Prediction Of Cardiovascular Events In Patients With Severe Dilated Cardiomyopathy Based On Machine Learning

Dilated cardiomyopathy(DCM)is the common cardiomyopathy in the world.The mortality of DCM still remains high,especially in the prediction of cardiovascular events in severe DCM patients.One of the important reasons is that the risk stratification in DCM is still unsatisfactory.For better risk stratification in severe DCM patients,there are mainly two approaches to solve this problem:(1)to explore the significant features that could be more effective to predict the cardiovascular events;(2)to establish a model used for predicting the cardiovascular events in every single patient.The present study was divided into two parts:1)The present study aimed to select and validate the significant features derived from CMR and radiomics features based on native T1 mapping schemes.A total of 45 severe DCM patients(LVEF < 35%)underwent 3.0-T CMR with T1 mapping was included in the present study,and the median follow-up time was 13 months(interquartile range,7–17 months).The analysis was performed using the Cox model with L1 regulization.After that,the selected features were input to the Cox model again,and we employed the time-dependent ROC and leave-one-out cross-validation.It turned out that the coefficient of ECV,the important feature derived from T1 mapping scheme,was 1.038 in Lasso-Cox.And the coefficients of 7 radiomics features based on native T1 mapping scheme were also non-zero.In addition,the predictive performance of the combination of ECV and those radiomics features was higher than ECV only.In conclusion,ECV and those radiomics features could identify the cardiovascular events in severe DCM patients;The combination of ECV and those radiomics features may more effectively predict the cardiovascular.2)The aim of this study was to establish and test the risk model utilising machine learning(ML)algorithms based on clinical features and CMR features to predict one-year cardiovascular events in severe DCM patients.The dataset used to establish the ML model was obtained from 98 severe DCM patients from two centres,and were discretized according to the clinical criterion or reference values in the pre-processing procedure.It turned out that the combination of Relief-F and Na?ve Bayes was the most effective way for prediction,the area under the ROC curves was 0.920(confidence interval: 0.868 – 0.972)in ten-fold crossvalidation.Furthermore,the predictive performance of ML was significantly higher than LVEF(0.504,confidence interval:0.354 – 0.654)and MAGGIC Score(0.599,confidence interval:0.469 – 0.729)(p < 0.01).In conclusion,ML could effectively predict the cardiovascular events in severe DCM patients in one-year follow-up and it may aim risk stratification and patient management in the future.