth lung infection prediction decision tree, and generate the first lung infection prediction model.

[0087] Furthermore, the machine learning-based lung infection prediction system for patients with acute brain injury is also used to: set a stop training L, where L represents the upper limit number of decision trees that can be merged; when the Yth training group number is greater than or equal to the convergence group number threshold, determine whether Y is greater than or equal to L; if Y is greater than or equal to L, generate the first lung infection prediction model, and configure the first prediction accuracy to be less than the prediction accuracy threshold; if Y is less than L, continue to execute the loop.

[0088] It should be noted that in the above embodiments, the description of each embodiment has its own emphasis, and for parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0089] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0090] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0091] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0092] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0093] Although preferred embodiments of the present invention have been described, additional changes and modifications may occur to these embodiments once those skilled in the art understand the basic inventive concepts.

[0094] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention belong to the scope of the present invention and its equivalent technologies, the present invention is also intended to include these changes and variations.

Claims

1. A method for predicting lung infection in patients with acute brain injury based on machine learning, characterized in that: include: Step 1: Based on the lung infection data of patients with acute brain injury, the set of monitoring indicators of patients with acute brain injury is sorted from large to small in terms of relevance to obtain a monitoring indicator sequence of patients with acute brain injury; Step 2: extracting the first k acute brain injury patient monitoring indicators of the acute brain injury patient monitoring indicator sequence as input variables, and training the acute brain injury patient lung infection prediction model in combination with the acute brain injury patient lung infection data to obtain a first lung infection prediction model, wherein the initial value of k is equal to 1, and k is an integer; Step 3: When the first prediction accuracy of the first lung infection prediction model is less than or equal to the prediction accuracy threshold, use k+1 to update the k value, and return to step 2 to execute the loop; Step 4: When the first prediction accuracy of the first lung infection prediction model is greater than the prediction accuracy threshold, the first lung infection prediction model is set as the target lung infection prediction model and built into the platform server to perform the lung infection prediction task for patients with acute brain injury.

2. The method for predicting lung infection in patients with acute brain injury based on machine learning as claimed in claim 1, characterized in that: Based on the lung infection data of patients with acute brain injury, the set of monitoring indicators of patients with acute brain injury is sorted from large to small in terms of relevance, and a sequence of monitoring indicators of patients with acute brain injury is obtained, including: Obtaining a first acute brain injury patient monitoring indicator of the acute brain injury patient monitoring indicator set; Taking the first acute brain injury patient monitoring indicator as an independent variable and the lung infection severity coefficient as a dependent variable, performing correlation analysis sample sorting on the lung infection data of the acute brain injury patient to obtain a characteristic value sequence of the first acute brain injury patient monitoring indicator and a calibration value sequence of the lung infection severity coefficient; Performing a Pearson correlation analysis on the first acute brain injury patient monitoring indicator characteristic value sequence and the pulmonary infection severity coefficient calibration value sequence to obtain a first Pearson correlation coefficient, and adding the first Pearson correlation coefficient to a Pearson correlation coefficient set, wherein the Pearson correlation coefficient set corresponds one-to-one to the acute brain injury patient monitoring indicator set; Constructing a Pearson correlation coefficient heat map according to the Pearson correlation coefficient set; The acute brain injury patient monitoring indicator set is sorted from large to small in correlation according to the Pearson correlation coefficient heat map to obtain the acute brain injury patient monitoring indicator sequence.

