A clinical–radiomics model based on noncontrast computed tomography to predict hemorrhagic transformation after stroke by machine learning: a multicenter study

Insights into Imaging

Table 3 Performance of five methods to select radiomics features in the training cohort

Selecting method	Dim	AUC	ACC	SEN	SPE	PPV	NPV
LASSO	41	0.931	0.823	0.788	0.853	0.819	0.829
Linear SVC	19	0.917	0.831	0.805	0.853	0.823	0.840
RFECV	116	0.936	0.840	0.810	0.865	0.836	0.845
RFE	12	0.911	0.832	0.802	0.857	0.825	0.839
Tree-based model	12	0.865	0.778	0.720	0.827	0.777	0.779

LASSO, Least Absolute Shrinkage and Selection Operator; Linear SVC, Linear Support Vector Classification; RFECV, Recursive Feature Elimination Cross Validation; RFE, Recursive Feature Elimination; Dim, dimension; AUC, area under the receiver operator characteristic curve; ACC, accuracy; SEN, sensitivity; SPE, specificity; PPV, positive predictive value; NPV, negative predictive value