Research On Intelligent Diagnosis Method Of Heart Disease Based On Multi-model Fusion

Hypertrophic cardiomyopathy(HCM)is a common heart disease and one of the main causes of sudden death in adolescents.Timely medical intervention can effectively prevent sudden death and reduce the clinical mortality rate of HCM patients.HCM has a clear genetic tendency and is highly correlated with gene mutations.Therefore,mining the mutation characteristics of pathogenic genes is of great significance for the prediction and treatment of HCM.However,high-dimensional gene sequencing data is the main challenge faced by clinical testing and data research.This study first preprocessed the gene sequencing data and baseline clinical data of 178 HCM patients,and conducted experiments based on different mutation frequency thresholds of HCM patients to determine the applicable threshold for HCM.On this basis,this study used decision tree,random forest,multiple perceptron and other methods to predict gene sequencing data and optimize the model to predict a variety of clinical key indicators.At the same time,this study introduced whole gene sequencing data from normal individuals,implemented classification and prediction of HCM using multiple models,and discussed the model bias under different positive and negative sample ratios,as well as the impact of normal human gene data on classification and prediction of HCM subtypes.In order to reduce model input and remove noise features,principal component analysis was used to achieve gene data dimensionality reduction.The impact of data dimensionality reduction on the prediction model was discussed under different optimizers and parameters.In order to further improve the accuracy of model prediction,this study also conducted comparative experiments using multiple model fusion methods to obtain an intelligent diagnostic model for HCM based on multiple model fusion,with accuracy,recall,and accuracy rates of 0.65,0.73,and 0.68,respectively.Because the pathogenesis of HCM is still unclear,and the pathogenic genes have not been fully excavated,this study uses principal component analysis,back-propagation calculation,random forest Gini index and other feature engineering methods to analyze the feasible feature extraction scheme of gene sequencing data.Based on this,a deep neural network is used to establish a classification model for predicting HCM patients and common subtypes of HCM,namely apical hypertrophic cardiomyopathy(AHCM)and obstructive cardiomyopathy(HOCM).The model can effectively classify HCM patients,with an area under the working characteristic curve(AUC)of 0.86,and an AUC of 0.69 and 0.77 for AHCM and HOCM,respectively.In addition,this study also analyzed the impact of mutation sites on model performance to identify potential pathogenic genes for HCM.The results indicate that some mutation sites play a more important role in the performance of the model,and a good prediction model can be established based on a small amount of mutation information from high weight gene sites.The AUC of HCM,AHCM,and HOCM are 0.76,0.83,and 0.85,respectively.Compared to existing methods,the HCM prediction model established in this study is based on high weight gene loci,which can predict patient risk in advance to take intervention measures,improve patient survival rate,and reduce input complexity and detection costs.