3. The method for predicting lung infection in patients with acute brain injury based on machine learning as claimed in claim 2, characterized in that: Taking the first acute brain injury patient monitoring indicator as the independent variable and the lung infection severity coefficient as the dependent variable, performing correlation analysis sample sorting on the lung infection data of the acute brain injury patient, and obtaining the first acute brain injury patient monitoring indicator characteristic value sequence and the lung infection severity coefficient calibration value sequence, including: Obtaining the first sample of the acute brain injury patient's lung infection data, wherein the first sample of the acute brain injury patient's lung infection data includes a first sample of the acute brain injury patient's monitoring indicator characteristic value set and a first sample of the lung infection severity coefficient calibration value, and the first sample of the lung infection severity coefficient calibration value is the doctor calibration data; Obtaining second sample acute brain injury patient lung infection data of the acute brain injury patient lung infection data, wherein the second sample acute brain injury patient lung infection data includes a second sample acute brain injury patient monitoring indicator characteristic value set and a second sample lung infection severity coefficient calibration value, and the second sample lung infection severity coefficient calibration value is doctor calibration data; Deleting the first acute brain injury patient monitoring indicator from the first sample acute brain injury patient monitoring indicator feature value set to obtain a first acute brain injury patient monitoring indicator feature value set to be analyzed; Deleting the first acute brain injury patient monitoring indicator from the second sample acute brain injury patient monitoring indicator feature value set to obtain a second acute brain injury patient monitoring indicator feature value set to be analyzed; When the high-dimensional distribution distance between the first set of acute brain injury patient monitoring indicator characteristic values ​​to be analyzed and the second set of acute brain injury patient monitoring indicator characteristic values ​​to be analyzed is less than or equal to the distribution distance threshold, the first set of acute brain injury patient monitoring indicator characteristic values ​​of the first sample set of acute brain injury patient monitoring indicator characteristic values ​​and the second set of acute brain injury patient monitoring indicator characteristic values ​​are added to the first acute brain injury patient monitoring indicator characteristic value sequence, and the first sample pulmonary infection severity coefficient calibration value and the second sample pulmonary infection severity coefficient calibration value are added to the pulmonary infection severity coefficient calibration value sequence, wherein the high-dimensional distribution distance is the N-dimensional Euclidean distance of the normalized values ​​of the first set of acute brain injury patient monitoring indicator characteristic values ​​to be analyzed and the second set of acute brain injury patient monitoring indicator characteristic values ​​to be analyzed, and N is the number of compared acute brain injury patient monitoring indicator attributes; Otherwise, updating the first sample acute brain injury patient lung infection data and / or the second sample acute brain injury patient lung infection data for cyclic analysis; The first acute brain injury patient monitoring indicator characteristic value sequence and the lung infection severity coefficient calibration value sequence are output until the first acute brain injury patient monitoring indicator characteristic value sequence meets the preset data amount.

4. The method for predicting lung infection in patients with acute brain injury based on machine learning as claimed in claim 1, characterized in that: The first k acute brain injury patient monitoring indicators of the acute brain injury patient monitoring indicator sequence are extracted as input variables, and the acute brain injury patient lung infection prediction model is trained in combination with the acute brain injury patient lung infection data to obtain a first lung infection prediction model, including: Based on the first k acute brain injury patient monitoring indicators of the acute brain injury patient monitoring indicator sequence, extracting a monitoring indicator record value data set from the acute brain injury patient lung infection data; Traversing the monitoring indicator record value data set to identify the probability of lung infection, and obtaining a lung infection probability identification data set; The lung infection probability identification data set is used as supervision, and the monitoring index record value data set is used as input data, and is set as a data set for constructing a lung infection prediction model for patients with acute brain injury; A data set is constructed based on the lung infection prediction model for patients with acute brain injury, and the first lung infection prediction model is trained.

5. The method for predicting lung infection in patients with acute brain injury based on machine learning as claimed in claim 4, characterized in that: Constructing a data set based on the lung infection prediction model for patients with acute brain injury, and training the first lung infection prediction model, including: The data set for constructing a prediction model for lung infection in patients with acute brain injury includes M groups of data for constructing a prediction model for lung infection in patients with acute brain injury, and a first training weight is configured for each group of data for constructing a prediction model for lung infection in patients with acute brain injury in the M groups of data for constructing a prediction model for lung infection in patients with acute brain injury, wherein the first training weight is equal to one-M; Construct a loss function for predicting lung infection in patients with acute brain injury: Among them, L(y i ,y i0 ) represents the training loss value of the i-th group, y i Represents the infection probability prediction value of the i-th group training, y i0 represents the infection probability supervision value of the i-th group training, δ represents the demarcation threshold, and w i The weight representing the i-th group of training data; According to the prediction loss function of lung infection in patients with acute brain injury, the M groups of lung infection prediction model construction data of patients with acute brain injury and the first training weight are retrieved to train a first regression decision tree to obtain a first lung infection prediction decision tree, wherein the first lung infection prediction decision tree has a first training group number whose loss value is greater than or equal to a loss threshold; When the first training group number is greater than or equal to the convergence group number threshold, M training accuracy rates of the M groups of acute brain injury patients' lung infection prediction model construction data in the first lung infection prediction decision tree are obtained, and a second training weight is configured for the M groups of acute brain injury patients' lung infection prediction model construction data according to the M training accuracy rates; According to the prediction loss function of lung infection in patients with acute brain injury, the M groups of lung infection prediction model construction data of patients with acute brain injury and the second training weight are retrieved to train a second regression decision tree to obtain a second lung infection prediction decision tree, wherein the second lung infection prediction decision tree has a second number of training groups whose loss value is greater than or equal to a loss threshold; Based on the second number of training groups, cyclic training is performed until the Qth number of training groups is less than the convergence group number threshold, and the first lung infection prediction decision tree, the second lung infection prediction decision tree, and the Qth lung infection prediction decision tree are merged to generate the first lung infection prediction model.

