Methods and related equipment for predicting the risk of heart failure in patients after surgery for acute myocardial infarction

Abstract

The present application relates to a method and related equipment for predicting the risk of heart failure after surgery for patients with acute myocardial infarction. The method includes: obtaining the first clinical indicator data of the target patient after surgery as the target clinical indicator data; the target patient refers to the patient with acute myocardial infarction whose risk of heart failure is to be predicted after surgery; constructing model input data according to the target clinical indicator data; using the KILLIP grade prediction model to process the model input data to obtain the KILLIP prediction grade of the target patient, and the KILLIP prediction grade of the target patient is used to determine the risk of heart failure after surgery for the target patient; the KILLIP grade prediction model is an artificial intelligence model trained using a first data set constructed based on the first clinical indicator data of multiple patients with acute myocardial infarction after surgery and their corresponding KILLIP grades. The present application can more accurately determine the risk of heart failure after surgery for patients with acute myocardial infarction.

Description

Technical Field

[0001] The present application relates to the intersection of artificial intelligence and medical technology, and in particular to a method and related equipment for predicting the risk of heart failure after surgery for patients with acute myocardial infarction. Background Art

[0002] Acute myocardial infarction (AMI), also known as heart attack, is one of the most deadly and morbid diseases worldwide. AMI often leads to the development of heart failure (HF), a debilitating condition associated with high morbidity and mortality. Identifying patients at risk for developing HF after AMI is critical to initiating preventive measures and optimizing treatment strategies.

[0003] The KILLIP classification is a simple tool for evaluating the clinical symptoms of heart failure after acute myocardial infarction, and has long been used to classify patients into different heart failure risk categories. However, this classification often relies on the subjective assessment of relevant personnel. Due to the limited computing power of humans, it is difficult to fully capture the complex interactions between the various factors that lead to the development of heart failure. Therefore, the subjective assessment of the patient's KILLIP level is not accurate. Summary of the invention

[0004] In view of the above-mentioned deficiencies or shortcomings, the present application provides a method and related equipment for predicting the risk of heart failure after surgery for patients with acute myocardial infarction.

[0005] According to a first aspect, the present application provides a method for predicting the risk of heart failure after surgery in patients with acute myocardial infarction. In some embodiments, the method comprises:

[0006] The first clinical indicator data of the target patient after surgery is obtained as the target clinical indicator data; the target patient refers to a patient with acute myocardial infarction whose risk of heart failure is to be predicted after surgery; the first clinical indicator data includes predetermined indicator values ​​of clinical indicators related to KILLIP classification prediction, and the clinical indicators related to KILLIP classification prediction include GRACE score, NT-proBNP, TIMI score, left ventricular ejection fraction, creatinine, creatinine clearance, age, length of hospital stay, interleukin-6 and high-sensitivity C-reactive protein;

[0007] Construct model input data based on target clinical indicator data;

[0008] The KILLIP level prediction model is used to process the model input data to obtain the KILLIP prediction level of the target patient. The KILLIP prediction level of the target patient is used to determine the risk of heart failure of the target patient after surgery. The KILLIP level prediction model is an artificial intelligence model trained using a first data set constructed based on the first clinical indicator data of multiple patients with acute myocardial infarction after surgery and their corresponding KILLIP levels.

[0009] In some embodiments, the operation of determining a clinical indicator associated with a KILLIP classification prediction comprises:

[0010] The second clinical indicator data and the corresponding KILLIP levels of multiple patients with acute myocardial infarction after surgery are collected to construct a second data set; the second clinical indicator data includes indicator values ​​of multiple candidate clinical indicators;

[0011] Using the second data set to perform model training and evaluation, an artificial intelligence model for predicting the KILLIP level is obtained;

[0012] Conduct interpretability analysis on the AI ​​model to determine the impact of each candidate clinical indicator on the KILLIP classification prediction;

[0013] According to the influence degree of each candidate clinical indicator on the KILLIP classification prediction, clinical indicators related to the KILLIP classification prediction are screened out from the multiple candidate clinical indicators.

[0014] In some embodiments, the second data set includes a training set and a test set;

[0015] The second dataset is used for model training and evaluation to obtain an artificial intelligence model for predicting the KILLIP level, including:

[0016] Using the training set, multiple machine learning models and multiple deep learning models are trained to predict the KILLIP level;

[0017] Use the test set to evaluate the trained machine learning models and deep learning models;

[0018] According to the model evaluation results, the parameters of each trained machine learning model and deep learning model are tuned by five-fold cross-validation combined with grid search, and then the model with the best performance is selected as the artificial intelligence model for predicting the KILLIP level.

[0019] In some embodiments, the second clinical indicator data and the corresponding KILLIP levels of multiple patients after acute myocardial infarction are collected to construct a second data set, including:

[0020] Preprocessing the collected second clinical indicator data of patients after acute myocardial infarction surgery to obtain a sample data set consisting of multiple pieces of preprocessed second clinical indicator data and their corresponding KILLIP levels; the preprocessing includes data cleaning and feature scaling;

[0021] Divide the sample data set into training set and test set;

[0022] Determine the minority class samples in the training set based on the data ratio corresponding to each KILLIP level in the training set.

[0023] Use the smote oversampling technique to generate new samples based on the minority class samples and add the new samples to the training set.

[0024] In some embodiments, a smote oversampling technique is used to generate new samples based on minority class samples, including:

[0025] Use the K nearest neighbor algorithm to find k nearest neighbor samples for each minority class sample;

[0026] For each minority class sample, one or more samples from the k nearest neighbor samples corresponding to the minority class sample are randomly selected according to a preset sampling ratio, and a new sample is generated by randomly selecting the connection between the minority class sample and the selected sample.

