Heart injury prediction model and enrocyclic drug heart injury prediction method

Abstract

The invention discloses a heart injury prediction model and an enrocyclic drug heart injury prediction method, and relates to the technical field of medical data analysis. The problems that in the prior art, heart injury prediction lags behind, accuracy is low, and individual pertinence is lacked are solved. According to the method, through multi-dimensional data acquisition and refined data quality verification and preprocessing, target features are mined in combination with medical knowledge and a machine learning method, and key factors influencing the heart injury of the Enroring drugs can be comprehensively captured; meanwhile, on the basis of prediction accuracy evaluation and error root positioning of historical data, an error compensation strategy is implemented in a targeted mode, a heart injury prediction model is optimized, the accuracy and reliability of heart injury prediction are remarkably improved, clinical changes are continuously adapted, and the prediction model has good generalization ability and timeliness; visual risk prompts are output, and clinical decisions and drug safety assessment are made based on prediction results, so that the drug cardiotoxicity risk can be reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of medical data analysis, and in particular to a heart damage prediction model and a method for predicting heart damage caused by encyclopedia drugs. Background Art

[0002] Encyclopedia drugs are widely used in clinical fields such as tumor treatment, but they are cardiotoxic and may cause heart damage in patients, seriously affecting their prognosis and quality of life. In existing technologies, the assessment of heart damage mostly relies on clinical symptom observation, electrocardiogram examination, myocardial enzyme spectrum detection, and echocardiography. However, these methods often have a lag, that is, they can only be discovered when heart damage occurs or has reached a certain extent, making it difficult to achieve true early warning and unable to fully capture the complex factors related to heart damage, which can easily lead to missed diagnosis or misdiagnosis. In addition, the clinical prediction of heart damage caused by encyclopedia drugs mainly relies on the doctor's experience and judgment, as well as simple evaluation methods based on single or a small number of medical indicators. However, traditional risk assessment methods are often based on group data, which makes it difficult to accurately predict the risk of individual patients. Doctors' experience is greatly affected by individual cognitive differences and the number of accumulated cases, making it difficult to ensure the consistency and accuracy of predictions.

[0003] Existing cardiac injury prediction models fail to fully consider the specific pharmacological and toxicological mechanisms of encyclopedia drugs and their interactions with individual factors, resulting in limited prediction accuracy. They are unable to meet the needs of clinical refined management and are unable to judge the risks of medication based on the patient's specific situation to decide whether to use the drug, how to adjust the dosage, or increase the monitoring frequency. Summary of the Invention

[0004] The purpose of the present invention is to provide a cardiac injury prediction model and a method for predicting cardiac injury caused by encyclopedia drugs. By extracting features from original data, characteristic variables with important value for predicting cardiac injury are mined, and a cardiac injury prediction model is constructed to achieve accurate prediction function. Based on the trained cardiac injury prediction model, cardiac injury prediction is achieved for patients using encyclopedia drugs, and the model is updated and optimized to improve the accuracy and timeliness of the prediction, thereby solving the problems raised in the above-mentioned background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A cardiac injury prediction model comprising:

[0007] a data input module configured to obtain multi-source medical data of the patient based on a multi-source data interface, and process and extract features of the multi-source medical data as input data for a cardiac injury prediction model;

[0008] a processing module configured to perform normalization processing on the input data, screen out a core feature subset closely related to cardiac damage caused by encyclopedia drugs from the input data, extract the feature variation patterns at different time scales, and form a structured feature matrix;

[0009] The prediction result output module is configured to input the feature matrix into the trained heart injury prediction model, calculate and output the probability and risk level of heart injury in patients, and automatically generate corresponding risk warnings and personalized clinical intervention recommendations based on the preset risk threshold.