6. The method for predicting lung infection in patients with acute brain injury based on machine learning as claimed in claim 5, characterized in that: Configuring a second training weight for the M groups of acute brain injury patients' lung infection prediction model construction data according to the M training accuracy rates includes: Subtract the M training accuracy rates from 1 to obtain M training error rates; Adding the M training error rates to obtain a training error rate sum value; The ratios of the M training error rates to the sum of the training error rates are calculated respectively to obtain M training weights, which are added to the second training weights.

7. The method for predicting lung infection in patients with acute brain injury based on machine learning as claimed in claim 5, characterized in that: The first lung infection prediction decision tree, the second lung infection prediction decision tree, and up to the Qth lung infection prediction decision tree are combined to generate the first lung infection prediction model, including: Obtaining a first verification accuracy rate of the first lung infection prediction decision tree, a second verification accuracy rate of the second lung infection prediction decision tree, and so on until the Qth verification accuracy rate of the Qth lung infection prediction decision tree; According to the first verification accuracy rate, the second verification accuracy rate, and so on, a weighted mean function of the outputs of the first lung infection prediction decision tree, the second lung infection prediction decision tree, and so on, is constructed; a fully connected layer is constructed; the first lung infection prediction decision tree, the second lung infection prediction decision tree, and so on, are merged to generate the first lung infection prediction model.

8. The method for predicting lung infection in patients with acute brain injury based on machine learning as claimed in claim 5, characterized in that: Also includes: Set the stop training L, where L represents the upper limit of the number of decision trees that can be merged; When the Yth training group number is greater than or equal to the convergence group number threshold, determining whether Y is greater than or equal to L; If Y is greater than or equal to L, the first lung infection prediction model is generated, and the first prediction accuracy is configured to be less than the prediction accuracy threshold; If Y is less than L, continue the loop.

9. A machine learning-based prediction system for lung infection in patients with acute brain injury, characterized in that: The steps for implementing the method for predicting lung infection in patients with acute brain injury based on machine learning according to any one of claims 1 to 8 include: A monitoring indicator correlation analysis module is used to sort the set of monitoring indicators of patients with acute brain injury from large to small in terms of correlation based on the lung infection data of patients with acute brain injury, so as to obtain a monitoring indicator sequence of patients with acute brain injury; A lung infection prediction model training module is used to extract the first k acute brain injury patient monitoring indicators of the acute brain injury patient monitoring indicator sequence as input variables, and train the acute brain injury patient lung infection prediction model in combination with the acute brain injury patient lung infection data to obtain a first lung infection prediction model, wherein the initial value of k is equal to 1, and k is an integer; An infection prediction model loop training module, used to update the k value using k+1 when the first prediction accuracy of the first lung infection prediction model is less than or equal to the prediction accuracy threshold, and return to step 2 to execute the loop; A lung infection prediction module is used to set the first lung infection prediction model as the target lung infection prediction model built into the platform server to perform the lung infection prediction task for patients with acute brain injury when the first prediction accuracy of the first lung infection prediction model is greater than the prediction accuracy threshold.

10. The machine learning-based lung infection prediction system for patients with acute brain injury according to claim 9, characterized in that: The monitoring indicator correlation analysis module is also used for: A first acute brain injury patient monitoring indicator obtaining unit, configured to obtain a first acute brain injury patient monitoring indicator of the acute brain injury patient monitoring indicator set; A correlation analysis sample sorting unit is used to perform correlation analysis sample sorting on the lung infection data of the acute brain injury patient with the first acute brain injury patient monitoring indicator as an independent variable and the lung infection severity coefficient as a dependent variable to obtain a characteristic value sequence of the first acute brain injury patient monitoring indicator and a calibration value sequence of the lung infection severity coefficient; A Pearson correlation analysis unit, configured to perform a Pearson correlation analysis on the first acute brain injury patient monitoring indicator characteristic value sequence and the pulmonary infection severity coefficient calibration value sequence to obtain a first Pearson correlation coefficient, and add the first Pearson correlation coefficient to a Pearson correlation coefficient set, wherein the Pearson correlation coefficient set corresponds one-to-one to the acute brain injury patient monitoring indicator set; A Pearson correlation coefficient heat map construction unit, used to construct a Pearson correlation coefficient heat map according to the Pearson correlation coefficient set; The acute brain injury patient monitoring indicator sorting unit is used to sort the acute brain injury patient monitoring indicator set from large to small correlation according to the Pearson correlation coefficient heat map to obtain the acute brain injury patient monitoring indicator sequence.

Images