[0027] In some embodiments, an interpretability analysis is performed on the trained artificial intelligence model to determine the degree of influence of each candidate clinical indicator on the KILLIP classification prediction, and clinical indicators related to the KILLIP classification prediction are screened out from the multiple candidate clinical indicators according to the degree of influence of each candidate clinical indicator on the KILLIP classification prediction, including:

[0028] Use the SHAP method to perform interpretability analysis on the trained artificial intelligence model and obtain the SHAP value of each candidate clinical indicator;

[0029] The candidate clinical indicators with shap values ​​greater than the preset threshold are obtained as clinical indicators related to KILLIP classification prediction.

[0030] In some embodiments, the multiple machine learning models are a random forest model and an XGboost model, the multiple deep learning models are an MLP model and a TabNet model; and the KILLIP level prediction model is a TabNet model.

[0031] According to a second aspect, the present application provides a device for predicting the risk of heart failure after surgery for patients with acute myocardial infarction. In some embodiments, the device comprises:

[0032] A target data acquisition module is used to acquire the first clinical indicator data of the target patient after surgery as the target clinical indicator data; the target patient refers to a patient with acute myocardial infarction whose risk of heart failure is to be predicted after surgery; the first clinical indicator data includes predetermined indicator values ​​of clinical indicators related to KILLIP classification prediction, and the clinical indicators related to KILLIP classification prediction include GRACE score, NT-proBNP, TIMI score, left ventricular ejection fraction, creatinine, creatinine clearance, age, length of hospital stay, interleukin-6 and high-sensitivity C-reactive protein;

[0033] A construction module, used to construct model input data according to target clinical indicator data;

[0034] The prediction module is used to process the model input data using the KILLIP level prediction model to obtain the KILLIP prediction level of the target patient. The KILLIP prediction level of the target patient is used to determine the risk of heart failure of the target patient after surgery. The KILLIP level prediction model is an artificial intelligence model trained using a first data set constructed based on the first clinical indicator data of multiple patients with acute myocardial infarction after surgery and their corresponding KILLIP levels.

[0035] According to a third aspect, the present application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the method for predicting the risk of heart failure after surgery for patients with acute myocardial infarction provided in any of the above-mentioned embodiments are implemented.

[0036] According to a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the method for predicting the risk of heart failure after surgery for patients with acute myocardial infarction provided in any of the above embodiments are implemented.

[0037] In the above-mentioned embodiment of the present application, the heart failure risk prediction device obtains the first clinical indicator data of the target patient, that is, the patient with acute myocardial infarction whose postoperative heart failure risk is to be predicted, as the target clinical indicator data, and then constructs the model input data according to the target clinical indicator data, and uses the pre-trained KILLIP level prediction model to process the model input data, thereby obtaining the KILLIP prediction level of the target patient, and finally the KILLIP prediction level of the target patient can be used to determine the postoperative heart failure risk of the target patient. Compared with the prior art, the present application can more accurately determine the postoperative heart failure risk of patients with acute myocardial infarction. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A schematic flow chart of a method for predicting the risk of heart failure after surgery for patients with acute myocardial infarction provided in an embodiment of the present application;

[0039] Figure 2 A schematic diagram of a process for determining clinical indicators related to KILLIP classification prediction provided in an embodiment of the present application;

[0040] Figure 3 A summary bar chart provided for an embodiment of the present application;

[0041] Figure 4-1 A correlation diagram of the GRACE score provided in the embodiment of the present application;

[0042] Figure 4-2 A correlation diagram of NT-pro BNP provided in the examples of the present application;

[0043] Figure 4-3 A correlation diagram of the TIMI score provided in the embodiment of the present application;

[0044] Figure 4-4 A correlation diagram of the left ventricular ejection fraction LVEF provided in an embodiment of the present application;

[0045] Figure 4-5 A correlation diagram of creatinine clearance provided in the embodiments of the present application;

[0046] Figure 4-6 A correlation diagram of creatinine provided in the embodiments of the present application;

[0047] Figure 4-7 A correlation diagram of age provided for an embodiment of the present application;

[0048] Figure 4-8 A correlation diagram of hospitalization days provided in an embodiment of the present application;

[0049] Figure 4-9 A correlation diagram of high-sensitivity C-reactive protein provided in the examples of the present application;

[0050] Figure 5 The index values ​​of candidate clinical indicators of multiple patients provided in the embodiments of the present application;

[0051] Figure 6-1 A force line diagram of patient 1 provided in an embodiment of the present application;

[0052] Figure 6-2 A force line diagram of patient 2 provided in an embodiment of the present application;

[0053] Figure 6-3 A force line diagram of patient 3 provided in an embodiment of the present application;

[0054] Figure 6-4 A force line diagram of patient 4 provided in an embodiment of the present application;

[0055] Figure 7 A structural block diagram of a device for predicting the risk of heart failure after surgery for patients with acute myocardial infarction provided in an embodiment of the present application;

[0056] Figure 8 This is a diagram of the internal structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solutions and advantages of the present application clearer, the embodiments of the present application will be further described in detail below in conjunction with the accompanying drawings. It should be clear that the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0058] When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. Instead, they are only examples of devices and methods consistent with some aspects of the present application as detailed in the attached claims.

[0059] In the description of the present application, it should be understood that the terms "first", "second", "third", etc. are only used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence, nor can they be understood as indicating or implying relative importance. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to the specific circumstances. In addition, in the description of the present application, unless otherwise specified, "multiple" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects associated before and after are in an "or" relationship.

[0060] In view of the deficiencies of the existing technology, the present application provides a method for predicting the risk of heart failure in patients with acute myocardial infarction after surgery. Compared with the existing technology, this method can more accurately determine the risk of heart failure in patients with acute myocardial infarction after surgery.