[0010] The present invention provides another technical solution, a method for predicting cardiac damage caused by encyclopedia drugs, comprising:

[0011] Patient data processing: Collect and preprocess the patient's multi-source medical data, extract key features from the pre-processed multi-source medical data, mine target features related to cardiac injury associated with encyclopedia drugs, and generate a target feature subset;

[0012] Model construction: Based on the prediction task and the characteristics of multi-source medical data, a prediction model architecture is constructed. The target feature subset is input into the prediction model architecture for training to obtain a heart injury prediction model that predicts the patient's risk of heart injury.

[0013] Prediction of cardiac injury caused by encyclopedia drugs: Based on the trained cardiac injury prediction model, cardiac injury is predicted for patients receiving encyclopedia drugs. The prediction is combined with the output of risk prompts for cardiac injury and presented in a visual form.

[0014] Furthermore, the multi-source medical data includes basic information, medical history, laboratory test indicators, imaging examination data and information on the use of encyclopedia drugs.

[0015] Furthermore, patient data processing includes:

[0016] Data collection: Establish a multi-source data access interface, connect to at least two different types of data sources, extract the patient's multi-source medical data from each data source based on the patient's unique identifier, and perform data quality verification based on the data type of the multi-source medical data;

[0017] Data preprocessing: Preprocess multi-source medical data that has passed data quality verification, including data cleaning and data standardization. Continuous data is divided into multiple discrete intervals based on data distribution characteristics to form binned data.

[0018] Feature extraction: based on the pre-processing results, key features are extracted from multi-source medical data. At the same time, combined with medical knowledge and clinical experience, the relevance of the key features to the heart injury related to the annular drug is analyzed, the target features that have a significant impact on heart injury are mined, different types of target features are combined to construct a composite feature reflecting the comprehensive function of the heart, and a target feature subset is formed.

[0019] Further, data quality verification includes:

[0020] Determine whether there is missing data in the key information field in the multi-source medical data. If there is missing data, determine whether to trigger the data completion process according to the preset missing threshold, and generate data missing prompt information to mark the corresponding data record;

[0021] Extract laboratory examination indicators, compare the test values with the normal reference range, and if they are outside the range, make a comprehensive judgment combined with the corresponding associated indicators to determine whether they are abnormal data;

[0022] Obtain the parameters in the imaging examination data, compare them with the standard anatomical structure and physiological parameters, and verify the accuracy of the data;

[0023] Compare the annular drug use information with the drug instructions and clinical drug specifications to determine whether the drug dosage and treatment course meet the regulations. If there is an anomaly, it is marked;

[0024] Check the consistency of the same information in multi-source medical data in different data sources or different time records. If the data is found to be inconsistent, trigger the manual review process.

[0025] Further, the target features include continuous target features and target features in the form of classification data:

[0026] For continuous target features, based on the numerical characteristics of the continuous target features after data standardization, a sliding window method is used to extract time series features, analyze the change trend of the time series features at different time scales, and select the feature dimensions that have a significant impact on heart injury prediction;

[0027] For target features in the form of classification data, if the number of feature values is small and there is no obvious order relationship, then a one-hot encoding method is used to map each category to a binary vector. If the feature has a certain order relationship, a label encoding process is used to assign a unique numerical label to each category, and the label value size is determined according to the logical order between categories.

[0028] Further, the annular drug heart injury prediction includes:

[0029] Data input: Obtain and preprocess multi-source medical data of the patient to be predicted, extract and filter target feature data that is consistent with the target feature subset during model training;

[0030] Data prediction: Input the target feature data into the trained heart injury prediction model to predict the likelihood of heart injury after the patient uses encyclopedia drugs and output the prediction results;

[0031] Result analysis: Analyze and interpret the predicted results, and make corresponding clinical decisions and drug safety assessments based on actual clinical conditions.

[0032] Furthermore, the prediction of cardiac injury caused by encyclopedia drugs is presented in a visual form, including: dynamically displaying the patient's current risk probability, key risk factors and comparison with patients in the same cohort; using the medication time as the horizontal axis, superimposing the predicted cardiac injury time window and the actual occurrence time point, and using colors to distinguish the prediction accuracy, and automatically generating intervention recommendations based on the risk level.

