Research And Implementation Of Machine Learning Based Clinical Assisted Decision Making For Cardiovascular Disease

The prevalence of cardiovascular disease is still on a continuous rise.Among them,the incidence of non-ST-segment elevation myocardial infarction(NSTEMI),a form of coronary heart disease among cardiovascular diseases,has been continuously increasing,while the inhospital mortality rate has not decreased.Because of the high accuracy of machine learning methods for classification problems,machine learning methods are used in this study as an important way to predict in-hospital mortality in patients with non-ST-segment elevation myocardial infarction.In this context,this thesis starts from three aspects: feature extraction,model building,and system design.By building a prediction model for in-hospital mortality of non-STsegment elevation myocardial infarction,and then constructing a clinical aid decision-making system for cardiovascular diseases,the system is used to assist physicians in diagnosing patients’ conditions and achieving accurate diagnosis and timely treatment.This project was supported by the Natural Science Foundation of Liaoning Province(2023-MS-054),and the main work and contributions are as follows:(1)In order to screen the high-dimensional non-ST-segment elevation myocardial infarction data,this thesis proposes a feature extraction approach based on the maximum redundancy minimum relevance algorithm to screen medical features such as patient age,systolic blood pressure and hemoglobin.Firstly,eight machine learning models such as logistic regression and decision tree are used as pre-selected models for subsequent experiments.Second,the best feature subset of the models was extracted by combining maximum redundancy minimum relevance with the above models.Finally,in order to verify the performance of the maximum redundancy minimum relevance algorithm on feature selection on medical feature datasets,this thesis compares and analyzes it with the wrapper and embedded methods in feature selection methods,and uses accuracy,precision,recall,F1,AUC and training time as evaluation indexes to objectively evaluate this thesis’ s feature selection method comprehensively.(2)In order to conduct a prediction study of in-hospital mortality in non-ST-segment elevation myocardial infarction,a Stacking-based two-layer integrated model is proposed as a prediction algorithm in this thesis.First,by improving the Stacking integration model,the training set of the base learner is improved from the original non-ST-segment elevation myocardial infarction patient dataset to the best feature subset of the base learner,which improves the variability and diversity of the base learner,thus enhancing the generalization ability of the integration model.Second,by comparing the training results of eight machine learning models on the original dataset,the extreme gradient boosting with the best integrated performance is selected as the meta-learner,and the other seven models are trained as the base learners.Finally,in order to verify the performance of the proposed algorithm,it is compared with the training results of the eight machine learning models on the best feature subset,and the accuracy,precision,recall,F1 and AUC are used as evaluation indexes to make an objective evaluation of the proposed method in this thesis.(3)In order to clinically validate the algorithm proposed in this thesis,and also to assist physicians in quickly and accurately diagnosing patients’ conditions,a cardiovascular disease clinical aid decision system is designed in this thesis.The system applies the WPF technique in computer software development technology.The algorithm proposed in this thesis is embedded in the system,and other clinically useful scores are integrated to assist physicians in making scientific diagnosis and analysis recommendations for patients’ conditions.The cardiovascular disease clinical aid decision-making system is currently in use at Liaoning Provincial People’s Hospital.