[0061] In some embodiments, the method can be applied to a heart failure risk prediction device, which can be implemented using an independent server or a server cluster. In this embodiment, the method for predicting the risk of heart failure in patients after surgery for acute myocardial infarction includes: Figure 1 The steps shown in the figure are described in detail below.

[0062] S110: Acquire first clinical indicator data of the target patient after surgery as target clinical indicator data.

[0063] Target patients are patients with acute myocardial infarction who are to be predicted for risk of heart failure after surgery.

[0064] The first clinical index data of the target patient after surgery can be collected during the target patient's postoperative hospitalization, including predetermined index values ​​of clinical indexes related to KILLIP classification prediction. In this embodiment, the clinical indexes related to KILLIP classification prediction include GRACE score, NT-proBNP, TIMI score, left ventricular ejection fraction, creatinine, creatinine clearance, age, length of hospital stay, interleukin-6, and high-sensitivity C-reactive protein.

[0065] S120: constructing a model input data according to target clinical indicator data;

[0066] This embodiment uses a pre-trained KILLIP level prediction model to predict the KILLIP level of the target patient. After the heart failure risk prediction device obtains the target clinical indicator data, it processes the target clinical indicator data according to the input data format corresponding to the KILLIP level prediction model to obtain model input data that meets the format requirements of the KILLIP level prediction model for input data.

[0067] The specific format requirements of the KILLIP level prediction model for input data are related to the type of artificial intelligence model it is. Different artificial intelligence models usually have different format requirements for input data. For example, the Random Forest model usually accepts numerical data, and its requirements for data preprocessing are not high. The XGBoost (eXtremeGradient Boosting) model requires numerical input, and usually requires data preprocessing such as missing value processing and feature encoding; the MLP (Multilayer Perceptron) model is usually used to process numerical data such as one-dimensional or multi-dimensional arrays. When processing non-numerical data, it needs to be converted into a numerical representation; the TabNet model is a neural network specifically used for tabular data. It can process data sets containing numerical and categorical features, and the input data usually needs to be preprocessed, such as missing value processing, categorical feature encoding, etc. Therefore, the specific format requirements of the KILLIP level prediction model for input data are not particularly limited in this embodiment.

[0068] S130: Use the KILLIP grade prediction model to process the model input data to obtain the KILLIP prediction grade of the target patient. The KILLIP prediction grade of the target patient is used to determine the risk of postoperative heart failure of the target patient.

[0069] The KILLIP level prediction model is a four-classification prediction model. After the model input data is input into the KILLIP level prediction model, the KILLIP prediction level of the target patient is determined based on the output data of the KILLIP level prediction model. In some embodiments, the output data of the KILLIP level prediction model can be a category label, and the category label is one of the four KILLIP levels. The KILLIP prediction level of the target patient is the KILLIP level indicated by the category label.

[0070] As shown above, the KILLIP classification is used to evaluate the clinical symptoms of heart failure in patients presenting acutely after myocardial infarction and has long been used to stratify patients into different heart failure risk categories.

[0071] There are four levels of KILLIP grading, and the descriptions of each level are as follows:

[0072] 1. KILLIP I: The patient has no signs or symptoms of heart failure and no rales in the lungs, but the pulmonary wedge pressure may be elevated, which is the compensatory stage of heart failure;

[0073] 2. KILLIP II: The patient has mild to moderate left heart failure, with rales covering less than 50% of the lung field; there may be a gallop rhythm of the first heart sound, sustained sinus tachycardia, and X-ray findings of pulmonary congestion;

[0074] 3. KILLIP III: The patient has severe heart failure, manifested as acute pulmonary edema, and the range of pulmonary rales is greater than 50% of the two lung fields;

[0075] 4. KILLIP Grade IV: The patient has cardiogenic shock, blood pressure below 12.0 kPa (90 mmHg), oliguria, cold and clammy skin, cyanosis, accelerated breathing and rapid pulse.

[0076] After determining the KILLIP prediction level corresponding to the target patient, the postoperative heart failure risk of the target patient can be estimated based on the KILLIP prediction level.

[0077] The KILLIP grade prediction model is an artificial intelligence model trained using a first data set constructed based on the first clinical indicator data of multiple patients after surgery for acute myocardial infarction and their corresponding KILLIP grades.

[0078] The KILLIP level prediction model can specifically be a machine learning model or a deep learning model. The model can be trained on the first data set using an existing model training method. Those skilled in the art should be aware of the specific training operations, which will not be elaborated in this embodiment.

[0079] Preferably, the KILLIP level prediction model can be a TabNet model.

[0080] In some embodiments, the operation of determining clinical indicators associated with KILLIP classification prediction, such as Figure 2 As shown, including:

[0081] S210: Collecting the second clinical indicator data of multiple patients after acute myocardial infarction surgery and their corresponding KILLIP levels to construct a second data set, wherein the second clinical indicator data includes indicator values ​​of multiple candidate clinical indicators.

[0082] Candidate clinical indicators refer to clinical indicators that may be related to KILLIP classification prediction. The above multiple candidate clinical indicators can be called a candidate clinical indicator set.

[0083] In some embodiments, the above-mentioned candidate clinical indicator set can be clinical indicators that are easy to collect during the patient's postoperative hospitalization. Such a setting can improve the convenience of data acquisition and reduce the cost of data acquisition. In addition, the clinical indicators during hospitalization are usually directly related to the patient's current condition. Selecting these clinical indicators as candidate clinical indicators can ensure the accuracy of KILLIP level prediction.