[0033] Furthermore, the prediction of cardiac damage caused by encyclopedia drugs also includes:

[0034] Historical data retrieval and comparison: Retrieve cardiac injury prediction data for patients using encyclopedia drugs from historical prediction data. At the same time, retrieve the actual clinical outcome data corresponding to the historical prediction data. Compare the historical prediction data with the actual clinical outcome data one by one to determine the accuracy of each prediction result.

[0035] Prediction accuracy evaluation: Based on the judgment results, the number of accurate and inaccurate prediction results is counted, and the accuracy of predicting cardiac injury in patients using encyclopedia drugs is calculated. The accuracy is compared with the preset accuracy threshold to evaluate whether the predictive performance of the cardiac injury prediction model meets the preset standard.

[0036] Furthermore, historical prediction data is compared with actual clinical outcome data one by one, including:

[0037] Read the prediction results of the heart injury prediction model in the historical prediction data center, extract the time window information of the patient's heart injury from the prediction results, and obtain the specific time when the patient's heart injury actually occurred from the actual clinical result data;

[0038] The actual time of cardiac injury occurrence is compared with the predicted time window information. If the actual time of cardiac injury occurrence is not within the predicted time window, it is determined that the prediction result has a deviation in the time dimension, that is, the prediction result is inaccurate;

[0039] Otherwise, it is determined that the prediction result is accurate in the time dimension, that is, the prediction result is accurate.

[0040] Furthermore, the prediction accuracy evaluation also includes:

[0041] If the prediction performance of the cardiac injury prediction model does not meet the standard, all historical prediction data judged to be inaccurate, the corresponding actual clinical outcome data, and the patient's multi-source medical data are obtained to construct an inaccurate prediction data subset;

[0042] Calculate the distribution of each target feature in the inaccurate prediction data subset in the events that lead to inaccurate predictions, identify the key features and target feature combinations that are misestimated or not fully captured, locate the root cause of the error that leads to inaccurate predictions, and develop and implement corresponding error compensation strategies;

[0043] The heart injury prediction model is optimized based on the error compensation strategy to obtain the optimized heart injury prediction model.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] 1. The cardiac injury prediction model and encyclopedia-related cardiac injury prediction method proposed in this invention can comprehensively capture the key factors affecting encyclopedia-related cardiac injury through multi-dimensional data collection and refined data quality verification and preprocessing, combining medical knowledge with machine learning methods to mine target features. At the same time, based on the prediction accuracy evaluation and error root location of historical data, a targeted error compensation strategy is implemented to optimize the cardiac injury prediction model, significantly improving the accuracy and reliability of cardiac injury prediction, constantly adapting to clinical changes, and giving the prediction model good generalization ability and timeliness.

[0046] 2. The cardiac injury prediction model and encyclopedia-type drug cardiac injury prediction method proposed in this invention output visual risk prompts to intuitively display the patient's risk probability, key risk factors, and comparison with the same cohort. They can also display the prediction window and actual time point in combination with the medication time, helping doctors to quickly understand the condition. Clinical decisions and drug safety assessments made based on the prediction results directly serve clinical treatment and help reduce the risk of drug cardiotoxicity. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a flow chart of the cardiac injury prediction model and the cardiac injury prediction method of encyclopedia drugs of the present invention. DETAILED DESCRIPTION

[0048] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0049] To address the technical issues of delayed cardiac injury prediction, low accuracy, and lack of individual targeting in the prior art, this embodiment provides the following technical solutions:

[0050] A cardiac injury prediction model comprising:

[0051] a data input module configured to obtain multi-source medical data of the patient based on a multi-source data interface, and process and extract features of the multi-source medical data as input data for a cardiac injury prediction model;

[0052] a processing module configured to perform normalization processing on the input data, screen out a core feature subset closely related to cardiac damage caused by encyclopedia drugs from the input data, extract the feature variation patterns at different time scales, and form a structured feature matrix;

[0053] The prediction result output module is configured to input the feature matrix into the trained heart injury prediction model, calculate and output the probability and risk level of heart injury in patients, and automatically generate corresponding risk warnings and personalized clinical intervention recommendations based on the preset risk threshold.

