Objective:To investigate the value of radiomics based on conventional MRI in predic-ting hemorrhagic transformation(HT)in acute cerebral infarction.Methods:In this retrospective stud-y,120 patients with acute cerebral infarction from January 2017 to December 2022 in our hospital were enrolled and divided into HT positive group(n=60)and HT negative group(n=60).All patients un-derwent routine head MRI scan and were randomly assigned to the training group(n=84)and the val-idation group(n=36)in a 7:3 ratio.Lesion ROI was delineated and texture features were extracted u-sing software.Minimum redundancy maximum relevance(mRMR)and least absolute shrinkage and selection operator(LASSO)regression analysis were used to screen features and establish radiomics signature.The performance and the clinical usefulness of the models was assessed by receiver operating characteristic(ROC)curve and decision curve analysis(DCA).Results:The area under curve(AUC)(95％CI)values of clinical model,conventional MRI model,combined sequential radiomics model,per-sonalized model 1 and personalized model 2 in the training and validation groups were 0.72(0.61～0.83)and 0.68(0.50～0.86),0.93(0.86～0.98)and 0.93(0.79～0.99),0.97(0.94～1.00)and 0.97(0.93～1.00),0.96(0.92～1.00)and 0.99(0.98～1.00),0.98(0.95～1.00)and 0.98(0.96～1.00),re-spectively.DCA demonstrated that personalized model 2 was better than personalized model 1 for clini-cal benefits.Conclusion:Compared with clinical model,the conventional MRI model,combined sequen-tial radiomics model,personalized model 1 and personalized model 2 have better diagnostic efficacy to predict hemorrhagic transformation in acute cerebral infarction.Although personalized model 1 and 2 have similar diagnostic performance,personalized model 2 has better clinical benefits.
NETL
NSTL
万方数据