[0084] In some embodiments, the candidate clinical indicator set includes sex, age, body mass index (BMI), GRACE risk score, TIMI risk score, history of hypertension, history of hyperlipidemia, history of diabetes, history of AF, old myocardial infarction, history of stent implantation, history of cerebrovascular disease, history of gastric ulcer, history of smoking, no quitting smoking, history of CABG, hospitalization days, etc. days), blood test indicators after admission and echocardiographic indicators; blood test indicators after admission included total white blood cell count (Leukocytes), neutrophils, lymphocytes, monocytes, hemoglobin, platelets, blood glucose, alanine aminotransferase (ALT), aspartate aminotransferase (AST), creatinine, creatinine clearance rate (CCR), total cholesterol (TC), low-density lipoprotein (LDL-C), high-density lipoprotein (HDL-C), triglycerides (TG), uric acid (UA), glycosylated hemoglobin (HbA1c), high-sensitivity C-reactive protein (hs-CRP), interleukin-6 (IL-6), NT-proBNP, and TNI peak (Peak troponin I); echocardiographic indicators included left atrial diameter (Left atrialdiameter), left ventricular ejection fraction (LVEF), and left ventricular end-diastolic dimension (LVEDD).

[0085] In some embodiments, the operation of collecting the second clinical indicator data of multiple patients after acute myocardial infarction and their corresponding KILLIP levels to construct a second data set includes:

[0086] (1) Preprocessing the collected second clinical indicator data of patients after acute myocardial infarction surgery to obtain a sample data set consisting of multiple pieces of preprocessed second clinical indicator data and their corresponding KILLIP levels;

[0087] Among them, preprocessing includes data cleaning and feature scaling. Data cleaning is used to delete patient sample data containing missing features, and feature scaling is used to normalize the feature values ​​of numerical features and scale them between 0 and 1 to ensure the stability and efficiency of model training. The above-mentioned features refer to candidate clinical indicators, and the feature values ​​are the indicator values ​​of candidate clinical indicators.

[0088] The sample data set includes multiple pieces of sample data (referred to as samples for short), and each piece of pre-processed second clinical indicator data and its corresponding KILLIP level is regarded as one piece of sample data.

[0089] (2) Divide the sample data set into training set and test set;

[0090] For a sample data set, it can be divided into a training set and a test set by random sampling. Specifically, the sample data set can be divided into multiple data groups, and different data sets correspond to different KILLIP levels. Then, samples are randomly selected from each data group according to a preset size ratio of the training set and the test set (such as 7:3). Finally, the divided data sets are saved as training sets and test sets, respectively.

[0091] (3) Determine the minority class samples in the training set according to the data ratio corresponding to each KILLIP level in the training set, use the smote oversampling technology to generate new samples based on the minority class samples, add the new samples to the training set, and obtain a class-balanced training set.

[0092] Data in real medical scenarios generally have the problem of unbalanced sample distribution, so the number of samples of one or some categories in the training set may be very small, or even far lower than the normal number of samples. When training a model on such a dataset, the trained model may be biased towards the majority class, and because the model fails to fully learn the patterns of the minority class during the training process, its generalization ability when facing real-world data will be affected, which may lead to poor prediction results for minority class samples in practical applications.

[0093] In view of the above problem, this embodiment first calculates the data ratio corresponding to each KILLIP level in the training set, indicating the ratio of the number of samples corresponding to each KILLIP level in the training set, and then determines the KILLIP level with a data ratio lower than a preset ratio threshold as a minority class, and the samples corresponding to the minority class are the minority class samples. Then, the smote oversampling technology is used to generate new samples based on the existing minority class samples until the training set reaches the required class balance level.

[0094] Among them, the operation of using the smote oversampling technology to generate new samples based on minority class samples may include: using the K nearest neighbor algorithm to find k nearest neighbor samples for each minority class sample; for each minority class sample, randomly selecting one or more samples from the k nearest neighbor samples corresponding to the minority class sample according to a preset sampling multiple, and randomly selecting according to the connection between the minority class sample and the selected sample to generate a new sample.

[0095] Specifically, for each minority class sample, the K nearest neighbor algorithm is used to identify its k nearest neighbor samples in the feature space, and the distance between samples can be measured using the Euclidean distance; when the neighbor samples (the number can be 1 or more) are identified, a neighbor is randomly selected, and the difference between the feature vector of the minority class sample and the selected neighbor sample is calculated, and then the difference is multiplied by a random number (between 0 and 1), and the product is added to the feature vector of the minority class sample to generate a new sample. The above sample generation process will be repeated until the class balance level of the training set reaches the class balance level required for practical applications.

[0096] Using the smote oversampling technology to generate new samples based on minority class samples can make the training set achieve category balance. Then, using the category-balanced training set for model training can reduce the bias of the trained model and improve the model's prediction effect on minority class samples in practical applications.

[0097] S220: Use the second data set to perform model training and evaluation to obtain an artificial intelligence model for predicting the KILLIP level.

[0098] The second data set includes a training set and a test set; the training set is a training set that has been processed by smote oversampling (i.e., it has achieved category balance). Accordingly, the second data set is used for model training and evaluation to obtain the operations of the artificial intelligence model for predicting the KILLIP level, including:

[0099] (1) Using the training set to train multiple machine learning models and multiple deep learning models for predicting KILLIP levels;

[0100] In some examples, the multiple machine learning models may be a random forest model and an XGboost model, and the multiple deep learning models may be an MLP model and a TabNet model.

[0101] First, the model is constructed. After each model is constructed, each model is trained on the training set. The model training method can be an existing one. In practical applications, the model training can be implemented based on the existing training method of the model to be trained. The following uses the random forest model, XGboost model, MLP model and TabNet model as examples.

[0102] Regarding the random forest model, for each tree, a sample set of the same size as the original training set is randomly selected with replacement from the original training set. At each node of each tree, the number of features specified by the maximum number of features is randomly selected, and then the best split method is found from these features. The node is split into two child nodes according to the selected best split method, and the above process is repeated recursively until the maximum depth of the tree is reached, the number of samples in the node is less than the minimum sample split, or the number of samples in the node is less than the minimum number of leaf node samples. Repeat the above steps n times to generate n decision trees. Use the obtained random forest model to predict the samples, and determine the final prediction category by voting (majority voting).