[0054] In this embodiment, multi-dimensional medical data such as patient basic information and examination indicators are comprehensively collected through a multi-source data interface. After cleaning, format unification, and feature extraction to improve processing efficiency, the processing module then standardizes the features, uses a pre-trained model to screen core features, and combines sliding windows to analyze time series changes to enhance sensitivity to cardiac injury risk. The prediction result output module quickly outputs risk probability and level, automatically generates early warnings and intervention recommendations, and assists clinicians in adjusting treatment plans, providing an accurate, efficient, and clinically valuable solution for predicting cardiac injury caused by encyclopedia drugs.

[0055] In order to better implement the prediction model of cardiac injury and solve the problems in the existing technology, please refer to Figure 1 This embodiment provides a method for predicting cardiac damage caused by encyclopedia drugs, comprising:

[0056] Patient data processing: Collect and preprocess the patient's multi-source medical data, extract key features from the pre-processed multi-source medical data, mine target features related to cardiac injury associated with encyclopedia drugs, and generate a target feature subset;

[0057] The multi-source medical data includes basic information (age, gender, weight, etc.), medical history (whether there is underlying heart disease or other chronic diseases, etc.), laboratory test indicators (such as myocardial enzyme spectrum, troponin, brain natriuretic peptide and other biomarkers reflecting cardiac function and damage, as well as liver and kidney function, blood routine and other related indicators), imaging examination data (cardiac echocardiography parameters, cardiac magnetic resonance imaging results, etc.), and encyclopedia drug use information (dose, medication course, medication interval, etc.);

[0058] Model construction: Based on the prediction task and the characteristics of multi-source medical data, a prediction model architecture is constructed. The target feature subset is input into the prediction model architecture for training to obtain a heart injury prediction model that predicts the patient's risk of heart injury.

[0059] Prediction of cardiac injury caused by encyclopedia drugs: Based on the trained cardiac injury prediction model, cardiac injury is predicted for patients receiving encyclopedia drugs. The prediction is combined with the output of risk warnings for cardiac injury, which are presented in the form of visual charts or concise text.

[0060] In this embodiment, a precise cardiac injury prediction system was constructed through multi-dimensional data integration and deep feature mining. Data preprocessing and target feature subset generation effectively captured the interactive effects of drug toxicity and individual characteristics, resolving the issue of existing models not fully considering pharmacological mechanisms and individual differences. The architecture was dynamically selected and optimized based on data characteristics, and predictions were made based on the trained cardiac injury prediction model. Clinically validated, this system can identify high-risk patients in advance. Visual risk prompts intuitively present individual risk probabilities, key factors, and time windows, providing doctors with precise medication adjustments and monitoring recommendations, addressing the issues of high lag and missed diagnosis rates in existing technologies.

[0061] In this embodiment, patient data processing includes:

[0062] Data collection: Establish a multi-source data access interface to connect to at least two different types of data sources, including electronic medical record systems, laboratory testing systems, and medical imaging systems. Extract the patient's multi-source medical data from each data source based on the patient's unique identifier, and perform data quality verification based on the data type of the multi-source medical data.

[0063] In this embodiment, data quality verification includes:

[0064] Determine whether there are missing data in key information fields in multi-source medical data, such as age and gender in basic information, basic heart disease records in medical history, myocardial enzyme spectrum test values ​​in laboratory examination indicators, key parameters in imaging examination data, and dosage and treatment course in encyclopedia drug use information. If there is missing data, determine whether to trigger the data completion process based on the preset missing threshold, generate a data missing prompt message, and mark the corresponding data record;

[0065] Based on medical expertise and clinical experience, reasonable value ranges and logical rules are set for each data field. Laboratory test indicators are extracted and the test values ​​are compared with the normal reference range. If the values ​​are out of range, a comprehensive judgment is made based on the corresponding related indicators to determine whether the data is abnormal.

