Table 3.

Results of models on validation data.

	ROC-AUC	DSC (core)	DSC (pen)	JI (core)	JI (pen)
LR	0.9757	0.8438	0.7874	0.7298	0.6494
	0.9776	0.848	0.7907	0.736	0.6538
RF	0.9825	0.8553	0.8172	0.7471	0.6908
	0.9841	0.8611	0.8269	0.7561	0.7048
XGB	0.983	0.8552	0.8185	0.7470	0.6927
	0.9844	0.8610	0.8275	0.7559	0.7057
SVM	0.9799	0.8467	0.8081	0.7341	0.678

Eight models in total were optimized. Three algorithms (Logistic Regression/LR, Random Forest/RF, XGBoost/XGB) were trained twice, one on Cerebral Blood Flow (CBF) and Delay Time (DT) data (top), and once on data from CBF, DT, Cerebral Blood Flow and Mean Transit Time (bottom). Support Vector Machine (SVM) was only trained for two maps due to excessive training times. Results on the validation data are shown for each model. The highest performance across all categories was obtained for XBG, trained on all four CTP maps.