[0103] Regarding the XGboost model, you can first initialize a decision tree weak learner, with the same initial prediction for all samples. Then, learn and iterate a new weak learner based on the prediction results of the weak learner. The specific iterative process is as follows:

[0104] a. For each iteration, the algorithm builds a new weak learner based on the residual of the current model.

[0105] b. Construct weak learners through greedy algorithms (such as feature selection in decision tree construction).

[0106] c. Fit the residuals of the current model to the newly constructed weak learner to improve the overall performance of the model.

[0107] d. Once constructed, the new weak learner is assigned a weight (i.e., learning rate) based on its performance in reducing the residual error and added to the existing model. The predictions of all weak learners are then summed up according to their weights to get the final prediction.

[0108] The above iteration process will continue until a preset termination condition is met, such as reaching a preset number of iterations or an error threshold.

[0109] Regarding the MLP model, after determining the network structure of the model (including the number of layers of the network, the number of nodes in each layer, and the activation function), the input data is sent to the network for training. The training process is as follows:

[0110] a. Forward propagation: The input data passes through each layer of the network, calculating the output of each neuron until the output layer.

[0111] b. Calculate the loss: In the output layer, the cross entropy loss function is used to calculate the loss between the predicted value and the true value.

[0112] c. Backpropagation: According to the chain rule, the gradient of the loss function with respect to each weight and bias is calculated (that is, the partial derivative of the loss function with respect to each parameter).

[0113] d. Parameter update: Use the Adam optimization algorithm to calculate the gradient and update the weights and biases of the network.

[0114] e. Iterative training: Repeat the above ad operations until the preset end conditions are met, such as reaching a predetermined number of iterations, the loss value no longer decreases significantly, or the performance on the validation set begins to decrease, etc.

[0115] Regarding the TabNet model, the entire workflow of TabNet is to input the model input data into the FeatureTransformer layer to extract features, then use the split layer to split the information, and then input it into the AttentiveTransformer layer, use the attention mechanism to learn the importance of features, retain important features and input them into the softmax layer to obtain the final prediction results. The training process can be shown as follows:

[0116] a. Forward propagation:

[0117] Data input: The input data is sent to the first feature conversion layer of TabNet through the batch normalization layer.

[0118] Feature selection: In each feature transformation block, an importance weight is assigned to each feature through the attention mechanism, and some features are selected for further processing.

[0119] Feature transformation: The selected features undergo linear transformation and nonlinear activation to generate new feature representations.

[0120] Decision layer: The outputs of all feature transformation blocks (usually the output of the last block) are fed into the decision layer to generate the final prediction.

[0121] b. Calculate the loss: In the output layer, the cross entropy loss function can be used to calculate the loss between the predicted value and the true value.

[0122] c. Back propagation: According to the chain rule, calculate the gradient of the loss function with respect to each weight and bias of TabNet (that is, the partial derivative of the loss function with respect to each parameter).

[0123] d. Parameter update: The Adam optimization algorithm can be used to calculate the gradient and update the weights and biases of the network.

[0124] e. Iterative training: Repeat the above ad operations until the preset end conditions are met, such as reaching a predetermined number of iterations, the loss value no longer decreases significantly, etc.

[0125] (2) Use the test set to evaluate the trained machine learning models and deep learning models;

[0126] Specifically, each trained model is evaluated on the test set. When evaluating the model, classification evaluation indicators such as recall, F1 score, and AUC can be used. Among them, recall refers to the proportion of samples that are actually positive that are correctly predicted to be positive. Recall focuses on the model's coverage of positive samples, indicating how many actual positive samples are correctly identified. F1 score is the harmonic mean of precision and recall, and is often used to comprehensively evaluate the performance of the model. F1 score can find a balance between precision and recall, and is suitable for class-imbalanced data sets; when the number of positive and negative samples is significantly different, it can provide a more comprehensive evaluation. AUC, or AUROC (area under the receiver operating characteristic curve), is the area under the ROC curve. The ROC curve depicts the false positive rate (FPR) and true positive rate (TPR) of the model at different thresholds, which is plotted by TPR against FPR. The AUC value is between 0 and 1. The larger the value, the better the performance of the model. When AUROC=0.5, the performance of the model is equivalent to random guessing and has no predictive ability. When AUROC>0.5, the model has a certain predictive ability. The closer the value is to 1, the stronger the ability. AUC can comprehensively evaluate the overall classification performance of the model.

[0127] (3) Based on the model evaluation results, the parameters of each trained machine learning model and deep learning model are tuned by a five-fold cross-validation combined with a grid search method, and then the model with the best performance is selected as the artificial intelligence model for predicting the KILLIP level.

[0128] In this embodiment, according to the results of model evaluation, the parameter space of each model is explored by using a five-fold cross validation combined with a grid search to find the best parameter combination.

[0129] Cross-validation is often used to evaluate model performance and prevent overfitting. K-fold cross-validation is to divide the data set into K subsets of equal size, perform K training and testing, use different subsets as test sets each time, and the remaining subsets as training sets, and take the average of K test results as the performance indicator of the model. Through the cross-validation method, the model is trained and validated multiple times on different subsets, which helps to detect and reduce overfitting; and all data are used as training sets and validation sets in a certain round, which improves data utilization efficiency; in addition, the average performance indicator obtained by multiple validations is used for evaluation, which is more stable and reliable. The use of five-fold cross-validation in this embodiment can ensure the accuracy and generalization of the model.