[0066] Obtain parameters from imaging examination data, compare them with standard anatomical structures and physiological parameters, and verify the accuracy of the data;

[0067] Compare the use information of encyclopedia drugs with the drug instructions and clinical medication specifications to determine whether the dosage and treatment course are in compliance with the regulations. If there are any abnormalities, mark them.

[0068] Verify the consistency of the same information in multi-source medical data across different data sources or at different times. For example, compare the patient's basic information between the electronic medical record system and the laboratory testing system to see if it is consistent, and check whether the changes in the weight and height data of the same patient at different time periods are within the normal physiological fluctuation range. If any inconsistency is found, a manual review process will be triggered.

[0069] Data preprocessing: Preprocess multi-source medical data that has passed data quality verification, including data cleaning and data standardization. Continuous data is divided into multiple discrete intervals based on data distribution characteristics to form binned data.

[0070] Feature extraction: Based on the preprocessing results, key features are extracted from multi-source medical data, such as heart disease history, cardiac markers, demographic data, previous cardiotoxicity, and chemotherapy regimens. At the same time, combined with medical knowledge and clinical experience, the correlation between the key features and cardiac damage associated with encyclopedia drugs is analyzed, and target features with significant impact on cardiac damage are discovered. Different types of target features are combined, such as combining drug dosage with patient weight to construct a unit weight drug dosage feature, and then combining different cardiac function indicators to construct a composite feature reflecting comprehensive cardiac function, forming a target feature subset.

[0071] In this embodiment, the target features include continuous target features and target features in the form of categorical data:

[0072] For continuous target features, a sliding window method was used to extract time series features based on the numerical characteristics of the continuous target features after data standardization. The changing trends of the time series features under different time spans were analyzed to screen out feature dimensions that have a significant impact on cardiac injury prediction.

[0073] For target features in the form of categorical data, if the number of feature values ​​is small and there is no obvious sequential relationship, one-hot encoding is used to map each category into a binary vector so that each category is independent and orthogonal to each other; if there is a certain sequential relationship between the features, label encoding is used to assign a unique numerical label to each category, and the label value is determined according to the logical order between the categories to preserve the sequential information of the features.

[0074] In this embodiment, the time series features extracted through the sliding window effectively capture the dynamic changes of continuous target features over time. These newly generated features, combined with the original features, form a more representative feature set, providing richer and more effective data support for the subsequent construction of a cardiac injury prediction model. Through the data quality verification process and target feature processing method, the reliability and applicability of multi-source medical data can be effectively improved, laying a solid data foundation for the accurate construction of a cardiac injury prediction model.

[0075] In this embodiment, the prediction of cardiac damage caused by encyclopedia drugs includes:

[0076] Data input: Obtain and preprocess multi-source medical data of the patient to be predicted, extract and filter target feature data that is consistent with the target feature subset during model training;

[0077] Data prediction: Input the target feature data into the trained heart injury prediction model to predict the possibility of heart injury after the patient uses encyclopedia drugs, and output the prediction results, such as the predicted probability value or risk level of heart injury in the patient;

[0078] Result analysis: The prediction results were analyzed and interpreted, and corresponding clinical decisions and drug safety assessments were made based on the actual clinical situation. For example, for patients predicted to have a higher risk of heart damage, closer cardiac function monitoring and adjustment of the encyclopedia drug use regimen are recommended. For patients predicted to have a lower risk, medication and monitoring can be carried out according to the conventional regimen.

[0079] In this embodiment, the prediction of cardiac injury caused by encyclopedia drugs is presented in a visual form, specifically including: dynamically displaying the patient's current risk probability, key risk factors, and comparison with patients in the same cohort; with the medication time as the horizontal axis, superimposing the predicted cardiac injury time window and the actual occurrence time point, and using colors to distinguish the prediction accuracy, green for accurate and red for inaccurate, and automatically generating intervention recommendations based on the risk level. For example, high-risk patients trigger the "cardiology consultation + weekly cardiac ultrasound" process, while moderate-risk patients are recommended to "monitor troponin every 2 weeks."