[0130] S230: Perform an explainability analysis on the artificial intelligence model used to predict the KILLIP level, determine the influence of each candidate clinical indicator on the KILLIP classification prediction, and screen out clinical indicators related to the KILLIP classification prediction from the multiple candidate clinical indicators based on the influence of each candidate clinical indicator on the KILLIP classification prediction.

[0131] This embodiment performs an explainability analysis on the artificial intelligence model finally selected for predicting the KILLIP level, and screens out clinical indicators related to the KILLIP level classification prediction from the candidate clinical indicator set based on the analysis results.

[0132] In some embodiments, an interpretability analysis is performed on the trained artificial intelligence model to determine the degree of influence of each candidate clinical indicator on the KILLIP classification prediction, and the operation of screening out clinical indicators related to the KILLIP classification prediction from the multiple candidate clinical indicators according to the degree of influence of each candidate clinical indicator on the KILLIP classification prediction includes: using the SHAP method to perform an interpretability analysis on the trained artificial intelligence model to obtain the SHAP value of each candidate clinical indicator; obtaining the candidate clinical indicator with a SHAP value greater than a preset threshold as the clinical indicator related to the KILLIP classification prediction.

[0133] This embodiment uses the shap interpretability method to perform interpretability analysis on the model, analyzes the impact of different candidate clinical indicators on KILLIP classification prediction, and thereby identifies clinical indicators related to KILLIP grade classification prediction. Among them, the shap interpretability method can be used to calculate the shap value of each candidate clinical indicator. A positive shap value indicates that the corresponding candidate clinical indicator has a high probability of causing the prediction result, while a negative shap value indicates that the candidate clinical indicator has a low probability of causing the prediction result. Furthermore, the contribution of each candidate clinical indicator to the model prediction can be evaluated based on the shap value of each candidate clinical indicator.

[0134] Specifically, we can first use the tree-based model TreeExplanier and the kernel model KernelExplanier to calculate the shap value of each sample in the test set, and then use the various visualization tools provided by shap, such as summary bar charts, summary dot plots, and correlation plots, to explain the model. For example, the summary graph can be used to show the overall impact (positive or negative impact) of each candidate clinical indicator on the model prediction and the distribution of these impacts.

[0135] In some examples, after performing shap interpretability analysis on the trained TabNet model, we can get the following: Figure 3 The summary bar chart shown, such as Figure 4-1 to Figure 4-9 The correlation graph is shown.

[0136] For the summary dot plot, the likelihood of a KILLIP grade increases as the shap value of the candidate clinical indicator increases. For each individual patient, there is a point for the shap value in the model, so there is a point for each candidate clinical indicator corresponding to each patient. The color of the point indicates the actual indicator value for each patient, with red indicating higher indicator values ​​and blue indicating lower indicator values. The points are stacked vertically to show density.

[0137] For a correlation graph, such as Figure 4-1 to Figure 4-9 Each of the graphs shown shows how a single candidate clinical indicator affects the output of the model, with the shap value represented by the y-axis and the actual value represented by the x-axis, with each point representing a single patient. For example, when the actual value of the GRACE score is lower, ≤150.0, the decision is biased toward the "KILLIP 1" class. When the shap value of a specific candidate clinical indicator exceeds zero, the decision is biased toward the "KILLIP 1" class.

[0138] Furthermore, we can also use the force graph to show how each candidate clinical indicator affects a single prediction. The following example shows the KILLIP prediction levels of four patients with different KILLIP levels, and uses the SHAP interpretability method to show the contribution of each candidate clinical indicator to the model prediction.

[0139] The four patients were Patient 1, Patient 2, Patient 3, and Patient 4, and the KILLIP grades of the four patients were 1 (Grade I), 2 (Grade II), 3 (Grade III), and 4 (Grade IV), respectively.

[0140] The index values ​​of candidate clinical indicators for each patient are as follows Figure 5As shown. Among them, gender, smoking history, smoking cessation, hypertension, diabetes, hyperlipidemia, cerebrovascular disease, gastric ulcer, old myocardial infarction, previous stent, previous bypass and atrial fibrillation are categorical features, and the remaining candidate clinical indicators are numerical features. The value of "gender" in the categorical feature is 1 for "male" and 0 for "female", and the values ​​of the remaining features are 1 for "yes" (or "yes") and 0 for "no" (or "no"). The corresponding figure for each patient is as follows Figures 6-1 to 6-4 As shown in the figure, the force graph corresponding to each patient represents the KILLIP prediction result of the patient and the contribution of each candidate clinical feature to the prediction result. Each bar in the force graph corresponds to a candidate clinical indicator. The bar has two colors, such as red and blue. Red indicates positive contribution and blue indicates negative contribution. Figure 6-2 For example, the color of each bar on the left side of the vertical line pointed by the arrow is red, and the color of each bar on the right side is blue. The longer the bar, the greater the contribution of the corresponding candidate clinical indicator to the prediction result. The top of the force graph shows the prediction result of the model, that is, the value of f(x), which represents the total contribution of all candidate clinical indicators.