[0080] In this embodiment, multi-source medical data of the patient to be predicted is obtained and preprocessed, and target feature data consistent with model training is extracted and screened to ensure uniform input dimensions and avoid feature deviation; the target feature data is then input into the trained model, and the cardiac injury probability value or risk level reflecting individual differences is output to achieve individualized risk quantification; complex prediction results are dynamically displayed through a visual interface to convert them into executable clinical pathways, which has been clinically verified to significantly improve decision-making efficiency and standardization.

[0081] In this embodiment, the prediction of cardiac damage caused by encyclopedia drugs also includes:

[0082] Historical data retrieval and comparison: Retrieve cardiac injury prediction data for patients using encyclopedia drugs from historical prediction data. At the same time, retrieve actual clinical outcome data corresponding to the historical prediction data, including detailed information such as whether the patient actually suffered cardiac injury and the extent of cardiac injury. Compare the historical prediction data with the actual clinical outcome data one by one to determine the accuracy of each prediction result.

[0083] Prediction accuracy evaluation: Based on the judgment results, the number of accurate and inaccurate prediction results is counted, and the accuracy of predicting cardiac injury in patients using encyclopedia drugs is calculated. The accuracy is compared with the preset accuracy threshold to evaluate whether the predictive performance of the cardiac injury prediction model meets the preset standard.

[0084] In this embodiment, the historical prediction data is compared with the actual clinical result data one by one, and the following steps are also included:

[0085] Read the prediction results of the heart injury prediction model in the historical prediction data center, extract the time window information of the patient's heart injury from the prediction results, and obtain the specific time when the patient's heart injury actually occurred from the actual clinical result data;

[0086] The actual time of cardiac injury occurrence is compared with the predicted time window information. If the actual time of cardiac injury occurrence is not within the predicted time window, it is determined that the prediction result has a deviation in the time dimension, that is, the prediction result is inaccurate;

[0087] Otherwise, it is determined that the prediction result is accurate in the time dimension, that is, the prediction result is accurate.

[0088] In this embodiment, the prediction accuracy evaluation also includes:

[0089] If the prediction performance of the cardiac injury prediction model does not meet the standard, all historical prediction data judged to be inaccurate, the corresponding actual clinical outcome data, and the patient's multi-source medical data are obtained to construct an inaccurate prediction data subset;

[0090] Calculate the distribution of each target feature in the inaccurate prediction data subset in the events that lead to inaccurate predictions, identify the key features and target feature combinations that are misestimated or not fully captured, locate the root cause of the error that leads to inaccurate predictions, and develop and implement corresponding error compensation strategies:

[0091] Feature engineering adjustments: Add or modify the feature extraction method in step 2, introduce new features found in the analysis that are critical for distinguishing inaccurate prediction events, or perform more sophisticated transformations on existing features (such as binning and interactive feature construction);

[0092] Model parameter fine-tuning: Adjust the parameters of the model architecture or algorithm used in step 3, such as changing the regularization strength, learning rate, number of network layers or nodes, to improve the model's ability to capture patterns in data inaccuracies;

[0093] Data augmentation or rebalancing: If analysis reveals that Data_ does not accurately represent a specific but important subgroup in the dataset, consider increasing the proportion or representation of that subgroup in the training data by oversampling, undersampling, or generating synthetic data when retraining the model, so that the model pays more attention to these "difficult examples";

[0094] Model structure optimization: In extreme cases, it may be necessary to consider changing the model architecture or introducing ensemble learning methods to achieve better generalization performance and robustness;

[0095] The heart injury prediction model is optimized based on the error compensation strategy to obtain the optimized heart injury prediction model.

[0096] In this embodiment, cardiac injury prediction results and corresponding actual clinical results are retrieved from historical prediction data. By comparing the prediction time window with the actual injury occurrence time, the prediction accuracy is judged from the time dimension. At the same time, the number of accurate and inaccurate times is counted to calculate the prediction accuracy and compare it with the preset threshold. If the prediction performance does not meet the standard, a subset of inaccurate prediction data is constructed, and the root cause of the error is located by analyzing the target feature distribution. A targeted error compensation strategy is implemented to enable the cardiac injury prediction model to continuously adapt to changes in clinical data, effectively solving the problem of difficult self-optimization of the model and long-term attenuation of prediction accuracy in the existing technology, and realizing continuous optimization of prediction performance and dynamic improvement of clinical value.

[0097] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A cardiac injury prediction model, characterized in that: include: a data input module configured to obtain multi-source medical data of the patient based on a multi-source data interface, and process and extract features of the multi-source medical data as input data for a cardiac injury prediction model; a processing module configured to perform normalization processing on the input data, screen out a core feature subset closely related to cardiac damage caused by encyclopedia drugs from the input data, extract the feature variation patterns at different time scales, and form a structured feature matrix; The prediction result output module is configured to input the feature matrix into the trained heart injury prediction model, calculate and output the probability and risk level of heart injury in patients, and automatically generate corresponding risk warnings and personalized clinical intervention recommendations based on the preset risk threshold.

2. A method for predicting cardiac damage caused by encyclopedia drugs, based on the cardiac damage prediction model of claim 1, characterized in that: include: patient Data processing: Collect and preprocess the patient's multi-source medical data, extract key features from the pre-processed multi-source medical data, mine target features related to cardiac injury associated with encyclopedia drugs, and generate a target feature subset. The multi-source medical data includes basic information, medical history, laboratory test indicators, imaging examination data, and encyclopedia drug use information; Model construction: Based on the prediction task and the characteristics of multi-source medical data, a prediction model architecture is constructed. The target feature subset is input into the prediction model architecture for training to obtain a heart injury prediction model that predicts the patient's risk of heart injury. Prediction of cardiac injury caused by encyclopedia drugs: Based on the trained cardiac injury prediction model, cardiac injury is predicted for patients receiving encyclopedia drugs. The prediction is combined with the output of risk prompts for cardiac injury and presented in a visual form.

3. The method for predicting cardiac damage caused by encyclopedia drugs according to claim 2, wherein: Patient data processing, including: Data collection: Establish a multi-source data access interface, connect to at least two different types of data sources, extract the patient's multi-source medical data from each data source based on the patient's unique identifier, and perform data quality verification based on the data type of the multi-source medical data; Data preprocessing: Preprocess multi-source medical data that has passed data quality verification, including data cleaning and data standardization. Continuous data is divided into multiple discrete intervals based on data distribution characteristics to form binned data. Feature extraction: Based on the preprocessing results, key features are extracted from multi-source medical data. At the same time, combined with medical knowledge and clinical experience, the correlation between these key features and cardiac damage associated with encyclopedia drugs is analyzed. Target features with significant impact on cardiac damage are discovered. Different types of target features are combined to construct composite features reflecting comprehensive cardiac function, forming a target feature subset.

4. The method for predicting cardiac damage caused by encyclopedia drugs according to claim 3, wherein: Data quality verification, including: Determine whether there is missing data in key information fields in multi-source medical data. If there is missing data, determine whether to trigger the data completion process based on the preset missing data threshold, generate data missing prompt information, and mark the corresponding data record; Extract laboratory test indicators and compare the test values ​​with the normal reference range. If the values ​​are out of range, make a comprehensive judgment based on the corresponding related indicators to determine whether the data is abnormal. Obtain parameters from imaging examination data, compare them with standard anatomical structures and physiological parameters, and verify the accuracy of the data; Compare the use information of encyclopedia drugs with the drug instructions and clinical medication specifications to determine whether the dosage and treatment course are in compliance with the regulations. If there are any abnormalities, mark them. Verify the consistency of the same information in multi-source medical data in different data sources or at different times. If data inconsistency is found, the manual review process will be triggered.