[0141] It should be noted that, with respect to the various steps included in the method for predicting the risk of heart failure after surgery for acute myocardial infarction patients provided in any of the above embodiments, unless otherwise clearly stated in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, at least a part of these steps may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0142] Based on the same inventive concept, the present application also provides a device for predicting the risk of heart failure in patients after surgery for acute myocardial infarction. Figure 7 As shown, the device for predicting the risk of heart failure after surgery for acute myocardial infarction patients includes the following modules:

[0143] The target data acquisition module 110 is used to acquire the first clinical indicator data of the target patient after surgery as the target clinical indicator data; the target patient refers to a patient with acute myocardial infarction whose risk of heart failure is to be predicted after surgery; the first clinical indicator data includes predetermined indicator values ​​of clinical indicators related to KILLIP classification prediction, and the clinical indicators related to KILLIP classification prediction include GRACE score, NT-proBNP, TIMI score, left ventricular ejection fraction, creatinine, creatinine clearance, age, length of hospital stay, interleukin-6 and high-sensitivity C-reactive protein;

[0144] A construction module 120, for constructing model input data according to target clinical indicator data;

[0145] The prediction module 130 is used to process the model input data using the KILLIP level prediction model to obtain the KILLIP prediction level of the target patient. The KILLIP prediction level of the target patient is used to determine the risk of heart failure of the target patient after surgery. The KILLIP level prediction model is an artificial intelligence model trained using a first data set constructed based on the first clinical indicator data of multiple patients with acute myocardial infarction after surgery and their corresponding KILLIP levels.

[0146] In some embodiments, the operation of determining a clinical indicator associated with a KILLIP classification prediction comprises:

[0147] The second clinical indicator data and the corresponding KILLIP levels of multiple patients with acute myocardial infarction after surgery are collected to construct a second data set; the second clinical indicator data includes indicator values ​​of multiple candidate clinical indicators;

[0148] Using the second data set to perform model training and evaluation, an artificial intelligence model for predicting the KILLIP level is obtained;

[0149] Conduct interpretability analysis on the AI ​​model to determine the impact of each candidate clinical indicator on the KILLIP classification prediction;

[0150] According to the influence degree of each candidate clinical indicator on the KILLIP classification prediction, clinical indicators related to the KILLIP classification prediction are screened out from the multiple candidate clinical indicators.

[0151] In some embodiments, the second data set includes a training set and a test set; accordingly, the second data set is used to perform model training and evaluation to obtain an artificial intelligence model for predicting the KILLIP level, including:

[0152] Using the training set, multiple machine learning models and multiple deep learning models are trained to predict the KILLIP level;

[0153] Use the test set to evaluate the trained machine learning models and deep learning models;

[0154] According to the model evaluation results, the parameters of each trained machine learning model and deep learning model are tuned by five-fold cross-validation combined with grid search, and then the model with the best performance is selected as the artificial intelligence model for predicting the KILLIP level.

[0155] In some embodiments, the second clinical indicator data and the corresponding KILLIP levels of multiple patients after acute myocardial infarction are collected to construct a second data set, including:

[0156] Preprocessing the collected second clinical indicator data of patients after acute myocardial infarction surgery to obtain a sample data set consisting of multiple pieces of preprocessed second clinical indicator data and their corresponding KILLIP levels; the preprocessing includes data cleaning and feature scaling;

[0157] Divide the sample data set into training set and test set;

[0158] The minority class samples in the training set are determined according to the data proportion corresponding to each KILLIP level in the training set. New samples are generated based on the minority class samples using the smote oversampling technology, and the new samples are added to the training set.

[0159] In some embodiments, a smote oversampling technique is used to generate new samples based on minority class samples, including:

[0160] Use the K nearest neighbor algorithm to find k nearest neighbor samples for each minority class sample;

[0161] For each minority class sample, one or more samples from the k nearest neighbor samples corresponding to the minority class sample are randomly selected according to a preset sampling ratio, and a new sample is generated by randomly selecting the connection between the minority class sample and the selected sample.

[0162] In some embodiments, an interpretability analysis is performed on the trained artificial intelligence model to determine the degree of influence of each candidate clinical indicator on the KILLIP classification prediction, and clinical indicators related to the KILLIP classification prediction are screened out from the multiple candidate clinical indicators according to the degree of influence of each candidate clinical indicator on the KILLIP classification prediction, including:

[0163] Use the SHAP method to perform interpretability analysis on the trained artificial intelligence model and obtain the SHAP value of each candidate clinical indicator;

[0164] The candidate clinical indicators with shap values ​​greater than the preset threshold are obtained as clinical indicators related to KILLIP classification prediction.

[0165] In some embodiments, the multiple machine learning models are a random forest model and an XGboost model, the multiple deep learning models are an MLP model and a TabNet model; and the KILLIP level prediction model is a TabNet model.

[0166] For the specific limitations of the device for predicting the risk of heart failure after surgery for patients with acute myocardial infarction, please refer to the limitations of the method for predicting the risk of heart failure after surgery for patients with acute myocardial infarction mentioned above, which will not be repeated here. Each module in the above-mentioned device for predicting the risk of heart failure after surgery for patients with acute myocardial infarction can be implemented in whole or in part through software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0167] The present application also provides a computer device. In some embodiments, the computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the clinical efficacy characteristic enhancement method provided in any of the above embodiments can be implemented.

[0168] Furthermore, in some embodiments, the internal structure diagram of the computer device may be as follows: Figure 8 As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data such as clinical indicators related to KILLIP classification prediction. The specific stored data can also refer to the definitions in the above method embodiments. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for predicting the risk of heart failure after surgery in patients with acute myocardial infarction is implemented.

[0169] Those skilled in the art will understand that Figure 8 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0170] The present application also provides a computer-readable storage medium. In some embodiments, a computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, the steps of the method for predicting the risk of heart failure after surgery for patients with acute myocardial infarction provided in any of the above embodiments are implemented.