5. The method for predicting cardiac damage caused by encyclopedia drugs according to claim 3, wherein: The target features include continuous target features and target features in the form of categorical data: For continuous target features, a sliding window method was used to extract time series features based on the numerical characteristics of the continuous target features after data standardization. The changing trends of the time series features under different time spans were analyzed to screen out feature dimensions that have a significant impact on cardiac injury prediction. For target features in the form of categorical data, if the number of feature values ​​is small and there is no obvious order relationship, one-hot encoding is used to map each category into a binary vector. If the features have a certain order relationship, label encoding is used to assign a unique numerical label to each category, and the label value size is determined according to the logical order between categories.

6. The method for predicting cardiac damage caused by encyclopedia drugs according to claim 2, wherein: Prediction of cardiac damage from encyclopedia drugs, including: Data input: Obtain and preprocess multi-source medical data of the patient to be predicted, extract and filter target feature data that is consistent with the target feature subset during model training; Data prediction: Input the target feature data into the trained heart injury prediction model to predict the likelihood of heart injury after the patient uses encyclopedia drugs and output the prediction results; Result analysis: Analyze and interpret the predicted results, and make corresponding clinical decisions and drug safety assessments based on actual clinical conditions.

7. The method for predicting cardiac damage caused by encyclopedia drugs according to claim 6, wherein: The prediction of cardiac injury caused by encyclopedia drugs is presented in a visual form, specifically including: dynamically displaying the patient's current risk probability, key risk factors and comparison with patients in the same cohort; using medication time as the horizontal axis, superimposing the predicted cardiac injury time window and the actual occurrence time point, and using colors to distinguish the prediction accuracy, and automatically generating intervention recommendations based on the risk level.

8. The method for predicting cardiac damage caused by encyclopedia drugs according to claim 2, wherein: Prediction of cardiac damage from encyclopedia drugs also includes: Historical data retrieval and comparison: Retrieve cardiac injury prediction data for patients using encyclopedia drugs from historical prediction data. At the same time, retrieve the actual clinical outcome data corresponding to the historical prediction data. Compare the historical prediction data with the actual clinical outcome data one by one to determine the accuracy of each prediction result. Prediction accuracy evaluation: Based on the judgment results, the number of accurate and inaccurate prediction results is counted, and the accuracy of predicting cardiac injury in patients using encyclopedia drugs is calculated. The accuracy is compared with the preset accuracy threshold to evaluate whether the predictive performance of the cardiac injury prediction model meets the preset standard.

9. The method for predicting cardiac damage caused by encyclopedia drugs according to claim 8, wherein: Compare historical prediction data with actual clinical outcome data, including: Read the prediction results of the heart injury prediction model in the historical prediction data center, extract the time window information of the patient's heart injury from the prediction results, and obtain the specific time when the patient's heart injury actually occurred from the actual clinical result data; The actual time of cardiac injury occurrence is compared with the predicted time window information. If the actual time of cardiac injury occurrence is not within the predicted time window, it is determined that the prediction result has a deviation in the time dimension, that is, the prediction result is inaccurate; Otherwise, it is determined that the prediction result is accurate in the time dimension, that is, the prediction result is accurate.

10. The method for predicting cardiac damage caused by encyclopedia drugs according to claim 9, wherein: Forecast accuracy evaluation also includes: If the prediction performance of the cardiac injury prediction model does not meet the standard, all historical prediction data judged to be inaccurate, the corresponding actual clinical outcome data, and the patient's multi-source medical data are obtained to construct an inaccurate prediction data subset; Calculate the distribution of each target feature in the inaccurate prediction data subset in the events that lead to inaccurate predictions, identify the key features and target feature combinations that are misestimated or not fully captured, locate the root cause of the error that leads to inaccurate predictions, and develop and implement corresponding error compensation strategies; The heart injury prediction model is optimized based on the error compensation strategy to obtain the optimized heart injury prediction model.

Images