[0171] In the above embodiments of the present application, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0172] Those of ordinary skill in the art can understand that all or part of the processes in the above method embodiments can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink), DRAM (SLDRAM), memory bus (Rambus), direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0173] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0174] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A method for predicting the risk of heart failure after surgery in patients with acute myocardial infarction, characterized in that: The method comprises: Obtaining the first clinical indicator data of the target patient after surgery as the target clinical indicator data; the target patient refers to an acute myocardial infarction patient whose risk of heart failure is to be predicted after surgery; the first clinical indicator data includes predetermined index values ​​of clinical indicators related to KILLIP classification prediction, and the clinical indicators related to KILLIP classification prediction include GRACE score, NT-proBNP, TIMI score, left ventricular ejection fraction, creatinine, creatinine clearance, age, length of hospital stay, interleukin-6 and high-sensitivity C-reactive protein; the clinical indicators related to KILLIP classification prediction are obtained by performing an interpretable analysis on a TabNet model trained with a second data set constructed based on second clinical indicator data including index values ​​of multiple candidate clinical indicators after surgery of multiple acute myocardial infarction patients and their corresponding KILLIP levels, and screening from multiple candidate clinical indicators based on the analysis results; the second data set includes a training set that has achieved category balance; Constructing a model input data according to the target clinical indicator data; The model input data is processed using a KILLIP level prediction model to obtain the KILLIP prediction level of the target patient, and the KILLIP prediction level of the target patient is used to determine the risk of heart failure of the target patient after surgery; the KILLIP level prediction model is a TabNet model trained using a first data set constructed based on the first clinical indicator data of multiple patients with acute myocardial infarction after surgery and their corresponding KILLIP levels.

2. The method according to claim 1, characterized in that The operation of determining the clinical indicator associated with the KILLIP classification prediction comprises: Collecting the second clinical indicator data of multiple patients after acute myocardial infarction and their corresponding KILLIP levels to construct a second data set; the second clinical indicator data includes indicator values ​​of multiple candidate clinical indicators; Using the second data set to perform model training and evaluation, to obtain an artificial intelligence model for predicting the KILLIP level; Performing an interpretability analysis on the artificial intelligence model to determine the degree of influence of each of the candidate clinical indicators on the KILLIP classification prediction; The clinical indicators related to the KILLIP classification prediction are screened out from the multiple candidate clinical indicators according to the influence degree of each candidate clinical indicator on the KILLIP classification prediction.

3. The method according to claim 2, characterized in that The second data set includes a training set and a test set; using the second data set to perform model training and evaluation to obtain an artificial intelligence model for predicting the KILLIP level, including: Using the training set to train multiple machine learning models for predicting KILLIP levels; Using the test set to perform model evaluation on each of the trained machine learning models; According to the model evaluation results, the parameters of each trained machine learning model are tuned by a five-fold cross-validation combined with a grid search, and then the model with the best performance is selected as the artificial intelligence model for predicting the KILLIP level.

4. The method according to claim 2, characterized in that The second data set was constructed by collecting the second clinical index data of multiple patients with acute myocardial infarction after surgery and their corresponding KILLIP levels, including: Preprocessing the collected second clinical indicator data of patients after surgery for acute myocardial infarction to obtain a sample data set consisting of a plurality of preprocessed second clinical indicator data and their corresponding KILLIP levels; the preprocessing includes data cleaning and feature scaling; Dividing the sample data set into a training set and a test set; Determine minority class samples in the training set according to the data proportions corresponding to each KILLIP level in the training set, and generate new samples based on the minority class samples using the smote oversampling technology; Add new samples to the training set.

5. The method according to claim 4, characterized in that The smote oversampling technique is used to generate new samples based on the minority class samples, including: Use the K nearest neighbor algorithm to find k nearest neighbor samples for each of the minority class samples; For each of the minority class samples, one or more samples of the k nearest neighbor samples corresponding to the minority class sample are randomly selected according to a preset sampling ratio, and a new sample is generated by randomly selecting a connection between the minority class sample and the selected sample.

6. The method according to claim 2, characterized in that Perform an interpretability analysis on the trained artificial intelligence model to determine the degree of influence of each of the candidate clinical indicators on the KILLIP classification prediction, and screen the clinical indicators related to the KILLIP classification prediction from the multiple candidate clinical indicators according to the degree of influence of each of the candidate clinical indicators on the KILLIP classification prediction, including: Use the SHAP method to perform interpretability analysis on the trained artificial intelligence model to obtain the SHAP value of each candidate clinical indicator; A candidate clinical indicator with a shap value greater than a preset threshold is obtained as the clinical indicator related to the KILLIP classification prediction.

7. The method according to claim 3, characterized in that The multiple machine learning models are a random forest model, an XGboost model, an MLP model and a TabNet model.

8. A device for predicting the risk of heart failure after surgery for patients with acute myocardial infarction, characterized in that: The device comprises: A target data acquisition module is used to acquire the first clinical indicator data of the target patient after surgery as the target clinical indicator data; the target patient refers to an acute myocardial infarction patient whose risk of heart failure is to be predicted after surgery; the first clinical indicator data includes predetermined index values ​​of clinical indicators related to KILLIP classification prediction, and the clinical indicators related to KILLIP classification prediction include GRACE score, NT-proBNP, TIMI score, left ventricular ejection fraction, creatinine, creatinine clearance, age, length of hospital stay, interleukin-6 and high-sensitivity C-reactive protein; the clinical indicators related to KILLIP classification prediction are obtained by performing an interpretable analysis on a TabNet model trained with a second data set constructed based on second clinical indicator data including index values ​​of multiple candidate clinical indicators after surgery of multiple acute myocardial infarction patients and their corresponding KILLIP levels, and are screened from multiple candidate clinical indicators based on the analysis results; the second data set includes a training set that has achieved class balance; A construction module, used for constructing model input data according to the target clinical indicator data; A prediction module is used to process the model input data using a KILLIP level prediction model to obtain the KILLIP prediction level of the target patient, wherein the KILLIP prediction level of the target patient is used to determine the risk of heart failure of the target patient after surgery; the KILLIP level prediction model is a TabNet model trained using a first data set constructed based on first clinical indicator data of multiple patients with acute myocardial infarction after surgery and their corresponding KILLIP levels.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

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

Images